xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b9765e96c225f468eaa0ee5e5b0626743b1d4e9c)
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_cycles.h>
28 #include <rte_flow_driver.h>
29 #include <rte_malloc.h>
30 #include <rte_ip.h>
31 
32 #include <mlx5_glue.h>
33 #include <mlx5_devx_cmds.h>
34 #include <mlx5_prm.h>
35 
36 #include "mlx5_defs.h"
37 #include "mlx5.h"
38 #include "mlx5_flow.h"
39 #include "mlx5_flow_os.h"
40 #include "mlx5_rxtx.h"
41 
42 /** Device flow drivers. */
43 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
44 
45 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
46 
47 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
48 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
49 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
50 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
51 #endif
52 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
53 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
54 };
55 
56 enum mlx5_expansion {
57 	MLX5_EXPANSION_ROOT,
58 	MLX5_EXPANSION_ROOT_OUTER,
59 	MLX5_EXPANSION_ROOT_ETH_VLAN,
60 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
61 	MLX5_EXPANSION_OUTER_ETH,
62 	MLX5_EXPANSION_OUTER_ETH_VLAN,
63 	MLX5_EXPANSION_OUTER_VLAN,
64 	MLX5_EXPANSION_OUTER_IPV4,
65 	MLX5_EXPANSION_OUTER_IPV4_UDP,
66 	MLX5_EXPANSION_OUTER_IPV4_TCP,
67 	MLX5_EXPANSION_OUTER_IPV6,
68 	MLX5_EXPANSION_OUTER_IPV6_UDP,
69 	MLX5_EXPANSION_OUTER_IPV6_TCP,
70 	MLX5_EXPANSION_VXLAN,
71 	MLX5_EXPANSION_VXLAN_GPE,
72 	MLX5_EXPANSION_GRE,
73 	MLX5_EXPANSION_MPLS,
74 	MLX5_EXPANSION_ETH,
75 	MLX5_EXPANSION_ETH_VLAN,
76 	MLX5_EXPANSION_VLAN,
77 	MLX5_EXPANSION_IPV4,
78 	MLX5_EXPANSION_IPV4_UDP,
79 	MLX5_EXPANSION_IPV4_TCP,
80 	MLX5_EXPANSION_IPV6,
81 	MLX5_EXPANSION_IPV6_UDP,
82 	MLX5_EXPANSION_IPV6_TCP,
83 };
84 
85 /** Supported expansion of items. */
86 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
87 	[MLX5_EXPANSION_ROOT] = {
88 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
89 						 MLX5_EXPANSION_IPV4,
90 						 MLX5_EXPANSION_IPV6),
91 		.type = RTE_FLOW_ITEM_TYPE_END,
92 	},
93 	[MLX5_EXPANSION_ROOT_OUTER] = {
94 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
95 						 MLX5_EXPANSION_OUTER_IPV4,
96 						 MLX5_EXPANSION_OUTER_IPV6),
97 		.type = RTE_FLOW_ITEM_TYPE_END,
98 	},
99 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
100 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
101 		.type = RTE_FLOW_ITEM_TYPE_END,
102 	},
103 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
104 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
105 		.type = RTE_FLOW_ITEM_TYPE_END,
106 	},
107 	[MLX5_EXPANSION_OUTER_ETH] = {
108 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
109 						 MLX5_EXPANSION_OUTER_IPV6,
110 						 MLX5_EXPANSION_MPLS),
111 		.type = RTE_FLOW_ITEM_TYPE_ETH,
112 		.rss_types = 0,
113 	},
114 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
115 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
116 		.type = RTE_FLOW_ITEM_TYPE_ETH,
117 		.rss_types = 0,
118 	},
119 	[MLX5_EXPANSION_OUTER_VLAN] = {
120 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
121 						 MLX5_EXPANSION_OUTER_IPV6),
122 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
123 	},
124 	[MLX5_EXPANSION_OUTER_IPV4] = {
125 		.next = RTE_FLOW_EXPAND_RSS_NEXT
126 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
127 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
128 			 MLX5_EXPANSION_GRE,
129 			 MLX5_EXPANSION_IPV4,
130 			 MLX5_EXPANSION_IPV6),
131 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
132 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
133 			ETH_RSS_NONFRAG_IPV4_OTHER,
134 	},
135 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
136 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
137 						 MLX5_EXPANSION_VXLAN_GPE),
138 		.type = RTE_FLOW_ITEM_TYPE_UDP,
139 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
140 	},
141 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
142 		.type = RTE_FLOW_ITEM_TYPE_TCP,
143 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
144 	},
145 	[MLX5_EXPANSION_OUTER_IPV6] = {
146 		.next = RTE_FLOW_EXPAND_RSS_NEXT
147 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
148 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
149 			 MLX5_EXPANSION_IPV4,
150 			 MLX5_EXPANSION_IPV6),
151 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
152 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
153 			ETH_RSS_NONFRAG_IPV6_OTHER,
154 	},
155 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
156 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
157 						 MLX5_EXPANSION_VXLAN_GPE),
158 		.type = RTE_FLOW_ITEM_TYPE_UDP,
159 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
160 	},
161 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
162 		.type = RTE_FLOW_ITEM_TYPE_TCP,
163 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
164 	},
165 	[MLX5_EXPANSION_VXLAN] = {
166 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
167 						 MLX5_EXPANSION_IPV4,
168 						 MLX5_EXPANSION_IPV6),
169 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
170 	},
171 	[MLX5_EXPANSION_VXLAN_GPE] = {
172 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173 						 MLX5_EXPANSION_IPV4,
174 						 MLX5_EXPANSION_IPV6),
175 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
176 	},
177 	[MLX5_EXPANSION_GRE] = {
178 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 		.type = RTE_FLOW_ITEM_TYPE_GRE,
180 	},
181 	[MLX5_EXPANSION_MPLS] = {
182 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 						 MLX5_EXPANSION_IPV6),
184 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
185 	},
186 	[MLX5_EXPANSION_ETH] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_ETH,
190 	},
191 	[MLX5_EXPANSION_ETH_VLAN] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 		.type = RTE_FLOW_ITEM_TYPE_ETH,
194 	},
195 	[MLX5_EXPANSION_VLAN] = {
196 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 						 MLX5_EXPANSION_IPV6),
198 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
199 	},
200 	[MLX5_EXPANSION_IPV4] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 						 MLX5_EXPANSION_IPV4_TCP),
203 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
204 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 			ETH_RSS_NONFRAG_IPV4_OTHER,
206 	},
207 	[MLX5_EXPANSION_IPV4_UDP] = {
208 		.type = RTE_FLOW_ITEM_TYPE_UDP,
209 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
210 	},
211 	[MLX5_EXPANSION_IPV4_TCP] = {
212 		.type = RTE_FLOW_ITEM_TYPE_TCP,
213 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
214 	},
215 	[MLX5_EXPANSION_IPV6] = {
216 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 						 MLX5_EXPANSION_IPV6_TCP),
218 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
219 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 			ETH_RSS_NONFRAG_IPV6_OTHER,
221 	},
222 	[MLX5_EXPANSION_IPV6_UDP] = {
223 		.type = RTE_FLOW_ITEM_TYPE_UDP,
224 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
225 	},
226 	[MLX5_EXPANSION_IPV6_TCP] = {
227 		.type = RTE_FLOW_ITEM_TYPE_TCP,
228 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
229 	},
230 };
231 
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 	.validate = mlx5_flow_validate,
234 	.create = mlx5_flow_create,
235 	.destroy = mlx5_flow_destroy,
236 	.flush = mlx5_flow_flush,
237 	.isolate = mlx5_flow_isolate,
238 	.query = mlx5_flow_query,
239 	.dev_dump = mlx5_flow_dev_dump,
240 	.get_aged_flows = mlx5_flow_get_aged_flows,
241 };
242 
243 /* Convert FDIR request to Generic flow. */
244 struct mlx5_fdir {
245 	struct rte_flow_attr attr;
246 	struct rte_flow_item items[4];
247 	struct rte_flow_item_eth l2;
248 	struct rte_flow_item_eth l2_mask;
249 	union {
250 		struct rte_flow_item_ipv4 ipv4;
251 		struct rte_flow_item_ipv6 ipv6;
252 	} l3;
253 	union {
254 		struct rte_flow_item_ipv4 ipv4;
255 		struct rte_flow_item_ipv6 ipv6;
256 	} l3_mask;
257 	union {
258 		struct rte_flow_item_udp udp;
259 		struct rte_flow_item_tcp tcp;
260 	} l4;
261 	union {
262 		struct rte_flow_item_udp udp;
263 		struct rte_flow_item_tcp tcp;
264 	} l4_mask;
265 	struct rte_flow_action actions[2];
266 	struct rte_flow_action_queue queue;
267 };
268 
269 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
270 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
271 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
272 };
273 
274 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
275 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
276 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
277 	{ 9, 10, 11 }, { 12, 13, 14 },
278 };
279 
280 /* Tunnel information. */
281 struct mlx5_flow_tunnel_info {
282 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
283 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
284 };
285 
286 static struct mlx5_flow_tunnel_info tunnels_info[] = {
287 	{
288 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
289 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
290 	},
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
293 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
297 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_GRE,
301 		.ptype = RTE_PTYPE_TUNNEL_GRE,
302 	},
303 	{
304 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
305 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
306 	},
307 	{
308 		.tunnel = MLX5_FLOW_LAYER_MPLS,
309 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
310 	},
311 	{
312 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
313 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
314 	},
315 	{
316 		.tunnel = MLX5_FLOW_LAYER_IPIP,
317 		.ptype = RTE_PTYPE_TUNNEL_IP,
318 	},
319 	{
320 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
321 		.ptype = RTE_PTYPE_TUNNEL_IP,
322 	},
323 	{
324 		.tunnel = MLX5_FLOW_LAYER_GTP,
325 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
326 	},
327 };
328 
329 /**
330  * Translate tag ID to register.
331  *
332  * @param[in] dev
333  *   Pointer to the Ethernet device structure.
334  * @param[in] feature
335  *   The feature that request the register.
336  * @param[in] id
337  *   The request register ID.
338  * @param[out] error
339  *   Error description in case of any.
340  *
341  * @return
342  *   The request register on success, a negative errno
343  *   value otherwise and rte_errno is set.
344  */
345 int
346 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
347 		     enum mlx5_feature_name feature,
348 		     uint32_t id,
349 		     struct rte_flow_error *error)
350 {
351 	struct mlx5_priv *priv = dev->data->dev_private;
352 	struct mlx5_dev_config *config = &priv->config;
353 	enum modify_reg start_reg;
354 	bool skip_mtr_reg = false;
355 
356 	switch (feature) {
357 	case MLX5_HAIRPIN_RX:
358 		return REG_B;
359 	case MLX5_HAIRPIN_TX:
360 		return REG_A;
361 	case MLX5_METADATA_RX:
362 		switch (config->dv_xmeta_en) {
363 		case MLX5_XMETA_MODE_LEGACY:
364 			return REG_B;
365 		case MLX5_XMETA_MODE_META16:
366 			return REG_C_0;
367 		case MLX5_XMETA_MODE_META32:
368 			return REG_C_1;
369 		}
370 		break;
371 	case MLX5_METADATA_TX:
372 		return REG_A;
373 	case MLX5_METADATA_FDB:
374 		switch (config->dv_xmeta_en) {
375 		case MLX5_XMETA_MODE_LEGACY:
376 			return REG_NONE;
377 		case MLX5_XMETA_MODE_META16:
378 			return REG_C_0;
379 		case MLX5_XMETA_MODE_META32:
380 			return REG_C_1;
381 		}
382 		break;
383 	case MLX5_FLOW_MARK:
384 		switch (config->dv_xmeta_en) {
385 		case MLX5_XMETA_MODE_LEGACY:
386 			return REG_NONE;
387 		case MLX5_XMETA_MODE_META16:
388 			return REG_C_1;
389 		case MLX5_XMETA_MODE_META32:
390 			return REG_C_0;
391 		}
392 		break;
393 	case MLX5_MTR_SFX:
394 		/*
395 		 * If meter color and flow match share one register, flow match
396 		 * should use the meter color register for match.
397 		 */
398 		if (priv->mtr_reg_share)
399 			return priv->mtr_color_reg;
400 		else
401 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
402 			       REG_C_3;
403 	case MLX5_MTR_COLOR:
404 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
405 		return priv->mtr_color_reg;
406 	case MLX5_COPY_MARK:
407 		/*
408 		 * Metadata COPY_MARK register using is in meter suffix sub
409 		 * flow while with meter. It's safe to share the same register.
410 		 */
411 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
412 	case MLX5_APP_TAG:
413 		/*
414 		 * If meter is enable, it will engage the register for color
415 		 * match and flow match. If meter color match is not using the
416 		 * REG_C_2, need to skip the REG_C_x be used by meter color
417 		 * match.
418 		 * If meter is disable, free to use all available registers.
419 		 */
420 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
421 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
422 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
423 		if (id > (REG_C_7 - start_reg))
424 			return rte_flow_error_set(error, EINVAL,
425 						  RTE_FLOW_ERROR_TYPE_ITEM,
426 						  NULL, "invalid tag id");
427 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
428 			return rte_flow_error_set(error, ENOTSUP,
429 						  RTE_FLOW_ERROR_TYPE_ITEM,
430 						  NULL, "unsupported tag id");
431 		/*
432 		 * This case means meter is using the REG_C_x great than 2.
433 		 * Take care not to conflict with meter color REG_C_x.
434 		 * If the available index REG_C_y >= REG_C_x, skip the
435 		 * color register.
436 		 */
437 		if (skip_mtr_reg && config->flow_mreg_c
438 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
439 			if (id >= (REG_C_7 - start_reg))
440 				return rte_flow_error_set(error, EINVAL,
441 						       RTE_FLOW_ERROR_TYPE_ITEM,
442 							NULL, "invalid tag id");
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->dev_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_rss_desc *rss_desc,
662 			    int tunnel __rte_unused, uint64_t layer_types,
663 			    uint64_t hash_fields)
664 {
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 	int rss_request_inner = rss_desc->level >= 2;
667 
668 	/* Check RSS hash level for tunnel. */
669 	if (tunnel && rss_request_inner)
670 		hash_fields |= IBV_RX_HASH_INNER;
671 	else if (tunnel || rss_request_inner)
672 		return 0;
673 #endif
674 	/* Check if requested layer matches RSS hash fields. */
675 	if (!(rss_desc->types & layer_types))
676 		return 0;
677 	return hash_fields;
678 }
679 
680 /**
681  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
682  * if several tunnel rules are used on this queue, the tunnel ptype will be
683  * cleared.
684  *
685  * @param rxq_ctrl
686  *   Rx queue to update.
687  */
688 static void
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
690 {
691 	unsigned int i;
692 	uint32_t tunnel_ptype = 0;
693 
694 	/* Look up for the ptype to use. */
695 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 		if (!rxq_ctrl->flow_tunnels_n[i])
697 			continue;
698 		if (!tunnel_ptype) {
699 			tunnel_ptype = tunnels_info[i].ptype;
700 		} else {
701 			tunnel_ptype = 0;
702 			break;
703 		}
704 	}
705 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
706 }
707 
708 /**
709  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
710  * flow.
711  *
712  * @param[in] dev
713  *   Pointer to the Ethernet device structure.
714  * @param[in] dev_handle
715  *   Pointer to device flow handle structure.
716  */
717 static void
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
719 		       struct mlx5_flow_handle *dev_handle)
720 {
721 	struct mlx5_priv *priv = dev->data->dev_private;
722 	const int mark = dev_handle->mark;
723 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
724 	struct mlx5_hrxq *hrxq;
725 	unsigned int i;
726 
727 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
728 		return;
729 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
730 			      dev_handle->rix_hrxq);
731 	if (!hrxq)
732 		return;
733 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
734 		int idx = hrxq->ind_table->queues[i];
735 		struct mlx5_rxq_ctrl *rxq_ctrl =
736 			container_of((*priv->rxqs)[idx],
737 				     struct mlx5_rxq_ctrl, rxq);
738 
739 		/*
740 		 * To support metadata register copy on Tx loopback,
741 		 * this must be always enabled (metadata may arive
742 		 * from other port - not from local flows only.
743 		 */
744 		if (priv->config.dv_flow_en &&
745 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
746 		    mlx5_flow_ext_mreg_supported(dev)) {
747 			rxq_ctrl->rxq.mark = 1;
748 			rxq_ctrl->flow_mark_n = 1;
749 		} else if (mark) {
750 			rxq_ctrl->rxq.mark = 1;
751 			rxq_ctrl->flow_mark_n++;
752 		}
753 		if (tunnel) {
754 			unsigned int j;
755 
756 			/* Increase the counter matching the flow. */
757 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
758 				if ((tunnels_info[j].tunnel &
759 				     dev_handle->layers) ==
760 				    tunnels_info[j].tunnel) {
761 					rxq_ctrl->flow_tunnels_n[j]++;
762 					break;
763 				}
764 			}
765 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
766 		}
767 	}
768 }
769 
770 /**
771  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
772  *
773  * @param[in] dev
774  *   Pointer to the Ethernet device structure.
775  * @param[in] flow
776  *   Pointer to flow structure.
777  */
778 static void
779 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
780 {
781 	struct mlx5_priv *priv = dev->data->dev_private;
782 	uint32_t handle_idx;
783 	struct mlx5_flow_handle *dev_handle;
784 
785 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
786 		       handle_idx, dev_handle, next)
787 		flow_drv_rxq_flags_set(dev, dev_handle);
788 }
789 
790 /**
791  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
792  * device flow if no other flow uses it with the same kind of request.
793  *
794  * @param dev
795  *   Pointer to Ethernet device.
796  * @param[in] dev_handle
797  *   Pointer to the device flow handle structure.
798  */
799 static void
800 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
801 			struct mlx5_flow_handle *dev_handle)
802 {
803 	struct mlx5_priv *priv = dev->data->dev_private;
804 	const int mark = dev_handle->mark;
805 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
806 	struct mlx5_hrxq *hrxq;
807 	unsigned int i;
808 
809 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
810 		return;
811 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
812 			      dev_handle->rix_hrxq);
813 	if (!hrxq)
814 		return;
815 	MLX5_ASSERT(dev->data->dev_started);
816 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
817 		int idx = hrxq->ind_table->queues[i];
818 		struct mlx5_rxq_ctrl *rxq_ctrl =
819 			container_of((*priv->rxqs)[idx],
820 				     struct mlx5_rxq_ctrl, rxq);
821 
822 		if (priv->config.dv_flow_en &&
823 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
824 		    mlx5_flow_ext_mreg_supported(dev)) {
825 			rxq_ctrl->rxq.mark = 1;
826 			rxq_ctrl->flow_mark_n = 1;
827 		} else if (mark) {
828 			rxq_ctrl->flow_mark_n--;
829 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
830 		}
831 		if (tunnel) {
832 			unsigned int j;
833 
834 			/* Decrease the counter matching the flow. */
835 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
836 				if ((tunnels_info[j].tunnel &
837 				     dev_handle->layers) ==
838 				    tunnels_info[j].tunnel) {
839 					rxq_ctrl->flow_tunnels_n[j]--;
840 					break;
841 				}
842 			}
843 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
844 		}
845 	}
846 }
847 
848 /**
849  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
850  * @p flow if no other flow uses it with the same kind of request.
851  *
852  * @param dev
853  *   Pointer to Ethernet device.
854  * @param[in] flow
855  *   Pointer to the flow.
856  */
857 static void
858 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
859 {
860 	struct mlx5_priv *priv = dev->data->dev_private;
861 	uint32_t handle_idx;
862 	struct mlx5_flow_handle *dev_handle;
863 
864 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
865 		       handle_idx, dev_handle, next)
866 		flow_drv_rxq_flags_trim(dev, dev_handle);
867 }
868 
869 /**
870  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
871  *
872  * @param dev
873  *   Pointer to Ethernet device.
874  */
875 static void
876 flow_rxq_flags_clear(struct rte_eth_dev *dev)
877 {
878 	struct mlx5_priv *priv = dev->data->dev_private;
879 	unsigned int i;
880 
881 	for (i = 0; i != priv->rxqs_n; ++i) {
882 		struct mlx5_rxq_ctrl *rxq_ctrl;
883 		unsigned int j;
884 
885 		if (!(*priv->rxqs)[i])
886 			continue;
887 		rxq_ctrl = container_of((*priv->rxqs)[i],
888 					struct mlx5_rxq_ctrl, rxq);
889 		rxq_ctrl->flow_mark_n = 0;
890 		rxq_ctrl->rxq.mark = 0;
891 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
892 			rxq_ctrl->flow_tunnels_n[j] = 0;
893 		rxq_ctrl->rxq.tunnel = 0;
894 	}
895 }
896 
897 /**
898  * Set the Rx queue dynamic metadata (mask and offset) for a flow
899  *
900  * @param[in] dev
901  *   Pointer to the Ethernet device structure.
902  */
903 void
904 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
905 {
906 	struct mlx5_priv *priv = dev->data->dev_private;
907 	struct mlx5_rxq_data *data;
908 	unsigned int i;
909 
910 	for (i = 0; i != priv->rxqs_n; ++i) {
911 		if (!(*priv->rxqs)[i])
912 			continue;
913 		data = (*priv->rxqs)[i];
914 		if (!rte_flow_dynf_metadata_avail()) {
915 			data->dynf_meta = 0;
916 			data->flow_meta_mask = 0;
917 			data->flow_meta_offset = -1;
918 		} else {
919 			data->dynf_meta = 1;
920 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
921 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
922 		}
923 	}
924 }
925 
926 /*
927  * return a pointer to the desired action in the list of actions.
928  *
929  * @param[in] actions
930  *   The list of actions to search the action in.
931  * @param[in] action
932  *   The action to find.
933  *
934  * @return
935  *   Pointer to the action in the list, if found. NULL otherwise.
936  */
937 const struct rte_flow_action *
938 mlx5_flow_find_action(const struct rte_flow_action *actions,
939 		      enum rte_flow_action_type action)
940 {
941 	if (actions == NULL)
942 		return NULL;
943 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
944 		if (actions->type == action)
945 			return actions;
946 	return NULL;
947 }
948 
949 /*
950  * Validate the flag action.
951  *
952  * @param[in] action_flags
953  *   Bit-fields that holds the actions detected until now.
954  * @param[in] attr
955  *   Attributes of flow that includes this action.
956  * @param[out] error
957  *   Pointer to error structure.
958  *
959  * @return
960  *   0 on success, a negative errno value otherwise and rte_errno is set.
961  */
962 int
963 mlx5_flow_validate_action_flag(uint64_t action_flags,
964 			       const struct rte_flow_attr *attr,
965 			       struct rte_flow_error *error)
966 {
967 	if (action_flags & MLX5_FLOW_ACTION_MARK)
968 		return rte_flow_error_set(error, EINVAL,
969 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
970 					  "can't mark and flag in same flow");
971 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
972 		return rte_flow_error_set(error, EINVAL,
973 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
974 					  "can't have 2 flag"
975 					  " actions in same flow");
976 	if (attr->egress)
977 		return rte_flow_error_set(error, ENOTSUP,
978 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
979 					  "flag action not supported for "
980 					  "egress");
981 	return 0;
982 }
983 
984 /*
985  * Validate the mark action.
986  *
987  * @param[in] action
988  *   Pointer to the queue action.
989  * @param[in] action_flags
990  *   Bit-fields that holds the actions detected until now.
991  * @param[in] attr
992  *   Attributes of flow that includes this action.
993  * @param[out] error
994  *   Pointer to error structure.
995  *
996  * @return
997  *   0 on success, a negative errno value otherwise and rte_errno is set.
998  */
999 int
1000 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1001 			       uint64_t action_flags,
1002 			       const struct rte_flow_attr *attr,
1003 			       struct rte_flow_error *error)
1004 {
1005 	const struct rte_flow_action_mark *mark = action->conf;
1006 
1007 	if (!mark)
1008 		return rte_flow_error_set(error, EINVAL,
1009 					  RTE_FLOW_ERROR_TYPE_ACTION,
1010 					  action,
1011 					  "configuration cannot be null");
1012 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1013 		return rte_flow_error_set(error, EINVAL,
1014 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1015 					  &mark->id,
1016 					  "mark id must in 0 <= id < "
1017 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1018 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1019 		return rte_flow_error_set(error, EINVAL,
1020 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1021 					  "can't flag and mark in same flow");
1022 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1023 		return rte_flow_error_set(error, EINVAL,
1024 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1025 					  "can't have 2 mark actions in same"
1026 					  " flow");
1027 	if (attr->egress)
1028 		return rte_flow_error_set(error, ENOTSUP,
1029 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1030 					  "mark action not supported for "
1031 					  "egress");
1032 	return 0;
1033 }
1034 
1035 /*
1036  * Validate the drop action.
1037  *
1038  * @param[in] action_flags
1039  *   Bit-fields that holds the actions detected until now.
1040  * @param[in] attr
1041  *   Attributes of flow that includes this action.
1042  * @param[out] error
1043  *   Pointer to error structure.
1044  *
1045  * @return
1046  *   0 on success, a negative errno value otherwise and rte_errno is set.
1047  */
1048 int
1049 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1050 			       const struct rte_flow_attr *attr,
1051 			       struct rte_flow_error *error)
1052 {
1053 	if (attr->egress)
1054 		return rte_flow_error_set(error, ENOTSUP,
1055 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1056 					  "drop action not supported for "
1057 					  "egress");
1058 	return 0;
1059 }
1060 
1061 /*
1062  * Validate the queue action.
1063  *
1064  * @param[in] action
1065  *   Pointer to the queue action.
1066  * @param[in] action_flags
1067  *   Bit-fields that holds the actions detected until now.
1068  * @param[in] dev
1069  *   Pointer to the Ethernet device structure.
1070  * @param[in] attr
1071  *   Attributes of flow that includes this action.
1072  * @param[out] error
1073  *   Pointer to error structure.
1074  *
1075  * @return
1076  *   0 on success, a negative errno value otherwise and rte_errno is set.
1077  */
1078 int
1079 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1080 				uint64_t action_flags,
1081 				struct rte_eth_dev *dev,
1082 				const struct rte_flow_attr *attr,
1083 				struct rte_flow_error *error)
1084 {
1085 	struct mlx5_priv *priv = dev->data->dev_private;
1086 	const struct rte_flow_action_queue *queue = action->conf;
1087 
1088 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1089 		return rte_flow_error_set(error, EINVAL,
1090 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1091 					  "can't have 2 fate actions in"
1092 					  " same flow");
1093 	if (!priv->rxqs_n)
1094 		return rte_flow_error_set(error, EINVAL,
1095 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1096 					  NULL, "No Rx queues configured");
1097 	if (queue->index >= priv->rxqs_n)
1098 		return rte_flow_error_set(error, EINVAL,
1099 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1100 					  &queue->index,
1101 					  "queue index out of range");
1102 	if (!(*priv->rxqs)[queue->index])
1103 		return rte_flow_error_set(error, EINVAL,
1104 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1105 					  &queue->index,
1106 					  "queue is not configured");
1107 	if (attr->egress)
1108 		return rte_flow_error_set(error, ENOTSUP,
1109 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1110 					  "queue action not supported for "
1111 					  "egress");
1112 	return 0;
1113 }
1114 
1115 /*
1116  * Validate the rss action.
1117  *
1118  * @param[in] action
1119  *   Pointer to the queue action.
1120  * @param[in] action_flags
1121  *   Bit-fields that holds the actions detected until now.
1122  * @param[in] dev
1123  *   Pointer to the Ethernet device structure.
1124  * @param[in] attr
1125  *   Attributes of flow that includes this action.
1126  * @param[in] item_flags
1127  *   Items that were detected.
1128  * @param[out] error
1129  *   Pointer to error structure.
1130  *
1131  * @return
1132  *   0 on success, a negative errno value otherwise and rte_errno is set.
1133  */
1134 int
1135 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1136 			      uint64_t action_flags,
1137 			      struct rte_eth_dev *dev,
1138 			      const struct rte_flow_attr *attr,
1139 			      uint64_t item_flags,
1140 			      struct rte_flow_error *error)
1141 {
1142 	struct mlx5_priv *priv = dev->data->dev_private;
1143 	const struct rte_flow_action_rss *rss = action->conf;
1144 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1145 	unsigned int i;
1146 
1147 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1148 		return rte_flow_error_set(error, EINVAL,
1149 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1150 					  "can't have 2 fate actions"
1151 					  " in same flow");
1152 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1153 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1154 		return rte_flow_error_set(error, ENOTSUP,
1155 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1156 					  &rss->func,
1157 					  "RSS hash function not supported");
1158 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1159 	if (rss->level > 2)
1160 #else
1161 	if (rss->level > 1)
1162 #endif
1163 		return rte_flow_error_set(error, ENOTSUP,
1164 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1165 					  &rss->level,
1166 					  "tunnel RSS is not supported");
1167 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1168 	if (rss->key_len == 0 && rss->key != NULL)
1169 		return rte_flow_error_set(error, ENOTSUP,
1170 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1171 					  &rss->key_len,
1172 					  "RSS hash key length 0");
1173 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1174 		return rte_flow_error_set(error, ENOTSUP,
1175 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1176 					  &rss->key_len,
1177 					  "RSS hash key too small");
1178 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1179 		return rte_flow_error_set(error, ENOTSUP,
1180 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1181 					  &rss->key_len,
1182 					  "RSS hash key too large");
1183 	if (rss->queue_num > priv->config.ind_table_max_size)
1184 		return rte_flow_error_set(error, ENOTSUP,
1185 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1186 					  &rss->queue_num,
1187 					  "number of queues too large");
1188 	if (rss->types & MLX5_RSS_HF_MASK)
1189 		return rte_flow_error_set(error, ENOTSUP,
1190 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1191 					  &rss->types,
1192 					  "some RSS protocols are not"
1193 					  " supported");
1194 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1195 	    !(rss->types & ETH_RSS_IP))
1196 		return rte_flow_error_set(error, EINVAL,
1197 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1198 					  "L3 partial RSS requested but L3 RSS"
1199 					  " type not specified");
1200 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1201 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1202 		return rte_flow_error_set(error, EINVAL,
1203 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204 					  "L4 partial RSS requested but L4 RSS"
1205 					  " type not specified");
1206 	if (!priv->rxqs_n)
1207 		return rte_flow_error_set(error, EINVAL,
1208 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1209 					  NULL, "No Rx queues configured");
1210 	if (!rss->queue_num)
1211 		return rte_flow_error_set(error, EINVAL,
1212 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1213 					  NULL, "No queues configured");
1214 	for (i = 0; i != rss->queue_num; ++i) {
1215 		if (rss->queue[i] >= priv->rxqs_n)
1216 			return rte_flow_error_set
1217 				(error, EINVAL,
1218 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1219 				 &rss->queue[i], "queue index out of range");
1220 		if (!(*priv->rxqs)[rss->queue[i]])
1221 			return rte_flow_error_set
1222 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1223 				 &rss->queue[i], "queue is not configured");
1224 	}
1225 	if (attr->egress)
1226 		return rte_flow_error_set(error, ENOTSUP,
1227 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1228 					  "rss action not supported for "
1229 					  "egress");
1230 	if (rss->level > 1 &&  !tunnel)
1231 		return rte_flow_error_set(error, EINVAL,
1232 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1233 					  "inner RSS is not supported for "
1234 					  "non-tunnel flows");
1235 	return 0;
1236 }
1237 
1238 /*
1239  * Validate the default miss action.
1240  *
1241  * @param[in] action_flags
1242  *   Bit-fields that holds the actions detected until now.
1243  * @param[out] error
1244  *   Pointer to error structure.
1245  *
1246  * @return
1247  *   0 on success, a negative errno value otherwise and rte_errno is set.
1248  */
1249 int
1250 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1251 				const struct rte_flow_attr *attr,
1252 				struct rte_flow_error *error)
1253 {
1254 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1255 		return rte_flow_error_set(error, EINVAL,
1256 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1257 					  "can't have 2 fate actions in"
1258 					  " same flow");
1259 	if (attr->egress)
1260 		return rte_flow_error_set(error, ENOTSUP,
1261 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1262 					  "default miss action not supported "
1263 					  "for egress");
1264 	if (attr->group)
1265 		return rte_flow_error_set(error, ENOTSUP,
1266 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1267 					  "only group 0 is supported");
1268 	if (attr->transfer)
1269 		return rte_flow_error_set(error, ENOTSUP,
1270 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1271 					  NULL, "transfer is not supported");
1272 	return 0;
1273 }
1274 
1275 /*
1276  * Validate the count action.
1277  *
1278  * @param[in] dev
1279  *   Pointer to the Ethernet device structure.
1280  * @param[in] attr
1281  *   Attributes of flow that includes this action.
1282  * @param[out] error
1283  *   Pointer to error structure.
1284  *
1285  * @return
1286  *   0 on success, a negative errno value otherwise and rte_errno is set.
1287  */
1288 int
1289 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1290 				const struct rte_flow_attr *attr,
1291 				struct rte_flow_error *error)
1292 {
1293 	if (attr->egress)
1294 		return rte_flow_error_set(error, ENOTSUP,
1295 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1296 					  "count action not supported for "
1297 					  "egress");
1298 	return 0;
1299 }
1300 
1301 /**
1302  * Verify the @p attributes will be correctly understood by the NIC and store
1303  * them in the @p flow if everything is correct.
1304  *
1305  * @param[in] dev
1306  *   Pointer to the Ethernet device structure.
1307  * @param[in] attributes
1308  *   Pointer to flow attributes
1309  * @param[out] error
1310  *   Pointer to error structure.
1311  *
1312  * @return
1313  *   0 on success, a negative errno value otherwise and rte_errno is set.
1314  */
1315 int
1316 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1317 			      const struct rte_flow_attr *attributes,
1318 			      struct rte_flow_error *error)
1319 {
1320 	struct mlx5_priv *priv = dev->data->dev_private;
1321 	uint32_t priority_max = priv->config.flow_prio - 1;
1322 
1323 	if (attributes->group)
1324 		return rte_flow_error_set(error, ENOTSUP,
1325 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1326 					  NULL, "groups is not supported");
1327 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1328 	    attributes->priority >= priority_max)
1329 		return rte_flow_error_set(error, ENOTSUP,
1330 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1331 					  NULL, "priority out of range");
1332 	if (attributes->egress)
1333 		return rte_flow_error_set(error, ENOTSUP,
1334 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1335 					  "egress is not supported");
1336 	if (attributes->transfer && !priv->config.dv_esw_en)
1337 		return rte_flow_error_set(error, ENOTSUP,
1338 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1339 					  NULL, "transfer is not supported");
1340 	if (!attributes->ingress)
1341 		return rte_flow_error_set(error, EINVAL,
1342 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1343 					  NULL,
1344 					  "ingress attribute is mandatory");
1345 	return 0;
1346 }
1347 
1348 /**
1349  * Validate ICMP6 item.
1350  *
1351  * @param[in] item
1352  *   Item specification.
1353  * @param[in] item_flags
1354  *   Bit-fields that holds the items detected until now.
1355  * @param[out] error
1356  *   Pointer to error structure.
1357  *
1358  * @return
1359  *   0 on success, a negative errno value otherwise and rte_errno is set.
1360  */
1361 int
1362 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1363 			       uint64_t item_flags,
1364 			       uint8_t target_protocol,
1365 			       struct rte_flow_error *error)
1366 {
1367 	const struct rte_flow_item_icmp6 *mask = item->mask;
1368 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1369 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1370 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1371 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1372 				      MLX5_FLOW_LAYER_OUTER_L4;
1373 	int ret;
1374 
1375 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1376 		return rte_flow_error_set(error, EINVAL,
1377 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1378 					  "protocol filtering not compatible"
1379 					  " with ICMP6 layer");
1380 	if (!(item_flags & l3m))
1381 		return rte_flow_error_set(error, EINVAL,
1382 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1383 					  "IPv6 is mandatory to filter on"
1384 					  " ICMP6");
1385 	if (item_flags & l4m)
1386 		return rte_flow_error_set(error, EINVAL,
1387 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1388 					  "multiple L4 layers not supported");
1389 	if (!mask)
1390 		mask = &rte_flow_item_icmp6_mask;
1391 	ret = mlx5_flow_item_acceptable
1392 		(item, (const uint8_t *)mask,
1393 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1394 		 sizeof(struct rte_flow_item_icmp6), error);
1395 	if (ret < 0)
1396 		return ret;
1397 	return 0;
1398 }
1399 
1400 /**
1401  * Validate ICMP item.
1402  *
1403  * @param[in] item
1404  *   Item specification.
1405  * @param[in] item_flags
1406  *   Bit-fields that holds the items detected until now.
1407  * @param[out] error
1408  *   Pointer to error structure.
1409  *
1410  * @return
1411  *   0 on success, a negative errno value otherwise and rte_errno is set.
1412  */
1413 int
1414 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1415 			     uint64_t item_flags,
1416 			     uint8_t target_protocol,
1417 			     struct rte_flow_error *error)
1418 {
1419 	const struct rte_flow_item_icmp *mask = item->mask;
1420 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1421 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1422 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1423 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1424 				      MLX5_FLOW_LAYER_OUTER_L4;
1425 	int ret;
1426 
1427 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1428 		return rte_flow_error_set(error, EINVAL,
1429 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1430 					  "protocol filtering not compatible"
1431 					  " with ICMP layer");
1432 	if (!(item_flags & l3m))
1433 		return rte_flow_error_set(error, EINVAL,
1434 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1435 					  "IPv4 is mandatory to filter"
1436 					  " on ICMP");
1437 	if (item_flags & l4m)
1438 		return rte_flow_error_set(error, EINVAL,
1439 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1440 					  "multiple L4 layers not supported");
1441 	if (!mask)
1442 		mask = &rte_flow_item_icmp_mask;
1443 	ret = mlx5_flow_item_acceptable
1444 		(item, (const uint8_t *)mask,
1445 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1446 		 sizeof(struct rte_flow_item_icmp), error);
1447 	if (ret < 0)
1448 		return ret;
1449 	return 0;
1450 }
1451 
1452 /**
1453  * Validate Ethernet item.
1454  *
1455  * @param[in] item
1456  *   Item specification.
1457  * @param[in] item_flags
1458  *   Bit-fields that holds the items detected until now.
1459  * @param[out] error
1460  *   Pointer to error structure.
1461  *
1462  * @return
1463  *   0 on success, a negative errno value otherwise and rte_errno is set.
1464  */
1465 int
1466 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1467 			    uint64_t item_flags,
1468 			    struct rte_flow_error *error)
1469 {
1470 	const struct rte_flow_item_eth *mask = item->mask;
1471 	const struct rte_flow_item_eth nic_mask = {
1472 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1473 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1474 		.type = RTE_BE16(0xffff),
1475 	};
1476 	int ret;
1477 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1478 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1479 				       MLX5_FLOW_LAYER_OUTER_L2;
1480 
1481 	if (item_flags & ethm)
1482 		return rte_flow_error_set(error, ENOTSUP,
1483 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1484 					  "multiple L2 layers not supported");
1485 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1486 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1487 		return rte_flow_error_set(error, EINVAL,
1488 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1489 					  "L2 layer should not follow "
1490 					  "L3 layers");
1491 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1492 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1493 		return rte_flow_error_set(error, EINVAL,
1494 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1495 					  "L2 layer should not follow VLAN");
1496 	if (!mask)
1497 		mask = &rte_flow_item_eth_mask;
1498 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1499 					(const uint8_t *)&nic_mask,
1500 					sizeof(struct rte_flow_item_eth),
1501 					error);
1502 	return ret;
1503 }
1504 
1505 /**
1506  * Validate VLAN item.
1507  *
1508  * @param[in] item
1509  *   Item specification.
1510  * @param[in] item_flags
1511  *   Bit-fields that holds the items detected until now.
1512  * @param[in] dev
1513  *   Ethernet device flow is being created on.
1514  * @param[out] error
1515  *   Pointer to error structure.
1516  *
1517  * @return
1518  *   0 on success, a negative errno value otherwise and rte_errno is set.
1519  */
1520 int
1521 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1522 			     uint64_t item_flags,
1523 			     struct rte_eth_dev *dev,
1524 			     struct rte_flow_error *error)
1525 {
1526 	const struct rte_flow_item_vlan *spec = item->spec;
1527 	const struct rte_flow_item_vlan *mask = item->mask;
1528 	const struct rte_flow_item_vlan nic_mask = {
1529 		.tci = RTE_BE16(UINT16_MAX),
1530 		.inner_type = RTE_BE16(UINT16_MAX),
1531 	};
1532 	uint16_t vlan_tag = 0;
1533 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1534 	int ret;
1535 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1536 					MLX5_FLOW_LAYER_INNER_L4) :
1537 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1538 					MLX5_FLOW_LAYER_OUTER_L4);
1539 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1540 					MLX5_FLOW_LAYER_OUTER_VLAN;
1541 
1542 	if (item_flags & vlanm)
1543 		return rte_flow_error_set(error, EINVAL,
1544 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1545 					  "multiple VLAN layers not supported");
1546 	else if ((item_flags & l34m) != 0)
1547 		return rte_flow_error_set(error, EINVAL,
1548 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1549 					  "VLAN cannot follow L3/L4 layer");
1550 	if (!mask)
1551 		mask = &rte_flow_item_vlan_mask;
1552 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1553 					(const uint8_t *)&nic_mask,
1554 					sizeof(struct rte_flow_item_vlan),
1555 					error);
1556 	if (ret)
1557 		return ret;
1558 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1559 		struct mlx5_priv *priv = dev->data->dev_private;
1560 
1561 		if (priv->vmwa_context) {
1562 			/*
1563 			 * Non-NULL context means we have a virtual machine
1564 			 * and SR-IOV enabled, we have to create VLAN interface
1565 			 * to make hypervisor to setup E-Switch vport
1566 			 * context correctly. We avoid creating the multiple
1567 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1568 			 */
1569 			return rte_flow_error_set(error, EINVAL,
1570 						  RTE_FLOW_ERROR_TYPE_ITEM,
1571 						  item,
1572 						  "VLAN tag mask is not"
1573 						  " supported in virtual"
1574 						  " environment");
1575 		}
1576 	}
1577 	if (spec) {
1578 		vlan_tag = spec->tci;
1579 		vlan_tag &= mask->tci;
1580 	}
1581 	/*
1582 	 * From verbs perspective an empty VLAN is equivalent
1583 	 * to a packet without VLAN layer.
1584 	 */
1585 	if (!vlan_tag)
1586 		return rte_flow_error_set(error, EINVAL,
1587 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1588 					  item->spec,
1589 					  "VLAN cannot be empty");
1590 	return 0;
1591 }
1592 
1593 /**
1594  * Validate IPV4 item.
1595  *
1596  * @param[in] item
1597  *   Item specification.
1598  * @param[in] item_flags
1599  *   Bit-fields that holds the items detected until now.
1600  * @param[in] acc_mask
1601  *   Acceptable mask, if NULL default internal default mask
1602  *   will be used to check whether item fields are supported.
1603  * @param[out] error
1604  *   Pointer to error structure.
1605  *
1606  * @return
1607  *   0 on success, a negative errno value otherwise and rte_errno is set.
1608  */
1609 int
1610 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1611 			     uint64_t item_flags,
1612 			     uint64_t last_item,
1613 			     uint16_t ether_type,
1614 			     const struct rte_flow_item_ipv4 *acc_mask,
1615 			     struct rte_flow_error *error)
1616 {
1617 	const struct rte_flow_item_ipv4 *mask = item->mask;
1618 	const struct rte_flow_item_ipv4 *spec = item->spec;
1619 	const struct rte_flow_item_ipv4 nic_mask = {
1620 		.hdr = {
1621 			.src_addr = RTE_BE32(0xffffffff),
1622 			.dst_addr = RTE_BE32(0xffffffff),
1623 			.type_of_service = 0xff,
1624 			.next_proto_id = 0xff,
1625 		},
1626 	};
1627 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1628 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1629 				      MLX5_FLOW_LAYER_OUTER_L3;
1630 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1631 				      MLX5_FLOW_LAYER_OUTER_L4;
1632 	int ret;
1633 	uint8_t next_proto = 0xFF;
1634 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1635 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1636 				  MLX5_FLOW_LAYER_INNER_VLAN);
1637 
1638 	if ((last_item & l2_vlan) && ether_type &&
1639 	    ether_type != RTE_ETHER_TYPE_IPV4)
1640 		return rte_flow_error_set(error, EINVAL,
1641 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1642 					  "IPv4 cannot follow L2/VLAN layer "
1643 					  "which ether type is not IPv4");
1644 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1645 		if (mask && spec)
1646 			next_proto = mask->hdr.next_proto_id &
1647 				     spec->hdr.next_proto_id;
1648 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1649 			return rte_flow_error_set(error, EINVAL,
1650 						  RTE_FLOW_ERROR_TYPE_ITEM,
1651 						  item,
1652 						  "multiple tunnel "
1653 						  "not supported");
1654 	}
1655 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1656 		return rte_flow_error_set(error, EINVAL,
1657 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1658 					  "wrong tunnel type - IPv6 specified "
1659 					  "but IPv4 item provided");
1660 	if (item_flags & l3m)
1661 		return rte_flow_error_set(error, ENOTSUP,
1662 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1663 					  "multiple L3 layers not supported");
1664 	else if (item_flags & l4m)
1665 		return rte_flow_error_set(error, EINVAL,
1666 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1667 					  "L3 cannot follow an L4 layer.");
1668 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1669 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1670 		return rte_flow_error_set(error, EINVAL,
1671 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1672 					  "L3 cannot follow an NVGRE layer.");
1673 	if (!mask)
1674 		mask = &rte_flow_item_ipv4_mask;
1675 	else if (mask->hdr.next_proto_id != 0 &&
1676 		 mask->hdr.next_proto_id != 0xff)
1677 		return rte_flow_error_set(error, EINVAL,
1678 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1679 					  "partial mask is not supported"
1680 					  " for protocol");
1681 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1682 					acc_mask ? (const uint8_t *)acc_mask
1683 						 : (const uint8_t *)&nic_mask,
1684 					sizeof(struct rte_flow_item_ipv4),
1685 					error);
1686 	if (ret < 0)
1687 		return ret;
1688 	return 0;
1689 }
1690 
1691 /**
1692  * Validate IPV6 item.
1693  *
1694  * @param[in] item
1695  *   Item specification.
1696  * @param[in] item_flags
1697  *   Bit-fields that holds the items detected until now.
1698  * @param[in] acc_mask
1699  *   Acceptable mask, if NULL default internal default mask
1700  *   will be used to check whether item fields are supported.
1701  * @param[out] error
1702  *   Pointer to error structure.
1703  *
1704  * @return
1705  *   0 on success, a negative errno value otherwise and rte_errno is set.
1706  */
1707 int
1708 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1709 			     uint64_t item_flags,
1710 			     uint64_t last_item,
1711 			     uint16_t ether_type,
1712 			     const struct rte_flow_item_ipv6 *acc_mask,
1713 			     struct rte_flow_error *error)
1714 {
1715 	const struct rte_flow_item_ipv6 *mask = item->mask;
1716 	const struct rte_flow_item_ipv6 *spec = item->spec;
1717 	const struct rte_flow_item_ipv6 nic_mask = {
1718 		.hdr = {
1719 			.src_addr =
1720 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1721 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1722 			.dst_addr =
1723 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1724 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1725 			.vtc_flow = RTE_BE32(0xffffffff),
1726 			.proto = 0xff,
1727 		},
1728 	};
1729 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1730 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1731 				      MLX5_FLOW_LAYER_OUTER_L3;
1732 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1733 				      MLX5_FLOW_LAYER_OUTER_L4;
1734 	int ret;
1735 	uint8_t next_proto = 0xFF;
1736 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1737 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1738 				  MLX5_FLOW_LAYER_INNER_VLAN);
1739 
1740 	if ((last_item & l2_vlan) && ether_type &&
1741 	    ether_type != RTE_ETHER_TYPE_IPV6)
1742 		return rte_flow_error_set(error, EINVAL,
1743 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1744 					  "IPv6 cannot follow L2/VLAN layer "
1745 					  "which ether type is not IPv6");
1746 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1747 		if (mask && spec)
1748 			next_proto = mask->hdr.proto & spec->hdr.proto;
1749 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1750 			return rte_flow_error_set(error, EINVAL,
1751 						  RTE_FLOW_ERROR_TYPE_ITEM,
1752 						  item,
1753 						  "multiple tunnel "
1754 						  "not supported");
1755 	}
1756 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1757 		return rte_flow_error_set(error, EINVAL,
1758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1759 					  "wrong tunnel type - IPv4 specified "
1760 					  "but IPv6 item provided");
1761 	if (item_flags & l3m)
1762 		return rte_flow_error_set(error, ENOTSUP,
1763 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1764 					  "multiple L3 layers not supported");
1765 	else if (item_flags & l4m)
1766 		return rte_flow_error_set(error, EINVAL,
1767 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1768 					  "L3 cannot follow an L4 layer.");
1769 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1770 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1771 		return rte_flow_error_set(error, EINVAL,
1772 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1773 					  "L3 cannot follow an NVGRE layer.");
1774 	if (!mask)
1775 		mask = &rte_flow_item_ipv6_mask;
1776 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1777 					acc_mask ? (const uint8_t *)acc_mask
1778 						 : (const uint8_t *)&nic_mask,
1779 					sizeof(struct rte_flow_item_ipv6),
1780 					error);
1781 	if (ret < 0)
1782 		return ret;
1783 	return 0;
1784 }
1785 
1786 /**
1787  * Validate UDP item.
1788  *
1789  * @param[in] item
1790  *   Item specification.
1791  * @param[in] item_flags
1792  *   Bit-fields that holds the items detected until now.
1793  * @param[in] target_protocol
1794  *   The next protocol in the previous item.
1795  * @param[in] flow_mask
1796  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1797  * @param[out] error
1798  *   Pointer to error structure.
1799  *
1800  * @return
1801  *   0 on success, a negative errno value otherwise and rte_errno is set.
1802  */
1803 int
1804 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1805 			    uint64_t item_flags,
1806 			    uint8_t target_protocol,
1807 			    struct rte_flow_error *error)
1808 {
1809 	const struct rte_flow_item_udp *mask = item->mask;
1810 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1811 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1812 				      MLX5_FLOW_LAYER_OUTER_L3;
1813 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1814 				      MLX5_FLOW_LAYER_OUTER_L4;
1815 	int ret;
1816 
1817 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1818 		return rte_flow_error_set(error, EINVAL,
1819 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1820 					  "protocol filtering not compatible"
1821 					  " with UDP layer");
1822 	if (!(item_flags & l3m))
1823 		return rte_flow_error_set(error, EINVAL,
1824 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1825 					  "L3 is mandatory to filter on L4");
1826 	if (item_flags & l4m)
1827 		return rte_flow_error_set(error, EINVAL,
1828 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1829 					  "multiple L4 layers not supported");
1830 	if (!mask)
1831 		mask = &rte_flow_item_udp_mask;
1832 	ret = mlx5_flow_item_acceptable
1833 		(item, (const uint8_t *)mask,
1834 		 (const uint8_t *)&rte_flow_item_udp_mask,
1835 		 sizeof(struct rte_flow_item_udp), error);
1836 	if (ret < 0)
1837 		return ret;
1838 	return 0;
1839 }
1840 
1841 /**
1842  * Validate TCP item.
1843  *
1844  * @param[in] item
1845  *   Item specification.
1846  * @param[in] item_flags
1847  *   Bit-fields that holds the items detected until now.
1848  * @param[in] target_protocol
1849  *   The next protocol in the previous item.
1850  * @param[out] error
1851  *   Pointer to error structure.
1852  *
1853  * @return
1854  *   0 on success, a negative errno value otherwise and rte_errno is set.
1855  */
1856 int
1857 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1858 			    uint64_t item_flags,
1859 			    uint8_t target_protocol,
1860 			    const struct rte_flow_item_tcp *flow_mask,
1861 			    struct rte_flow_error *error)
1862 {
1863 	const struct rte_flow_item_tcp *mask = item->mask;
1864 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1865 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1866 				      MLX5_FLOW_LAYER_OUTER_L3;
1867 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1868 				      MLX5_FLOW_LAYER_OUTER_L4;
1869 	int ret;
1870 
1871 	MLX5_ASSERT(flow_mask);
1872 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1873 		return rte_flow_error_set(error, EINVAL,
1874 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1875 					  "protocol filtering not compatible"
1876 					  " with TCP layer");
1877 	if (!(item_flags & l3m))
1878 		return rte_flow_error_set(error, EINVAL,
1879 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1880 					  "L3 is mandatory to filter on L4");
1881 	if (item_flags & l4m)
1882 		return rte_flow_error_set(error, EINVAL,
1883 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1884 					  "multiple L4 layers not supported");
1885 	if (!mask)
1886 		mask = &rte_flow_item_tcp_mask;
1887 	ret = mlx5_flow_item_acceptable
1888 		(item, (const uint8_t *)mask,
1889 		 (const uint8_t *)flow_mask,
1890 		 sizeof(struct rte_flow_item_tcp), error);
1891 	if (ret < 0)
1892 		return ret;
1893 	return 0;
1894 }
1895 
1896 /**
1897  * Validate VXLAN item.
1898  *
1899  * @param[in] item
1900  *   Item specification.
1901  * @param[in] item_flags
1902  *   Bit-fields that holds the items detected until now.
1903  * @param[in] target_protocol
1904  *   The next protocol in the previous item.
1905  * @param[out] error
1906  *   Pointer to error structure.
1907  *
1908  * @return
1909  *   0 on success, a negative errno value otherwise and rte_errno is set.
1910  */
1911 int
1912 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1913 			      uint64_t item_flags,
1914 			      struct rte_flow_error *error)
1915 {
1916 	const struct rte_flow_item_vxlan *spec = item->spec;
1917 	const struct rte_flow_item_vxlan *mask = item->mask;
1918 	int ret;
1919 	union vni {
1920 		uint32_t vlan_id;
1921 		uint8_t vni[4];
1922 	} id = { .vlan_id = 0, };
1923 
1924 
1925 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1926 		return rte_flow_error_set(error, ENOTSUP,
1927 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1928 					  "multiple tunnel layers not"
1929 					  " supported");
1930 	/*
1931 	 * Verify only UDPv4 is present as defined in
1932 	 * https://tools.ietf.org/html/rfc7348
1933 	 */
1934 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1935 		return rte_flow_error_set(error, EINVAL,
1936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1937 					  "no outer UDP layer found");
1938 	if (!mask)
1939 		mask = &rte_flow_item_vxlan_mask;
1940 	ret = mlx5_flow_item_acceptable
1941 		(item, (const uint8_t *)mask,
1942 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1943 		 sizeof(struct rte_flow_item_vxlan),
1944 		 error);
1945 	if (ret < 0)
1946 		return ret;
1947 	if (spec) {
1948 		memcpy(&id.vni[1], spec->vni, 3);
1949 		memcpy(&id.vni[1], mask->vni, 3);
1950 	}
1951 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1952 		return rte_flow_error_set(error, ENOTSUP,
1953 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1954 					  "VXLAN tunnel must be fully defined");
1955 	return 0;
1956 }
1957 
1958 /**
1959  * Validate VXLAN_GPE item.
1960  *
1961  * @param[in] item
1962  *   Item specification.
1963  * @param[in] item_flags
1964  *   Bit-fields that holds the items detected until now.
1965  * @param[in] priv
1966  *   Pointer to the private data structure.
1967  * @param[in] target_protocol
1968  *   The next protocol in the previous item.
1969  * @param[out] error
1970  *   Pointer to error structure.
1971  *
1972  * @return
1973  *   0 on success, a negative errno value otherwise and rte_errno is set.
1974  */
1975 int
1976 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1977 				  uint64_t item_flags,
1978 				  struct rte_eth_dev *dev,
1979 				  struct rte_flow_error *error)
1980 {
1981 	struct mlx5_priv *priv = dev->data->dev_private;
1982 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1983 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1984 	int ret;
1985 	union vni {
1986 		uint32_t vlan_id;
1987 		uint8_t vni[4];
1988 	} id = { .vlan_id = 0, };
1989 
1990 	if (!priv->config.l3_vxlan_en)
1991 		return rte_flow_error_set(error, ENOTSUP,
1992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 					  "L3 VXLAN is not enabled by device"
1994 					  " parameter and/or not configured in"
1995 					  " firmware");
1996 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1997 		return rte_flow_error_set(error, ENOTSUP,
1998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 					  "multiple tunnel layers not"
2000 					  " supported");
2001 	/*
2002 	 * Verify only UDPv4 is present as defined in
2003 	 * https://tools.ietf.org/html/rfc7348
2004 	 */
2005 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2006 		return rte_flow_error_set(error, EINVAL,
2007 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2008 					  "no outer UDP layer found");
2009 	if (!mask)
2010 		mask = &rte_flow_item_vxlan_gpe_mask;
2011 	ret = mlx5_flow_item_acceptable
2012 		(item, (const uint8_t *)mask,
2013 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2014 		 sizeof(struct rte_flow_item_vxlan_gpe),
2015 		 error);
2016 	if (ret < 0)
2017 		return ret;
2018 	if (spec) {
2019 		if (spec->protocol)
2020 			return rte_flow_error_set(error, ENOTSUP,
2021 						  RTE_FLOW_ERROR_TYPE_ITEM,
2022 						  item,
2023 						  "VxLAN-GPE protocol"
2024 						  " not supported");
2025 		memcpy(&id.vni[1], spec->vni, 3);
2026 		memcpy(&id.vni[1], mask->vni, 3);
2027 	}
2028 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2029 		return rte_flow_error_set(error, ENOTSUP,
2030 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2031 					  "VXLAN-GPE tunnel must be fully"
2032 					  " defined");
2033 	return 0;
2034 }
2035 /**
2036  * Validate GRE Key item.
2037  *
2038  * @param[in] item
2039  *   Item specification.
2040  * @param[in] item_flags
2041  *   Bit flags to mark detected items.
2042  * @param[in] gre_item
2043  *   Pointer to gre_item
2044  * @param[out] error
2045  *   Pointer to error structure.
2046  *
2047  * @return
2048  *   0 on success, a negative errno value otherwise and rte_errno is set.
2049  */
2050 int
2051 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2052 				uint64_t item_flags,
2053 				const struct rte_flow_item *gre_item,
2054 				struct rte_flow_error *error)
2055 {
2056 	const rte_be32_t *mask = item->mask;
2057 	int ret = 0;
2058 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2059 	const struct rte_flow_item_gre *gre_spec;
2060 	const struct rte_flow_item_gre *gre_mask;
2061 
2062 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2063 		return rte_flow_error_set(error, ENOTSUP,
2064 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2065 					  "Multiple GRE key not support");
2066 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2067 		return rte_flow_error_set(error, ENOTSUP,
2068 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2069 					  "No preceding GRE header");
2070 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2071 		return rte_flow_error_set(error, ENOTSUP,
2072 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2073 					  "GRE key following a wrong item");
2074 	gre_mask = gre_item->mask;
2075 	if (!gre_mask)
2076 		gre_mask = &rte_flow_item_gre_mask;
2077 	gre_spec = gre_item->spec;
2078 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2079 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2080 		return rte_flow_error_set(error, EINVAL,
2081 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2082 					  "Key bit must be on");
2083 
2084 	if (!mask)
2085 		mask = &gre_key_default_mask;
2086 	ret = mlx5_flow_item_acceptable
2087 		(item, (const uint8_t *)mask,
2088 		 (const uint8_t *)&gre_key_default_mask,
2089 		 sizeof(rte_be32_t), error);
2090 	return ret;
2091 }
2092 
2093 /**
2094  * Validate GRE item.
2095  *
2096  * @param[in] item
2097  *   Item specification.
2098  * @param[in] item_flags
2099  *   Bit flags to mark detected items.
2100  * @param[in] target_protocol
2101  *   The next protocol in the previous item.
2102  * @param[out] error
2103  *   Pointer to error structure.
2104  *
2105  * @return
2106  *   0 on success, a negative errno value otherwise and rte_errno is set.
2107  */
2108 int
2109 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2110 			    uint64_t item_flags,
2111 			    uint8_t target_protocol,
2112 			    struct rte_flow_error *error)
2113 {
2114 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2115 	const struct rte_flow_item_gre *mask = item->mask;
2116 	int ret;
2117 	const struct rte_flow_item_gre nic_mask = {
2118 		.c_rsvd0_ver = RTE_BE16(0xB000),
2119 		.protocol = RTE_BE16(UINT16_MAX),
2120 	};
2121 
2122 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2123 		return rte_flow_error_set(error, EINVAL,
2124 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2125 					  "protocol filtering not compatible"
2126 					  " with this GRE layer");
2127 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2128 		return rte_flow_error_set(error, ENOTSUP,
2129 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2130 					  "multiple tunnel layers not"
2131 					  " supported");
2132 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2133 		return rte_flow_error_set(error, ENOTSUP,
2134 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2135 					  "L3 Layer is missing");
2136 	if (!mask)
2137 		mask = &rte_flow_item_gre_mask;
2138 	ret = mlx5_flow_item_acceptable
2139 		(item, (const uint8_t *)mask,
2140 		 (const uint8_t *)&nic_mask,
2141 		 sizeof(struct rte_flow_item_gre), error);
2142 	if (ret < 0)
2143 		return ret;
2144 #ifndef HAVE_MLX5DV_DR
2145 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2146 	if (spec && (spec->protocol & mask->protocol))
2147 		return rte_flow_error_set(error, ENOTSUP,
2148 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2149 					  "without MPLS support the"
2150 					  " specification cannot be used for"
2151 					  " filtering");
2152 #endif
2153 #endif
2154 	return 0;
2155 }
2156 
2157 /**
2158  * Validate Geneve item.
2159  *
2160  * @param[in] item
2161  *   Item specification.
2162  * @param[in] itemFlags
2163  *   Bit-fields that holds the items detected until now.
2164  * @param[in] enPriv
2165  *   Pointer to the private data structure.
2166  * @param[out] error
2167  *   Pointer to error structure.
2168  *
2169  * @return
2170  *   0 on success, a negative errno value otherwise and rte_errno is set.
2171  */
2172 
2173 int
2174 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2175 			       uint64_t item_flags,
2176 			       struct rte_eth_dev *dev,
2177 			       struct rte_flow_error *error)
2178 {
2179 	struct mlx5_priv *priv = dev->data->dev_private;
2180 	const struct rte_flow_item_geneve *spec = item->spec;
2181 	const struct rte_flow_item_geneve *mask = item->mask;
2182 	int ret;
2183 	uint16_t gbhdr;
2184 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2185 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2186 	const struct rte_flow_item_geneve nic_mask = {
2187 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2188 		.vni = "\xff\xff\xff",
2189 		.protocol = RTE_BE16(UINT16_MAX),
2190 	};
2191 
2192 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2193 		return rte_flow_error_set(error, ENOTSUP,
2194 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2195 					  "L3 Geneve is not enabled by device"
2196 					  " parameter and/or not configured in"
2197 					  " firmware");
2198 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2199 		return rte_flow_error_set(error, ENOTSUP,
2200 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2201 					  "multiple tunnel layers not"
2202 					  " supported");
2203 	/*
2204 	 * Verify only UDPv4 is present as defined in
2205 	 * https://tools.ietf.org/html/rfc7348
2206 	 */
2207 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2208 		return rte_flow_error_set(error, EINVAL,
2209 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2210 					  "no outer UDP layer found");
2211 	if (!mask)
2212 		mask = &rte_flow_item_geneve_mask;
2213 	ret = mlx5_flow_item_acceptable
2214 				  (item, (const uint8_t *)mask,
2215 				   (const uint8_t *)&nic_mask,
2216 				   sizeof(struct rte_flow_item_geneve), error);
2217 	if (ret)
2218 		return ret;
2219 	if (spec) {
2220 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2221 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2222 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2223 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2224 			return rte_flow_error_set(error, ENOTSUP,
2225 						  RTE_FLOW_ERROR_TYPE_ITEM,
2226 						  item,
2227 						  "Geneve protocol unsupported"
2228 						  " fields are being used");
2229 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2230 			return rte_flow_error_set
2231 					(error, ENOTSUP,
2232 					 RTE_FLOW_ERROR_TYPE_ITEM,
2233 					 item,
2234 					 "Unsupported Geneve options length");
2235 	}
2236 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2237 		return rte_flow_error_set
2238 				    (error, ENOTSUP,
2239 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 				     "Geneve tunnel must be fully defined");
2241 	return 0;
2242 }
2243 
2244 /**
2245  * Validate MPLS item.
2246  *
2247  * @param[in] dev
2248  *   Pointer to the rte_eth_dev structure.
2249  * @param[in] item
2250  *   Item specification.
2251  * @param[in] item_flags
2252  *   Bit-fields that holds the items detected until now.
2253  * @param[in] prev_layer
2254  *   The protocol layer indicated in previous item.
2255  * @param[out] error
2256  *   Pointer to error structure.
2257  *
2258  * @return
2259  *   0 on success, a negative errno value otherwise and rte_errno is set.
2260  */
2261 int
2262 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2263 			     const struct rte_flow_item *item __rte_unused,
2264 			     uint64_t item_flags __rte_unused,
2265 			     uint64_t prev_layer __rte_unused,
2266 			     struct rte_flow_error *error)
2267 {
2268 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2269 	const struct rte_flow_item_mpls *mask = item->mask;
2270 	struct mlx5_priv *priv = dev->data->dev_private;
2271 	int ret;
2272 
2273 	if (!priv->config.mpls_en)
2274 		return rte_flow_error_set(error, ENOTSUP,
2275 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 					  "MPLS not supported or"
2277 					  " disabled in firmware"
2278 					  " configuration.");
2279 	/* MPLS over IP, UDP, GRE is allowed */
2280 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2281 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2282 			    MLX5_FLOW_LAYER_GRE)))
2283 		return rte_flow_error_set(error, EINVAL,
2284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 					  "protocol filtering not compatible"
2286 					  " with MPLS layer");
2287 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2288 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2289 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2290 		return rte_flow_error_set(error, ENOTSUP,
2291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2292 					  "multiple tunnel layers not"
2293 					  " supported");
2294 	if (!mask)
2295 		mask = &rte_flow_item_mpls_mask;
2296 	ret = mlx5_flow_item_acceptable
2297 		(item, (const uint8_t *)mask,
2298 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2299 		 sizeof(struct rte_flow_item_mpls), error);
2300 	if (ret < 0)
2301 		return ret;
2302 	return 0;
2303 #else
2304 	return rte_flow_error_set(error, ENOTSUP,
2305 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2306 				  "MPLS is not supported by Verbs, please"
2307 				  " update.");
2308 #endif
2309 }
2310 
2311 /**
2312  * Validate NVGRE item.
2313  *
2314  * @param[in] item
2315  *   Item specification.
2316  * @param[in] item_flags
2317  *   Bit flags to mark detected items.
2318  * @param[in] target_protocol
2319  *   The next protocol in the previous item.
2320  * @param[out] error
2321  *   Pointer to error structure.
2322  *
2323  * @return
2324  *   0 on success, a negative errno value otherwise and rte_errno is set.
2325  */
2326 int
2327 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2328 			      uint64_t item_flags,
2329 			      uint8_t target_protocol,
2330 			      struct rte_flow_error *error)
2331 {
2332 	const struct rte_flow_item_nvgre *mask = item->mask;
2333 	int ret;
2334 
2335 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2336 		return rte_flow_error_set(error, EINVAL,
2337 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2338 					  "protocol filtering not compatible"
2339 					  " with this GRE layer");
2340 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2341 		return rte_flow_error_set(error, ENOTSUP,
2342 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2343 					  "multiple tunnel layers not"
2344 					  " supported");
2345 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2346 		return rte_flow_error_set(error, ENOTSUP,
2347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2348 					  "L3 Layer is missing");
2349 	if (!mask)
2350 		mask = &rte_flow_item_nvgre_mask;
2351 	ret = mlx5_flow_item_acceptable
2352 		(item, (const uint8_t *)mask,
2353 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2354 		 sizeof(struct rte_flow_item_nvgre), error);
2355 	if (ret < 0)
2356 		return ret;
2357 	return 0;
2358 }
2359 
2360 /* Allocate unique ID for the split Q/RSS subflows. */
2361 static uint32_t
2362 flow_qrss_get_id(struct rte_eth_dev *dev)
2363 {
2364 	struct mlx5_priv *priv = dev->data->dev_private;
2365 	uint32_t qrss_id, ret;
2366 
2367 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2368 	if (ret)
2369 		return 0;
2370 	MLX5_ASSERT(qrss_id);
2371 	return qrss_id;
2372 }
2373 
2374 /* Free unique ID for the split Q/RSS subflows. */
2375 static void
2376 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2377 {
2378 	struct mlx5_priv *priv = dev->data->dev_private;
2379 
2380 	if (qrss_id)
2381 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2382 }
2383 
2384 /**
2385  * Release resource related QUEUE/RSS action split.
2386  *
2387  * @param dev
2388  *   Pointer to Ethernet device.
2389  * @param flow
2390  *   Flow to release id's from.
2391  */
2392 static void
2393 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2394 			     struct rte_flow *flow)
2395 {
2396 	struct mlx5_priv *priv = dev->data->dev_private;
2397 	uint32_t handle_idx;
2398 	struct mlx5_flow_handle *dev_handle;
2399 
2400 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2401 		       handle_idx, dev_handle, next)
2402 		if (dev_handle->split_flow_id)
2403 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2404 }
2405 
2406 static int
2407 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2408 		   const struct rte_flow_attr *attr __rte_unused,
2409 		   const struct rte_flow_item items[] __rte_unused,
2410 		   const struct rte_flow_action actions[] __rte_unused,
2411 		   bool external __rte_unused,
2412 		   int hairpin __rte_unused,
2413 		   struct rte_flow_error *error)
2414 {
2415 	return rte_flow_error_set(error, ENOTSUP,
2416 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2417 }
2418 
2419 static struct mlx5_flow *
2420 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2421 		  const struct rte_flow_attr *attr __rte_unused,
2422 		  const struct rte_flow_item items[] __rte_unused,
2423 		  const struct rte_flow_action actions[] __rte_unused,
2424 		  struct rte_flow_error *error)
2425 {
2426 	rte_flow_error_set(error, ENOTSUP,
2427 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2428 	return NULL;
2429 }
2430 
2431 static int
2432 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2433 		    struct mlx5_flow *dev_flow __rte_unused,
2434 		    const struct rte_flow_attr *attr __rte_unused,
2435 		    const struct rte_flow_item items[] __rte_unused,
2436 		    const struct rte_flow_action actions[] __rte_unused,
2437 		    struct rte_flow_error *error)
2438 {
2439 	return rte_flow_error_set(error, ENOTSUP,
2440 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2441 }
2442 
2443 static int
2444 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2445 		struct rte_flow *flow __rte_unused,
2446 		struct rte_flow_error *error)
2447 {
2448 	return rte_flow_error_set(error, ENOTSUP,
2449 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2450 }
2451 
2452 static void
2453 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2454 		 struct rte_flow *flow __rte_unused)
2455 {
2456 }
2457 
2458 static void
2459 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2460 		  struct rte_flow *flow __rte_unused)
2461 {
2462 }
2463 
2464 static int
2465 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2466 		struct rte_flow *flow __rte_unused,
2467 		const struct rte_flow_action *actions __rte_unused,
2468 		void *data __rte_unused,
2469 		struct rte_flow_error *error)
2470 {
2471 	return rte_flow_error_set(error, ENOTSUP,
2472 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2473 }
2474 
2475 /* Void driver to protect from null pointer reference. */
2476 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2477 	.validate = flow_null_validate,
2478 	.prepare = flow_null_prepare,
2479 	.translate = flow_null_translate,
2480 	.apply = flow_null_apply,
2481 	.remove = flow_null_remove,
2482 	.destroy = flow_null_destroy,
2483 	.query = flow_null_query,
2484 };
2485 
2486 /**
2487  * Select flow driver type according to flow attributes and device
2488  * configuration.
2489  *
2490  * @param[in] dev
2491  *   Pointer to the dev structure.
2492  * @param[in] attr
2493  *   Pointer to the flow attributes.
2494  *
2495  * @return
2496  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2497  */
2498 static enum mlx5_flow_drv_type
2499 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2500 {
2501 	struct mlx5_priv *priv = dev->data->dev_private;
2502 	/* The OS can determine first a specific flow type (DV, VERBS) */
2503 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2504 
2505 	if (type != MLX5_FLOW_TYPE_MAX)
2506 		return type;
2507 	/* If no OS specific type - continue with DV/VERBS selection */
2508 	if (attr->transfer && priv->config.dv_esw_en)
2509 		type = MLX5_FLOW_TYPE_DV;
2510 	if (!attr->transfer)
2511 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2512 						 MLX5_FLOW_TYPE_VERBS;
2513 	return type;
2514 }
2515 
2516 #define flow_get_drv_ops(type) flow_drv_ops[type]
2517 
2518 /**
2519  * Flow driver validation API. This abstracts calling driver specific functions.
2520  * The type of flow driver is determined according to flow attributes.
2521  *
2522  * @param[in] dev
2523  *   Pointer to the dev structure.
2524  * @param[in] attr
2525  *   Pointer to the flow attributes.
2526  * @param[in] items
2527  *   Pointer to the list of items.
2528  * @param[in] actions
2529  *   Pointer to the list of actions.
2530  * @param[in] external
2531  *   This flow rule is created by request external to PMD.
2532  * @param[in] hairpin
2533  *   Number of hairpin TX actions, 0 means classic flow.
2534  * @param[out] error
2535  *   Pointer to the error structure.
2536  *
2537  * @return
2538  *   0 on success, a negative errno value otherwise and rte_errno is set.
2539  */
2540 static inline int
2541 flow_drv_validate(struct rte_eth_dev *dev,
2542 		  const struct rte_flow_attr *attr,
2543 		  const struct rte_flow_item items[],
2544 		  const struct rte_flow_action actions[],
2545 		  bool external, int hairpin, struct rte_flow_error *error)
2546 {
2547 	const struct mlx5_flow_driver_ops *fops;
2548 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2549 
2550 	fops = flow_get_drv_ops(type);
2551 	return fops->validate(dev, attr, items, actions, external,
2552 			      hairpin, error);
2553 }
2554 
2555 /**
2556  * Flow driver preparation API. This abstracts calling driver specific
2557  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2558  * calculates the size of memory required for device flow, allocates the memory,
2559  * initializes the device flow and returns the pointer.
2560  *
2561  * @note
2562  *   This function initializes device flow structure such as dv or verbs in
2563  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2564  *   rest. For example, adding returning device flow to flow->dev_flow list and
2565  *   setting backward reference to the flow should be done out of this function.
2566  *   layers field is not filled either.
2567  *
2568  * @param[in] dev
2569  *   Pointer to the dev structure.
2570  * @param[in] attr
2571  *   Pointer to the flow attributes.
2572  * @param[in] items
2573  *   Pointer to the list of items.
2574  * @param[in] actions
2575  *   Pointer to the list of actions.
2576  * @param[in] flow_idx
2577  *   This memory pool index to the flow.
2578  * @param[out] error
2579  *   Pointer to the error structure.
2580  *
2581  * @return
2582  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2583  */
2584 static inline struct mlx5_flow *
2585 flow_drv_prepare(struct rte_eth_dev *dev,
2586 		 const struct rte_flow *flow,
2587 		 const struct rte_flow_attr *attr,
2588 		 const struct rte_flow_item items[],
2589 		 const struct rte_flow_action actions[],
2590 		 uint32_t flow_idx,
2591 		 struct rte_flow_error *error)
2592 {
2593 	const struct mlx5_flow_driver_ops *fops;
2594 	enum mlx5_flow_drv_type type = flow->drv_type;
2595 	struct mlx5_flow *mlx5_flow = NULL;
2596 
2597 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2598 	fops = flow_get_drv_ops(type);
2599 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2600 	if (mlx5_flow)
2601 		mlx5_flow->flow_idx = flow_idx;
2602 	return mlx5_flow;
2603 }
2604 
2605 /**
2606  * Flow driver translation API. This abstracts calling driver specific
2607  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2608  * translates a generic flow into a driver flow. flow_drv_prepare() must
2609  * precede.
2610  *
2611  * @note
2612  *   dev_flow->layers could be filled as a result of parsing during translation
2613  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2614  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2615  *   flow->actions could be overwritten even though all the expanded dev_flows
2616  *   have the same actions.
2617  *
2618  * @param[in] dev
2619  *   Pointer to the rte dev structure.
2620  * @param[in, out] dev_flow
2621  *   Pointer to the mlx5 flow.
2622  * @param[in] attr
2623  *   Pointer to the flow attributes.
2624  * @param[in] items
2625  *   Pointer to the list of items.
2626  * @param[in] actions
2627  *   Pointer to the list of actions.
2628  * @param[out] error
2629  *   Pointer to the error structure.
2630  *
2631  * @return
2632  *   0 on success, a negative errno value otherwise and rte_errno is set.
2633  */
2634 static inline int
2635 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2636 		   const struct rte_flow_attr *attr,
2637 		   const struct rte_flow_item items[],
2638 		   const struct rte_flow_action actions[],
2639 		   struct rte_flow_error *error)
2640 {
2641 	const struct mlx5_flow_driver_ops *fops;
2642 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2643 
2644 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2645 	fops = flow_get_drv_ops(type);
2646 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2647 }
2648 
2649 /**
2650  * Flow driver apply API. This abstracts calling driver specific functions.
2651  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2652  * translated driver flows on to device. flow_drv_translate() must precede.
2653  *
2654  * @param[in] dev
2655  *   Pointer to Ethernet device structure.
2656  * @param[in, out] flow
2657  *   Pointer to flow structure.
2658  * @param[out] error
2659  *   Pointer to error structure.
2660  *
2661  * @return
2662  *   0 on success, a negative errno value otherwise and rte_errno is set.
2663  */
2664 static inline int
2665 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2666 	       struct rte_flow_error *error)
2667 {
2668 	const struct mlx5_flow_driver_ops *fops;
2669 	enum mlx5_flow_drv_type type = flow->drv_type;
2670 
2671 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2672 	fops = flow_get_drv_ops(type);
2673 	return fops->apply(dev, flow, error);
2674 }
2675 
2676 /**
2677  * Flow driver remove API. This abstracts calling driver specific functions.
2678  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2679  * on device. All the resources of the flow should be freed by calling
2680  * flow_drv_destroy().
2681  *
2682  * @param[in] dev
2683  *   Pointer to Ethernet device.
2684  * @param[in, out] flow
2685  *   Pointer to flow structure.
2686  */
2687 static inline void
2688 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2689 {
2690 	const struct mlx5_flow_driver_ops *fops;
2691 	enum mlx5_flow_drv_type type = flow->drv_type;
2692 
2693 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2694 	fops = flow_get_drv_ops(type);
2695 	fops->remove(dev, flow);
2696 }
2697 
2698 /**
2699  * Flow driver destroy API. This abstracts calling driver specific functions.
2700  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2701  * on device and releases resources of the flow.
2702  *
2703  * @param[in] dev
2704  *   Pointer to Ethernet device.
2705  * @param[in, out] flow
2706  *   Pointer to flow structure.
2707  */
2708 static inline void
2709 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2710 {
2711 	const struct mlx5_flow_driver_ops *fops;
2712 	enum mlx5_flow_drv_type type = flow->drv_type;
2713 
2714 	flow_mreg_split_qrss_release(dev, flow);
2715 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2716 	fops = flow_get_drv_ops(type);
2717 	fops->destroy(dev, flow);
2718 }
2719 
2720 /**
2721  * Get RSS action from the action list.
2722  *
2723  * @param[in] actions
2724  *   Pointer to the list of actions.
2725  *
2726  * @return
2727  *   Pointer to the RSS action if exist, else return NULL.
2728  */
2729 static const struct rte_flow_action_rss*
2730 flow_get_rss_action(const struct rte_flow_action actions[])
2731 {
2732 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2733 		switch (actions->type) {
2734 		case RTE_FLOW_ACTION_TYPE_RSS:
2735 			return (const struct rte_flow_action_rss *)
2736 			       actions->conf;
2737 		default:
2738 			break;
2739 		}
2740 	}
2741 	return NULL;
2742 }
2743 
2744 static unsigned int
2745 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2746 {
2747 	const struct rte_flow_item *item;
2748 	unsigned int has_vlan = 0;
2749 
2750 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2751 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2752 			has_vlan = 1;
2753 			break;
2754 		}
2755 	}
2756 	if (has_vlan)
2757 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2758 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2759 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2760 			       MLX5_EXPANSION_ROOT_OUTER;
2761 }
2762 
2763 /**
2764  *  Get layer flags from the prefix flow.
2765  *
2766  *  Some flows may be split to several subflows, the prefix subflow gets the
2767  *  match items and the suffix sub flow gets the actions.
2768  *  Some actions need the user defined match item flags to get the detail for
2769  *  the action.
2770  *  This function helps the suffix flow to get the item layer flags from prefix
2771  *  subflow.
2772  *
2773  * @param[in] dev_flow
2774  *   Pointer the created preifx subflow.
2775  *
2776  * @return
2777  *   The layers get from prefix subflow.
2778  */
2779 static inline uint64_t
2780 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2781 {
2782 	uint64_t layers = 0;
2783 
2784 	/*
2785 	 * Layers bits could be localization, but usually the compiler will
2786 	 * help to do the optimization work for source code.
2787 	 * If no decap actions, use the layers directly.
2788 	 */
2789 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2790 		return dev_flow->handle->layers;
2791 	/* Convert L3 layers with decap action. */
2792 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2793 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2794 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2795 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2796 	/* Convert L4 layers with decap action.  */
2797 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2798 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2799 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2800 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2801 	return layers;
2802 }
2803 
2804 /**
2805  * Get metadata split action information.
2806  *
2807  * @param[in] actions
2808  *   Pointer to the list of actions.
2809  * @param[out] qrss
2810  *   Pointer to the return pointer.
2811  * @param[out] qrss_type
2812  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2813  *   if no QUEUE/RSS is found.
2814  * @param[out] encap_idx
2815  *   Pointer to the index of the encap action if exists, otherwise the last
2816  *   action index.
2817  *
2818  * @return
2819  *   Total number of actions.
2820  */
2821 static int
2822 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2823 				       const struct rte_flow_action **qrss,
2824 				       int *encap_idx)
2825 {
2826 	const struct rte_flow_action_raw_encap *raw_encap;
2827 	int actions_n = 0;
2828 	int raw_decap_idx = -1;
2829 
2830 	*encap_idx = -1;
2831 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2832 		switch (actions->type) {
2833 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2834 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2835 			*encap_idx = actions_n;
2836 			break;
2837 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2838 			raw_decap_idx = actions_n;
2839 			break;
2840 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2841 			raw_encap = actions->conf;
2842 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2843 				*encap_idx = raw_decap_idx != -1 ?
2844 						      raw_decap_idx : actions_n;
2845 			break;
2846 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2847 		case RTE_FLOW_ACTION_TYPE_RSS:
2848 			*qrss = actions;
2849 			break;
2850 		default:
2851 			break;
2852 		}
2853 		actions_n++;
2854 	}
2855 	if (*encap_idx == -1)
2856 		*encap_idx = actions_n;
2857 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2858 	return actions_n + 1;
2859 }
2860 
2861 /**
2862  * Check meter action from the action list.
2863  *
2864  * @param[in] actions
2865  *   Pointer to the list of actions.
2866  * @param[out] mtr
2867  *   Pointer to the meter exist flag.
2868  *
2869  * @return
2870  *   Total number of actions.
2871  */
2872 static int
2873 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2874 {
2875 	int actions_n = 0;
2876 
2877 	MLX5_ASSERT(mtr);
2878 	*mtr = 0;
2879 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2880 		switch (actions->type) {
2881 		case RTE_FLOW_ACTION_TYPE_METER:
2882 			*mtr = 1;
2883 			break;
2884 		default:
2885 			break;
2886 		}
2887 		actions_n++;
2888 	}
2889 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2890 	return actions_n + 1;
2891 }
2892 
2893 /**
2894  * Check if the flow should be splited due to hairpin.
2895  * The reason for the split is that in current HW we can't
2896  * support encap on Rx, so if a flow have encap we move it
2897  * to Tx.
2898  *
2899  * @param dev
2900  *   Pointer to Ethernet device.
2901  * @param[in] attr
2902  *   Flow rule attributes.
2903  * @param[in] actions
2904  *   Associated actions (list terminated by the END action).
2905  *
2906  * @return
2907  *   > 0 the number of actions and the flow should be split,
2908  *   0 when no split required.
2909  */
2910 static int
2911 flow_check_hairpin_split(struct rte_eth_dev *dev,
2912 			 const struct rte_flow_attr *attr,
2913 			 const struct rte_flow_action actions[])
2914 {
2915 	int queue_action = 0;
2916 	int action_n = 0;
2917 	int encap = 0;
2918 	const struct rte_flow_action_queue *queue;
2919 	const struct rte_flow_action_rss *rss;
2920 	const struct rte_flow_action_raw_encap *raw_encap;
2921 
2922 	if (!attr->ingress)
2923 		return 0;
2924 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2925 		switch (actions->type) {
2926 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2927 			queue = actions->conf;
2928 			if (queue == NULL)
2929 				return 0;
2930 			if (mlx5_rxq_get_type(dev, queue->index) !=
2931 			    MLX5_RXQ_TYPE_HAIRPIN)
2932 				return 0;
2933 			queue_action = 1;
2934 			action_n++;
2935 			break;
2936 		case RTE_FLOW_ACTION_TYPE_RSS:
2937 			rss = actions->conf;
2938 			if (rss == NULL || rss->queue_num == 0)
2939 				return 0;
2940 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2941 			    MLX5_RXQ_TYPE_HAIRPIN)
2942 				return 0;
2943 			queue_action = 1;
2944 			action_n++;
2945 			break;
2946 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2947 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2948 			encap = 1;
2949 			action_n++;
2950 			break;
2951 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2952 			raw_encap = actions->conf;
2953 			if (raw_encap->size >
2954 			    (sizeof(struct rte_flow_item_eth) +
2955 			     sizeof(struct rte_flow_item_ipv4)))
2956 				encap = 1;
2957 			action_n++;
2958 			break;
2959 		default:
2960 			action_n++;
2961 			break;
2962 		}
2963 	}
2964 	if (encap == 1 && queue_action)
2965 		return action_n;
2966 	return 0;
2967 }
2968 
2969 /* Declare flow create/destroy prototype in advance. */
2970 static uint32_t
2971 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2972 		 const struct rte_flow_attr *attr,
2973 		 const struct rte_flow_item items[],
2974 		 const struct rte_flow_action actions[],
2975 		 bool external, struct rte_flow_error *error);
2976 
2977 static void
2978 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2979 		  uint32_t flow_idx);
2980 
2981 /**
2982  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2983  *
2984  * As mark_id is unique, if there's already a registered flow for the mark_id,
2985  * return by increasing the reference counter of the resource. Otherwise, create
2986  * the resource (mcp_res) and flow.
2987  *
2988  * Flow looks like,
2989  *   - If ingress port is ANY and reg_c[1] is mark_id,
2990  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2991  *
2992  * For default flow (zero mark_id), flow is like,
2993  *   - If ingress port is ANY,
2994  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2995  *
2996  * @param dev
2997  *   Pointer to Ethernet device.
2998  * @param mark_id
2999  *   ID of MARK action, zero means default flow for META.
3000  * @param[out] error
3001  *   Perform verbose error reporting if not NULL.
3002  *
3003  * @return
3004  *   Associated resource on success, NULL otherwise and rte_errno is set.
3005  */
3006 static struct mlx5_flow_mreg_copy_resource *
3007 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3008 			  struct rte_flow_error *error)
3009 {
3010 	struct mlx5_priv *priv = dev->data->dev_private;
3011 	struct rte_flow_attr attr = {
3012 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3013 		.ingress = 1,
3014 	};
3015 	struct mlx5_rte_flow_item_tag tag_spec = {
3016 		.data = mark_id,
3017 	};
3018 	struct rte_flow_item items[] = {
3019 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3020 	};
3021 	struct rte_flow_action_mark ftag = {
3022 		.id = mark_id,
3023 	};
3024 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3025 		.dst = REG_B,
3026 		.src = 0,
3027 	};
3028 	struct rte_flow_action_jump jump = {
3029 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3030 	};
3031 	struct rte_flow_action actions[] = {
3032 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3033 	};
3034 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3035 	uint32_t idx = 0;
3036 	int ret;
3037 
3038 	/* Fill the register fileds in the flow. */
3039 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3040 	if (ret < 0)
3041 		return NULL;
3042 	tag_spec.id = ret;
3043 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3044 	if (ret < 0)
3045 		return NULL;
3046 	cp_mreg.src = ret;
3047 	/* Check if already registered. */
3048 	MLX5_ASSERT(priv->mreg_cp_tbl);
3049 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3050 	if (mcp_res) {
3051 		/* For non-default rule. */
3052 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3053 			mcp_res->refcnt++;
3054 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3055 			    mcp_res->refcnt == 1);
3056 		return mcp_res;
3057 	}
3058 	/* Provide the full width of FLAG specific value. */
3059 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3060 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3061 	/* Build a new flow. */
3062 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3063 		items[0] = (struct rte_flow_item){
3064 			.type = (enum rte_flow_item_type)
3065 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3066 			.spec = &tag_spec,
3067 		};
3068 		items[1] = (struct rte_flow_item){
3069 			.type = RTE_FLOW_ITEM_TYPE_END,
3070 		};
3071 		actions[0] = (struct rte_flow_action){
3072 			.type = (enum rte_flow_action_type)
3073 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3074 			.conf = &ftag,
3075 		};
3076 		actions[1] = (struct rte_flow_action){
3077 			.type = (enum rte_flow_action_type)
3078 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3079 			.conf = &cp_mreg,
3080 		};
3081 		actions[2] = (struct rte_flow_action){
3082 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3083 			.conf = &jump,
3084 		};
3085 		actions[3] = (struct rte_flow_action){
3086 			.type = RTE_FLOW_ACTION_TYPE_END,
3087 		};
3088 	} else {
3089 		/* Default rule, wildcard match. */
3090 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3091 		items[0] = (struct rte_flow_item){
3092 			.type = RTE_FLOW_ITEM_TYPE_END,
3093 		};
3094 		actions[0] = (struct rte_flow_action){
3095 			.type = (enum rte_flow_action_type)
3096 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3097 			.conf = &cp_mreg,
3098 		};
3099 		actions[1] = (struct rte_flow_action){
3100 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3101 			.conf = &jump,
3102 		};
3103 		actions[2] = (struct rte_flow_action){
3104 			.type = RTE_FLOW_ACTION_TYPE_END,
3105 		};
3106 	}
3107 	/* Build a new entry. */
3108 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3109 	if (!mcp_res) {
3110 		rte_errno = ENOMEM;
3111 		return NULL;
3112 	}
3113 	mcp_res->idx = idx;
3114 	/*
3115 	 * The copy Flows are not included in any list. There
3116 	 * ones are referenced from other Flows and can not
3117 	 * be applied, removed, deleted in ardbitrary order
3118 	 * by list traversing.
3119 	 */
3120 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3121 					 actions, false, error);
3122 	if (!mcp_res->rix_flow)
3123 		goto error;
3124 	mcp_res->refcnt++;
3125 	mcp_res->hlist_ent.key = mark_id;
3126 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3127 				&mcp_res->hlist_ent);
3128 	MLX5_ASSERT(!ret);
3129 	if (ret)
3130 		goto error;
3131 	return mcp_res;
3132 error:
3133 	if (mcp_res->rix_flow)
3134 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3135 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3136 	return NULL;
3137 }
3138 
3139 /**
3140  * Release flow in RX_CP_TBL.
3141  *
3142  * @param dev
3143  *   Pointer to Ethernet device.
3144  * @flow
3145  *   Parent flow for wich copying is provided.
3146  */
3147 static void
3148 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3149 			  struct rte_flow *flow)
3150 {
3151 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3152 	struct mlx5_priv *priv = dev->data->dev_private;
3153 
3154 	if (!flow->rix_mreg_copy)
3155 		return;
3156 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3157 				 flow->rix_mreg_copy);
3158 	if (!mcp_res || !priv->mreg_cp_tbl)
3159 		return;
3160 	if (flow->copy_applied) {
3161 		MLX5_ASSERT(mcp_res->appcnt);
3162 		flow->copy_applied = 0;
3163 		--mcp_res->appcnt;
3164 		if (!mcp_res->appcnt) {
3165 			struct rte_flow *mcp_flow = mlx5_ipool_get
3166 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3167 					mcp_res->rix_flow);
3168 
3169 			if (mcp_flow)
3170 				flow_drv_remove(dev, mcp_flow);
3171 		}
3172 	}
3173 	/*
3174 	 * We do not check availability of metadata registers here,
3175 	 * because copy resources are not allocated in this case.
3176 	 */
3177 	if (--mcp_res->refcnt)
3178 		return;
3179 	MLX5_ASSERT(mcp_res->rix_flow);
3180 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3181 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3182 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3183 	flow->rix_mreg_copy = 0;
3184 }
3185 
3186 /**
3187  * Start flow in RX_CP_TBL.
3188  *
3189  * @param dev
3190  *   Pointer to Ethernet device.
3191  * @flow
3192  *   Parent flow for wich copying is provided.
3193  *
3194  * @return
3195  *   0 on success, a negative errno value otherwise and rte_errno is set.
3196  */
3197 static int
3198 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3199 			    struct rte_flow *flow)
3200 {
3201 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3202 	struct mlx5_priv *priv = dev->data->dev_private;
3203 	int ret;
3204 
3205 	if (!flow->rix_mreg_copy || flow->copy_applied)
3206 		return 0;
3207 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3208 				 flow->rix_mreg_copy);
3209 	if (!mcp_res)
3210 		return 0;
3211 	if (!mcp_res->appcnt) {
3212 		struct rte_flow *mcp_flow = mlx5_ipool_get
3213 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3214 				mcp_res->rix_flow);
3215 
3216 		if (mcp_flow) {
3217 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3218 			if (ret)
3219 				return ret;
3220 		}
3221 	}
3222 	++mcp_res->appcnt;
3223 	flow->copy_applied = 1;
3224 	return 0;
3225 }
3226 
3227 /**
3228  * Stop flow in RX_CP_TBL.
3229  *
3230  * @param dev
3231  *   Pointer to Ethernet device.
3232  * @flow
3233  *   Parent flow for wich copying is provided.
3234  */
3235 static void
3236 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3237 			   struct rte_flow *flow)
3238 {
3239 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3240 	struct mlx5_priv *priv = dev->data->dev_private;
3241 
3242 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3243 		return;
3244 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3245 				 flow->rix_mreg_copy);
3246 	if (!mcp_res)
3247 		return;
3248 	MLX5_ASSERT(mcp_res->appcnt);
3249 	--mcp_res->appcnt;
3250 	flow->copy_applied = 0;
3251 	if (!mcp_res->appcnt) {
3252 		struct rte_flow *mcp_flow = mlx5_ipool_get
3253 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3254 				mcp_res->rix_flow);
3255 
3256 		if (mcp_flow)
3257 			flow_drv_remove(dev, mcp_flow);
3258 	}
3259 }
3260 
3261 /**
3262  * Remove the default copy action from RX_CP_TBL.
3263  *
3264  * @param dev
3265  *   Pointer to Ethernet device.
3266  */
3267 static void
3268 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3269 {
3270 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3271 	struct mlx5_priv *priv = dev->data->dev_private;
3272 
3273 	/* Check if default flow is registered. */
3274 	if (!priv->mreg_cp_tbl)
3275 		return;
3276 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3277 					    MLX5_DEFAULT_COPY_ID);
3278 	if (!mcp_res)
3279 		return;
3280 	MLX5_ASSERT(mcp_res->rix_flow);
3281 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3282 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3283 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3284 }
3285 
3286 /**
3287  * Add the default copy action in in RX_CP_TBL.
3288  *
3289  * @param dev
3290  *   Pointer to Ethernet device.
3291  * @param[out] error
3292  *   Perform verbose error reporting if not NULL.
3293  *
3294  * @return
3295  *   0 for success, negative value otherwise and rte_errno is set.
3296  */
3297 static int
3298 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3299 				  struct rte_flow_error *error)
3300 {
3301 	struct mlx5_priv *priv = dev->data->dev_private;
3302 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3303 
3304 	/* Check whether extensive metadata feature is engaged. */
3305 	if (!priv->config.dv_flow_en ||
3306 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3307 	    !mlx5_flow_ext_mreg_supported(dev) ||
3308 	    !priv->sh->dv_regc0_mask)
3309 		return 0;
3310 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3311 	if (!mcp_res)
3312 		return -rte_errno;
3313 	return 0;
3314 }
3315 
3316 /**
3317  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3318  *
3319  * All the flow having Q/RSS action should be split by
3320  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3321  * performs the following,
3322  *   - CQE->flow_tag := reg_c[1] (MARK)
3323  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3324  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3325  * but there should be a flow per each MARK ID set by MARK action.
3326  *
3327  * For the aforementioned reason, if there's a MARK action in flow's action
3328  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3329  * the MARK ID to CQE's flow_tag like,
3330  *   - If reg_c[1] is mark_id,
3331  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3332  *
3333  * For SET_META action which stores value in reg_c[0], as the destination is
3334  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3335  * MARK ID means the default flow. The default flow looks like,
3336  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3337  *
3338  * @param dev
3339  *   Pointer to Ethernet device.
3340  * @param flow
3341  *   Pointer to flow structure.
3342  * @param[in] actions
3343  *   Pointer to the list of actions.
3344  * @param[out] error
3345  *   Perform verbose error reporting if not NULL.
3346  *
3347  * @return
3348  *   0 on success, negative value otherwise and rte_errno is set.
3349  */
3350 static int
3351 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3352 			    struct rte_flow *flow,
3353 			    const struct rte_flow_action *actions,
3354 			    struct rte_flow_error *error)
3355 {
3356 	struct mlx5_priv *priv = dev->data->dev_private;
3357 	struct mlx5_dev_config *config = &priv->config;
3358 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3359 	const struct rte_flow_action_mark *mark;
3360 
3361 	/* Check whether extensive metadata feature is engaged. */
3362 	if (!config->dv_flow_en ||
3363 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3364 	    !mlx5_flow_ext_mreg_supported(dev) ||
3365 	    !priv->sh->dv_regc0_mask)
3366 		return 0;
3367 	/* Find MARK action. */
3368 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3369 		switch (actions->type) {
3370 		case RTE_FLOW_ACTION_TYPE_FLAG:
3371 			mcp_res = flow_mreg_add_copy_action
3372 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3373 			if (!mcp_res)
3374 				return -rte_errno;
3375 			flow->rix_mreg_copy = mcp_res->idx;
3376 			if (dev->data->dev_started) {
3377 				mcp_res->appcnt++;
3378 				flow->copy_applied = 1;
3379 			}
3380 			return 0;
3381 		case RTE_FLOW_ACTION_TYPE_MARK:
3382 			mark = (const struct rte_flow_action_mark *)
3383 				actions->conf;
3384 			mcp_res =
3385 				flow_mreg_add_copy_action(dev, mark->id, error);
3386 			if (!mcp_res)
3387 				return -rte_errno;
3388 			flow->rix_mreg_copy = mcp_res->idx;
3389 			if (dev->data->dev_started) {
3390 				mcp_res->appcnt++;
3391 				flow->copy_applied = 1;
3392 			}
3393 			return 0;
3394 		default:
3395 			break;
3396 		}
3397 	}
3398 	return 0;
3399 }
3400 
3401 #define MLX5_MAX_SPLIT_ACTIONS 24
3402 #define MLX5_MAX_SPLIT_ITEMS 24
3403 
3404 /**
3405  * Split the hairpin flow.
3406  * Since HW can't support encap on Rx we move the encap to Tx.
3407  * If the count action is after the encap then we also
3408  * move the count action. in this case the count will also measure
3409  * the outer bytes.
3410  *
3411  * @param dev
3412  *   Pointer to Ethernet device.
3413  * @param[in] actions
3414  *   Associated actions (list terminated by the END action).
3415  * @param[out] actions_rx
3416  *   Rx flow actions.
3417  * @param[out] actions_tx
3418  *   Tx flow actions..
3419  * @param[out] pattern_tx
3420  *   The pattern items for the Tx flow.
3421  * @param[out] flow_id
3422  *   The flow ID connected to this flow.
3423  *
3424  * @return
3425  *   0 on success.
3426  */
3427 static int
3428 flow_hairpin_split(struct rte_eth_dev *dev,
3429 		   const struct rte_flow_action actions[],
3430 		   struct rte_flow_action actions_rx[],
3431 		   struct rte_flow_action actions_tx[],
3432 		   struct rte_flow_item pattern_tx[],
3433 		   uint32_t *flow_id)
3434 {
3435 	struct mlx5_priv *priv = dev->data->dev_private;
3436 	const struct rte_flow_action_raw_encap *raw_encap;
3437 	const struct rte_flow_action_raw_decap *raw_decap;
3438 	struct mlx5_rte_flow_action_set_tag *set_tag;
3439 	struct rte_flow_action *tag_action;
3440 	struct mlx5_rte_flow_item_tag *tag_item;
3441 	struct rte_flow_item *item;
3442 	char *addr;
3443 	int encap = 0;
3444 
3445 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3446 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3447 		switch (actions->type) {
3448 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3449 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3450 			rte_memcpy(actions_tx, actions,
3451 			       sizeof(struct rte_flow_action));
3452 			actions_tx++;
3453 			break;
3454 		case RTE_FLOW_ACTION_TYPE_COUNT:
3455 			if (encap) {
3456 				rte_memcpy(actions_tx, actions,
3457 					   sizeof(struct rte_flow_action));
3458 				actions_tx++;
3459 			} else {
3460 				rte_memcpy(actions_rx, actions,
3461 					   sizeof(struct rte_flow_action));
3462 				actions_rx++;
3463 			}
3464 			break;
3465 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3466 			raw_encap = actions->conf;
3467 			if (raw_encap->size >
3468 			    (sizeof(struct rte_flow_item_eth) +
3469 			     sizeof(struct rte_flow_item_ipv4))) {
3470 				memcpy(actions_tx, actions,
3471 				       sizeof(struct rte_flow_action));
3472 				actions_tx++;
3473 				encap = 1;
3474 			} else {
3475 				rte_memcpy(actions_rx, actions,
3476 					   sizeof(struct rte_flow_action));
3477 				actions_rx++;
3478 			}
3479 			break;
3480 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3481 			raw_decap = actions->conf;
3482 			if (raw_decap->size <
3483 			    (sizeof(struct rte_flow_item_eth) +
3484 			     sizeof(struct rte_flow_item_ipv4))) {
3485 				memcpy(actions_tx, actions,
3486 				       sizeof(struct rte_flow_action));
3487 				actions_tx++;
3488 			} else {
3489 				rte_memcpy(actions_rx, actions,
3490 					   sizeof(struct rte_flow_action));
3491 				actions_rx++;
3492 			}
3493 			break;
3494 		default:
3495 			rte_memcpy(actions_rx, actions,
3496 				   sizeof(struct rte_flow_action));
3497 			actions_rx++;
3498 			break;
3499 		}
3500 	}
3501 	/* Add set meta action and end action for the Rx flow. */
3502 	tag_action = actions_rx;
3503 	tag_action->type = (enum rte_flow_action_type)
3504 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3505 	actions_rx++;
3506 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3507 	actions_rx++;
3508 	set_tag = (void *)actions_rx;
3509 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3510 	MLX5_ASSERT(set_tag->id > REG_NONE);
3511 	set_tag->data = *flow_id;
3512 	tag_action->conf = set_tag;
3513 	/* Create Tx item list. */
3514 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3515 	addr = (void *)&pattern_tx[2];
3516 	item = pattern_tx;
3517 	item->type = (enum rte_flow_item_type)
3518 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3519 	tag_item = (void *)addr;
3520 	tag_item->data = *flow_id;
3521 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3522 	MLX5_ASSERT(set_tag->id > REG_NONE);
3523 	item->spec = tag_item;
3524 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3525 	tag_item = (void *)addr;
3526 	tag_item->data = UINT32_MAX;
3527 	tag_item->id = UINT16_MAX;
3528 	item->mask = tag_item;
3529 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3530 	item->last = NULL;
3531 	item++;
3532 	item->type = RTE_FLOW_ITEM_TYPE_END;
3533 	return 0;
3534 }
3535 
3536 /**
3537  * The last stage of splitting chain, just creates the subflow
3538  * without any modification.
3539  *
3540  * @param[in] dev
3541  *   Pointer to Ethernet device.
3542  * @param[in] flow
3543  *   Parent flow structure pointer.
3544  * @param[in, out] sub_flow
3545  *   Pointer to return the created subflow, may be NULL.
3546  * @param[in] prefix_layers
3547  *   Prefix subflow layers, may be 0.
3548  * @param[in] attr
3549  *   Flow rule attributes.
3550  * @param[in] items
3551  *   Pattern specification (list terminated by the END pattern item).
3552  * @param[in] actions
3553  *   Associated actions (list terminated by the END action).
3554  * @param[in] external
3555  *   This flow rule is created by request external to PMD.
3556  * @param[in] flow_idx
3557  *   This memory pool index to the flow.
3558  * @param[out] error
3559  *   Perform verbose error reporting if not NULL.
3560  * @return
3561  *   0 on success, negative value otherwise
3562  */
3563 static int
3564 flow_create_split_inner(struct rte_eth_dev *dev,
3565 			struct rte_flow *flow,
3566 			struct mlx5_flow **sub_flow,
3567 			uint64_t prefix_layers,
3568 			const struct rte_flow_attr *attr,
3569 			const struct rte_flow_item items[],
3570 			const struct rte_flow_action actions[],
3571 			bool external, uint32_t flow_idx,
3572 			struct rte_flow_error *error)
3573 {
3574 	struct mlx5_flow *dev_flow;
3575 
3576 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3577 		flow_idx, error);
3578 	if (!dev_flow)
3579 		return -rte_errno;
3580 	dev_flow->flow = flow;
3581 	dev_flow->external = external;
3582 	/* Subflow object was created, we must include one in the list. */
3583 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3584 		      dev_flow->handle, next);
3585 	/*
3586 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3587 	 * flow may need some user defined item layer flags.
3588 	 */
3589 	if (prefix_layers)
3590 		dev_flow->handle->layers = prefix_layers;
3591 	if (sub_flow)
3592 		*sub_flow = dev_flow;
3593 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3594 }
3595 
3596 /**
3597  * Split the meter flow.
3598  *
3599  * As meter flow will split to three sub flow, other than meter
3600  * action, the other actions make sense to only meter accepts
3601  * the packet. If it need to be dropped, no other additional
3602  * actions should be take.
3603  *
3604  * One kind of special action which decapsulates the L3 tunnel
3605  * header will be in the prefix sub flow, as not to take the
3606  * L3 tunnel header into account.
3607  *
3608  * @param dev
3609  *   Pointer to Ethernet device.
3610  * @param[in] items
3611  *   Pattern specification (list terminated by the END pattern item).
3612  * @param[out] sfx_items
3613  *   Suffix flow match items (list terminated by the END pattern item).
3614  * @param[in] actions
3615  *   Associated actions (list terminated by the END action).
3616  * @param[out] actions_sfx
3617  *   Suffix flow actions.
3618  * @param[out] actions_pre
3619  *   Prefix flow actions.
3620  * @param[out] pattern_sfx
3621  *   The pattern items for the suffix flow.
3622  * @param[out] tag_sfx
3623  *   Pointer to suffix flow tag.
3624  *
3625  * @return
3626  *   0 on success.
3627  */
3628 static int
3629 flow_meter_split_prep(struct rte_eth_dev *dev,
3630 		 const struct rte_flow_item items[],
3631 		 struct rte_flow_item sfx_items[],
3632 		 const struct rte_flow_action actions[],
3633 		 struct rte_flow_action actions_sfx[],
3634 		 struct rte_flow_action actions_pre[])
3635 {
3636 	struct rte_flow_action *tag_action = NULL;
3637 	struct rte_flow_item *tag_item;
3638 	struct mlx5_rte_flow_action_set_tag *set_tag;
3639 	struct rte_flow_error error;
3640 	const struct rte_flow_action_raw_encap *raw_encap;
3641 	const struct rte_flow_action_raw_decap *raw_decap;
3642 	struct mlx5_rte_flow_item_tag *tag_spec;
3643 	struct mlx5_rte_flow_item_tag *tag_mask;
3644 	uint32_t tag_id;
3645 	bool copy_vlan = false;
3646 
3647 	/* Prepare the actions for prefix and suffix flow. */
3648 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3649 		struct rte_flow_action **action_cur = NULL;
3650 
3651 		switch (actions->type) {
3652 		case RTE_FLOW_ACTION_TYPE_METER:
3653 			/* Add the extra tag action first. */
3654 			tag_action = actions_pre;
3655 			tag_action->type = (enum rte_flow_action_type)
3656 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3657 			actions_pre++;
3658 			action_cur = &actions_pre;
3659 			break;
3660 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3661 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3662 			action_cur = &actions_pre;
3663 			break;
3664 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3665 			raw_encap = actions->conf;
3666 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3667 				action_cur = &actions_pre;
3668 			break;
3669 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3670 			raw_decap = actions->conf;
3671 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3672 				action_cur = &actions_pre;
3673 			break;
3674 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3675 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3676 			copy_vlan = true;
3677 			break;
3678 		default:
3679 			break;
3680 		}
3681 		if (!action_cur)
3682 			action_cur = &actions_sfx;
3683 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3684 		(*action_cur)++;
3685 	}
3686 	/* Add end action to the actions. */
3687 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3688 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3689 	actions_pre++;
3690 	/* Set the tag. */
3691 	set_tag = (void *)actions_pre;
3692 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3693 	/*
3694 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3695 	 */
3696 	tag_id = flow_qrss_get_id(dev);
3697 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3698 	assert(tag_action);
3699 	tag_action->conf = set_tag;
3700 	/* Prepare the suffix subflow items. */
3701 	tag_item = sfx_items++;
3702 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3703 		int item_type = items->type;
3704 
3705 		switch (item_type) {
3706 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3707 			memcpy(sfx_items, items, sizeof(*sfx_items));
3708 			sfx_items++;
3709 			break;
3710 		case RTE_FLOW_ITEM_TYPE_VLAN:
3711 			if (copy_vlan) {
3712 				memcpy(sfx_items, items, sizeof(*sfx_items));
3713 				/*
3714 				 * Convert to internal match item, it is used
3715 				 * for vlan push and set vid.
3716 				 */
3717 				sfx_items->type = (enum rte_flow_item_type)
3718 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3719 				sfx_items++;
3720 			}
3721 			break;
3722 		default:
3723 			break;
3724 		}
3725 	}
3726 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3727 	sfx_items++;
3728 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3729 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3730 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3731 	tag_mask = tag_spec + 1;
3732 	tag_mask->data = 0xffffff00;
3733 	tag_item->type = (enum rte_flow_item_type)
3734 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3735 	tag_item->spec = tag_spec;
3736 	tag_item->last = NULL;
3737 	tag_item->mask = tag_mask;
3738 	return tag_id;
3739 }
3740 
3741 /**
3742  * Split action list having QUEUE/RSS for metadata register copy.
3743  *
3744  * Once Q/RSS action is detected in user's action list, the flow action
3745  * should be split in order to copy metadata registers, which will happen in
3746  * RX_CP_TBL like,
3747  *   - CQE->flow_tag := reg_c[1] (MARK)
3748  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3749  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3750  * This is because the last action of each flow must be a terminal action
3751  * (QUEUE, RSS or DROP).
3752  *
3753  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3754  * stored and kept in the mlx5_flow structure per each sub_flow.
3755  *
3756  * The Q/RSS action is replaced with,
3757  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3758  * And the following JUMP action is added at the end,
3759  *   - JUMP, to RX_CP_TBL.
3760  *
3761  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3762  * flow_create_split_metadata() routine. The flow will look like,
3763  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3764  *
3765  * @param dev
3766  *   Pointer to Ethernet device.
3767  * @param[out] split_actions
3768  *   Pointer to store split actions to jump to CP_TBL.
3769  * @param[in] actions
3770  *   Pointer to the list of original flow actions.
3771  * @param[in] qrss
3772  *   Pointer to the Q/RSS action.
3773  * @param[in] actions_n
3774  *   Number of original actions.
3775  * @param[out] error
3776  *   Perform verbose error reporting if not NULL.
3777  *
3778  * @return
3779  *   non-zero unique flow_id on success, otherwise 0 and
3780  *   error/rte_error are set.
3781  */
3782 static uint32_t
3783 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3784 			  struct rte_flow_action *split_actions,
3785 			  const struct rte_flow_action *actions,
3786 			  const struct rte_flow_action *qrss,
3787 			  int actions_n, struct rte_flow_error *error)
3788 {
3789 	struct mlx5_rte_flow_action_set_tag *set_tag;
3790 	struct rte_flow_action_jump *jump;
3791 	const int qrss_idx = qrss - actions;
3792 	uint32_t flow_id = 0;
3793 	int ret = 0;
3794 
3795 	/*
3796 	 * Given actions will be split
3797 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3798 	 * - Add jump to mreg CP_TBL.
3799 	 * As a result, there will be one more action.
3800 	 */
3801 	++actions_n;
3802 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3803 	set_tag = (void *)(split_actions + actions_n);
3804 	/*
3805 	 * If tag action is not set to void(it means we are not the meter
3806 	 * suffix flow), add the tag action. Since meter suffix flow already
3807 	 * has the tag added.
3808 	 */
3809 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3810 		/*
3811 		 * Allocate the new subflow ID. This one is unique within
3812 		 * device and not shared with representors. Otherwise,
3813 		 * we would have to resolve multi-thread access synch
3814 		 * issue. Each flow on the shared device is appended
3815 		 * with source vport identifier, so the resulting
3816 		 * flows will be unique in the shared (by master and
3817 		 * representors) domain even if they have coinciding
3818 		 * IDs.
3819 		 */
3820 		flow_id = flow_qrss_get_id(dev);
3821 		if (!flow_id)
3822 			return rte_flow_error_set(error, ENOMEM,
3823 						  RTE_FLOW_ERROR_TYPE_ACTION,
3824 						  NULL, "can't allocate id "
3825 						  "for split Q/RSS subflow");
3826 		/* Internal SET_TAG action to set flow ID. */
3827 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3828 			.data = flow_id,
3829 		};
3830 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3831 		if (ret < 0)
3832 			return ret;
3833 		set_tag->id = ret;
3834 		/* Construct new actions array. */
3835 		/* Replace QUEUE/RSS action. */
3836 		split_actions[qrss_idx] = (struct rte_flow_action){
3837 			.type = (enum rte_flow_action_type)
3838 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3839 			.conf = set_tag,
3840 		};
3841 	}
3842 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3843 	jump = (void *)(set_tag + 1);
3844 	*jump = (struct rte_flow_action_jump){
3845 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3846 	};
3847 	split_actions[actions_n - 2] = (struct rte_flow_action){
3848 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3849 		.conf = jump,
3850 	};
3851 	split_actions[actions_n - 1] = (struct rte_flow_action){
3852 		.type = RTE_FLOW_ACTION_TYPE_END,
3853 	};
3854 	return flow_id;
3855 }
3856 
3857 /**
3858  * Extend the given action list for Tx metadata copy.
3859  *
3860  * Copy the given action list to the ext_actions and add flow metadata register
3861  * copy action in order to copy reg_a set by WQE to reg_c[0].
3862  *
3863  * @param[out] ext_actions
3864  *   Pointer to the extended action list.
3865  * @param[in] actions
3866  *   Pointer to the list of actions.
3867  * @param[in] actions_n
3868  *   Number of actions in the list.
3869  * @param[out] error
3870  *   Perform verbose error reporting if not NULL.
3871  * @param[in] encap_idx
3872  *   The encap action inndex.
3873  *
3874  * @return
3875  *   0 on success, negative value otherwise
3876  */
3877 static int
3878 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3879 		       struct rte_flow_action *ext_actions,
3880 		       const struct rte_flow_action *actions,
3881 		       int actions_n, struct rte_flow_error *error,
3882 		       int encap_idx)
3883 {
3884 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3885 		(struct mlx5_flow_action_copy_mreg *)
3886 			(ext_actions + actions_n + 1);
3887 	int ret;
3888 
3889 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3890 	if (ret < 0)
3891 		return ret;
3892 	cp_mreg->dst = ret;
3893 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3894 	if (ret < 0)
3895 		return ret;
3896 	cp_mreg->src = ret;
3897 	if (encap_idx != 0)
3898 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3899 	if (encap_idx == actions_n - 1) {
3900 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3901 			.type = (enum rte_flow_action_type)
3902 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3903 			.conf = cp_mreg,
3904 		};
3905 		ext_actions[actions_n] = (struct rte_flow_action){
3906 			.type = RTE_FLOW_ACTION_TYPE_END,
3907 		};
3908 	} else {
3909 		ext_actions[encap_idx] = (struct rte_flow_action){
3910 			.type = (enum rte_flow_action_type)
3911 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3912 			.conf = cp_mreg,
3913 		};
3914 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3915 				sizeof(*ext_actions) * (actions_n - encap_idx));
3916 	}
3917 	return 0;
3918 }
3919 
3920 /**
3921  * The splitting for metadata feature.
3922  *
3923  * - Q/RSS action on NIC Rx should be split in order to pass by
3924  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3925  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3926  *
3927  * - All the actions on NIC Tx should have a mreg copy action to
3928  *   copy reg_a from WQE to reg_c[0].
3929  *
3930  * @param dev
3931  *   Pointer to Ethernet device.
3932  * @param[in] flow
3933  *   Parent flow structure pointer.
3934  * @param[in] prefix_layers
3935  *   Prefix flow layer flags.
3936  * @param[in] attr
3937  *   Flow rule attributes.
3938  * @param[in] items
3939  *   Pattern specification (list terminated by the END pattern item).
3940  * @param[in] actions
3941  *   Associated actions (list terminated by the END action).
3942  * @param[in] external
3943  *   This flow rule is created by request external to PMD.
3944  * @param[in] flow_idx
3945  *   This memory pool index to the flow.
3946  * @param[out] error
3947  *   Perform verbose error reporting if not NULL.
3948  * @return
3949  *   0 on success, negative value otherwise
3950  */
3951 static int
3952 flow_create_split_metadata(struct rte_eth_dev *dev,
3953 			   struct rte_flow *flow,
3954 			   uint64_t prefix_layers,
3955 			   const struct rte_flow_attr *attr,
3956 			   const struct rte_flow_item items[],
3957 			   const struct rte_flow_action actions[],
3958 			   bool external, uint32_t flow_idx,
3959 			   struct rte_flow_error *error)
3960 {
3961 	struct mlx5_priv *priv = dev->data->dev_private;
3962 	struct mlx5_dev_config *config = &priv->config;
3963 	const struct rte_flow_action *qrss = NULL;
3964 	struct rte_flow_action *ext_actions = NULL;
3965 	struct mlx5_flow *dev_flow = NULL;
3966 	uint32_t qrss_id = 0;
3967 	int mtr_sfx = 0;
3968 	size_t act_size;
3969 	int actions_n;
3970 	int encap_idx;
3971 	int ret;
3972 
3973 	/* Check whether extensive metadata feature is engaged. */
3974 	if (!config->dv_flow_en ||
3975 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3976 	    !mlx5_flow_ext_mreg_supported(dev))
3977 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3978 					       attr, items, actions, external,
3979 					       flow_idx, error);
3980 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3981 							   &encap_idx);
3982 	if (qrss) {
3983 		/* Exclude hairpin flows from splitting. */
3984 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3985 			const struct rte_flow_action_queue *queue;
3986 
3987 			queue = qrss->conf;
3988 			if (mlx5_rxq_get_type(dev, queue->index) ==
3989 			    MLX5_RXQ_TYPE_HAIRPIN)
3990 				qrss = NULL;
3991 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3992 			const struct rte_flow_action_rss *rss;
3993 
3994 			rss = qrss->conf;
3995 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3996 			    MLX5_RXQ_TYPE_HAIRPIN)
3997 				qrss = NULL;
3998 		}
3999 	}
4000 	if (qrss) {
4001 		/* Check if it is in meter suffix table. */
4002 		mtr_sfx = attr->group == (attr->transfer ?
4003 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4004 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4005 		/*
4006 		 * Q/RSS action on NIC Rx should be split in order to pass by
4007 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4008 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4009 		 */
4010 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4011 			   sizeof(struct rte_flow_action_set_tag) +
4012 			   sizeof(struct rte_flow_action_jump);
4013 		ext_actions = rte_zmalloc(__func__, act_size, 0);
4014 		if (!ext_actions)
4015 			return rte_flow_error_set(error, ENOMEM,
4016 						  RTE_FLOW_ERROR_TYPE_ACTION,
4017 						  NULL, "no memory to split "
4018 						  "metadata flow");
4019 		/*
4020 		 * If we are the suffix flow of meter, tag already exist.
4021 		 * Set the tag action to void.
4022 		 */
4023 		if (mtr_sfx)
4024 			ext_actions[qrss - actions].type =
4025 						RTE_FLOW_ACTION_TYPE_VOID;
4026 		else
4027 			ext_actions[qrss - actions].type =
4028 						(enum rte_flow_action_type)
4029 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4030 		/*
4031 		 * Create the new actions list with removed Q/RSS action
4032 		 * and appended set tag and jump to register copy table
4033 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4034 		 * in advance, because it is needed for set tag action.
4035 		 */
4036 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4037 						    qrss, actions_n, error);
4038 		if (!mtr_sfx && !qrss_id) {
4039 			ret = -rte_errno;
4040 			goto exit;
4041 		}
4042 	} else if (attr->egress && !attr->transfer) {
4043 		/*
4044 		 * All the actions on NIC Tx should have a metadata register
4045 		 * copy action to copy reg_a from WQE to reg_c[meta]
4046 		 */
4047 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4048 			   sizeof(struct mlx5_flow_action_copy_mreg);
4049 		ext_actions = rte_zmalloc(__func__, act_size, 0);
4050 		if (!ext_actions)
4051 			return rte_flow_error_set(error, ENOMEM,
4052 						  RTE_FLOW_ERROR_TYPE_ACTION,
4053 						  NULL, "no memory to split "
4054 						  "metadata flow");
4055 		/* Create the action list appended with copy register. */
4056 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4057 					     actions_n, error, encap_idx);
4058 		if (ret < 0)
4059 			goto exit;
4060 	}
4061 	/* Add the unmodified original or prefix subflow. */
4062 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4063 				      items, ext_actions ? ext_actions :
4064 				      actions, external, flow_idx, error);
4065 	if (ret < 0)
4066 		goto exit;
4067 	MLX5_ASSERT(dev_flow);
4068 	if (qrss) {
4069 		const struct rte_flow_attr q_attr = {
4070 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4071 			.ingress = 1,
4072 		};
4073 		/* Internal PMD action to set register. */
4074 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4075 			.data = qrss_id,
4076 			.id = 0,
4077 		};
4078 		struct rte_flow_item q_items[] = {
4079 			{
4080 				.type = (enum rte_flow_item_type)
4081 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4082 				.spec = &q_tag_spec,
4083 				.last = NULL,
4084 				.mask = NULL,
4085 			},
4086 			{
4087 				.type = RTE_FLOW_ITEM_TYPE_END,
4088 			},
4089 		};
4090 		struct rte_flow_action q_actions[] = {
4091 			{
4092 				.type = qrss->type,
4093 				.conf = qrss->conf,
4094 			},
4095 			{
4096 				.type = RTE_FLOW_ACTION_TYPE_END,
4097 			},
4098 		};
4099 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4100 
4101 		/*
4102 		 * Configure the tag item only if there is no meter subflow.
4103 		 * Since tag is already marked in the meter suffix subflow
4104 		 * we can just use the meter suffix items as is.
4105 		 */
4106 		if (qrss_id) {
4107 			/* Not meter subflow. */
4108 			MLX5_ASSERT(!mtr_sfx);
4109 			/*
4110 			 * Put unique id in prefix flow due to it is destroyed
4111 			 * after suffix flow and id will be freed after there
4112 			 * is no actual flows with this id and identifier
4113 			 * reallocation becomes possible (for example, for
4114 			 * other flows in other threads).
4115 			 */
4116 			dev_flow->handle->split_flow_id = qrss_id;
4117 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4118 						   error);
4119 			if (ret < 0)
4120 				goto exit;
4121 			q_tag_spec.id = ret;
4122 		}
4123 		dev_flow = NULL;
4124 		/* Add suffix subflow to execute Q/RSS. */
4125 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4126 					      &q_attr, mtr_sfx ? items :
4127 					      q_items, q_actions,
4128 					      external, flow_idx, error);
4129 		if (ret < 0)
4130 			goto exit;
4131 		/* qrss ID should be freed if failed. */
4132 		qrss_id = 0;
4133 		MLX5_ASSERT(dev_flow);
4134 	}
4135 
4136 exit:
4137 	/*
4138 	 * We do not destroy the partially created sub_flows in case of error.
4139 	 * These ones are included into parent flow list and will be destroyed
4140 	 * by flow_drv_destroy.
4141 	 */
4142 	flow_qrss_free_id(dev, qrss_id);
4143 	rte_free(ext_actions);
4144 	return ret;
4145 }
4146 
4147 /**
4148  * The splitting for meter feature.
4149  *
4150  * - The meter flow will be split to two flows as prefix and
4151  *   suffix flow. The packets make sense only it pass the prefix
4152  *   meter action.
4153  *
4154  * - Reg_C_5 is used for the packet to match betweend prefix and
4155  *   suffix flow.
4156  *
4157  * @param dev
4158  *   Pointer to Ethernet device.
4159  * @param[in] flow
4160  *   Parent flow structure pointer.
4161  * @param[in] attr
4162  *   Flow rule attributes.
4163  * @param[in] items
4164  *   Pattern specification (list terminated by the END pattern item).
4165  * @param[in] actions
4166  *   Associated actions (list terminated by the END action).
4167  * @param[in] external
4168  *   This flow rule is created by request external to PMD.
4169  * @param[in] flow_idx
4170  *   This memory pool index to the flow.
4171  * @param[out] error
4172  *   Perform verbose error reporting if not NULL.
4173  * @return
4174  *   0 on success, negative value otherwise
4175  */
4176 static int
4177 flow_create_split_meter(struct rte_eth_dev *dev,
4178 			   struct rte_flow *flow,
4179 			   const struct rte_flow_attr *attr,
4180 			   const struct rte_flow_item items[],
4181 			   const struct rte_flow_action actions[],
4182 			   bool external, uint32_t flow_idx,
4183 			   struct rte_flow_error *error)
4184 {
4185 	struct mlx5_priv *priv = dev->data->dev_private;
4186 	struct rte_flow_action *sfx_actions = NULL;
4187 	struct rte_flow_action *pre_actions = NULL;
4188 	struct rte_flow_item *sfx_items = NULL;
4189 	struct mlx5_flow *dev_flow = NULL;
4190 	struct rte_flow_attr sfx_attr = *attr;
4191 	uint32_t mtr = 0;
4192 	uint32_t mtr_tag_id = 0;
4193 	size_t act_size;
4194 	size_t item_size;
4195 	int actions_n = 0;
4196 	int ret;
4197 
4198 	if (priv->mtr_en)
4199 		actions_n = flow_check_meter_action(actions, &mtr);
4200 	if (mtr) {
4201 		/* The five prefix actions: meter, decap, encap, tag, end. */
4202 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4203 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4204 		/* tag, vlan, port id, end. */
4205 #define METER_SUFFIX_ITEM 4
4206 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4207 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4208 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4209 		if (!sfx_actions)
4210 			return rte_flow_error_set(error, ENOMEM,
4211 						  RTE_FLOW_ERROR_TYPE_ACTION,
4212 						  NULL, "no memory to split "
4213 						  "meter flow");
4214 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4215 			     act_size);
4216 		pre_actions = sfx_actions + actions_n;
4217 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4218 						   actions, sfx_actions,
4219 						   pre_actions);
4220 		if (!mtr_tag_id) {
4221 			ret = -rte_errno;
4222 			goto exit;
4223 		}
4224 		/* Add the prefix subflow. */
4225 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4226 					      items, pre_actions, external,
4227 					      flow_idx, error);
4228 		if (ret) {
4229 			ret = -rte_errno;
4230 			goto exit;
4231 		}
4232 		dev_flow->handle->split_flow_id = mtr_tag_id;
4233 		/* Setting the sfx group atrr. */
4234 		sfx_attr.group = sfx_attr.transfer ?
4235 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4236 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4237 	}
4238 	/* Add the prefix subflow. */
4239 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4240 					 flow_get_prefix_layer_flags(dev_flow) :
4241 					 0, &sfx_attr,
4242 					 sfx_items ? sfx_items : items,
4243 					 sfx_actions ? sfx_actions : actions,
4244 					 external, flow_idx, error);
4245 exit:
4246 	if (sfx_actions)
4247 		rte_free(sfx_actions);
4248 	return ret;
4249 }
4250 
4251 /**
4252  * Split the flow to subflow set. The splitters might be linked
4253  * in the chain, like this:
4254  * flow_create_split_outer() calls:
4255  *   flow_create_split_meter() calls:
4256  *     flow_create_split_metadata(meter_subflow_0) calls:
4257  *       flow_create_split_inner(metadata_subflow_0)
4258  *       flow_create_split_inner(metadata_subflow_1)
4259  *       flow_create_split_inner(metadata_subflow_2)
4260  *     flow_create_split_metadata(meter_subflow_1) calls:
4261  *       flow_create_split_inner(metadata_subflow_0)
4262  *       flow_create_split_inner(metadata_subflow_1)
4263  *       flow_create_split_inner(metadata_subflow_2)
4264  *
4265  * This provide flexible way to add new levels of flow splitting.
4266  * The all of successfully created subflows are included to the
4267  * parent flow dev_flow list.
4268  *
4269  * @param dev
4270  *   Pointer to Ethernet device.
4271  * @param[in] flow
4272  *   Parent flow structure pointer.
4273  * @param[in] attr
4274  *   Flow rule attributes.
4275  * @param[in] items
4276  *   Pattern specification (list terminated by the END pattern item).
4277  * @param[in] actions
4278  *   Associated actions (list terminated by the END action).
4279  * @param[in] external
4280  *   This flow rule is created by request external to PMD.
4281  * @param[in] flow_idx
4282  *   This memory pool index to the flow.
4283  * @param[out] error
4284  *   Perform verbose error reporting if not NULL.
4285  * @return
4286  *   0 on success, negative value otherwise
4287  */
4288 static int
4289 flow_create_split_outer(struct rte_eth_dev *dev,
4290 			struct rte_flow *flow,
4291 			const struct rte_flow_attr *attr,
4292 			const struct rte_flow_item items[],
4293 			const struct rte_flow_action actions[],
4294 			bool external, uint32_t flow_idx,
4295 			struct rte_flow_error *error)
4296 {
4297 	int ret;
4298 
4299 	ret = flow_create_split_meter(dev, flow, attr, items,
4300 					 actions, external, flow_idx, error);
4301 	MLX5_ASSERT(ret <= 0);
4302 	return ret;
4303 }
4304 
4305 /**
4306  * Create a flow and add it to @p list.
4307  *
4308  * @param dev
4309  *   Pointer to Ethernet device.
4310  * @param list
4311  *   Pointer to a TAILQ flow list. If this parameter NULL,
4312  *   no list insertion occurred, flow is just created,
4313  *   this is caller's responsibility to track the
4314  *   created flow.
4315  * @param[in] attr
4316  *   Flow rule attributes.
4317  * @param[in] items
4318  *   Pattern specification (list terminated by the END pattern item).
4319  * @param[in] actions
4320  *   Associated actions (list terminated by the END action).
4321  * @param[in] external
4322  *   This flow rule is created by request external to PMD.
4323  * @param[out] error
4324  *   Perform verbose error reporting if not NULL.
4325  *
4326  * @return
4327  *   A flow index on success, 0 otherwise and rte_errno is set.
4328  */
4329 static uint32_t
4330 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4331 		 const struct rte_flow_attr *attr,
4332 		 const struct rte_flow_item items[],
4333 		 const struct rte_flow_action actions[],
4334 		 bool external, struct rte_flow_error *error)
4335 {
4336 	struct mlx5_priv *priv = dev->data->dev_private;
4337 	struct rte_flow *flow = NULL;
4338 	struct mlx5_flow *dev_flow;
4339 	const struct rte_flow_action_rss *rss;
4340 	union {
4341 		struct rte_flow_expand_rss buf;
4342 		uint8_t buffer[2048];
4343 	} expand_buffer;
4344 	union {
4345 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4346 		uint8_t buffer[2048];
4347 	} actions_rx;
4348 	union {
4349 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4350 		uint8_t buffer[2048];
4351 	} actions_hairpin_tx;
4352 	union {
4353 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4354 		uint8_t buffer[2048];
4355 	} items_tx;
4356 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4357 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4358 					      priv->rss_desc)[!!priv->flow_idx];
4359 	const struct rte_flow_action *p_actions_rx = actions;
4360 	uint32_t i;
4361 	uint32_t idx = 0;
4362 	int hairpin_flow;
4363 	uint32_t hairpin_id = 0;
4364 	struct rte_flow_attr attr_tx = { .priority = 0 };
4365 	int ret;
4366 
4367 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4368 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4369 				external, hairpin_flow, error);
4370 	if (ret < 0)
4371 		return 0;
4372 	if (hairpin_flow > 0) {
4373 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4374 			rte_errno = EINVAL;
4375 			return 0;
4376 		}
4377 		flow_hairpin_split(dev, actions, actions_rx.actions,
4378 				   actions_hairpin_tx.actions, items_tx.items,
4379 				   &hairpin_id);
4380 		p_actions_rx = actions_rx.actions;
4381 	}
4382 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4383 	if (!flow) {
4384 		rte_errno = ENOMEM;
4385 		goto error_before_flow;
4386 	}
4387 	flow->drv_type = flow_get_drv_type(dev, attr);
4388 	if (hairpin_id != 0)
4389 		flow->hairpin_flow_id = hairpin_id;
4390 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4391 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4392 	memset(rss_desc, 0, sizeof(*rss_desc));
4393 	rss = flow_get_rss_action(p_actions_rx);
4394 	if (rss) {
4395 		/*
4396 		 * The following information is required by
4397 		 * mlx5_flow_hashfields_adjust() in advance.
4398 		 */
4399 		rss_desc->level = rss->level;
4400 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4401 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4402 	}
4403 	flow->dev_handles = 0;
4404 	if (rss && rss->types) {
4405 		unsigned int graph_root;
4406 
4407 		graph_root = find_graph_root(items, rss->level);
4408 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4409 					  items, rss->types,
4410 					  mlx5_support_expansion,
4411 					  graph_root);
4412 		MLX5_ASSERT(ret > 0 &&
4413 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4414 	} else {
4415 		buf->entries = 1;
4416 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4417 	}
4418 	/*
4419 	 * Record the start index when there is a nested call. All sub-flows
4420 	 * need to be translated before another calling.
4421 	 * No need to use ping-pong buffer to save memory here.
4422 	 */
4423 	if (priv->flow_idx) {
4424 		MLX5_ASSERT(!priv->flow_nested_idx);
4425 		priv->flow_nested_idx = priv->flow_idx;
4426 	}
4427 	for (i = 0; i < buf->entries; ++i) {
4428 		/*
4429 		 * The splitter may create multiple dev_flows,
4430 		 * depending on configuration. In the simplest
4431 		 * case it just creates unmodified original flow.
4432 		 */
4433 		ret = flow_create_split_outer(dev, flow, attr,
4434 					      buf->entry[i].pattern,
4435 					      p_actions_rx, external, idx,
4436 					      error);
4437 		if (ret < 0)
4438 			goto error;
4439 	}
4440 	/* Create the tx flow. */
4441 	if (hairpin_flow) {
4442 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4443 		attr_tx.ingress = 0;
4444 		attr_tx.egress = 1;
4445 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4446 					 actions_hairpin_tx.actions,
4447 					 idx, error);
4448 		if (!dev_flow)
4449 			goto error;
4450 		dev_flow->flow = flow;
4451 		dev_flow->external = 0;
4452 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4453 			      dev_flow->handle, next);
4454 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4455 					 items_tx.items,
4456 					 actions_hairpin_tx.actions, error);
4457 		if (ret < 0)
4458 			goto error;
4459 	}
4460 	/*
4461 	 * Update the metadata register copy table. If extensive
4462 	 * metadata feature is enabled and registers are supported
4463 	 * we might create the extra rte_flow for each unique
4464 	 * MARK/FLAG action ID.
4465 	 *
4466 	 * The table is updated for ingress Flows only, because
4467 	 * the egress Flows belong to the different device and
4468 	 * copy table should be updated in peer NIC Rx domain.
4469 	 */
4470 	if (attr->ingress &&
4471 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4472 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4473 		if (ret)
4474 			goto error;
4475 	}
4476 	/*
4477 	 * If the flow is external (from application) OR device is started, then
4478 	 * the flow will be applied immediately.
4479 	 */
4480 	if (external || dev->data->dev_started) {
4481 		ret = flow_drv_apply(dev, flow, error);
4482 		if (ret < 0)
4483 			goto error;
4484 	}
4485 	if (list)
4486 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4487 			     flow, next);
4488 	flow_rxq_flags_set(dev, flow);
4489 	/* Nested flow creation index recovery. */
4490 	priv->flow_idx = priv->flow_nested_idx;
4491 	if (priv->flow_nested_idx)
4492 		priv->flow_nested_idx = 0;
4493 	return idx;
4494 error:
4495 	MLX5_ASSERT(flow);
4496 	ret = rte_errno; /* Save rte_errno before cleanup. */
4497 	flow_mreg_del_copy_action(dev, flow);
4498 	flow_drv_destroy(dev, flow);
4499 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4500 	rte_errno = ret; /* Restore rte_errno. */
4501 error_before_flow:
4502 	ret = rte_errno;
4503 	if (hairpin_id)
4504 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4505 				     hairpin_id);
4506 	rte_errno = ret;
4507 	priv->flow_idx = priv->flow_nested_idx;
4508 	if (priv->flow_nested_idx)
4509 		priv->flow_nested_idx = 0;
4510 	return 0;
4511 }
4512 
4513 /**
4514  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4515  * incoming packets to table 1.
4516  *
4517  * Other flow rules, requested for group n, will be created in
4518  * e-switch table n+1.
4519  * Jump action to e-switch group n will be created to group n+1.
4520  *
4521  * Used when working in switchdev mode, to utilise advantages of table 1
4522  * and above.
4523  *
4524  * @param dev
4525  *   Pointer to Ethernet device.
4526  *
4527  * @return
4528  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4529  */
4530 struct rte_flow *
4531 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4532 {
4533 	const struct rte_flow_attr attr = {
4534 		.group = 0,
4535 		.priority = 0,
4536 		.ingress = 1,
4537 		.egress = 0,
4538 		.transfer = 1,
4539 	};
4540 	const struct rte_flow_item pattern = {
4541 		.type = RTE_FLOW_ITEM_TYPE_END,
4542 	};
4543 	struct rte_flow_action_jump jump = {
4544 		.group = 1,
4545 	};
4546 	const struct rte_flow_action actions[] = {
4547 		{
4548 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4549 			.conf = &jump,
4550 		},
4551 		{
4552 			.type = RTE_FLOW_ACTION_TYPE_END,
4553 		},
4554 	};
4555 	struct mlx5_priv *priv = dev->data->dev_private;
4556 	struct rte_flow_error error;
4557 
4558 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4559 						   &attr, &pattern,
4560 						   actions, false, &error);
4561 }
4562 
4563 /**
4564  * Validate a flow supported by the NIC.
4565  *
4566  * @see rte_flow_validate()
4567  * @see rte_flow_ops
4568  */
4569 int
4570 mlx5_flow_validate(struct rte_eth_dev *dev,
4571 		   const struct rte_flow_attr *attr,
4572 		   const struct rte_flow_item items[],
4573 		   const struct rte_flow_action actions[],
4574 		   struct rte_flow_error *error)
4575 {
4576 	int hairpin_flow;
4577 
4578 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4579 	return flow_drv_validate(dev, attr, items, actions,
4580 				true, hairpin_flow, error);
4581 }
4582 
4583 /**
4584  * Create a flow.
4585  *
4586  * @see rte_flow_create()
4587  * @see rte_flow_ops
4588  */
4589 struct rte_flow *
4590 mlx5_flow_create(struct rte_eth_dev *dev,
4591 		 const struct rte_flow_attr *attr,
4592 		 const struct rte_flow_item items[],
4593 		 const struct rte_flow_action actions[],
4594 		 struct rte_flow_error *error)
4595 {
4596 	struct mlx5_priv *priv = dev->data->dev_private;
4597 
4598 	/*
4599 	 * If the device is not started yet, it is not allowed to created a
4600 	 * flow from application. PMD default flows and traffic control flows
4601 	 * are not affected.
4602 	 */
4603 	if (unlikely(!dev->data->dev_started)) {
4604 		DRV_LOG(DEBUG, "port %u is not started when "
4605 			"inserting a flow", dev->data->port_id);
4606 		rte_flow_error_set(error, ENODEV,
4607 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4608 				   NULL,
4609 				   "port not started");
4610 		return NULL;
4611 	}
4612 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4613 				  attr, items, actions, true, error);
4614 }
4615 
4616 /**
4617  * Destroy a flow in a list.
4618  *
4619  * @param dev
4620  *   Pointer to Ethernet device.
4621  * @param list
4622  *   Pointer to the Indexed flow list. If this parameter NULL,
4623  *   there is no flow removal from the list. Be noted that as
4624  *   flow is add to the indexed list, memory of the indexed
4625  *   list points to maybe changed as flow destroyed.
4626  * @param[in] flow_idx
4627  *   Index of flow to destroy.
4628  */
4629 static void
4630 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4631 		  uint32_t flow_idx)
4632 {
4633 	struct mlx5_priv *priv = dev->data->dev_private;
4634 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4635 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4636 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4637 
4638 	if (!flow)
4639 		return;
4640 	/*
4641 	 * Update RX queue flags only if port is started, otherwise it is
4642 	 * already clean.
4643 	 */
4644 	if (dev->data->dev_started)
4645 		flow_rxq_flags_trim(dev, flow);
4646 	if (flow->hairpin_flow_id)
4647 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4648 				     flow->hairpin_flow_id);
4649 	flow_drv_destroy(dev, flow);
4650 	if (list)
4651 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4652 			     flow_idx, flow, next);
4653 	flow_mreg_del_copy_action(dev, flow);
4654 	if (flow->fdir) {
4655 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4656 			if (priv_fdir_flow->rix_flow == flow_idx)
4657 				break;
4658 		}
4659 		if (priv_fdir_flow) {
4660 			LIST_REMOVE(priv_fdir_flow, next);
4661 			rte_free(priv_fdir_flow->fdir);
4662 			rte_free(priv_fdir_flow);
4663 		}
4664 	}
4665 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4666 }
4667 
4668 /**
4669  * Destroy all flows.
4670  *
4671  * @param dev
4672  *   Pointer to Ethernet device.
4673  * @param list
4674  *   Pointer to the Indexed flow list.
4675  * @param active
4676  *   If flushing is called avtively.
4677  */
4678 void
4679 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4680 {
4681 	uint32_t num_flushed = 0;
4682 
4683 	while (*list) {
4684 		flow_list_destroy(dev, list, *list);
4685 		num_flushed++;
4686 	}
4687 	if (active) {
4688 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4689 			dev->data->port_id, num_flushed);
4690 	}
4691 }
4692 
4693 /**
4694  * Remove all flows.
4695  *
4696  * @param dev
4697  *   Pointer to Ethernet device.
4698  * @param list
4699  *   Pointer to the Indexed flow list.
4700  */
4701 void
4702 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4703 {
4704 	struct mlx5_priv *priv = dev->data->dev_private;
4705 	struct rte_flow *flow = NULL;
4706 	uint32_t idx;
4707 
4708 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4709 		      flow, next) {
4710 		flow_drv_remove(dev, flow);
4711 		flow_mreg_stop_copy_action(dev, flow);
4712 	}
4713 	flow_mreg_del_default_copy_action(dev);
4714 	flow_rxq_flags_clear(dev);
4715 }
4716 
4717 /**
4718  * Add all flows.
4719  *
4720  * @param dev
4721  *   Pointer to Ethernet device.
4722  * @param list
4723  *   Pointer to the Indexed flow list.
4724  *
4725  * @return
4726  *   0 on success, a negative errno value otherwise and rte_errno is set.
4727  */
4728 int
4729 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4730 {
4731 	struct mlx5_priv *priv = dev->data->dev_private;
4732 	struct rte_flow *flow = NULL;
4733 	struct rte_flow_error error;
4734 	uint32_t idx;
4735 	int ret = 0;
4736 
4737 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4738 	ret = flow_mreg_add_default_copy_action(dev, &error);
4739 	if (ret < 0)
4740 		return -rte_errno;
4741 	/* Apply Flows created by application. */
4742 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4743 		      flow, next) {
4744 		ret = flow_mreg_start_copy_action(dev, flow);
4745 		if (ret < 0)
4746 			goto error;
4747 		ret = flow_drv_apply(dev, flow, &error);
4748 		if (ret < 0)
4749 			goto error;
4750 		flow_rxq_flags_set(dev, flow);
4751 	}
4752 	return 0;
4753 error:
4754 	ret = rte_errno; /* Save rte_errno before cleanup. */
4755 	mlx5_flow_stop(dev, list);
4756 	rte_errno = ret; /* Restore rte_errno. */
4757 	return -rte_errno;
4758 }
4759 
4760 /**
4761  * Stop all default actions for flows.
4762  *
4763  * @param dev
4764  *   Pointer to Ethernet device.
4765  */
4766 void
4767 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4768 {
4769 	flow_mreg_del_default_copy_action(dev);
4770 	flow_rxq_flags_clear(dev);
4771 }
4772 
4773 /**
4774  * Start all default actions for flows.
4775  *
4776  * @param dev
4777  *   Pointer to Ethernet device.
4778  * @return
4779  *   0 on success, a negative errno value otherwise and rte_errno is set.
4780  */
4781 int
4782 mlx5_flow_start_default(struct rte_eth_dev *dev)
4783 {
4784 	struct rte_flow_error error;
4785 
4786 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4787 	return flow_mreg_add_default_copy_action(dev, &error);
4788 }
4789 
4790 /**
4791  * Allocate intermediate resources for flow creation.
4792  *
4793  * @param dev
4794  *   Pointer to Ethernet device.
4795  */
4796 void
4797 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4798 {
4799 	struct mlx5_priv *priv = dev->data->dev_private;
4800 
4801 	if (!priv->inter_flows) {
4802 		priv->inter_flows = rte_calloc(__func__, 1,
4803 				    MLX5_NUM_MAX_DEV_FLOWS *
4804 				    sizeof(struct mlx5_flow) +
4805 				    (sizeof(struct mlx5_flow_rss_desc) +
4806 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4807 		if (!priv->inter_flows) {
4808 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4809 			return;
4810 		}
4811 	}
4812 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4813 			 [MLX5_NUM_MAX_DEV_FLOWS];
4814 	/* Reset the index. */
4815 	priv->flow_idx = 0;
4816 	priv->flow_nested_idx = 0;
4817 }
4818 
4819 /**
4820  * Free intermediate resources for flows.
4821  *
4822  * @param dev
4823  *   Pointer to Ethernet device.
4824  */
4825 void
4826 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4827 {
4828 	struct mlx5_priv *priv = dev->data->dev_private;
4829 
4830 	rte_free(priv->inter_flows);
4831 	priv->inter_flows = NULL;
4832 }
4833 
4834 /**
4835  * Verify the flow list is empty
4836  *
4837  * @param dev
4838  *  Pointer to Ethernet device.
4839  *
4840  * @return the number of flows not released.
4841  */
4842 int
4843 mlx5_flow_verify(struct rte_eth_dev *dev)
4844 {
4845 	struct mlx5_priv *priv = dev->data->dev_private;
4846 	struct rte_flow *flow;
4847 	uint32_t idx;
4848 	int ret = 0;
4849 
4850 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4851 		      flow, next) {
4852 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4853 			dev->data->port_id, (void *)flow);
4854 		++ret;
4855 	}
4856 	return ret;
4857 }
4858 
4859 /**
4860  * Enable default hairpin egress flow.
4861  *
4862  * @param dev
4863  *   Pointer to Ethernet device.
4864  * @param queue
4865  *   The queue index.
4866  *
4867  * @return
4868  *   0 on success, a negative errno value otherwise and rte_errno is set.
4869  */
4870 int
4871 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4872 			    uint32_t queue)
4873 {
4874 	struct mlx5_priv *priv = dev->data->dev_private;
4875 	const struct rte_flow_attr attr = {
4876 		.egress = 1,
4877 		.priority = 0,
4878 	};
4879 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4880 		.queue = queue,
4881 	};
4882 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4883 		.queue = UINT32_MAX,
4884 	};
4885 	struct rte_flow_item items[] = {
4886 		{
4887 			.type = (enum rte_flow_item_type)
4888 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4889 			.spec = &queue_spec,
4890 			.last = NULL,
4891 			.mask = &queue_mask,
4892 		},
4893 		{
4894 			.type = RTE_FLOW_ITEM_TYPE_END,
4895 		},
4896 	};
4897 	struct rte_flow_action_jump jump = {
4898 		.group = MLX5_HAIRPIN_TX_TABLE,
4899 	};
4900 	struct rte_flow_action actions[2];
4901 	uint32_t flow_idx;
4902 	struct rte_flow_error error;
4903 
4904 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4905 	actions[0].conf = &jump;
4906 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4907 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4908 				&attr, items, actions, false, &error);
4909 	if (!flow_idx) {
4910 		DRV_LOG(DEBUG,
4911 			"Failed to create ctrl flow: rte_errno(%d),"
4912 			" type(%d), message(%s)",
4913 			rte_errno, error.type,
4914 			error.message ? error.message : " (no stated reason)");
4915 		return -rte_errno;
4916 	}
4917 	return 0;
4918 }
4919 
4920 /**
4921  * Enable a control flow configured from the control plane.
4922  *
4923  * @param dev
4924  *   Pointer to Ethernet device.
4925  * @param eth_spec
4926  *   An Ethernet flow spec to apply.
4927  * @param eth_mask
4928  *   An Ethernet flow mask to apply.
4929  * @param vlan_spec
4930  *   A VLAN flow spec to apply.
4931  * @param vlan_mask
4932  *   A VLAN flow mask to apply.
4933  *
4934  * @return
4935  *   0 on success, a negative errno value otherwise and rte_errno is set.
4936  */
4937 int
4938 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4939 		    struct rte_flow_item_eth *eth_spec,
4940 		    struct rte_flow_item_eth *eth_mask,
4941 		    struct rte_flow_item_vlan *vlan_spec,
4942 		    struct rte_flow_item_vlan *vlan_mask)
4943 {
4944 	struct mlx5_priv *priv = dev->data->dev_private;
4945 	const struct rte_flow_attr attr = {
4946 		.ingress = 1,
4947 		.priority = MLX5_FLOW_PRIO_RSVD,
4948 	};
4949 	struct rte_flow_item items[] = {
4950 		{
4951 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4952 			.spec = eth_spec,
4953 			.last = NULL,
4954 			.mask = eth_mask,
4955 		},
4956 		{
4957 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4958 					      RTE_FLOW_ITEM_TYPE_END,
4959 			.spec = vlan_spec,
4960 			.last = NULL,
4961 			.mask = vlan_mask,
4962 		},
4963 		{
4964 			.type = RTE_FLOW_ITEM_TYPE_END,
4965 		},
4966 	};
4967 	uint16_t queue[priv->reta_idx_n];
4968 	struct rte_flow_action_rss action_rss = {
4969 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4970 		.level = 0,
4971 		.types = priv->rss_conf.rss_hf,
4972 		.key_len = priv->rss_conf.rss_key_len,
4973 		.queue_num = priv->reta_idx_n,
4974 		.key = priv->rss_conf.rss_key,
4975 		.queue = queue,
4976 	};
4977 	struct rte_flow_action actions[] = {
4978 		{
4979 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4980 			.conf = &action_rss,
4981 		},
4982 		{
4983 			.type = RTE_FLOW_ACTION_TYPE_END,
4984 		},
4985 	};
4986 	uint32_t flow_idx;
4987 	struct rte_flow_error error;
4988 	unsigned int i;
4989 
4990 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4991 		return 0;
4992 	}
4993 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4994 		action_rss.types = 0;
4995 	for (i = 0; i != priv->reta_idx_n; ++i)
4996 		queue[i] = (*priv->reta_idx)[i];
4997 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4998 				&attr, items, actions, false, &error);
4999 	if (!flow_idx)
5000 		return -rte_errno;
5001 	return 0;
5002 }
5003 
5004 /**
5005  * Enable a flow control configured from the control plane.
5006  *
5007  * @param dev
5008  *   Pointer to Ethernet device.
5009  * @param eth_spec
5010  *   An Ethernet flow spec to apply.
5011  * @param eth_mask
5012  *   An Ethernet flow mask to apply.
5013  *
5014  * @return
5015  *   0 on success, a negative errno value otherwise and rte_errno is set.
5016  */
5017 int
5018 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5019 	       struct rte_flow_item_eth *eth_spec,
5020 	       struct rte_flow_item_eth *eth_mask)
5021 {
5022 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5023 }
5024 
5025 /**
5026  * Create default miss flow rule matching lacp traffic
5027  *
5028  * @param dev
5029  *   Pointer to Ethernet device.
5030  * @param eth_spec
5031  *   An Ethernet flow spec to apply.
5032  *
5033  * @return
5034  *   0 on success, a negative errno value otherwise and rte_errno is set.
5035  */
5036 int
5037 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5038 {
5039 	struct mlx5_priv *priv = dev->data->dev_private;
5040 	/*
5041 	 * The LACP matching is done by only using ether type since using
5042 	 * a multicast dst mac causes kernel to give low priority to this flow.
5043 	 */
5044 	static const struct rte_flow_item_eth lacp_spec = {
5045 		.type = RTE_BE16(0x8809),
5046 	};
5047 	static const struct rte_flow_item_eth lacp_mask = {
5048 		.type = 0xffff,
5049 	};
5050 	const struct rte_flow_attr attr = {
5051 		.ingress = 1,
5052 	};
5053 	struct rte_flow_item items[] = {
5054 		{
5055 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5056 			.spec = &lacp_spec,
5057 			.mask = &lacp_mask,
5058 		},
5059 		{
5060 			.type = RTE_FLOW_ITEM_TYPE_END,
5061 		},
5062 	};
5063 	struct rte_flow_action actions[] = {
5064 		{
5065 			.type = (enum rte_flow_action_type)
5066 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5067 		},
5068 		{
5069 			.type = RTE_FLOW_ACTION_TYPE_END,
5070 		},
5071 	};
5072 	struct rte_flow_error error;
5073 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5074 				&attr, items, actions, false, &error);
5075 
5076 	if (!flow_idx)
5077 		return -rte_errno;
5078 	return 0;
5079 }
5080 
5081 /**
5082  * Destroy a flow.
5083  *
5084  * @see rte_flow_destroy()
5085  * @see rte_flow_ops
5086  */
5087 int
5088 mlx5_flow_destroy(struct rte_eth_dev *dev,
5089 		  struct rte_flow *flow,
5090 		  struct rte_flow_error *error __rte_unused)
5091 {
5092 	struct mlx5_priv *priv = dev->data->dev_private;
5093 
5094 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5095 	return 0;
5096 }
5097 
5098 /**
5099  * Destroy all flows.
5100  *
5101  * @see rte_flow_flush()
5102  * @see rte_flow_ops
5103  */
5104 int
5105 mlx5_flow_flush(struct rte_eth_dev *dev,
5106 		struct rte_flow_error *error __rte_unused)
5107 {
5108 	struct mlx5_priv *priv = dev->data->dev_private;
5109 
5110 	mlx5_flow_list_flush(dev, &priv->flows, false);
5111 	return 0;
5112 }
5113 
5114 /**
5115  * Isolated mode.
5116  *
5117  * @see rte_flow_isolate()
5118  * @see rte_flow_ops
5119  */
5120 int
5121 mlx5_flow_isolate(struct rte_eth_dev *dev,
5122 		  int enable,
5123 		  struct rte_flow_error *error)
5124 {
5125 	struct mlx5_priv *priv = dev->data->dev_private;
5126 
5127 	if (dev->data->dev_started) {
5128 		rte_flow_error_set(error, EBUSY,
5129 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5130 				   NULL,
5131 				   "port must be stopped first");
5132 		return -rte_errno;
5133 	}
5134 	priv->isolated = !!enable;
5135 	if (enable)
5136 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
5137 	else
5138 		dev->dev_ops = &mlx5_os_dev_ops;
5139 	return 0;
5140 }
5141 
5142 /**
5143  * Query a flow.
5144  *
5145  * @see rte_flow_query()
5146  * @see rte_flow_ops
5147  */
5148 static int
5149 flow_drv_query(struct rte_eth_dev *dev,
5150 	       uint32_t flow_idx,
5151 	       const struct rte_flow_action *actions,
5152 	       void *data,
5153 	       struct rte_flow_error *error)
5154 {
5155 	struct mlx5_priv *priv = dev->data->dev_private;
5156 	const struct mlx5_flow_driver_ops *fops;
5157 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5158 					       [MLX5_IPOOL_RTE_FLOW],
5159 					       flow_idx);
5160 	enum mlx5_flow_drv_type ftype;
5161 
5162 	if (!flow) {
5163 		return rte_flow_error_set(error, ENOENT,
5164 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5165 			  NULL,
5166 			  "invalid flow handle");
5167 	}
5168 	ftype = flow->drv_type;
5169 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5170 	fops = flow_get_drv_ops(ftype);
5171 
5172 	return fops->query(dev, flow, actions, data, error);
5173 }
5174 
5175 /**
5176  * Query a flow.
5177  *
5178  * @see rte_flow_query()
5179  * @see rte_flow_ops
5180  */
5181 int
5182 mlx5_flow_query(struct rte_eth_dev *dev,
5183 		struct rte_flow *flow,
5184 		const struct rte_flow_action *actions,
5185 		void *data,
5186 		struct rte_flow_error *error)
5187 {
5188 	int ret;
5189 
5190 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5191 			     error);
5192 	if (ret < 0)
5193 		return ret;
5194 	return 0;
5195 }
5196 
5197 /**
5198  * Convert a flow director filter to a generic flow.
5199  *
5200  * @param dev
5201  *   Pointer to Ethernet device.
5202  * @param fdir_filter
5203  *   Flow director filter to add.
5204  * @param attributes
5205  *   Generic flow parameters structure.
5206  *
5207  * @return
5208  *   0 on success, a negative errno value otherwise and rte_errno is set.
5209  */
5210 static int
5211 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5212 			 const struct rte_eth_fdir_filter *fdir_filter,
5213 			 struct mlx5_fdir *attributes)
5214 {
5215 	struct mlx5_priv *priv = dev->data->dev_private;
5216 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5217 	const struct rte_eth_fdir_masks *mask =
5218 		&dev->data->dev_conf.fdir_conf.mask;
5219 
5220 	/* Validate queue number. */
5221 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5222 		DRV_LOG(ERR, "port %u invalid queue number %d",
5223 			dev->data->port_id, fdir_filter->action.rx_queue);
5224 		rte_errno = EINVAL;
5225 		return -rte_errno;
5226 	}
5227 	attributes->attr.ingress = 1;
5228 	attributes->items[0] = (struct rte_flow_item) {
5229 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5230 		.spec = &attributes->l2,
5231 		.mask = &attributes->l2_mask,
5232 	};
5233 	switch (fdir_filter->action.behavior) {
5234 	case RTE_ETH_FDIR_ACCEPT:
5235 		attributes->actions[0] = (struct rte_flow_action){
5236 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5237 			.conf = &attributes->queue,
5238 		};
5239 		break;
5240 	case RTE_ETH_FDIR_REJECT:
5241 		attributes->actions[0] = (struct rte_flow_action){
5242 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5243 		};
5244 		break;
5245 	default:
5246 		DRV_LOG(ERR, "port %u invalid behavior %d",
5247 			dev->data->port_id,
5248 			fdir_filter->action.behavior);
5249 		rte_errno = ENOTSUP;
5250 		return -rte_errno;
5251 	}
5252 	attributes->queue.index = fdir_filter->action.rx_queue;
5253 	/* Handle L3. */
5254 	switch (fdir_filter->input.flow_type) {
5255 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5256 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5257 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5258 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5259 			.src_addr = input->flow.ip4_flow.src_ip,
5260 			.dst_addr = input->flow.ip4_flow.dst_ip,
5261 			.time_to_live = input->flow.ip4_flow.ttl,
5262 			.type_of_service = input->flow.ip4_flow.tos,
5263 		};
5264 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5265 			.src_addr = mask->ipv4_mask.src_ip,
5266 			.dst_addr = mask->ipv4_mask.dst_ip,
5267 			.time_to_live = mask->ipv4_mask.ttl,
5268 			.type_of_service = mask->ipv4_mask.tos,
5269 			.next_proto_id = mask->ipv4_mask.proto,
5270 		};
5271 		attributes->items[1] = (struct rte_flow_item){
5272 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5273 			.spec = &attributes->l3,
5274 			.mask = &attributes->l3_mask,
5275 		};
5276 		break;
5277 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5278 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5279 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5280 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5281 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5282 			.proto = input->flow.ipv6_flow.proto,
5283 		};
5284 
5285 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5286 		       input->flow.ipv6_flow.src_ip,
5287 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5288 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5289 		       input->flow.ipv6_flow.dst_ip,
5290 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5291 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5292 		       mask->ipv6_mask.src_ip,
5293 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5294 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5295 		       mask->ipv6_mask.dst_ip,
5296 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5297 		attributes->items[1] = (struct rte_flow_item){
5298 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5299 			.spec = &attributes->l3,
5300 			.mask = &attributes->l3_mask,
5301 		};
5302 		break;
5303 	default:
5304 		DRV_LOG(ERR, "port %u invalid flow type%d",
5305 			dev->data->port_id, fdir_filter->input.flow_type);
5306 		rte_errno = ENOTSUP;
5307 		return -rte_errno;
5308 	}
5309 	/* Handle L4. */
5310 	switch (fdir_filter->input.flow_type) {
5311 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5312 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5313 			.src_port = input->flow.udp4_flow.src_port,
5314 			.dst_port = input->flow.udp4_flow.dst_port,
5315 		};
5316 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5317 			.src_port = mask->src_port_mask,
5318 			.dst_port = mask->dst_port_mask,
5319 		};
5320 		attributes->items[2] = (struct rte_flow_item){
5321 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5322 			.spec = &attributes->l4,
5323 			.mask = &attributes->l4_mask,
5324 		};
5325 		break;
5326 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5327 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5328 			.src_port = input->flow.tcp4_flow.src_port,
5329 			.dst_port = input->flow.tcp4_flow.dst_port,
5330 		};
5331 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5332 			.src_port = mask->src_port_mask,
5333 			.dst_port = mask->dst_port_mask,
5334 		};
5335 		attributes->items[2] = (struct rte_flow_item){
5336 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5337 			.spec = &attributes->l4,
5338 			.mask = &attributes->l4_mask,
5339 		};
5340 		break;
5341 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5342 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5343 			.src_port = input->flow.udp6_flow.src_port,
5344 			.dst_port = input->flow.udp6_flow.dst_port,
5345 		};
5346 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5347 			.src_port = mask->src_port_mask,
5348 			.dst_port = mask->dst_port_mask,
5349 		};
5350 		attributes->items[2] = (struct rte_flow_item){
5351 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5352 			.spec = &attributes->l4,
5353 			.mask = &attributes->l4_mask,
5354 		};
5355 		break;
5356 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5357 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5358 			.src_port = input->flow.tcp6_flow.src_port,
5359 			.dst_port = input->flow.tcp6_flow.dst_port,
5360 		};
5361 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5362 			.src_port = mask->src_port_mask,
5363 			.dst_port = mask->dst_port_mask,
5364 		};
5365 		attributes->items[2] = (struct rte_flow_item){
5366 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5367 			.spec = &attributes->l4,
5368 			.mask = &attributes->l4_mask,
5369 		};
5370 		break;
5371 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5372 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5373 		break;
5374 	default:
5375 		DRV_LOG(ERR, "port %u invalid flow type%d",
5376 			dev->data->port_id, fdir_filter->input.flow_type);
5377 		rte_errno = ENOTSUP;
5378 		return -rte_errno;
5379 	}
5380 	return 0;
5381 }
5382 
5383 #define FLOW_FDIR_CMP(f1, f2, fld) \
5384 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5385 
5386 /**
5387  * Compare two FDIR flows. If items and actions are identical, the two flows are
5388  * regarded as same.
5389  *
5390  * @param dev
5391  *   Pointer to Ethernet device.
5392  * @param f1
5393  *   FDIR flow to compare.
5394  * @param f2
5395  *   FDIR flow to compare.
5396  *
5397  * @return
5398  *   Zero on match, 1 otherwise.
5399  */
5400 static int
5401 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5402 {
5403 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5404 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5405 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5406 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5407 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5408 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5409 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5410 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5411 		return 1;
5412 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5413 	    FLOW_FDIR_CMP(f1, f2, queue))
5414 		return 1;
5415 	return 0;
5416 }
5417 
5418 /**
5419  * Search device flow list to find out a matched FDIR flow.
5420  *
5421  * @param dev
5422  *   Pointer to Ethernet device.
5423  * @param fdir_flow
5424  *   FDIR flow to lookup.
5425  *
5426  * @return
5427  *   Index of flow if found, 0 otherwise.
5428  */
5429 static uint32_t
5430 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5431 {
5432 	struct mlx5_priv *priv = dev->data->dev_private;
5433 	uint32_t flow_idx = 0;
5434 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5435 
5436 	MLX5_ASSERT(fdir_flow);
5437 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5438 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5439 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5440 				dev->data->port_id, flow_idx);
5441 			flow_idx = priv_fdir_flow->rix_flow;
5442 			break;
5443 		}
5444 	}
5445 	return flow_idx;
5446 }
5447 
5448 /**
5449  * Add new flow director filter and store it in list.
5450  *
5451  * @param dev
5452  *   Pointer to Ethernet device.
5453  * @param fdir_filter
5454  *   Flow director filter to add.
5455  *
5456  * @return
5457  *   0 on success, a negative errno value otherwise and rte_errno is set.
5458  */
5459 static int
5460 flow_fdir_filter_add(struct rte_eth_dev *dev,
5461 		     const struct rte_eth_fdir_filter *fdir_filter)
5462 {
5463 	struct mlx5_priv *priv = dev->data->dev_private;
5464 	struct mlx5_fdir *fdir_flow;
5465 	struct rte_flow *flow;
5466 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5467 	uint32_t flow_idx;
5468 	int ret;
5469 
5470 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5471 	if (!fdir_flow) {
5472 		rte_errno = ENOMEM;
5473 		return -rte_errno;
5474 	}
5475 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5476 	if (ret)
5477 		goto error;
5478 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5479 	if (flow_idx) {
5480 		rte_errno = EEXIST;
5481 		goto error;
5482 	}
5483 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5484 				     0);
5485 	if (!priv_fdir_flow) {
5486 		rte_errno = ENOMEM;
5487 		goto error;
5488 	}
5489 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5490 				    fdir_flow->items, fdir_flow->actions, true,
5491 				    NULL);
5492 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5493 	if (!flow)
5494 		goto error;
5495 	flow->fdir = 1;
5496 	priv_fdir_flow->fdir = fdir_flow;
5497 	priv_fdir_flow->rix_flow = flow_idx;
5498 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5499 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5500 		dev->data->port_id, (void *)flow);
5501 	return 0;
5502 error:
5503 	rte_free(priv_fdir_flow);
5504 	rte_free(fdir_flow);
5505 	return -rte_errno;
5506 }
5507 
5508 /**
5509  * Delete specific filter.
5510  *
5511  * @param dev
5512  *   Pointer to Ethernet device.
5513  * @param fdir_filter
5514  *   Filter to be deleted.
5515  *
5516  * @return
5517  *   0 on success, a negative errno value otherwise and rte_errno is set.
5518  */
5519 static int
5520 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5521 			const struct rte_eth_fdir_filter *fdir_filter)
5522 {
5523 	struct mlx5_priv *priv = dev->data->dev_private;
5524 	uint32_t flow_idx;
5525 	struct mlx5_fdir fdir_flow = {
5526 		.attr.group = 0,
5527 	};
5528 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5529 	int ret;
5530 
5531 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5532 	if (ret)
5533 		return -rte_errno;
5534 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5535 		/* Find the fdir in priv list */
5536 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5537 			break;
5538 	}
5539 	if (!priv_fdir_flow)
5540 		return 0;
5541 	LIST_REMOVE(priv_fdir_flow, next);
5542 	flow_idx = priv_fdir_flow->rix_flow;
5543 	flow_list_destroy(dev, &priv->flows, flow_idx);
5544 	rte_free(priv_fdir_flow->fdir);
5545 	rte_free(priv_fdir_flow);
5546 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5547 		dev->data->port_id, flow_idx);
5548 	return 0;
5549 }
5550 
5551 /**
5552  * Update queue for specific filter.
5553  *
5554  * @param dev
5555  *   Pointer to Ethernet device.
5556  * @param fdir_filter
5557  *   Filter to be updated.
5558  *
5559  * @return
5560  *   0 on success, a negative errno value otherwise and rte_errno is set.
5561  */
5562 static int
5563 flow_fdir_filter_update(struct rte_eth_dev *dev,
5564 			const struct rte_eth_fdir_filter *fdir_filter)
5565 {
5566 	int ret;
5567 
5568 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5569 	if (ret)
5570 		return ret;
5571 	return flow_fdir_filter_add(dev, fdir_filter);
5572 }
5573 
5574 /**
5575  * Flush all filters.
5576  *
5577  * @param dev
5578  *   Pointer to Ethernet device.
5579  */
5580 static void
5581 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5582 {
5583 	struct mlx5_priv *priv = dev->data->dev_private;
5584 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5585 
5586 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5587 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5588 		LIST_REMOVE(priv_fdir_flow, next);
5589 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5590 		rte_free(priv_fdir_flow->fdir);
5591 		rte_free(priv_fdir_flow);
5592 	}
5593 }
5594 
5595 /**
5596  * Get flow director information.
5597  *
5598  * @param dev
5599  *   Pointer to Ethernet device.
5600  * @param[out] fdir_info
5601  *   Resulting flow director information.
5602  */
5603 static void
5604 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5605 {
5606 	struct rte_eth_fdir_masks *mask =
5607 		&dev->data->dev_conf.fdir_conf.mask;
5608 
5609 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5610 	fdir_info->guarant_spc = 0;
5611 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5612 	fdir_info->max_flexpayload = 0;
5613 	fdir_info->flow_types_mask[0] = 0;
5614 	fdir_info->flex_payload_unit = 0;
5615 	fdir_info->max_flex_payload_segment_num = 0;
5616 	fdir_info->flex_payload_limit = 0;
5617 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5618 }
5619 
5620 /**
5621  * Deal with flow director operations.
5622  *
5623  * @param dev
5624  *   Pointer to Ethernet device.
5625  * @param filter_op
5626  *   Operation to perform.
5627  * @param arg
5628  *   Pointer to operation-specific structure.
5629  *
5630  * @return
5631  *   0 on success, a negative errno value otherwise and rte_errno is set.
5632  */
5633 static int
5634 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5635 		    void *arg)
5636 {
5637 	enum rte_fdir_mode fdir_mode =
5638 		dev->data->dev_conf.fdir_conf.mode;
5639 
5640 	if (filter_op == RTE_ETH_FILTER_NOP)
5641 		return 0;
5642 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5643 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5644 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5645 			dev->data->port_id, fdir_mode);
5646 		rte_errno = EINVAL;
5647 		return -rte_errno;
5648 	}
5649 	switch (filter_op) {
5650 	case RTE_ETH_FILTER_ADD:
5651 		return flow_fdir_filter_add(dev, arg);
5652 	case RTE_ETH_FILTER_UPDATE:
5653 		return flow_fdir_filter_update(dev, arg);
5654 	case RTE_ETH_FILTER_DELETE:
5655 		return flow_fdir_filter_delete(dev, arg);
5656 	case RTE_ETH_FILTER_FLUSH:
5657 		flow_fdir_filter_flush(dev);
5658 		break;
5659 	case RTE_ETH_FILTER_INFO:
5660 		flow_fdir_info_get(dev, arg);
5661 		break;
5662 	default:
5663 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5664 			dev->data->port_id, filter_op);
5665 		rte_errno = EINVAL;
5666 		return -rte_errno;
5667 	}
5668 	return 0;
5669 }
5670 
5671 /**
5672  * Manage filter operations.
5673  *
5674  * @param dev
5675  *   Pointer to Ethernet device structure.
5676  * @param filter_type
5677  *   Filter type.
5678  * @param filter_op
5679  *   Operation to perform.
5680  * @param arg
5681  *   Pointer to operation-specific structure.
5682  *
5683  * @return
5684  *   0 on success, a negative errno value otherwise and rte_errno is set.
5685  */
5686 int
5687 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5688 		     enum rte_filter_type filter_type,
5689 		     enum rte_filter_op filter_op,
5690 		     void *arg)
5691 {
5692 	switch (filter_type) {
5693 	case RTE_ETH_FILTER_GENERIC:
5694 		if (filter_op != RTE_ETH_FILTER_GET) {
5695 			rte_errno = EINVAL;
5696 			return -rte_errno;
5697 		}
5698 		*(const void **)arg = &mlx5_flow_ops;
5699 		return 0;
5700 	case RTE_ETH_FILTER_FDIR:
5701 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5702 	default:
5703 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5704 			dev->data->port_id, filter_type);
5705 		rte_errno = ENOTSUP;
5706 		return -rte_errno;
5707 	}
5708 	return 0;
5709 }
5710 
5711 /**
5712  * Create the needed meter and suffix tables.
5713  *
5714  * @param[in] dev
5715  *   Pointer to Ethernet device.
5716  * @param[in] fm
5717  *   Pointer to the flow meter.
5718  *
5719  * @return
5720  *   Pointer to table set on success, NULL otherwise.
5721  */
5722 struct mlx5_meter_domains_infos *
5723 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5724 			  const struct mlx5_flow_meter *fm)
5725 {
5726 	const struct mlx5_flow_driver_ops *fops;
5727 
5728 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5729 	return fops->create_mtr_tbls(dev, fm);
5730 }
5731 
5732 /**
5733  * Destroy the meter table set.
5734  *
5735  * @param[in] dev
5736  *   Pointer to Ethernet device.
5737  * @param[in] tbl
5738  *   Pointer to the meter table set.
5739  *
5740  * @return
5741  *   0 on success.
5742  */
5743 int
5744 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5745 			   struct mlx5_meter_domains_infos *tbls)
5746 {
5747 	const struct mlx5_flow_driver_ops *fops;
5748 
5749 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5750 	return fops->destroy_mtr_tbls(dev, tbls);
5751 }
5752 
5753 /**
5754  * Create policer rules.
5755  *
5756  * @param[in] dev
5757  *   Pointer to Ethernet device.
5758  * @param[in] fm
5759  *   Pointer to flow meter structure.
5760  * @param[in] attr
5761  *   Pointer to flow attributes.
5762  *
5763  * @return
5764  *   0 on success, -1 otherwise.
5765  */
5766 int
5767 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5768 			       struct mlx5_flow_meter *fm,
5769 			       const struct rte_flow_attr *attr)
5770 {
5771 	const struct mlx5_flow_driver_ops *fops;
5772 
5773 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5774 	return fops->create_policer_rules(dev, fm, attr);
5775 }
5776 
5777 /**
5778  * Destroy policer rules.
5779  *
5780  * @param[in] fm
5781  *   Pointer to flow meter structure.
5782  * @param[in] attr
5783  *   Pointer to flow attributes.
5784  *
5785  * @return
5786  *   0 on success, -1 otherwise.
5787  */
5788 int
5789 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5790 				struct mlx5_flow_meter *fm,
5791 				const struct rte_flow_attr *attr)
5792 {
5793 	const struct mlx5_flow_driver_ops *fops;
5794 
5795 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5796 	return fops->destroy_policer_rules(dev, fm, attr);
5797 }
5798 
5799 /**
5800  * Allocate a counter.
5801  *
5802  * @param[in] dev
5803  *   Pointer to Ethernet device structure.
5804  *
5805  * @return
5806  *   Index to allocated counter  on success, 0 otherwise.
5807  */
5808 uint32_t
5809 mlx5_counter_alloc(struct rte_eth_dev *dev)
5810 {
5811 	const struct mlx5_flow_driver_ops *fops;
5812 	struct rte_flow_attr attr = { .transfer = 0 };
5813 
5814 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5815 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5816 		return fops->counter_alloc(dev);
5817 	}
5818 	DRV_LOG(ERR,
5819 		"port %u counter allocate is not supported.",
5820 		 dev->data->port_id);
5821 	return 0;
5822 }
5823 
5824 /**
5825  * Free a counter.
5826  *
5827  * @param[in] dev
5828  *   Pointer to Ethernet device structure.
5829  * @param[in] cnt
5830  *   Index to counter to be free.
5831  */
5832 void
5833 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5834 {
5835 	const struct mlx5_flow_driver_ops *fops;
5836 	struct rte_flow_attr attr = { .transfer = 0 };
5837 
5838 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5839 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5840 		fops->counter_free(dev, cnt);
5841 		return;
5842 	}
5843 	DRV_LOG(ERR,
5844 		"port %u counter free is not supported.",
5845 		 dev->data->port_id);
5846 }
5847 
5848 /**
5849  * Query counter statistics.
5850  *
5851  * @param[in] dev
5852  *   Pointer to Ethernet device structure.
5853  * @param[in] cnt
5854  *   Index to counter to query.
5855  * @param[in] clear
5856  *   Set to clear counter statistics.
5857  * @param[out] pkts
5858  *   The counter hits packets number to save.
5859  * @param[out] bytes
5860  *   The counter hits bytes number to save.
5861  *
5862  * @return
5863  *   0 on success, a negative errno value otherwise.
5864  */
5865 int
5866 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5867 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5868 {
5869 	const struct mlx5_flow_driver_ops *fops;
5870 	struct rte_flow_attr attr = { .transfer = 0 };
5871 
5872 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5873 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5874 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5875 	}
5876 	DRV_LOG(ERR,
5877 		"port %u counter query is not supported.",
5878 		 dev->data->port_id);
5879 	return -ENOTSUP;
5880 }
5881 
5882 #define MLX5_POOL_QUERY_FREQ_US 1000000
5883 
5884 /**
5885  * Get number of all validate pools.
5886  *
5887  * @param[in] sh
5888  *   Pointer to mlx5_dev_ctx_shared object.
5889  *
5890  * @return
5891  *   The number of all validate pools.
5892  */
5893 static uint32_t
5894 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
5895 {
5896 	int i;
5897 	uint32_t pools_n = 0;
5898 
5899 	for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
5900 		pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
5901 	return pools_n;
5902 }
5903 
5904 /**
5905  * Set the periodic procedure for triggering asynchronous batch queries for all
5906  * the counter pools.
5907  *
5908  * @param[in] sh
5909  *   Pointer to mlx5_dev_ctx_shared object.
5910  */
5911 void
5912 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
5913 {
5914 	uint32_t pools_n, us;
5915 
5916 	pools_n = mlx5_get_all_valid_pool_count(sh);
5917 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5918 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5919 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5920 		sh->cmng.query_thread_on = 0;
5921 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5922 	} else {
5923 		sh->cmng.query_thread_on = 1;
5924 	}
5925 }
5926 
5927 /**
5928  * The periodic procedure for triggering asynchronous batch queries for all the
5929  * counter pools. This function is probably called by the host thread.
5930  *
5931  * @param[in] arg
5932  *   The parameter for the alarm process.
5933  */
5934 void
5935 mlx5_flow_query_alarm(void *arg)
5936 {
5937 	struct mlx5_dev_ctx_shared *sh = arg;
5938 	struct mlx5_devx_obj *dcs;
5939 	uint16_t offset;
5940 	int ret;
5941 	uint8_t batch = sh->cmng.batch;
5942 	uint8_t age = sh->cmng.age;
5943 	uint16_t pool_index = sh->cmng.pool_index;
5944 	struct mlx5_pools_container *cont;
5945 	struct mlx5_flow_counter_pool *pool;
5946 	int cont_loop = MLX5_CCONT_TYPE_MAX;
5947 
5948 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5949 		goto set_alarm;
5950 next_container:
5951 	cont = MLX5_CNT_CONTAINER(sh, batch, age);
5952 	rte_spinlock_lock(&cont->resize_sl);
5953 	if (!cont->pools) {
5954 		rte_spinlock_unlock(&cont->resize_sl);
5955 		/* Check if all the containers are empty. */
5956 		if (unlikely(--cont_loop == 0))
5957 			goto set_alarm;
5958 		batch ^= 0x1;
5959 		pool_index = 0;
5960 		if (batch == 0 && pool_index == 0) {
5961 			age ^= 0x1;
5962 			sh->cmng.batch = batch;
5963 			sh->cmng.age = age;
5964 		}
5965 		goto next_container;
5966 	}
5967 	pool = cont->pools[pool_index];
5968 	rte_spinlock_unlock(&cont->resize_sl);
5969 	if (pool->raw_hw)
5970 		/* There is a pool query in progress. */
5971 		goto set_alarm;
5972 	pool->raw_hw =
5973 		LIST_FIRST(&sh->cmng.free_stat_raws);
5974 	if (!pool->raw_hw)
5975 		/* No free counter statistics raw memory. */
5976 		goto set_alarm;
5977 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5978 							      (&pool->a64_dcs);
5979 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5980 	/*
5981 	 * Identify the counters released between query trigger and query
5982 	 * handle more effiecntly. The counter released in this gap period
5983 	 * should wait for a new round of query as the new arrived packets
5984 	 * will not be taken into account.
5985 	 */
5986 	pool->query_gen++;
5987 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5988 					       offset, NULL, NULL,
5989 					       pool->raw_hw->mem_mng->dm->id,
5990 					       (void *)(uintptr_t)
5991 					       (pool->raw_hw->data + offset),
5992 					       sh->devx_comp,
5993 					       (uint64_t)(uintptr_t)pool);
5994 	if (ret) {
5995 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5996 			" %d", pool->min_dcs->id);
5997 		pool->raw_hw = NULL;
5998 		goto set_alarm;
5999 	}
6000 	pool->raw_hw->min_dcs_id = dcs->id;
6001 	LIST_REMOVE(pool->raw_hw, next);
6002 	sh->cmng.pending_queries++;
6003 	pool_index++;
6004 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6005 		batch ^= 0x1;
6006 		pool_index = 0;
6007 		if (batch == 0 && pool_index == 0)
6008 			age ^= 0x1;
6009 	}
6010 set_alarm:
6011 	sh->cmng.batch = batch;
6012 	sh->cmng.pool_index = pool_index;
6013 	sh->cmng.age = age;
6014 	mlx5_set_query_alarm(sh);
6015 }
6016 
6017 /**
6018  * Check and callback event for new aged flow in the counter pool
6019  *
6020  * @param[in] sh
6021  *   Pointer to mlx5_dev_ctx_shared object.
6022  * @param[in] pool
6023  *   Pointer to Current counter pool.
6024  */
6025 static void
6026 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6027 		   struct mlx5_flow_counter_pool *pool)
6028 {
6029 	struct mlx5_priv *priv;
6030 	struct mlx5_flow_counter *cnt;
6031 	struct mlx5_age_info *age_info;
6032 	struct mlx5_age_param *age_param;
6033 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6034 	struct mlx5_counter_stats_raw *prev = pool->raw;
6035 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6036 	uint32_t i;
6037 
6038 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6039 		cnt = MLX5_POOL_GET_CNT(pool, i);
6040 		age_param = MLX5_CNT_TO_AGE(cnt);
6041 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6042 			continue;
6043 		if (cur->data[i].hits != prev->data[i].hits) {
6044 			age_param->expire = curr + age_param->timeout;
6045 			continue;
6046 		}
6047 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6048 			continue;
6049 		/**
6050 		 * Hold the lock first, or if between the
6051 		 * state AGE_TMOUT and tailq operation the
6052 		 * release happened, the release procedure
6053 		 * may delete a non-existent tailq node.
6054 		 */
6055 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6056 		age_info = GET_PORT_AGE_INFO(priv);
6057 		rte_spinlock_lock(&age_info->aged_sl);
6058 		/* If the cpmset fails, release happens. */
6059 		if (rte_atomic16_cmpset((volatile uint16_t *)
6060 					&age_param->state,
6061 					AGE_CANDIDATE,
6062 					AGE_TMOUT) ==
6063 					AGE_CANDIDATE) {
6064 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6065 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6066 		}
6067 		rte_spinlock_unlock(&age_info->aged_sl);
6068 	}
6069 	for (i = 0; i < sh->max_port; i++) {
6070 		age_info = &sh->port[i].age_info;
6071 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6072 			continue;
6073 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6074 			_rte_eth_dev_callback_process
6075 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
6076 				RTE_ETH_EVENT_FLOW_AGED, NULL);
6077 		age_info->flags = 0;
6078 	}
6079 }
6080 
6081 /**
6082  * Handler for the HW respond about ready values from an asynchronous batch
6083  * query. This function is probably called by the host thread.
6084  *
6085  * @param[in] sh
6086  *   The pointer to the shared device context.
6087  * @param[in] async_id
6088  *   The Devx async ID.
6089  * @param[in] status
6090  *   The status of the completion.
6091  */
6092 void
6093 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6094 				  uint64_t async_id, int status)
6095 {
6096 	struct mlx5_flow_counter_pool *pool =
6097 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6098 	struct mlx5_counter_stats_raw *raw_to_free;
6099 	uint8_t age = !!IS_AGE_POOL(pool);
6100 	uint8_t query_gen = pool->query_gen ^ 1;
6101 	struct mlx5_pools_container *cont =
6102 		MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6103 
6104 	if (unlikely(status)) {
6105 		raw_to_free = pool->raw_hw;
6106 	} else {
6107 		raw_to_free = pool->raw;
6108 		if (IS_AGE_POOL(pool))
6109 			mlx5_flow_aging_check(sh, pool);
6110 		rte_spinlock_lock(&pool->sl);
6111 		pool->raw = pool->raw_hw;
6112 		rte_spinlock_unlock(&pool->sl);
6113 		/* Be sure the new raw counters data is updated in memory. */
6114 		rte_cio_wmb();
6115 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6116 			rte_spinlock_lock(&cont->csl);
6117 			TAILQ_CONCAT(&cont->counters,
6118 				     &pool->counters[query_gen], next);
6119 			rte_spinlock_unlock(&cont->csl);
6120 		}
6121 	}
6122 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6123 	pool->raw_hw = NULL;
6124 	sh->cmng.pending_queries--;
6125 }
6126 
6127 /**
6128  * Translate the rte_flow group index to HW table value.
6129  *
6130  * @param[in] attributes
6131  *   Pointer to flow attributes
6132  * @param[in] external
6133  *   Value is part of flow rule created by request external to PMD.
6134  * @param[in] group
6135  *   rte_flow group index value.
6136  * @param[out] fdb_def_rule
6137  *   Whether fdb jump to table 1 is configured.
6138  * @param[out] table
6139  *   HW table value.
6140  * @param[out] error
6141  *   Pointer to error structure.
6142  *
6143  * @return
6144  *   0 on success, a negative errno value otherwise and rte_errno is set.
6145  */
6146 int
6147 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6148 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6149 			 struct rte_flow_error *error)
6150 {
6151 	if (attributes->transfer && external && fdb_def_rule) {
6152 		if (group == UINT32_MAX)
6153 			return rte_flow_error_set
6154 						(error, EINVAL,
6155 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6156 						 NULL,
6157 						 "group index not supported");
6158 		*table = group + 1;
6159 	} else {
6160 		*table = group;
6161 	}
6162 	return 0;
6163 }
6164 
6165 /**
6166  * Discover availability of metadata reg_c's.
6167  *
6168  * Iteratively use test flows to check availability.
6169  *
6170  * @param[in] dev
6171  *   Pointer to the Ethernet device structure.
6172  *
6173  * @return
6174  *   0 on success, a negative errno value otherwise and rte_errno is set.
6175  */
6176 int
6177 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6178 {
6179 	struct mlx5_priv *priv = dev->data->dev_private;
6180 	struct mlx5_dev_config *config = &priv->config;
6181 	enum modify_reg idx;
6182 	int n = 0;
6183 
6184 	/* reg_c[0] and reg_c[1] are reserved. */
6185 	config->flow_mreg_c[n++] = REG_C_0;
6186 	config->flow_mreg_c[n++] = REG_C_1;
6187 	/* Discover availability of other reg_c's. */
6188 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6189 		struct rte_flow_attr attr = {
6190 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6191 			.priority = MLX5_FLOW_PRIO_RSVD,
6192 			.ingress = 1,
6193 		};
6194 		struct rte_flow_item items[] = {
6195 			[0] = {
6196 				.type = RTE_FLOW_ITEM_TYPE_END,
6197 			},
6198 		};
6199 		struct rte_flow_action actions[] = {
6200 			[0] = {
6201 				.type = (enum rte_flow_action_type)
6202 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6203 				.conf = &(struct mlx5_flow_action_copy_mreg){
6204 					.src = REG_C_1,
6205 					.dst = idx,
6206 				},
6207 			},
6208 			[1] = {
6209 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6210 				.conf = &(struct rte_flow_action_jump){
6211 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6212 				},
6213 			},
6214 			[2] = {
6215 				.type = RTE_FLOW_ACTION_TYPE_END,
6216 			},
6217 		};
6218 		uint32_t flow_idx;
6219 		struct rte_flow *flow;
6220 		struct rte_flow_error error;
6221 
6222 		if (!config->dv_flow_en)
6223 			break;
6224 		/* Create internal flow, validation skips copy action. */
6225 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6226 					    actions, false, &error);
6227 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6228 				      flow_idx);
6229 		if (!flow)
6230 			continue;
6231 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6232 			config->flow_mreg_c[n++] = idx;
6233 		flow_list_destroy(dev, NULL, flow_idx);
6234 	}
6235 	for (; n < MLX5_MREG_C_NUM; ++n)
6236 		config->flow_mreg_c[n] = REG_NONE;
6237 	return 0;
6238 }
6239 
6240 /**
6241  * Dump flow raw hw data to file
6242  *
6243  * @param[in] dev
6244  *    The pointer to Ethernet device.
6245  * @param[in] file
6246  *   A pointer to a file for output.
6247  * @param[out] error
6248  *   Perform verbose error reporting if not NULL. PMDs initialize this
6249  *   structure in case of error only.
6250  * @return
6251  *   0 on success, a nagative value otherwise.
6252  */
6253 int
6254 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6255 		   FILE *file,
6256 		   struct rte_flow_error *error __rte_unused)
6257 {
6258 	struct mlx5_priv *priv = dev->data->dev_private;
6259 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6260 
6261 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6262 				       sh->tx_domain, file);
6263 }
6264 
6265 /**
6266  * Get aged-out flows.
6267  *
6268  * @param[in] dev
6269  *   Pointer to the Ethernet device structure.
6270  * @param[in] context
6271  *   The address of an array of pointers to the aged-out flows contexts.
6272  * @param[in] nb_countexts
6273  *   The length of context array pointers.
6274  * @param[out] error
6275  *   Perform verbose error reporting if not NULL. Initialized in case of
6276  *   error only.
6277  *
6278  * @return
6279  *   how many contexts get in success, otherwise negative errno value.
6280  *   if nb_contexts is 0, return the amount of all aged contexts.
6281  *   if nb_contexts is not 0 , return the amount of aged flows reported
6282  *   in the context array.
6283  */
6284 int
6285 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6286 			uint32_t nb_contexts, struct rte_flow_error *error)
6287 {
6288 	const struct mlx5_flow_driver_ops *fops;
6289 	struct rte_flow_attr attr = { .transfer = 0 };
6290 
6291 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6292 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6293 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6294 						    error);
6295 	}
6296 	DRV_LOG(ERR,
6297 		"port %u get aged flows is not supported.",
6298 		 dev->data->port_id);
6299 	return -ENOTSUP;
6300 }
6301