xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 83c2047c5fe82fb25674c3053d4871838e7455cb)
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 #include <mlx5_malloc.h>
36 
37 #include "mlx5_defs.h"
38 #include "mlx5.h"
39 #include "mlx5_flow.h"
40 #include "mlx5_flow_os.h"
41 #include "mlx5_rxtx.h"
42 
43 /** Device flow drivers. */
44 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
45 
46 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
47 
48 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
49 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
50 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
51 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
52 #endif
53 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
54 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
55 };
56 
57 enum mlx5_expansion {
58 	MLX5_EXPANSION_ROOT,
59 	MLX5_EXPANSION_ROOT_OUTER,
60 	MLX5_EXPANSION_ROOT_ETH_VLAN,
61 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
62 	MLX5_EXPANSION_OUTER_ETH,
63 	MLX5_EXPANSION_OUTER_ETH_VLAN,
64 	MLX5_EXPANSION_OUTER_VLAN,
65 	MLX5_EXPANSION_OUTER_IPV4,
66 	MLX5_EXPANSION_OUTER_IPV4_UDP,
67 	MLX5_EXPANSION_OUTER_IPV4_TCP,
68 	MLX5_EXPANSION_OUTER_IPV6,
69 	MLX5_EXPANSION_OUTER_IPV6_UDP,
70 	MLX5_EXPANSION_OUTER_IPV6_TCP,
71 	MLX5_EXPANSION_VXLAN,
72 	MLX5_EXPANSION_VXLAN_GPE,
73 	MLX5_EXPANSION_GRE,
74 	MLX5_EXPANSION_MPLS,
75 	MLX5_EXPANSION_ETH,
76 	MLX5_EXPANSION_ETH_VLAN,
77 	MLX5_EXPANSION_VLAN,
78 	MLX5_EXPANSION_IPV4,
79 	MLX5_EXPANSION_IPV4_UDP,
80 	MLX5_EXPANSION_IPV4_TCP,
81 	MLX5_EXPANSION_IPV6,
82 	MLX5_EXPANSION_IPV6_UDP,
83 	MLX5_EXPANSION_IPV6_TCP,
84 };
85 
86 /** Supported expansion of items. */
87 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
88 	[MLX5_EXPANSION_ROOT] = {
89 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
90 						 MLX5_EXPANSION_IPV4,
91 						 MLX5_EXPANSION_IPV6),
92 		.type = RTE_FLOW_ITEM_TYPE_END,
93 	},
94 	[MLX5_EXPANSION_ROOT_OUTER] = {
95 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
96 						 MLX5_EXPANSION_OUTER_IPV4,
97 						 MLX5_EXPANSION_OUTER_IPV6),
98 		.type = RTE_FLOW_ITEM_TYPE_END,
99 	},
100 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
101 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
102 		.type = RTE_FLOW_ITEM_TYPE_END,
103 	},
104 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
105 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
106 		.type = RTE_FLOW_ITEM_TYPE_END,
107 	},
108 	[MLX5_EXPANSION_OUTER_ETH] = {
109 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
110 						 MLX5_EXPANSION_OUTER_IPV6,
111 						 MLX5_EXPANSION_MPLS),
112 		.type = RTE_FLOW_ITEM_TYPE_ETH,
113 		.rss_types = 0,
114 	},
115 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
116 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
117 		.type = RTE_FLOW_ITEM_TYPE_ETH,
118 		.rss_types = 0,
119 	},
120 	[MLX5_EXPANSION_OUTER_VLAN] = {
121 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
122 						 MLX5_EXPANSION_OUTER_IPV6),
123 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
124 	},
125 	[MLX5_EXPANSION_OUTER_IPV4] = {
126 		.next = RTE_FLOW_EXPAND_RSS_NEXT
127 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
128 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
129 			 MLX5_EXPANSION_GRE,
130 			 MLX5_EXPANSION_IPV4,
131 			 MLX5_EXPANSION_IPV6),
132 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
133 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
134 			ETH_RSS_NONFRAG_IPV4_OTHER,
135 	},
136 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
137 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
138 						 MLX5_EXPANSION_VXLAN_GPE),
139 		.type = RTE_FLOW_ITEM_TYPE_UDP,
140 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
141 	},
142 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
143 		.type = RTE_FLOW_ITEM_TYPE_TCP,
144 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
145 	},
146 	[MLX5_EXPANSION_OUTER_IPV6] = {
147 		.next = RTE_FLOW_EXPAND_RSS_NEXT
148 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
149 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
150 			 MLX5_EXPANSION_IPV4,
151 			 MLX5_EXPANSION_IPV6),
152 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
153 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
154 			ETH_RSS_NONFRAG_IPV6_OTHER,
155 	},
156 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
157 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
158 						 MLX5_EXPANSION_VXLAN_GPE),
159 		.type = RTE_FLOW_ITEM_TYPE_UDP,
160 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
161 	},
162 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
163 		.type = RTE_FLOW_ITEM_TYPE_TCP,
164 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
165 	},
166 	[MLX5_EXPANSION_VXLAN] = {
167 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
168 						 MLX5_EXPANSION_IPV4,
169 						 MLX5_EXPANSION_IPV6),
170 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
171 	},
172 	[MLX5_EXPANSION_VXLAN_GPE] = {
173 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
174 						 MLX5_EXPANSION_IPV4,
175 						 MLX5_EXPANSION_IPV6),
176 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
177 	},
178 	[MLX5_EXPANSION_GRE] = {
179 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
180 		.type = RTE_FLOW_ITEM_TYPE_GRE,
181 	},
182 	[MLX5_EXPANSION_MPLS] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
184 						 MLX5_EXPANSION_IPV6),
185 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
186 	},
187 	[MLX5_EXPANSION_ETH] = {
188 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
189 						 MLX5_EXPANSION_IPV6),
190 		.type = RTE_FLOW_ITEM_TYPE_ETH,
191 	},
192 	[MLX5_EXPANSION_ETH_VLAN] = {
193 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
194 		.type = RTE_FLOW_ITEM_TYPE_ETH,
195 	},
196 	[MLX5_EXPANSION_VLAN] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
198 						 MLX5_EXPANSION_IPV6),
199 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
200 	},
201 	[MLX5_EXPANSION_IPV4] = {
202 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
203 						 MLX5_EXPANSION_IPV4_TCP),
204 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
205 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
206 			ETH_RSS_NONFRAG_IPV4_OTHER,
207 	},
208 	[MLX5_EXPANSION_IPV4_UDP] = {
209 		.type = RTE_FLOW_ITEM_TYPE_UDP,
210 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
211 	},
212 	[MLX5_EXPANSION_IPV4_TCP] = {
213 		.type = RTE_FLOW_ITEM_TYPE_TCP,
214 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
215 	},
216 	[MLX5_EXPANSION_IPV6] = {
217 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
218 						 MLX5_EXPANSION_IPV6_TCP),
219 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
220 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
221 			ETH_RSS_NONFRAG_IPV6_OTHER,
222 	},
223 	[MLX5_EXPANSION_IPV6_UDP] = {
224 		.type = RTE_FLOW_ITEM_TYPE_UDP,
225 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
226 	},
227 	[MLX5_EXPANSION_IPV6_TCP] = {
228 		.type = RTE_FLOW_ITEM_TYPE_TCP,
229 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
230 	},
231 };
232 
233 static const struct rte_flow_ops mlx5_flow_ops = {
234 	.validate = mlx5_flow_validate,
235 	.create = mlx5_flow_create,
236 	.destroy = mlx5_flow_destroy,
237 	.flush = mlx5_flow_flush,
238 	.isolate = mlx5_flow_isolate,
239 	.query = mlx5_flow_query,
240 	.dev_dump = mlx5_flow_dev_dump,
241 	.get_aged_flows = mlx5_flow_get_aged_flows,
242 };
243 
244 /* Convert FDIR request to Generic flow. */
245 struct mlx5_fdir {
246 	struct rte_flow_attr attr;
247 	struct rte_flow_item items[4];
248 	struct rte_flow_item_eth l2;
249 	struct rte_flow_item_eth l2_mask;
250 	union {
251 		struct rte_flow_item_ipv4 ipv4;
252 		struct rte_flow_item_ipv6 ipv6;
253 	} l3;
254 	union {
255 		struct rte_flow_item_ipv4 ipv4;
256 		struct rte_flow_item_ipv6 ipv6;
257 	} l3_mask;
258 	union {
259 		struct rte_flow_item_udp udp;
260 		struct rte_flow_item_tcp tcp;
261 	} l4;
262 	union {
263 		struct rte_flow_item_udp udp;
264 		struct rte_flow_item_tcp tcp;
265 	} l4_mask;
266 	struct rte_flow_action actions[2];
267 	struct rte_flow_action_queue queue;
268 };
269 
270 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
271 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
272 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
273 };
274 
275 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
276 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
277 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
278 	{ 9, 10, 11 }, { 12, 13, 14 },
279 };
280 
281 /* Tunnel information. */
282 struct mlx5_flow_tunnel_info {
283 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
284 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
285 };
286 
287 static struct mlx5_flow_tunnel_info tunnels_info[] = {
288 	{
289 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
290 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
291 	},
292 	{
293 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
294 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
295 	},
296 	{
297 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
298 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
299 	},
300 	{
301 		.tunnel = MLX5_FLOW_LAYER_GRE,
302 		.ptype = RTE_PTYPE_TUNNEL_GRE,
303 	},
304 	{
305 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
306 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
307 	},
308 	{
309 		.tunnel = MLX5_FLOW_LAYER_MPLS,
310 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
311 	},
312 	{
313 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
314 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
315 	},
316 	{
317 		.tunnel = MLX5_FLOW_LAYER_IPIP,
318 		.ptype = RTE_PTYPE_TUNNEL_IP,
319 	},
320 	{
321 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
322 		.ptype = RTE_PTYPE_TUNNEL_IP,
323 	},
324 	{
325 		.tunnel = MLX5_FLOW_LAYER_GTP,
326 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
327 	},
328 };
329 
330 /**
331  * Translate tag ID to register.
332  *
333  * @param[in] dev
334  *   Pointer to the Ethernet device structure.
335  * @param[in] feature
336  *   The feature that request the register.
337  * @param[in] id
338  *   The request register ID.
339  * @param[out] error
340  *   Error description in case of any.
341  *
342  * @return
343  *   The request register on success, a negative errno
344  *   value otherwise and rte_errno is set.
345  */
346 int
347 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
348 		     enum mlx5_feature_name feature,
349 		     uint32_t id,
350 		     struct rte_flow_error *error)
351 {
352 	struct mlx5_priv *priv = dev->data->dev_private;
353 	struct mlx5_dev_config *config = &priv->config;
354 	enum modify_reg start_reg;
355 	bool skip_mtr_reg = false;
356 
357 	switch (feature) {
358 	case MLX5_HAIRPIN_RX:
359 		return REG_B;
360 	case MLX5_HAIRPIN_TX:
361 		return REG_A;
362 	case MLX5_METADATA_RX:
363 		switch (config->dv_xmeta_en) {
364 		case MLX5_XMETA_MODE_LEGACY:
365 			return REG_B;
366 		case MLX5_XMETA_MODE_META16:
367 			return REG_C_0;
368 		case MLX5_XMETA_MODE_META32:
369 			return REG_C_1;
370 		}
371 		break;
372 	case MLX5_METADATA_TX:
373 		return REG_A;
374 	case MLX5_METADATA_FDB:
375 		switch (config->dv_xmeta_en) {
376 		case MLX5_XMETA_MODE_LEGACY:
377 			return REG_NONE;
378 		case MLX5_XMETA_MODE_META16:
379 			return REG_C_0;
380 		case MLX5_XMETA_MODE_META32:
381 			return REG_C_1;
382 		}
383 		break;
384 	case MLX5_FLOW_MARK:
385 		switch (config->dv_xmeta_en) {
386 		case MLX5_XMETA_MODE_LEGACY:
387 			return REG_NONE;
388 		case MLX5_XMETA_MODE_META16:
389 			return REG_C_1;
390 		case MLX5_XMETA_MODE_META32:
391 			return REG_C_0;
392 		}
393 		break;
394 	case MLX5_MTR_SFX:
395 		/*
396 		 * If meter color and flow match share one register, flow match
397 		 * should use the meter color register for match.
398 		 */
399 		if (priv->mtr_reg_share)
400 			return priv->mtr_color_reg;
401 		else
402 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
403 			       REG_C_3;
404 	case MLX5_MTR_COLOR:
405 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
406 		return priv->mtr_color_reg;
407 	case MLX5_COPY_MARK:
408 		/*
409 		 * Metadata COPY_MARK register using is in meter suffix sub
410 		 * flow while with meter. It's safe to share the same register.
411 		 */
412 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
413 	case MLX5_APP_TAG:
414 		/*
415 		 * If meter is enable, it will engage the register for color
416 		 * match and flow match. If meter color match is not using the
417 		 * REG_C_2, need to skip the REG_C_x be used by meter color
418 		 * match.
419 		 * If meter is disable, free to use all available registers.
420 		 */
421 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
422 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
423 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
424 		if (id > (REG_C_7 - start_reg))
425 			return rte_flow_error_set(error, EINVAL,
426 						  RTE_FLOW_ERROR_TYPE_ITEM,
427 						  NULL, "invalid tag id");
428 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
429 			return rte_flow_error_set(error, ENOTSUP,
430 						  RTE_FLOW_ERROR_TYPE_ITEM,
431 						  NULL, "unsupported tag id");
432 		/*
433 		 * This case means meter is using the REG_C_x great than 2.
434 		 * Take care not to conflict with meter color REG_C_x.
435 		 * If the available index REG_C_y >= REG_C_x, skip the
436 		 * color register.
437 		 */
438 		if (skip_mtr_reg && config->flow_mreg_c
439 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
440 			if (id >= (REG_C_7 - start_reg))
441 				return rte_flow_error_set(error, EINVAL,
442 						       RTE_FLOW_ERROR_TYPE_ITEM,
443 							NULL, "invalid tag id");
444 			if (config->flow_mreg_c
445 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
446 				return config->flow_mreg_c
447 					       [id + 1 + start_reg - REG_C_0];
448 			return rte_flow_error_set(error, ENOTSUP,
449 						  RTE_FLOW_ERROR_TYPE_ITEM,
450 						  NULL, "unsupported tag id");
451 		}
452 		return config->flow_mreg_c[id + start_reg - REG_C_0];
453 	}
454 	MLX5_ASSERT(false);
455 	return rte_flow_error_set(error, EINVAL,
456 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
457 				  NULL, "invalid feature name");
458 }
459 
460 /**
461  * Check extensive flow metadata register support.
462  *
463  * @param dev
464  *   Pointer to rte_eth_dev structure.
465  *
466  * @return
467  *   True if device supports extensive flow metadata register, otherwise false.
468  */
469 bool
470 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
471 {
472 	struct mlx5_priv *priv = dev->data->dev_private;
473 	struct mlx5_dev_config *config = &priv->config;
474 
475 	/*
476 	 * Having available reg_c can be regarded inclusively as supporting
477 	 * extensive flow metadata register, which could mean,
478 	 * - metadata register copy action by modify header.
479 	 * - 16 modify header actions is supported.
480 	 * - reg_c's are preserved across different domain (FDB and NIC) on
481 	 *   packet loopback by flow lookup miss.
482 	 */
483 	return config->flow_mreg_c[2] != REG_NONE;
484 }
485 
486 /**
487  * Discover the maximum number of priority available.
488  *
489  * @param[in] dev
490  *   Pointer to the Ethernet device structure.
491  *
492  * @return
493  *   number of supported flow priority on success, a negative errno
494  *   value otherwise and rte_errno is set.
495  */
496 int
497 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
498 {
499 	struct mlx5_priv *priv = dev->data->dev_private;
500 	struct {
501 		struct ibv_flow_attr attr;
502 		struct ibv_flow_spec_eth eth;
503 		struct ibv_flow_spec_action_drop drop;
504 	} flow_attr = {
505 		.attr = {
506 			.num_of_specs = 2,
507 			.port = (uint8_t)priv->dev_port,
508 		},
509 		.eth = {
510 			.type = IBV_FLOW_SPEC_ETH,
511 			.size = sizeof(struct ibv_flow_spec_eth),
512 		},
513 		.drop = {
514 			.size = sizeof(struct ibv_flow_spec_action_drop),
515 			.type = IBV_FLOW_SPEC_ACTION_DROP,
516 		},
517 	};
518 	struct ibv_flow *flow;
519 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
520 	uint16_t vprio[] = { 8, 16 };
521 	int i;
522 	int priority = 0;
523 
524 	if (!drop) {
525 		rte_errno = ENOTSUP;
526 		return -rte_errno;
527 	}
528 	for (i = 0; i != RTE_DIM(vprio); i++) {
529 		flow_attr.attr.priority = vprio[i] - 1;
530 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
531 		if (!flow)
532 			break;
533 		claim_zero(mlx5_glue->destroy_flow(flow));
534 		priority = vprio[i];
535 	}
536 	mlx5_hrxq_drop_release(dev);
537 	switch (priority) {
538 	case 8:
539 		priority = RTE_DIM(priority_map_3);
540 		break;
541 	case 16:
542 		priority = RTE_DIM(priority_map_5);
543 		break;
544 	default:
545 		rte_errno = ENOTSUP;
546 		DRV_LOG(ERR,
547 			"port %u verbs maximum priority: %d expected 8/16",
548 			dev->data->port_id, priority);
549 		return -rte_errno;
550 	}
551 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
552 		dev->data->port_id, priority);
553 	return priority;
554 }
555 
556 /**
557  * Adjust flow priority based on the highest layer and the request priority.
558  *
559  * @param[in] dev
560  *   Pointer to the Ethernet device structure.
561  * @param[in] priority
562  *   The rule base priority.
563  * @param[in] subpriority
564  *   The priority based on the items.
565  *
566  * @return
567  *   The new priority.
568  */
569 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
570 				   uint32_t subpriority)
571 {
572 	uint32_t res = 0;
573 	struct mlx5_priv *priv = dev->data->dev_private;
574 
575 	switch (priv->config.flow_prio) {
576 	case RTE_DIM(priority_map_3):
577 		res = priority_map_3[priority][subpriority];
578 		break;
579 	case RTE_DIM(priority_map_5):
580 		res = priority_map_5[priority][subpriority];
581 		break;
582 	}
583 	return  res;
584 }
585 
586 /**
587  * Verify the @p item specifications (spec, last, mask) are compatible with the
588  * NIC capabilities.
589  *
590  * @param[in] item
591  *   Item specification.
592  * @param[in] mask
593  *   @p item->mask or flow default bit-masks.
594  * @param[in] nic_mask
595  *   Bit-masks covering supported fields by the NIC to compare with user mask.
596  * @param[in] size
597  *   Bit-masks size in bytes.
598  * @param[out] error
599  *   Pointer to error structure.
600  *
601  * @return
602  *   0 on success, a negative errno value otherwise and rte_errno is set.
603  */
604 int
605 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
606 			  const uint8_t *mask,
607 			  const uint8_t *nic_mask,
608 			  unsigned int size,
609 			  struct rte_flow_error *error)
610 {
611 	unsigned int i;
612 
613 	MLX5_ASSERT(nic_mask);
614 	for (i = 0; i < size; ++i)
615 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
616 			return rte_flow_error_set(error, ENOTSUP,
617 						  RTE_FLOW_ERROR_TYPE_ITEM,
618 						  item,
619 						  "mask enables non supported"
620 						  " bits");
621 	if (!item->spec && (item->mask || item->last))
622 		return rte_flow_error_set(error, EINVAL,
623 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
624 					  "mask/last without a spec is not"
625 					  " supported");
626 	if (item->spec && item->last) {
627 		uint8_t spec[size];
628 		uint8_t last[size];
629 		unsigned int i;
630 		int ret;
631 
632 		for (i = 0; i < size; ++i) {
633 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
634 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
635 		}
636 		ret = memcmp(spec, last, size);
637 		if (ret != 0)
638 			return rte_flow_error_set(error, EINVAL,
639 						  RTE_FLOW_ERROR_TYPE_ITEM,
640 						  item,
641 						  "range is not valid");
642 	}
643 	return 0;
644 }
645 
646 /**
647  * Adjust the hash fields according to the @p flow information.
648  *
649  * @param[in] dev_flow.
650  *   Pointer to the mlx5_flow.
651  * @param[in] tunnel
652  *   1 when the hash field is for a tunnel item.
653  * @param[in] layer_types
654  *   ETH_RSS_* types.
655  * @param[in] hash_fields
656  *   Item hash fields.
657  *
658  * @return
659  *   The hash fields that should be used.
660  */
661 uint64_t
662 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
663 			    int tunnel __rte_unused, uint64_t layer_types,
664 			    uint64_t hash_fields)
665 {
666 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
667 	int rss_request_inner = rss_desc->level >= 2;
668 
669 	/* Check RSS hash level for tunnel. */
670 	if (tunnel && rss_request_inner)
671 		hash_fields |= IBV_RX_HASH_INNER;
672 	else if (tunnel || rss_request_inner)
673 		return 0;
674 #endif
675 	/* Check if requested layer matches RSS hash fields. */
676 	if (!(rss_desc->types & layer_types))
677 		return 0;
678 	return hash_fields;
679 }
680 
681 /**
682  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
683  * if several tunnel rules are used on this queue, the tunnel ptype will be
684  * cleared.
685  *
686  * @param rxq_ctrl
687  *   Rx queue to update.
688  */
689 static void
690 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
691 {
692 	unsigned int i;
693 	uint32_t tunnel_ptype = 0;
694 
695 	/* Look up for the ptype to use. */
696 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
697 		if (!rxq_ctrl->flow_tunnels_n[i])
698 			continue;
699 		if (!tunnel_ptype) {
700 			tunnel_ptype = tunnels_info[i].ptype;
701 		} else {
702 			tunnel_ptype = 0;
703 			break;
704 		}
705 	}
706 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
707 }
708 
709 /**
710  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
711  * flow.
712  *
713  * @param[in] dev
714  *   Pointer to the Ethernet device structure.
715  * @param[in] dev_handle
716  *   Pointer to device flow handle structure.
717  */
718 static void
719 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
720 		       struct mlx5_flow_handle *dev_handle)
721 {
722 	struct mlx5_priv *priv = dev->data->dev_private;
723 	const int mark = dev_handle->mark;
724 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
725 	struct mlx5_hrxq *hrxq;
726 	unsigned int i;
727 
728 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
729 		return;
730 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
731 			      dev_handle->rix_hrxq);
732 	if (!hrxq)
733 		return;
734 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
735 		int idx = hrxq->ind_table->queues[i];
736 		struct mlx5_rxq_ctrl *rxq_ctrl =
737 			container_of((*priv->rxqs)[idx],
738 				     struct mlx5_rxq_ctrl, rxq);
739 
740 		/*
741 		 * To support metadata register copy on Tx loopback,
742 		 * this must be always enabled (metadata may arive
743 		 * from other port - not from local flows only.
744 		 */
745 		if (priv->config.dv_flow_en &&
746 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
747 		    mlx5_flow_ext_mreg_supported(dev)) {
748 			rxq_ctrl->rxq.mark = 1;
749 			rxq_ctrl->flow_mark_n = 1;
750 		} else if (mark) {
751 			rxq_ctrl->rxq.mark = 1;
752 			rxq_ctrl->flow_mark_n++;
753 		}
754 		if (tunnel) {
755 			unsigned int j;
756 
757 			/* Increase the counter matching the flow. */
758 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
759 				if ((tunnels_info[j].tunnel &
760 				     dev_handle->layers) ==
761 				    tunnels_info[j].tunnel) {
762 					rxq_ctrl->flow_tunnels_n[j]++;
763 					break;
764 				}
765 			}
766 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
767 		}
768 	}
769 }
770 
771 /**
772  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
773  *
774  * @param[in] dev
775  *   Pointer to the Ethernet device structure.
776  * @param[in] flow
777  *   Pointer to flow structure.
778  */
779 static void
780 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
781 {
782 	struct mlx5_priv *priv = dev->data->dev_private;
783 	uint32_t handle_idx;
784 	struct mlx5_flow_handle *dev_handle;
785 
786 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
787 		       handle_idx, dev_handle, next)
788 		flow_drv_rxq_flags_set(dev, dev_handle);
789 }
790 
791 /**
792  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
793  * device flow if no other flow uses it with the same kind of request.
794  *
795  * @param dev
796  *   Pointer to Ethernet device.
797  * @param[in] dev_handle
798  *   Pointer to the device flow handle structure.
799  */
800 static void
801 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
802 			struct mlx5_flow_handle *dev_handle)
803 {
804 	struct mlx5_priv *priv = dev->data->dev_private;
805 	const int mark = dev_handle->mark;
806 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
807 	struct mlx5_hrxq *hrxq;
808 	unsigned int i;
809 
810 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
811 		return;
812 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
813 			      dev_handle->rix_hrxq);
814 	if (!hrxq)
815 		return;
816 	MLX5_ASSERT(dev->data->dev_started);
817 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
818 		int idx = hrxq->ind_table->queues[i];
819 		struct mlx5_rxq_ctrl *rxq_ctrl =
820 			container_of((*priv->rxqs)[idx],
821 				     struct mlx5_rxq_ctrl, rxq);
822 
823 		if (priv->config.dv_flow_en &&
824 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
825 		    mlx5_flow_ext_mreg_supported(dev)) {
826 			rxq_ctrl->rxq.mark = 1;
827 			rxq_ctrl->flow_mark_n = 1;
828 		} else if (mark) {
829 			rxq_ctrl->flow_mark_n--;
830 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
831 		}
832 		if (tunnel) {
833 			unsigned int j;
834 
835 			/* Decrease the counter matching the flow. */
836 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
837 				if ((tunnels_info[j].tunnel &
838 				     dev_handle->layers) ==
839 				    tunnels_info[j].tunnel) {
840 					rxq_ctrl->flow_tunnels_n[j]--;
841 					break;
842 				}
843 			}
844 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
845 		}
846 	}
847 }
848 
849 /**
850  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
851  * @p flow if no other flow uses it with the same kind of request.
852  *
853  * @param dev
854  *   Pointer to Ethernet device.
855  * @param[in] flow
856  *   Pointer to the flow.
857  */
858 static void
859 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
860 {
861 	struct mlx5_priv *priv = dev->data->dev_private;
862 	uint32_t handle_idx;
863 	struct mlx5_flow_handle *dev_handle;
864 
865 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
866 		       handle_idx, dev_handle, next)
867 		flow_drv_rxq_flags_trim(dev, dev_handle);
868 }
869 
870 /**
871  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
872  *
873  * @param dev
874  *   Pointer to Ethernet device.
875  */
876 static void
877 flow_rxq_flags_clear(struct rte_eth_dev *dev)
878 {
879 	struct mlx5_priv *priv = dev->data->dev_private;
880 	unsigned int i;
881 
882 	for (i = 0; i != priv->rxqs_n; ++i) {
883 		struct mlx5_rxq_ctrl *rxq_ctrl;
884 		unsigned int j;
885 
886 		if (!(*priv->rxqs)[i])
887 			continue;
888 		rxq_ctrl = container_of((*priv->rxqs)[i],
889 					struct mlx5_rxq_ctrl, rxq);
890 		rxq_ctrl->flow_mark_n = 0;
891 		rxq_ctrl->rxq.mark = 0;
892 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
893 			rxq_ctrl->flow_tunnels_n[j] = 0;
894 		rxq_ctrl->rxq.tunnel = 0;
895 	}
896 }
897 
898 /**
899  * Set the Rx queue dynamic metadata (mask and offset) for a flow
900  *
901  * @param[in] dev
902  *   Pointer to the Ethernet device structure.
903  */
904 void
905 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
906 {
907 	struct mlx5_priv *priv = dev->data->dev_private;
908 	struct mlx5_rxq_data *data;
909 	unsigned int i;
910 
911 	for (i = 0; i != priv->rxqs_n; ++i) {
912 		if (!(*priv->rxqs)[i])
913 			continue;
914 		data = (*priv->rxqs)[i];
915 		if (!rte_flow_dynf_metadata_avail()) {
916 			data->dynf_meta = 0;
917 			data->flow_meta_mask = 0;
918 			data->flow_meta_offset = -1;
919 		} else {
920 			data->dynf_meta = 1;
921 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
922 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
923 		}
924 	}
925 }
926 
927 /*
928  * return a pointer to the desired action in the list of actions.
929  *
930  * @param[in] actions
931  *   The list of actions to search the action in.
932  * @param[in] action
933  *   The action to find.
934  *
935  * @return
936  *   Pointer to the action in the list, if found. NULL otherwise.
937  */
938 const struct rte_flow_action *
939 mlx5_flow_find_action(const struct rte_flow_action *actions,
940 		      enum rte_flow_action_type action)
941 {
942 	if (actions == NULL)
943 		return NULL;
944 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
945 		if (actions->type == action)
946 			return actions;
947 	return NULL;
948 }
949 
950 /*
951  * Validate the flag action.
952  *
953  * @param[in] action_flags
954  *   Bit-fields that holds the actions detected until now.
955  * @param[in] attr
956  *   Attributes of flow that includes this action.
957  * @param[out] error
958  *   Pointer to error structure.
959  *
960  * @return
961  *   0 on success, a negative errno value otherwise and rte_errno is set.
962  */
963 int
964 mlx5_flow_validate_action_flag(uint64_t action_flags,
965 			       const struct rte_flow_attr *attr,
966 			       struct rte_flow_error *error)
967 {
968 	if (action_flags & MLX5_FLOW_ACTION_MARK)
969 		return rte_flow_error_set(error, EINVAL,
970 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
971 					  "can't mark and flag in same flow");
972 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
973 		return rte_flow_error_set(error, EINVAL,
974 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
975 					  "can't have 2 flag"
976 					  " actions in same flow");
977 	if (attr->egress)
978 		return rte_flow_error_set(error, ENOTSUP,
979 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
980 					  "flag action not supported for "
981 					  "egress");
982 	return 0;
983 }
984 
985 /*
986  * Validate the mark action.
987  *
988  * @param[in] action
989  *   Pointer to the queue action.
990  * @param[in] action_flags
991  *   Bit-fields that holds the actions detected until now.
992  * @param[in] attr
993  *   Attributes of flow that includes this action.
994  * @param[out] error
995  *   Pointer to error structure.
996  *
997  * @return
998  *   0 on success, a negative errno value otherwise and rte_errno is set.
999  */
1000 int
1001 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1002 			       uint64_t action_flags,
1003 			       const struct rte_flow_attr *attr,
1004 			       struct rte_flow_error *error)
1005 {
1006 	const struct rte_flow_action_mark *mark = action->conf;
1007 
1008 	if (!mark)
1009 		return rte_flow_error_set(error, EINVAL,
1010 					  RTE_FLOW_ERROR_TYPE_ACTION,
1011 					  action,
1012 					  "configuration cannot be null");
1013 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1014 		return rte_flow_error_set(error, EINVAL,
1015 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1016 					  &mark->id,
1017 					  "mark id must in 0 <= id < "
1018 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1019 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1020 		return rte_flow_error_set(error, EINVAL,
1021 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1022 					  "can't flag and mark in same flow");
1023 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1024 		return rte_flow_error_set(error, EINVAL,
1025 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1026 					  "can't have 2 mark actions in same"
1027 					  " flow");
1028 	if (attr->egress)
1029 		return rte_flow_error_set(error, ENOTSUP,
1030 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1031 					  "mark action not supported for "
1032 					  "egress");
1033 	return 0;
1034 }
1035 
1036 /*
1037  * Validate the drop action.
1038  *
1039  * @param[in] action_flags
1040  *   Bit-fields that holds the actions detected until now.
1041  * @param[in] attr
1042  *   Attributes of flow that includes this action.
1043  * @param[out] error
1044  *   Pointer to error structure.
1045  *
1046  * @return
1047  *   0 on success, a negative errno value otherwise and rte_errno is set.
1048  */
1049 int
1050 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1051 			       const struct rte_flow_attr *attr,
1052 			       struct rte_flow_error *error)
1053 {
1054 	if (attr->egress)
1055 		return rte_flow_error_set(error, ENOTSUP,
1056 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1057 					  "drop action not supported for "
1058 					  "egress");
1059 	return 0;
1060 }
1061 
1062 /*
1063  * Validate the queue action.
1064  *
1065  * @param[in] action
1066  *   Pointer to the queue action.
1067  * @param[in] action_flags
1068  *   Bit-fields that holds the actions detected until now.
1069  * @param[in] dev
1070  *   Pointer to the Ethernet device structure.
1071  * @param[in] attr
1072  *   Attributes of flow that includes this action.
1073  * @param[out] error
1074  *   Pointer to error structure.
1075  *
1076  * @return
1077  *   0 on success, a negative errno value otherwise and rte_errno is set.
1078  */
1079 int
1080 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1081 				uint64_t action_flags,
1082 				struct rte_eth_dev *dev,
1083 				const struct rte_flow_attr *attr,
1084 				struct rte_flow_error *error)
1085 {
1086 	struct mlx5_priv *priv = dev->data->dev_private;
1087 	const struct rte_flow_action_queue *queue = action->conf;
1088 
1089 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1090 		return rte_flow_error_set(error, EINVAL,
1091 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1092 					  "can't have 2 fate actions in"
1093 					  " same flow");
1094 	if (!priv->rxqs_n)
1095 		return rte_flow_error_set(error, EINVAL,
1096 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1097 					  NULL, "No Rx queues configured");
1098 	if (queue->index >= priv->rxqs_n)
1099 		return rte_flow_error_set(error, EINVAL,
1100 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1101 					  &queue->index,
1102 					  "queue index out of range");
1103 	if (!(*priv->rxqs)[queue->index])
1104 		return rte_flow_error_set(error, EINVAL,
1105 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1106 					  &queue->index,
1107 					  "queue is not configured");
1108 	if (attr->egress)
1109 		return rte_flow_error_set(error, ENOTSUP,
1110 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1111 					  "queue action not supported for "
1112 					  "egress");
1113 	return 0;
1114 }
1115 
1116 /*
1117  * Validate the rss action.
1118  *
1119  * @param[in] action
1120  *   Pointer to the queue action.
1121  * @param[in] action_flags
1122  *   Bit-fields that holds the actions detected until now.
1123  * @param[in] dev
1124  *   Pointer to the Ethernet device structure.
1125  * @param[in] attr
1126  *   Attributes of flow that includes this action.
1127  * @param[in] item_flags
1128  *   Items that were detected.
1129  * @param[out] error
1130  *   Pointer to error structure.
1131  *
1132  * @return
1133  *   0 on success, a negative errno value otherwise and rte_errno is set.
1134  */
1135 int
1136 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1137 			      uint64_t action_flags,
1138 			      struct rte_eth_dev *dev,
1139 			      const struct rte_flow_attr *attr,
1140 			      uint64_t item_flags,
1141 			      struct rte_flow_error *error)
1142 {
1143 	struct mlx5_priv *priv = dev->data->dev_private;
1144 	const struct rte_flow_action_rss *rss = action->conf;
1145 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1146 	unsigned int i;
1147 
1148 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1149 		return rte_flow_error_set(error, EINVAL,
1150 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1151 					  "can't have 2 fate actions"
1152 					  " in same flow");
1153 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1154 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1155 		return rte_flow_error_set(error, ENOTSUP,
1156 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1157 					  &rss->func,
1158 					  "RSS hash function not supported");
1159 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1160 	if (rss->level > 2)
1161 #else
1162 	if (rss->level > 1)
1163 #endif
1164 		return rte_flow_error_set(error, ENOTSUP,
1165 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1166 					  &rss->level,
1167 					  "tunnel RSS is not supported");
1168 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1169 	if (rss->key_len == 0 && rss->key != NULL)
1170 		return rte_flow_error_set(error, ENOTSUP,
1171 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1172 					  &rss->key_len,
1173 					  "RSS hash key length 0");
1174 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1175 		return rte_flow_error_set(error, ENOTSUP,
1176 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177 					  &rss->key_len,
1178 					  "RSS hash key too small");
1179 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1180 		return rte_flow_error_set(error, ENOTSUP,
1181 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1182 					  &rss->key_len,
1183 					  "RSS hash key too large");
1184 	if (rss->queue_num > priv->config.ind_table_max_size)
1185 		return rte_flow_error_set(error, ENOTSUP,
1186 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1187 					  &rss->queue_num,
1188 					  "number of queues too large");
1189 	if (rss->types & MLX5_RSS_HF_MASK)
1190 		return rte_flow_error_set(error, ENOTSUP,
1191 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1192 					  &rss->types,
1193 					  "some RSS protocols are not"
1194 					  " supported");
1195 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1196 	    !(rss->types & ETH_RSS_IP))
1197 		return rte_flow_error_set(error, EINVAL,
1198 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1199 					  "L3 partial RSS requested but L3 RSS"
1200 					  " type not specified");
1201 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1202 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1203 		return rte_flow_error_set(error, EINVAL,
1204 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1205 					  "L4 partial RSS requested but L4 RSS"
1206 					  " type not specified");
1207 	if (!priv->rxqs_n)
1208 		return rte_flow_error_set(error, EINVAL,
1209 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1210 					  NULL, "No Rx queues configured");
1211 	if (!rss->queue_num)
1212 		return rte_flow_error_set(error, EINVAL,
1213 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1214 					  NULL, "No queues configured");
1215 	for (i = 0; i != rss->queue_num; ++i) {
1216 		if (rss->queue[i] >= priv->rxqs_n)
1217 			return rte_flow_error_set
1218 				(error, EINVAL,
1219 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1220 				 &rss->queue[i], "queue index out of range");
1221 		if (!(*priv->rxqs)[rss->queue[i]])
1222 			return rte_flow_error_set
1223 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1224 				 &rss->queue[i], "queue is not configured");
1225 	}
1226 	if (attr->egress)
1227 		return rte_flow_error_set(error, ENOTSUP,
1228 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1229 					  "rss action not supported for "
1230 					  "egress");
1231 	if (rss->level > 1 && !tunnel)
1232 		return rte_flow_error_set(error, EINVAL,
1233 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1234 					  "inner RSS is not supported for "
1235 					  "non-tunnel flows");
1236 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1237 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1238 		return rte_flow_error_set(error, EINVAL,
1239 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1240 					  "RSS on eCPRI is not supported now");
1241 	}
1242 	return 0;
1243 }
1244 
1245 /*
1246  * Validate the default miss action.
1247  *
1248  * @param[in] action_flags
1249  *   Bit-fields that holds the actions detected until now.
1250  * @param[out] error
1251  *   Pointer to error structure.
1252  *
1253  * @return
1254  *   0 on success, a negative errno value otherwise and rte_errno is set.
1255  */
1256 int
1257 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1258 				const struct rte_flow_attr *attr,
1259 				struct rte_flow_error *error)
1260 {
1261 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1262 		return rte_flow_error_set(error, EINVAL,
1263 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1264 					  "can't have 2 fate actions in"
1265 					  " same flow");
1266 	if (attr->egress)
1267 		return rte_flow_error_set(error, ENOTSUP,
1268 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1269 					  "default miss action not supported "
1270 					  "for egress");
1271 	if (attr->group)
1272 		return rte_flow_error_set(error, ENOTSUP,
1273 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1274 					  "only group 0 is supported");
1275 	if (attr->transfer)
1276 		return rte_flow_error_set(error, ENOTSUP,
1277 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1278 					  NULL, "transfer is not supported");
1279 	return 0;
1280 }
1281 
1282 /*
1283  * Validate the count action.
1284  *
1285  * @param[in] dev
1286  *   Pointer to the Ethernet device structure.
1287  * @param[in] attr
1288  *   Attributes of flow that includes this action.
1289  * @param[out] error
1290  *   Pointer to error structure.
1291  *
1292  * @return
1293  *   0 on success, a negative errno value otherwise and rte_errno is set.
1294  */
1295 int
1296 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1297 				const struct rte_flow_attr *attr,
1298 				struct rte_flow_error *error)
1299 {
1300 	if (attr->egress)
1301 		return rte_flow_error_set(error, ENOTSUP,
1302 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1303 					  "count action not supported for "
1304 					  "egress");
1305 	return 0;
1306 }
1307 
1308 /**
1309  * Verify the @p attributes will be correctly understood by the NIC and store
1310  * them in the @p flow if everything is correct.
1311  *
1312  * @param[in] dev
1313  *   Pointer to the Ethernet device structure.
1314  * @param[in] attributes
1315  *   Pointer to flow attributes
1316  * @param[out] error
1317  *   Pointer to error structure.
1318  *
1319  * @return
1320  *   0 on success, a negative errno value otherwise and rte_errno is set.
1321  */
1322 int
1323 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1324 			      const struct rte_flow_attr *attributes,
1325 			      struct rte_flow_error *error)
1326 {
1327 	struct mlx5_priv *priv = dev->data->dev_private;
1328 	uint32_t priority_max = priv->config.flow_prio - 1;
1329 
1330 	if (attributes->group)
1331 		return rte_flow_error_set(error, ENOTSUP,
1332 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1333 					  NULL, "groups is not supported");
1334 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1335 	    attributes->priority >= priority_max)
1336 		return rte_flow_error_set(error, ENOTSUP,
1337 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1338 					  NULL, "priority out of range");
1339 	if (attributes->egress)
1340 		return rte_flow_error_set(error, ENOTSUP,
1341 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1342 					  "egress is not supported");
1343 	if (attributes->transfer && !priv->config.dv_esw_en)
1344 		return rte_flow_error_set(error, ENOTSUP,
1345 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1346 					  NULL, "transfer is not supported");
1347 	if (!attributes->ingress)
1348 		return rte_flow_error_set(error, EINVAL,
1349 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1350 					  NULL,
1351 					  "ingress attribute is mandatory");
1352 	return 0;
1353 }
1354 
1355 /**
1356  * Validate ICMP6 item.
1357  *
1358  * @param[in] item
1359  *   Item specification.
1360  * @param[in] item_flags
1361  *   Bit-fields that holds the items detected until now.
1362  * @param[out] error
1363  *   Pointer to error structure.
1364  *
1365  * @return
1366  *   0 on success, a negative errno value otherwise and rte_errno is set.
1367  */
1368 int
1369 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1370 			       uint64_t item_flags,
1371 			       uint8_t target_protocol,
1372 			       struct rte_flow_error *error)
1373 {
1374 	const struct rte_flow_item_icmp6 *mask = item->mask;
1375 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1376 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1377 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1378 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1379 				      MLX5_FLOW_LAYER_OUTER_L4;
1380 	int ret;
1381 
1382 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1383 		return rte_flow_error_set(error, EINVAL,
1384 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1385 					  "protocol filtering not compatible"
1386 					  " with ICMP6 layer");
1387 	if (!(item_flags & l3m))
1388 		return rte_flow_error_set(error, EINVAL,
1389 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 					  "IPv6 is mandatory to filter on"
1391 					  " ICMP6");
1392 	if (item_flags & l4m)
1393 		return rte_flow_error_set(error, EINVAL,
1394 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1395 					  "multiple L4 layers not supported");
1396 	if (!mask)
1397 		mask = &rte_flow_item_icmp6_mask;
1398 	ret = mlx5_flow_item_acceptable
1399 		(item, (const uint8_t *)mask,
1400 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1401 		 sizeof(struct rte_flow_item_icmp6), error);
1402 	if (ret < 0)
1403 		return ret;
1404 	return 0;
1405 }
1406 
1407 /**
1408  * Validate ICMP item.
1409  *
1410  * @param[in] item
1411  *   Item specification.
1412  * @param[in] item_flags
1413  *   Bit-fields that holds the items detected until now.
1414  * @param[out] error
1415  *   Pointer to error structure.
1416  *
1417  * @return
1418  *   0 on success, a negative errno value otherwise and rte_errno is set.
1419  */
1420 int
1421 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1422 			     uint64_t item_flags,
1423 			     uint8_t target_protocol,
1424 			     struct rte_flow_error *error)
1425 {
1426 	const struct rte_flow_item_icmp *mask = item->mask;
1427 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1428 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1429 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1430 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1431 				      MLX5_FLOW_LAYER_OUTER_L4;
1432 	int ret;
1433 
1434 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1435 		return rte_flow_error_set(error, EINVAL,
1436 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1437 					  "protocol filtering not compatible"
1438 					  " with ICMP layer");
1439 	if (!(item_flags & l3m))
1440 		return rte_flow_error_set(error, EINVAL,
1441 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1442 					  "IPv4 is mandatory to filter"
1443 					  " on ICMP");
1444 	if (item_flags & l4m)
1445 		return rte_flow_error_set(error, EINVAL,
1446 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1447 					  "multiple L4 layers not supported");
1448 	if (!mask)
1449 		mask = &rte_flow_item_icmp_mask;
1450 	ret = mlx5_flow_item_acceptable
1451 		(item, (const uint8_t *)mask,
1452 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1453 		 sizeof(struct rte_flow_item_icmp), error);
1454 	if (ret < 0)
1455 		return ret;
1456 	return 0;
1457 }
1458 
1459 /**
1460  * Validate Ethernet item.
1461  *
1462  * @param[in] item
1463  *   Item specification.
1464  * @param[in] item_flags
1465  *   Bit-fields that holds the items detected until now.
1466  * @param[out] error
1467  *   Pointer to error structure.
1468  *
1469  * @return
1470  *   0 on success, a negative errno value otherwise and rte_errno is set.
1471  */
1472 int
1473 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1474 			    uint64_t item_flags,
1475 			    struct rte_flow_error *error)
1476 {
1477 	const struct rte_flow_item_eth *mask = item->mask;
1478 	const struct rte_flow_item_eth nic_mask = {
1479 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1480 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1481 		.type = RTE_BE16(0xffff),
1482 	};
1483 	int ret;
1484 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1485 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1486 				       MLX5_FLOW_LAYER_OUTER_L2;
1487 
1488 	if (item_flags & ethm)
1489 		return rte_flow_error_set(error, ENOTSUP,
1490 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1491 					  "multiple L2 layers not supported");
1492 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1493 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1494 		return rte_flow_error_set(error, EINVAL,
1495 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1496 					  "L2 layer should not follow "
1497 					  "L3 layers");
1498 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1499 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1502 					  "L2 layer should not follow VLAN");
1503 	if (!mask)
1504 		mask = &rte_flow_item_eth_mask;
1505 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1506 					(const uint8_t *)&nic_mask,
1507 					sizeof(struct rte_flow_item_eth),
1508 					error);
1509 	return ret;
1510 }
1511 
1512 /**
1513  * Validate VLAN item.
1514  *
1515  * @param[in] item
1516  *   Item specification.
1517  * @param[in] item_flags
1518  *   Bit-fields that holds the items detected until now.
1519  * @param[in] dev
1520  *   Ethernet device flow is being created on.
1521  * @param[out] error
1522  *   Pointer to error structure.
1523  *
1524  * @return
1525  *   0 on success, a negative errno value otherwise and rte_errno is set.
1526  */
1527 int
1528 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1529 			     uint64_t item_flags,
1530 			     struct rte_eth_dev *dev,
1531 			     struct rte_flow_error *error)
1532 {
1533 	const struct rte_flow_item_vlan *spec = item->spec;
1534 	const struct rte_flow_item_vlan *mask = item->mask;
1535 	const struct rte_flow_item_vlan nic_mask = {
1536 		.tci = RTE_BE16(UINT16_MAX),
1537 		.inner_type = RTE_BE16(UINT16_MAX),
1538 	};
1539 	uint16_t vlan_tag = 0;
1540 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1541 	int ret;
1542 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1543 					MLX5_FLOW_LAYER_INNER_L4) :
1544 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1545 					MLX5_FLOW_LAYER_OUTER_L4);
1546 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1547 					MLX5_FLOW_LAYER_OUTER_VLAN;
1548 
1549 	if (item_flags & vlanm)
1550 		return rte_flow_error_set(error, EINVAL,
1551 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1552 					  "multiple VLAN layers not supported");
1553 	else if ((item_flags & l34m) != 0)
1554 		return rte_flow_error_set(error, EINVAL,
1555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1556 					  "VLAN cannot follow L3/L4 layer");
1557 	if (!mask)
1558 		mask = &rte_flow_item_vlan_mask;
1559 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1560 					(const uint8_t *)&nic_mask,
1561 					sizeof(struct rte_flow_item_vlan),
1562 					error);
1563 	if (ret)
1564 		return ret;
1565 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1566 		struct mlx5_priv *priv = dev->data->dev_private;
1567 
1568 		if (priv->vmwa_context) {
1569 			/*
1570 			 * Non-NULL context means we have a virtual machine
1571 			 * and SR-IOV enabled, we have to create VLAN interface
1572 			 * to make hypervisor to setup E-Switch vport
1573 			 * context correctly. We avoid creating the multiple
1574 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1575 			 */
1576 			return rte_flow_error_set(error, EINVAL,
1577 						  RTE_FLOW_ERROR_TYPE_ITEM,
1578 						  item,
1579 						  "VLAN tag mask is not"
1580 						  " supported in virtual"
1581 						  " environment");
1582 		}
1583 	}
1584 	if (spec) {
1585 		vlan_tag = spec->tci;
1586 		vlan_tag &= mask->tci;
1587 	}
1588 	/*
1589 	 * From verbs perspective an empty VLAN is equivalent
1590 	 * to a packet without VLAN layer.
1591 	 */
1592 	if (!vlan_tag)
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1595 					  item->spec,
1596 					  "VLAN cannot be empty");
1597 	return 0;
1598 }
1599 
1600 /**
1601  * Validate IPV4 item.
1602  *
1603  * @param[in] item
1604  *   Item specification.
1605  * @param[in] item_flags
1606  *   Bit-fields that holds the items detected until now.
1607  * @param[in] last_item
1608  *   Previous validated item in the pattern items.
1609  * @param[in] ether_type
1610  *   Type in the ethernet layer header (including dot1q).
1611  * @param[in] acc_mask
1612  *   Acceptable mask, if NULL default internal default mask
1613  *   will be used to check whether item fields are supported.
1614  * @param[out] error
1615  *   Pointer to error structure.
1616  *
1617  * @return
1618  *   0 on success, a negative errno value otherwise and rte_errno is set.
1619  */
1620 int
1621 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1622 			     uint64_t item_flags,
1623 			     uint64_t last_item,
1624 			     uint16_t ether_type,
1625 			     const struct rte_flow_item_ipv4 *acc_mask,
1626 			     struct rte_flow_error *error)
1627 {
1628 	const struct rte_flow_item_ipv4 *mask = item->mask;
1629 	const struct rte_flow_item_ipv4 *spec = item->spec;
1630 	const struct rte_flow_item_ipv4 nic_mask = {
1631 		.hdr = {
1632 			.src_addr = RTE_BE32(0xffffffff),
1633 			.dst_addr = RTE_BE32(0xffffffff),
1634 			.type_of_service = 0xff,
1635 			.next_proto_id = 0xff,
1636 		},
1637 	};
1638 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1639 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1640 				      MLX5_FLOW_LAYER_OUTER_L3;
1641 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1642 				      MLX5_FLOW_LAYER_OUTER_L4;
1643 	int ret;
1644 	uint8_t next_proto = 0xFF;
1645 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1646 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1647 				  MLX5_FLOW_LAYER_INNER_VLAN);
1648 
1649 	if ((last_item & l2_vlan) && ether_type &&
1650 	    ether_type != RTE_ETHER_TYPE_IPV4)
1651 		return rte_flow_error_set(error, EINVAL,
1652 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1653 					  "IPv4 cannot follow L2/VLAN layer "
1654 					  "which ether type is not IPv4");
1655 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1656 		if (mask && spec)
1657 			next_proto = mask->hdr.next_proto_id &
1658 				     spec->hdr.next_proto_id;
1659 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1660 			return rte_flow_error_set(error, EINVAL,
1661 						  RTE_FLOW_ERROR_TYPE_ITEM,
1662 						  item,
1663 						  "multiple tunnel "
1664 						  "not supported");
1665 	}
1666 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1667 		return rte_flow_error_set(error, EINVAL,
1668 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1669 					  "wrong tunnel type - IPv6 specified "
1670 					  "but IPv4 item provided");
1671 	if (item_flags & l3m)
1672 		return rte_flow_error_set(error, ENOTSUP,
1673 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1674 					  "multiple L3 layers not supported");
1675 	else if (item_flags & l4m)
1676 		return rte_flow_error_set(error, EINVAL,
1677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1678 					  "L3 cannot follow an L4 layer.");
1679 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1680 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 					  "L3 cannot follow an NVGRE layer.");
1684 	if (!mask)
1685 		mask = &rte_flow_item_ipv4_mask;
1686 	else if (mask->hdr.next_proto_id != 0 &&
1687 		 mask->hdr.next_proto_id != 0xff)
1688 		return rte_flow_error_set(error, EINVAL,
1689 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1690 					  "partial mask is not supported"
1691 					  " for protocol");
1692 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1693 					acc_mask ? (const uint8_t *)acc_mask
1694 						 : (const uint8_t *)&nic_mask,
1695 					sizeof(struct rte_flow_item_ipv4),
1696 					error);
1697 	if (ret < 0)
1698 		return ret;
1699 	return 0;
1700 }
1701 
1702 /**
1703  * Validate IPV6 item.
1704  *
1705  * @param[in] item
1706  *   Item specification.
1707  * @param[in] item_flags
1708  *   Bit-fields that holds the items detected until now.
1709  * @param[in] last_item
1710  *   Previous validated item in the pattern items.
1711  * @param[in] ether_type
1712  *   Type in the ethernet layer header (including dot1q).
1713  * @param[in] acc_mask
1714  *   Acceptable mask, if NULL default internal default mask
1715  *   will be used to check whether item fields are supported.
1716  * @param[out] error
1717  *   Pointer to error structure.
1718  *
1719  * @return
1720  *   0 on success, a negative errno value otherwise and rte_errno is set.
1721  */
1722 int
1723 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1724 			     uint64_t item_flags,
1725 			     uint64_t last_item,
1726 			     uint16_t ether_type,
1727 			     const struct rte_flow_item_ipv6 *acc_mask,
1728 			     struct rte_flow_error *error)
1729 {
1730 	const struct rte_flow_item_ipv6 *mask = item->mask;
1731 	const struct rte_flow_item_ipv6 *spec = item->spec;
1732 	const struct rte_flow_item_ipv6 nic_mask = {
1733 		.hdr = {
1734 			.src_addr =
1735 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1736 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1737 			.dst_addr =
1738 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1739 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1740 			.vtc_flow = RTE_BE32(0xffffffff),
1741 			.proto = 0xff,
1742 		},
1743 	};
1744 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1745 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1746 				      MLX5_FLOW_LAYER_OUTER_L3;
1747 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1748 				      MLX5_FLOW_LAYER_OUTER_L4;
1749 	int ret;
1750 	uint8_t next_proto = 0xFF;
1751 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1752 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1753 				  MLX5_FLOW_LAYER_INNER_VLAN);
1754 
1755 	if ((last_item & l2_vlan) && ether_type &&
1756 	    ether_type != RTE_ETHER_TYPE_IPV6)
1757 		return rte_flow_error_set(error, EINVAL,
1758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1759 					  "IPv6 cannot follow L2/VLAN layer "
1760 					  "which ether type is not IPv6");
1761 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1762 		if (mask && spec)
1763 			next_proto = mask->hdr.proto & spec->hdr.proto;
1764 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1765 			return rte_flow_error_set(error, EINVAL,
1766 						  RTE_FLOW_ERROR_TYPE_ITEM,
1767 						  item,
1768 						  "multiple tunnel "
1769 						  "not supported");
1770 	}
1771 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1772 		return rte_flow_error_set(error, EINVAL,
1773 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1774 					  "wrong tunnel type - IPv4 specified "
1775 					  "but IPv6 item provided");
1776 	if (item_flags & l3m)
1777 		return rte_flow_error_set(error, ENOTSUP,
1778 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1779 					  "multiple L3 layers not supported");
1780 	else if (item_flags & l4m)
1781 		return rte_flow_error_set(error, EINVAL,
1782 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1783 					  "L3 cannot follow an L4 layer.");
1784 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1785 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1786 		return rte_flow_error_set(error, EINVAL,
1787 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1788 					  "L3 cannot follow an NVGRE layer.");
1789 	if (!mask)
1790 		mask = &rte_flow_item_ipv6_mask;
1791 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1792 					acc_mask ? (const uint8_t *)acc_mask
1793 						 : (const uint8_t *)&nic_mask,
1794 					sizeof(struct rte_flow_item_ipv6),
1795 					error);
1796 	if (ret < 0)
1797 		return ret;
1798 	return 0;
1799 }
1800 
1801 /**
1802  * Validate UDP item.
1803  *
1804  * @param[in] item
1805  *   Item specification.
1806  * @param[in] item_flags
1807  *   Bit-fields that holds the items detected until now.
1808  * @param[in] target_protocol
1809  *   The next protocol in the previous item.
1810  * @param[in] flow_mask
1811  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1812  * @param[out] error
1813  *   Pointer to error structure.
1814  *
1815  * @return
1816  *   0 on success, a negative errno value otherwise and rte_errno is set.
1817  */
1818 int
1819 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1820 			    uint64_t item_flags,
1821 			    uint8_t target_protocol,
1822 			    struct rte_flow_error *error)
1823 {
1824 	const struct rte_flow_item_udp *mask = item->mask;
1825 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1826 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1827 				      MLX5_FLOW_LAYER_OUTER_L3;
1828 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1829 				      MLX5_FLOW_LAYER_OUTER_L4;
1830 	int ret;
1831 
1832 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1833 		return rte_flow_error_set(error, EINVAL,
1834 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1835 					  "protocol filtering not compatible"
1836 					  " with UDP layer");
1837 	if (!(item_flags & l3m))
1838 		return rte_flow_error_set(error, EINVAL,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "L3 is mandatory to filter on L4");
1841 	if (item_flags & l4m)
1842 		return rte_flow_error_set(error, EINVAL,
1843 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1844 					  "multiple L4 layers not supported");
1845 	if (!mask)
1846 		mask = &rte_flow_item_udp_mask;
1847 	ret = mlx5_flow_item_acceptable
1848 		(item, (const uint8_t *)mask,
1849 		 (const uint8_t *)&rte_flow_item_udp_mask,
1850 		 sizeof(struct rte_flow_item_udp), error);
1851 	if (ret < 0)
1852 		return ret;
1853 	return 0;
1854 }
1855 
1856 /**
1857  * Validate TCP item.
1858  *
1859  * @param[in] item
1860  *   Item specification.
1861  * @param[in] item_flags
1862  *   Bit-fields that holds the items detected until now.
1863  * @param[in] target_protocol
1864  *   The next protocol in the previous item.
1865  * @param[out] error
1866  *   Pointer to error structure.
1867  *
1868  * @return
1869  *   0 on success, a negative errno value otherwise and rte_errno is set.
1870  */
1871 int
1872 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1873 			    uint64_t item_flags,
1874 			    uint8_t target_protocol,
1875 			    const struct rte_flow_item_tcp *flow_mask,
1876 			    struct rte_flow_error *error)
1877 {
1878 	const struct rte_flow_item_tcp *mask = item->mask;
1879 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1880 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1881 				      MLX5_FLOW_LAYER_OUTER_L3;
1882 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1883 				      MLX5_FLOW_LAYER_OUTER_L4;
1884 	int ret;
1885 
1886 	MLX5_ASSERT(flow_mask);
1887 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1888 		return rte_flow_error_set(error, EINVAL,
1889 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1890 					  "protocol filtering not compatible"
1891 					  " with TCP layer");
1892 	if (!(item_flags & l3m))
1893 		return rte_flow_error_set(error, EINVAL,
1894 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1895 					  "L3 is mandatory to filter on L4");
1896 	if (item_flags & l4m)
1897 		return rte_flow_error_set(error, EINVAL,
1898 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1899 					  "multiple L4 layers not supported");
1900 	if (!mask)
1901 		mask = &rte_flow_item_tcp_mask;
1902 	ret = mlx5_flow_item_acceptable
1903 		(item, (const uint8_t *)mask,
1904 		 (const uint8_t *)flow_mask,
1905 		 sizeof(struct rte_flow_item_tcp), error);
1906 	if (ret < 0)
1907 		return ret;
1908 	return 0;
1909 }
1910 
1911 /**
1912  * Validate VXLAN item.
1913  *
1914  * @param[in] item
1915  *   Item specification.
1916  * @param[in] item_flags
1917  *   Bit-fields that holds the items detected until now.
1918  * @param[in] target_protocol
1919  *   The next protocol in the previous item.
1920  * @param[out] error
1921  *   Pointer to error structure.
1922  *
1923  * @return
1924  *   0 on success, a negative errno value otherwise and rte_errno is set.
1925  */
1926 int
1927 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1928 			      uint64_t item_flags,
1929 			      struct rte_flow_error *error)
1930 {
1931 	const struct rte_flow_item_vxlan *spec = item->spec;
1932 	const struct rte_flow_item_vxlan *mask = item->mask;
1933 	int ret;
1934 	union vni {
1935 		uint32_t vlan_id;
1936 		uint8_t vni[4];
1937 	} id = { .vlan_id = 0, };
1938 
1939 
1940 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1941 		return rte_flow_error_set(error, ENOTSUP,
1942 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1943 					  "multiple tunnel layers not"
1944 					  " supported");
1945 	/*
1946 	 * Verify only UDPv4 is present as defined in
1947 	 * https://tools.ietf.org/html/rfc7348
1948 	 */
1949 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1950 		return rte_flow_error_set(error, EINVAL,
1951 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1952 					  "no outer UDP layer found");
1953 	if (!mask)
1954 		mask = &rte_flow_item_vxlan_mask;
1955 	ret = mlx5_flow_item_acceptable
1956 		(item, (const uint8_t *)mask,
1957 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1958 		 sizeof(struct rte_flow_item_vxlan),
1959 		 error);
1960 	if (ret < 0)
1961 		return ret;
1962 	if (spec) {
1963 		memcpy(&id.vni[1], spec->vni, 3);
1964 		memcpy(&id.vni[1], mask->vni, 3);
1965 	}
1966 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1967 		return rte_flow_error_set(error, ENOTSUP,
1968 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1969 					  "VXLAN tunnel must be fully defined");
1970 	return 0;
1971 }
1972 
1973 /**
1974  * Validate VXLAN_GPE item.
1975  *
1976  * @param[in] item
1977  *   Item specification.
1978  * @param[in] item_flags
1979  *   Bit-fields that holds the items detected until now.
1980  * @param[in] priv
1981  *   Pointer to the private data structure.
1982  * @param[in] target_protocol
1983  *   The next protocol in the previous item.
1984  * @param[out] error
1985  *   Pointer to error structure.
1986  *
1987  * @return
1988  *   0 on success, a negative errno value otherwise and rte_errno is set.
1989  */
1990 int
1991 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1992 				  uint64_t item_flags,
1993 				  struct rte_eth_dev *dev,
1994 				  struct rte_flow_error *error)
1995 {
1996 	struct mlx5_priv *priv = dev->data->dev_private;
1997 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1998 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1999 	int ret;
2000 	union vni {
2001 		uint32_t vlan_id;
2002 		uint8_t vni[4];
2003 	} id = { .vlan_id = 0, };
2004 
2005 	if (!priv->config.l3_vxlan_en)
2006 		return rte_flow_error_set(error, ENOTSUP,
2007 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2008 					  "L3 VXLAN is not enabled by device"
2009 					  " parameter and/or not configured in"
2010 					  " firmware");
2011 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2012 		return rte_flow_error_set(error, ENOTSUP,
2013 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2014 					  "multiple tunnel layers not"
2015 					  " supported");
2016 	/*
2017 	 * Verify only UDPv4 is present as defined in
2018 	 * https://tools.ietf.org/html/rfc7348
2019 	 */
2020 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2021 		return rte_flow_error_set(error, EINVAL,
2022 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2023 					  "no outer UDP layer found");
2024 	if (!mask)
2025 		mask = &rte_flow_item_vxlan_gpe_mask;
2026 	ret = mlx5_flow_item_acceptable
2027 		(item, (const uint8_t *)mask,
2028 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2029 		 sizeof(struct rte_flow_item_vxlan_gpe),
2030 		 error);
2031 	if (ret < 0)
2032 		return ret;
2033 	if (spec) {
2034 		if (spec->protocol)
2035 			return rte_flow_error_set(error, ENOTSUP,
2036 						  RTE_FLOW_ERROR_TYPE_ITEM,
2037 						  item,
2038 						  "VxLAN-GPE protocol"
2039 						  " not supported");
2040 		memcpy(&id.vni[1], spec->vni, 3);
2041 		memcpy(&id.vni[1], mask->vni, 3);
2042 	}
2043 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2044 		return rte_flow_error_set(error, ENOTSUP,
2045 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2046 					  "VXLAN-GPE tunnel must be fully"
2047 					  " defined");
2048 	return 0;
2049 }
2050 /**
2051  * Validate GRE Key item.
2052  *
2053  * @param[in] item
2054  *   Item specification.
2055  * @param[in] item_flags
2056  *   Bit flags to mark detected items.
2057  * @param[in] gre_item
2058  *   Pointer to gre_item
2059  * @param[out] error
2060  *   Pointer to error structure.
2061  *
2062  * @return
2063  *   0 on success, a negative errno value otherwise and rte_errno is set.
2064  */
2065 int
2066 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2067 				uint64_t item_flags,
2068 				const struct rte_flow_item *gre_item,
2069 				struct rte_flow_error *error)
2070 {
2071 	const rte_be32_t *mask = item->mask;
2072 	int ret = 0;
2073 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2074 	const struct rte_flow_item_gre *gre_spec;
2075 	const struct rte_flow_item_gre *gre_mask;
2076 
2077 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2078 		return rte_flow_error_set(error, ENOTSUP,
2079 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2080 					  "Multiple GRE key not support");
2081 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2082 		return rte_flow_error_set(error, ENOTSUP,
2083 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2084 					  "No preceding GRE header");
2085 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2086 		return rte_flow_error_set(error, ENOTSUP,
2087 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2088 					  "GRE key following a wrong item");
2089 	gre_mask = gre_item->mask;
2090 	if (!gre_mask)
2091 		gre_mask = &rte_flow_item_gre_mask;
2092 	gre_spec = gre_item->spec;
2093 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2094 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2095 		return rte_flow_error_set(error, EINVAL,
2096 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2097 					  "Key bit must be on");
2098 
2099 	if (!mask)
2100 		mask = &gre_key_default_mask;
2101 	ret = mlx5_flow_item_acceptable
2102 		(item, (const uint8_t *)mask,
2103 		 (const uint8_t *)&gre_key_default_mask,
2104 		 sizeof(rte_be32_t), error);
2105 	return ret;
2106 }
2107 
2108 /**
2109  * Validate GRE item.
2110  *
2111  * @param[in] item
2112  *   Item specification.
2113  * @param[in] item_flags
2114  *   Bit flags to mark detected items.
2115  * @param[in] target_protocol
2116  *   The next protocol in the previous item.
2117  * @param[out] error
2118  *   Pointer to error structure.
2119  *
2120  * @return
2121  *   0 on success, a negative errno value otherwise and rte_errno is set.
2122  */
2123 int
2124 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2125 			    uint64_t item_flags,
2126 			    uint8_t target_protocol,
2127 			    struct rte_flow_error *error)
2128 {
2129 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2130 	const struct rte_flow_item_gre *mask = item->mask;
2131 	int ret;
2132 	const struct rte_flow_item_gre nic_mask = {
2133 		.c_rsvd0_ver = RTE_BE16(0xB000),
2134 		.protocol = RTE_BE16(UINT16_MAX),
2135 	};
2136 
2137 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2138 		return rte_flow_error_set(error, EINVAL,
2139 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2140 					  "protocol filtering not compatible"
2141 					  " with this GRE layer");
2142 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2143 		return rte_flow_error_set(error, ENOTSUP,
2144 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2145 					  "multiple tunnel layers not"
2146 					  " supported");
2147 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2148 		return rte_flow_error_set(error, ENOTSUP,
2149 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2150 					  "L3 Layer is missing");
2151 	if (!mask)
2152 		mask = &rte_flow_item_gre_mask;
2153 	ret = mlx5_flow_item_acceptable
2154 		(item, (const uint8_t *)mask,
2155 		 (const uint8_t *)&nic_mask,
2156 		 sizeof(struct rte_flow_item_gre), error);
2157 	if (ret < 0)
2158 		return ret;
2159 #ifndef HAVE_MLX5DV_DR
2160 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2161 	if (spec && (spec->protocol & mask->protocol))
2162 		return rte_flow_error_set(error, ENOTSUP,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "without MPLS support the"
2165 					  " specification cannot be used for"
2166 					  " filtering");
2167 #endif
2168 #endif
2169 	return 0;
2170 }
2171 
2172 /**
2173  * Validate Geneve item.
2174  *
2175  * @param[in] item
2176  *   Item specification.
2177  * @param[in] itemFlags
2178  *   Bit-fields that holds the items detected until now.
2179  * @param[in] enPriv
2180  *   Pointer to the private data structure.
2181  * @param[out] error
2182  *   Pointer to error structure.
2183  *
2184  * @return
2185  *   0 on success, a negative errno value otherwise and rte_errno is set.
2186  */
2187 
2188 int
2189 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2190 			       uint64_t item_flags,
2191 			       struct rte_eth_dev *dev,
2192 			       struct rte_flow_error *error)
2193 {
2194 	struct mlx5_priv *priv = dev->data->dev_private;
2195 	const struct rte_flow_item_geneve *spec = item->spec;
2196 	const struct rte_flow_item_geneve *mask = item->mask;
2197 	int ret;
2198 	uint16_t gbhdr;
2199 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2200 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2201 	const struct rte_flow_item_geneve nic_mask = {
2202 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2203 		.vni = "\xff\xff\xff",
2204 		.protocol = RTE_BE16(UINT16_MAX),
2205 	};
2206 
2207 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2208 		return rte_flow_error_set(error, ENOTSUP,
2209 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2210 					  "L3 Geneve is not enabled by device"
2211 					  " parameter and/or not configured in"
2212 					  " firmware");
2213 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2214 		return rte_flow_error_set(error, ENOTSUP,
2215 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2216 					  "multiple tunnel layers not"
2217 					  " supported");
2218 	/*
2219 	 * Verify only UDPv4 is present as defined in
2220 	 * https://tools.ietf.org/html/rfc7348
2221 	 */
2222 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2223 		return rte_flow_error_set(error, EINVAL,
2224 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2225 					  "no outer UDP layer found");
2226 	if (!mask)
2227 		mask = &rte_flow_item_geneve_mask;
2228 	ret = mlx5_flow_item_acceptable
2229 				  (item, (const uint8_t *)mask,
2230 				   (const uint8_t *)&nic_mask,
2231 				   sizeof(struct rte_flow_item_geneve), error);
2232 	if (ret)
2233 		return ret;
2234 	if (spec) {
2235 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2236 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2237 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2238 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2239 			return rte_flow_error_set(error, ENOTSUP,
2240 						  RTE_FLOW_ERROR_TYPE_ITEM,
2241 						  item,
2242 						  "Geneve protocol unsupported"
2243 						  " fields are being used");
2244 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2245 			return rte_flow_error_set
2246 					(error, ENOTSUP,
2247 					 RTE_FLOW_ERROR_TYPE_ITEM,
2248 					 item,
2249 					 "Unsupported Geneve options length");
2250 	}
2251 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2252 		return rte_flow_error_set
2253 				    (error, ENOTSUP,
2254 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2255 				     "Geneve tunnel must be fully defined");
2256 	return 0;
2257 }
2258 
2259 /**
2260  * Validate MPLS item.
2261  *
2262  * @param[in] dev
2263  *   Pointer to the rte_eth_dev structure.
2264  * @param[in] item
2265  *   Item specification.
2266  * @param[in] item_flags
2267  *   Bit-fields that holds the items detected until now.
2268  * @param[in] prev_layer
2269  *   The protocol layer indicated in previous item.
2270  * @param[out] error
2271  *   Pointer to error structure.
2272  *
2273  * @return
2274  *   0 on success, a negative errno value otherwise and rte_errno is set.
2275  */
2276 int
2277 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2278 			     const struct rte_flow_item *item __rte_unused,
2279 			     uint64_t item_flags __rte_unused,
2280 			     uint64_t prev_layer __rte_unused,
2281 			     struct rte_flow_error *error)
2282 {
2283 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2284 	const struct rte_flow_item_mpls *mask = item->mask;
2285 	struct mlx5_priv *priv = dev->data->dev_private;
2286 	int ret;
2287 
2288 	if (!priv->config.mpls_en)
2289 		return rte_flow_error_set(error, ENOTSUP,
2290 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2291 					  "MPLS not supported or"
2292 					  " disabled in firmware"
2293 					  " configuration.");
2294 	/* MPLS over IP, UDP, GRE is allowed */
2295 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2296 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2297 			    MLX5_FLOW_LAYER_GRE)))
2298 		return rte_flow_error_set(error, EINVAL,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "protocol filtering not compatible"
2301 					  " with MPLS layer");
2302 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2303 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2304 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2305 		return rte_flow_error_set(error, ENOTSUP,
2306 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2307 					  "multiple tunnel layers not"
2308 					  " supported");
2309 	if (!mask)
2310 		mask = &rte_flow_item_mpls_mask;
2311 	ret = mlx5_flow_item_acceptable
2312 		(item, (const uint8_t *)mask,
2313 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2314 		 sizeof(struct rte_flow_item_mpls), error);
2315 	if (ret < 0)
2316 		return ret;
2317 	return 0;
2318 #else
2319 	return rte_flow_error_set(error, ENOTSUP,
2320 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2321 				  "MPLS is not supported by Verbs, please"
2322 				  " update.");
2323 #endif
2324 }
2325 
2326 /**
2327  * Validate NVGRE item.
2328  *
2329  * @param[in] item
2330  *   Item specification.
2331  * @param[in] item_flags
2332  *   Bit flags to mark detected items.
2333  * @param[in] target_protocol
2334  *   The next protocol in the previous item.
2335  * @param[out] error
2336  *   Pointer to error structure.
2337  *
2338  * @return
2339  *   0 on success, a negative errno value otherwise and rte_errno is set.
2340  */
2341 int
2342 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2343 			      uint64_t item_flags,
2344 			      uint8_t target_protocol,
2345 			      struct rte_flow_error *error)
2346 {
2347 	const struct rte_flow_item_nvgre *mask = item->mask;
2348 	int ret;
2349 
2350 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2351 		return rte_flow_error_set(error, EINVAL,
2352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2353 					  "protocol filtering not compatible"
2354 					  " with this GRE layer");
2355 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2356 		return rte_flow_error_set(error, ENOTSUP,
2357 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2358 					  "multiple tunnel layers not"
2359 					  " supported");
2360 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2361 		return rte_flow_error_set(error, ENOTSUP,
2362 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2363 					  "L3 Layer is missing");
2364 	if (!mask)
2365 		mask = &rte_flow_item_nvgre_mask;
2366 	ret = mlx5_flow_item_acceptable
2367 		(item, (const uint8_t *)mask,
2368 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2369 		 sizeof(struct rte_flow_item_nvgre), error);
2370 	if (ret < 0)
2371 		return ret;
2372 	return 0;
2373 }
2374 
2375 /**
2376  * Validate eCPRI item.
2377  *
2378  * @param[in] item
2379  *   Item specification.
2380  * @param[in] item_flags
2381  *   Bit-fields that holds the items detected until now.
2382  * @param[in] last_item
2383  *   Previous validated item in the pattern items.
2384  * @param[in] ether_type
2385  *   Type in the ethernet layer header (including dot1q).
2386  * @param[in] acc_mask
2387  *   Acceptable mask, if NULL default internal default mask
2388  *   will be used to check whether item fields are supported.
2389  * @param[out] error
2390  *   Pointer to error structure.
2391  *
2392  * @return
2393  *   0 on success, a negative errno value otherwise and rte_errno is set.
2394  */
2395 int
2396 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2397 			      uint64_t item_flags,
2398 			      uint64_t last_item,
2399 			      uint16_t ether_type,
2400 			      const struct rte_flow_item_ecpri *acc_mask,
2401 			      struct rte_flow_error *error)
2402 {
2403 	const struct rte_flow_item_ecpri *mask = item->mask;
2404 	const struct rte_flow_item_ecpri nic_mask = {
2405 		.hdr = {
2406 			.common = {
2407 				.u32 =
2408 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2409 					.type = 0xFF,
2410 					}).u32),
2411 			},
2412 			.dummy[0] = 0xFFFFFFFF,
2413 		},
2414 	};
2415 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2416 					MLX5_FLOW_LAYER_OUTER_VLAN);
2417 	struct rte_flow_item_ecpri mask_lo;
2418 
2419 	if ((last_item & outer_l2_vlan) && ether_type &&
2420 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2421 		return rte_flow_error_set(error, EINVAL,
2422 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2423 					  "eCPRI cannot follow L2/VLAN layer "
2424 					  "which ether type is not 0xAEFE.");
2425 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2426 		return rte_flow_error_set(error, EINVAL,
2427 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2428 					  "eCPRI with tunnel is not supported "
2429 					  "right now.");
2430 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2431 		return rte_flow_error_set(error, ENOTSUP,
2432 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2433 					  "multiple L3 layers not supported");
2434 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2435 		return rte_flow_error_set(error, EINVAL,
2436 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2437 					  "eCPRI cannot follow a TCP layer.");
2438 	/* In specification, eCPRI could be over UDP layer. */
2439 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2440 		return rte_flow_error_set(error, EINVAL,
2441 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2442 					  "eCPRI over UDP layer is not yet "
2443 					  "supported right now.");
2444 	/* Mask for type field in common header could be zero. */
2445 	if (!mask)
2446 		mask = &rte_flow_item_ecpri_mask;
2447 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2448 	/* Input mask is in big-endian format. */
2449 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2450 		return rte_flow_error_set(error, EINVAL,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2452 					  "partial mask is not supported "
2453 					  "for protocol");
2454 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2455 		return rte_flow_error_set(error, EINVAL,
2456 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2457 					  "message header mask must be after "
2458 					  "a type mask");
2459 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2460 					 acc_mask ? (const uint8_t *)acc_mask
2461 						  : (const uint8_t *)&nic_mask,
2462 					 sizeof(struct rte_flow_item_ecpri),
2463 					 error);
2464 }
2465 
2466 /* Allocate unique ID for the split Q/RSS subflows. */
2467 static uint32_t
2468 flow_qrss_get_id(struct rte_eth_dev *dev)
2469 {
2470 	struct mlx5_priv *priv = dev->data->dev_private;
2471 	uint32_t qrss_id, ret;
2472 
2473 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2474 	if (ret)
2475 		return 0;
2476 	MLX5_ASSERT(qrss_id);
2477 	return qrss_id;
2478 }
2479 
2480 /* Free unique ID for the split Q/RSS subflows. */
2481 static void
2482 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2483 {
2484 	struct mlx5_priv *priv = dev->data->dev_private;
2485 
2486 	if (qrss_id)
2487 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2488 }
2489 
2490 /**
2491  * Release resource related QUEUE/RSS action split.
2492  *
2493  * @param dev
2494  *   Pointer to Ethernet device.
2495  * @param flow
2496  *   Flow to release id's from.
2497  */
2498 static void
2499 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2500 			     struct rte_flow *flow)
2501 {
2502 	struct mlx5_priv *priv = dev->data->dev_private;
2503 	uint32_t handle_idx;
2504 	struct mlx5_flow_handle *dev_handle;
2505 
2506 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2507 		       handle_idx, dev_handle, next)
2508 		if (dev_handle->split_flow_id)
2509 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2510 }
2511 
2512 static int
2513 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2514 		   const struct rte_flow_attr *attr __rte_unused,
2515 		   const struct rte_flow_item items[] __rte_unused,
2516 		   const struct rte_flow_action actions[] __rte_unused,
2517 		   bool external __rte_unused,
2518 		   int hairpin __rte_unused,
2519 		   struct rte_flow_error *error)
2520 {
2521 	return rte_flow_error_set(error, ENOTSUP,
2522 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2523 }
2524 
2525 static struct mlx5_flow *
2526 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2527 		  const struct rte_flow_attr *attr __rte_unused,
2528 		  const struct rte_flow_item items[] __rte_unused,
2529 		  const struct rte_flow_action actions[] __rte_unused,
2530 		  struct rte_flow_error *error)
2531 {
2532 	rte_flow_error_set(error, ENOTSUP,
2533 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2534 	return NULL;
2535 }
2536 
2537 static int
2538 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2539 		    struct mlx5_flow *dev_flow __rte_unused,
2540 		    const struct rte_flow_attr *attr __rte_unused,
2541 		    const struct rte_flow_item items[] __rte_unused,
2542 		    const struct rte_flow_action actions[] __rte_unused,
2543 		    struct rte_flow_error *error)
2544 {
2545 	return rte_flow_error_set(error, ENOTSUP,
2546 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2547 }
2548 
2549 static int
2550 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2551 		struct rte_flow *flow __rte_unused,
2552 		struct rte_flow_error *error)
2553 {
2554 	return rte_flow_error_set(error, ENOTSUP,
2555 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2556 }
2557 
2558 static void
2559 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2560 		 struct rte_flow *flow __rte_unused)
2561 {
2562 }
2563 
2564 static void
2565 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2566 		  struct rte_flow *flow __rte_unused)
2567 {
2568 }
2569 
2570 static int
2571 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2572 		struct rte_flow *flow __rte_unused,
2573 		const struct rte_flow_action *actions __rte_unused,
2574 		void *data __rte_unused,
2575 		struct rte_flow_error *error)
2576 {
2577 	return rte_flow_error_set(error, ENOTSUP,
2578 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2579 }
2580 
2581 /* Void driver to protect from null pointer reference. */
2582 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2583 	.validate = flow_null_validate,
2584 	.prepare = flow_null_prepare,
2585 	.translate = flow_null_translate,
2586 	.apply = flow_null_apply,
2587 	.remove = flow_null_remove,
2588 	.destroy = flow_null_destroy,
2589 	.query = flow_null_query,
2590 };
2591 
2592 /**
2593  * Select flow driver type according to flow attributes and device
2594  * configuration.
2595  *
2596  * @param[in] dev
2597  *   Pointer to the dev structure.
2598  * @param[in] attr
2599  *   Pointer to the flow attributes.
2600  *
2601  * @return
2602  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2603  */
2604 static enum mlx5_flow_drv_type
2605 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2606 {
2607 	struct mlx5_priv *priv = dev->data->dev_private;
2608 	/* The OS can determine first a specific flow type (DV, VERBS) */
2609 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2610 
2611 	if (type != MLX5_FLOW_TYPE_MAX)
2612 		return type;
2613 	/* If no OS specific type - continue with DV/VERBS selection */
2614 	if (attr->transfer && priv->config.dv_esw_en)
2615 		type = MLX5_FLOW_TYPE_DV;
2616 	if (!attr->transfer)
2617 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2618 						 MLX5_FLOW_TYPE_VERBS;
2619 	return type;
2620 }
2621 
2622 #define flow_get_drv_ops(type) flow_drv_ops[type]
2623 
2624 /**
2625  * Flow driver validation API. This abstracts calling driver specific functions.
2626  * The type of flow driver is determined according to flow attributes.
2627  *
2628  * @param[in] dev
2629  *   Pointer to the dev structure.
2630  * @param[in] attr
2631  *   Pointer to the flow attributes.
2632  * @param[in] items
2633  *   Pointer to the list of items.
2634  * @param[in] actions
2635  *   Pointer to the list of actions.
2636  * @param[in] external
2637  *   This flow rule is created by request external to PMD.
2638  * @param[in] hairpin
2639  *   Number of hairpin TX actions, 0 means classic flow.
2640  * @param[out] error
2641  *   Pointer to the error structure.
2642  *
2643  * @return
2644  *   0 on success, a negative errno value otherwise and rte_errno is set.
2645  */
2646 static inline int
2647 flow_drv_validate(struct rte_eth_dev *dev,
2648 		  const struct rte_flow_attr *attr,
2649 		  const struct rte_flow_item items[],
2650 		  const struct rte_flow_action actions[],
2651 		  bool external, int hairpin, struct rte_flow_error *error)
2652 {
2653 	const struct mlx5_flow_driver_ops *fops;
2654 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2655 
2656 	fops = flow_get_drv_ops(type);
2657 	return fops->validate(dev, attr, items, actions, external,
2658 			      hairpin, error);
2659 }
2660 
2661 /**
2662  * Flow driver preparation API. This abstracts calling driver specific
2663  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2664  * calculates the size of memory required for device flow, allocates the memory,
2665  * initializes the device flow and returns the pointer.
2666  *
2667  * @note
2668  *   This function initializes device flow structure such as dv or verbs in
2669  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2670  *   rest. For example, adding returning device flow to flow->dev_flow list and
2671  *   setting backward reference to the flow should be done out of this function.
2672  *   layers field is not filled either.
2673  *
2674  * @param[in] dev
2675  *   Pointer to the dev structure.
2676  * @param[in] attr
2677  *   Pointer to the flow attributes.
2678  * @param[in] items
2679  *   Pointer to the list of items.
2680  * @param[in] actions
2681  *   Pointer to the list of actions.
2682  * @param[in] flow_idx
2683  *   This memory pool index to the flow.
2684  * @param[out] error
2685  *   Pointer to the error structure.
2686  *
2687  * @return
2688  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2689  */
2690 static inline struct mlx5_flow *
2691 flow_drv_prepare(struct rte_eth_dev *dev,
2692 		 const struct rte_flow *flow,
2693 		 const struct rte_flow_attr *attr,
2694 		 const struct rte_flow_item items[],
2695 		 const struct rte_flow_action actions[],
2696 		 uint32_t flow_idx,
2697 		 struct rte_flow_error *error)
2698 {
2699 	const struct mlx5_flow_driver_ops *fops;
2700 	enum mlx5_flow_drv_type type = flow->drv_type;
2701 	struct mlx5_flow *mlx5_flow = NULL;
2702 
2703 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2704 	fops = flow_get_drv_ops(type);
2705 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2706 	if (mlx5_flow)
2707 		mlx5_flow->flow_idx = flow_idx;
2708 	return mlx5_flow;
2709 }
2710 
2711 /**
2712  * Flow driver translation API. This abstracts calling driver specific
2713  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2714  * translates a generic flow into a driver flow. flow_drv_prepare() must
2715  * precede.
2716  *
2717  * @note
2718  *   dev_flow->layers could be filled as a result of parsing during translation
2719  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2720  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2721  *   flow->actions could be overwritten even though all the expanded dev_flows
2722  *   have the same actions.
2723  *
2724  * @param[in] dev
2725  *   Pointer to the rte dev structure.
2726  * @param[in, out] dev_flow
2727  *   Pointer to the mlx5 flow.
2728  * @param[in] attr
2729  *   Pointer to the flow attributes.
2730  * @param[in] items
2731  *   Pointer to the list of items.
2732  * @param[in] actions
2733  *   Pointer to the list of actions.
2734  * @param[out] error
2735  *   Pointer to the error structure.
2736  *
2737  * @return
2738  *   0 on success, a negative errno value otherwise and rte_errno is set.
2739  */
2740 static inline int
2741 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2742 		   const struct rte_flow_attr *attr,
2743 		   const struct rte_flow_item items[],
2744 		   const struct rte_flow_action actions[],
2745 		   struct rte_flow_error *error)
2746 {
2747 	const struct mlx5_flow_driver_ops *fops;
2748 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2749 
2750 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2751 	fops = flow_get_drv_ops(type);
2752 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2753 }
2754 
2755 /**
2756  * Flow driver apply API. This abstracts calling driver specific functions.
2757  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2758  * translated driver flows on to device. flow_drv_translate() must precede.
2759  *
2760  * @param[in] dev
2761  *   Pointer to Ethernet device structure.
2762  * @param[in, out] flow
2763  *   Pointer to flow structure.
2764  * @param[out] error
2765  *   Pointer to error structure.
2766  *
2767  * @return
2768  *   0 on success, a negative errno value otherwise and rte_errno is set.
2769  */
2770 static inline int
2771 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2772 	       struct rte_flow_error *error)
2773 {
2774 	const struct mlx5_flow_driver_ops *fops;
2775 	enum mlx5_flow_drv_type type = flow->drv_type;
2776 
2777 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2778 	fops = flow_get_drv_ops(type);
2779 	return fops->apply(dev, flow, error);
2780 }
2781 
2782 /**
2783  * Flow driver remove API. This abstracts calling driver specific functions.
2784  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2785  * on device. All the resources of the flow should be freed by calling
2786  * flow_drv_destroy().
2787  *
2788  * @param[in] dev
2789  *   Pointer to Ethernet device.
2790  * @param[in, out] flow
2791  *   Pointer to flow structure.
2792  */
2793 static inline void
2794 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2795 {
2796 	const struct mlx5_flow_driver_ops *fops;
2797 	enum mlx5_flow_drv_type type = flow->drv_type;
2798 
2799 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2800 	fops = flow_get_drv_ops(type);
2801 	fops->remove(dev, flow);
2802 }
2803 
2804 /**
2805  * Flow driver destroy API. This abstracts calling driver specific functions.
2806  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2807  * on device and releases resources of the flow.
2808  *
2809  * @param[in] dev
2810  *   Pointer to Ethernet device.
2811  * @param[in, out] flow
2812  *   Pointer to flow structure.
2813  */
2814 static inline void
2815 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2816 {
2817 	const struct mlx5_flow_driver_ops *fops;
2818 	enum mlx5_flow_drv_type type = flow->drv_type;
2819 
2820 	flow_mreg_split_qrss_release(dev, flow);
2821 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2822 	fops = flow_get_drv_ops(type);
2823 	fops->destroy(dev, flow);
2824 }
2825 
2826 /**
2827  * Get RSS action from the action list.
2828  *
2829  * @param[in] actions
2830  *   Pointer to the list of actions.
2831  *
2832  * @return
2833  *   Pointer to the RSS action if exist, else return NULL.
2834  */
2835 static const struct rte_flow_action_rss*
2836 flow_get_rss_action(const struct rte_flow_action actions[])
2837 {
2838 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2839 		switch (actions->type) {
2840 		case RTE_FLOW_ACTION_TYPE_RSS:
2841 			return (const struct rte_flow_action_rss *)
2842 			       actions->conf;
2843 		default:
2844 			break;
2845 		}
2846 	}
2847 	return NULL;
2848 }
2849 
2850 static unsigned int
2851 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2852 {
2853 	const struct rte_flow_item *item;
2854 	unsigned int has_vlan = 0;
2855 
2856 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2857 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2858 			has_vlan = 1;
2859 			break;
2860 		}
2861 	}
2862 	if (has_vlan)
2863 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2864 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2865 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2866 			       MLX5_EXPANSION_ROOT_OUTER;
2867 }
2868 
2869 /**
2870  *  Get layer flags from the prefix flow.
2871  *
2872  *  Some flows may be split to several subflows, the prefix subflow gets the
2873  *  match items and the suffix sub flow gets the actions.
2874  *  Some actions need the user defined match item flags to get the detail for
2875  *  the action.
2876  *  This function helps the suffix flow to get the item layer flags from prefix
2877  *  subflow.
2878  *
2879  * @param[in] dev_flow
2880  *   Pointer the created preifx subflow.
2881  *
2882  * @return
2883  *   The layers get from prefix subflow.
2884  */
2885 static inline uint64_t
2886 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2887 {
2888 	uint64_t layers = 0;
2889 
2890 	/*
2891 	 * Layers bits could be localization, but usually the compiler will
2892 	 * help to do the optimization work for source code.
2893 	 * If no decap actions, use the layers directly.
2894 	 */
2895 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2896 		return dev_flow->handle->layers;
2897 	/* Convert L3 layers with decap action. */
2898 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2899 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2900 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2901 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2902 	/* Convert L4 layers with decap action.  */
2903 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2904 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2905 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2906 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2907 	return layers;
2908 }
2909 
2910 /**
2911  * Get metadata split action information.
2912  *
2913  * @param[in] actions
2914  *   Pointer to the list of actions.
2915  * @param[out] qrss
2916  *   Pointer to the return pointer.
2917  * @param[out] qrss_type
2918  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2919  *   if no QUEUE/RSS is found.
2920  * @param[out] encap_idx
2921  *   Pointer to the index of the encap action if exists, otherwise the last
2922  *   action index.
2923  *
2924  * @return
2925  *   Total number of actions.
2926  */
2927 static int
2928 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2929 				       const struct rte_flow_action **qrss,
2930 				       int *encap_idx)
2931 {
2932 	const struct rte_flow_action_raw_encap *raw_encap;
2933 	int actions_n = 0;
2934 	int raw_decap_idx = -1;
2935 
2936 	*encap_idx = -1;
2937 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2938 		switch (actions->type) {
2939 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2940 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2941 			*encap_idx = actions_n;
2942 			break;
2943 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2944 			raw_decap_idx = actions_n;
2945 			break;
2946 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2947 			raw_encap = actions->conf;
2948 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2949 				*encap_idx = raw_decap_idx != -1 ?
2950 						      raw_decap_idx : actions_n;
2951 			break;
2952 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2953 		case RTE_FLOW_ACTION_TYPE_RSS:
2954 			*qrss = actions;
2955 			break;
2956 		default:
2957 			break;
2958 		}
2959 		actions_n++;
2960 	}
2961 	if (*encap_idx == -1)
2962 		*encap_idx = actions_n;
2963 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2964 	return actions_n + 1;
2965 }
2966 
2967 /**
2968  * Check meter action from the action list.
2969  *
2970  * @param[in] actions
2971  *   Pointer to the list of actions.
2972  * @param[out] mtr
2973  *   Pointer to the meter exist flag.
2974  *
2975  * @return
2976  *   Total number of actions.
2977  */
2978 static int
2979 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2980 {
2981 	int actions_n = 0;
2982 
2983 	MLX5_ASSERT(mtr);
2984 	*mtr = 0;
2985 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2986 		switch (actions->type) {
2987 		case RTE_FLOW_ACTION_TYPE_METER:
2988 			*mtr = 1;
2989 			break;
2990 		default:
2991 			break;
2992 		}
2993 		actions_n++;
2994 	}
2995 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2996 	return actions_n + 1;
2997 }
2998 
2999 /**
3000  * Check if the flow should be splited due to hairpin.
3001  * The reason for the split is that in current HW we can't
3002  * support encap on Rx, so if a flow have encap we move it
3003  * to Tx.
3004  *
3005  * @param dev
3006  *   Pointer to Ethernet device.
3007  * @param[in] attr
3008  *   Flow rule attributes.
3009  * @param[in] actions
3010  *   Associated actions (list terminated by the END action).
3011  *
3012  * @return
3013  *   > 0 the number of actions and the flow should be split,
3014  *   0 when no split required.
3015  */
3016 static int
3017 flow_check_hairpin_split(struct rte_eth_dev *dev,
3018 			 const struct rte_flow_attr *attr,
3019 			 const struct rte_flow_action actions[])
3020 {
3021 	int queue_action = 0;
3022 	int action_n = 0;
3023 	int encap = 0;
3024 	const struct rte_flow_action_queue *queue;
3025 	const struct rte_flow_action_rss *rss;
3026 	const struct rte_flow_action_raw_encap *raw_encap;
3027 
3028 	if (!attr->ingress)
3029 		return 0;
3030 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3031 		switch (actions->type) {
3032 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3033 			queue = actions->conf;
3034 			if (queue == NULL)
3035 				return 0;
3036 			if (mlx5_rxq_get_type(dev, queue->index) !=
3037 			    MLX5_RXQ_TYPE_HAIRPIN)
3038 				return 0;
3039 			queue_action = 1;
3040 			action_n++;
3041 			break;
3042 		case RTE_FLOW_ACTION_TYPE_RSS:
3043 			rss = actions->conf;
3044 			if (rss == NULL || rss->queue_num == 0)
3045 				return 0;
3046 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
3047 			    MLX5_RXQ_TYPE_HAIRPIN)
3048 				return 0;
3049 			queue_action = 1;
3050 			action_n++;
3051 			break;
3052 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3053 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3054 			encap = 1;
3055 			action_n++;
3056 			break;
3057 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3058 			raw_encap = actions->conf;
3059 			if (raw_encap->size >
3060 			    (sizeof(struct rte_flow_item_eth) +
3061 			     sizeof(struct rte_flow_item_ipv4)))
3062 				encap = 1;
3063 			action_n++;
3064 			break;
3065 		default:
3066 			action_n++;
3067 			break;
3068 		}
3069 	}
3070 	if (encap == 1 && queue_action)
3071 		return action_n;
3072 	return 0;
3073 }
3074 
3075 /* Declare flow create/destroy prototype in advance. */
3076 static uint32_t
3077 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3078 		 const struct rte_flow_attr *attr,
3079 		 const struct rte_flow_item items[],
3080 		 const struct rte_flow_action actions[],
3081 		 bool external, struct rte_flow_error *error);
3082 
3083 static void
3084 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3085 		  uint32_t flow_idx);
3086 
3087 /**
3088  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3089  *
3090  * As mark_id is unique, if there's already a registered flow for the mark_id,
3091  * return by increasing the reference counter of the resource. Otherwise, create
3092  * the resource (mcp_res) and flow.
3093  *
3094  * Flow looks like,
3095  *   - If ingress port is ANY and reg_c[1] is mark_id,
3096  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3097  *
3098  * For default flow (zero mark_id), flow is like,
3099  *   - If ingress port is ANY,
3100  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3101  *
3102  * @param dev
3103  *   Pointer to Ethernet device.
3104  * @param mark_id
3105  *   ID of MARK action, zero means default flow for META.
3106  * @param[out] error
3107  *   Perform verbose error reporting if not NULL.
3108  *
3109  * @return
3110  *   Associated resource on success, NULL otherwise and rte_errno is set.
3111  */
3112 static struct mlx5_flow_mreg_copy_resource *
3113 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3114 			  struct rte_flow_error *error)
3115 {
3116 	struct mlx5_priv *priv = dev->data->dev_private;
3117 	struct rte_flow_attr attr = {
3118 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3119 		.ingress = 1,
3120 	};
3121 	struct mlx5_rte_flow_item_tag tag_spec = {
3122 		.data = mark_id,
3123 	};
3124 	struct rte_flow_item items[] = {
3125 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3126 	};
3127 	struct rte_flow_action_mark ftag = {
3128 		.id = mark_id,
3129 	};
3130 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3131 		.dst = REG_B,
3132 		.src = 0,
3133 	};
3134 	struct rte_flow_action_jump jump = {
3135 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3136 	};
3137 	struct rte_flow_action actions[] = {
3138 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3139 	};
3140 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3141 	uint32_t idx = 0;
3142 	int ret;
3143 
3144 	/* Fill the register fileds in the flow. */
3145 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3146 	if (ret < 0)
3147 		return NULL;
3148 	tag_spec.id = ret;
3149 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3150 	if (ret < 0)
3151 		return NULL;
3152 	cp_mreg.src = ret;
3153 	/* Check if already registered. */
3154 	MLX5_ASSERT(priv->mreg_cp_tbl);
3155 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3156 	if (mcp_res) {
3157 		/* For non-default rule. */
3158 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3159 			mcp_res->refcnt++;
3160 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3161 			    mcp_res->refcnt == 1);
3162 		return mcp_res;
3163 	}
3164 	/* Provide the full width of FLAG specific value. */
3165 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3166 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3167 	/* Build a new flow. */
3168 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3169 		items[0] = (struct rte_flow_item){
3170 			.type = (enum rte_flow_item_type)
3171 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3172 			.spec = &tag_spec,
3173 		};
3174 		items[1] = (struct rte_flow_item){
3175 			.type = RTE_FLOW_ITEM_TYPE_END,
3176 		};
3177 		actions[0] = (struct rte_flow_action){
3178 			.type = (enum rte_flow_action_type)
3179 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3180 			.conf = &ftag,
3181 		};
3182 		actions[1] = (struct rte_flow_action){
3183 			.type = (enum rte_flow_action_type)
3184 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3185 			.conf = &cp_mreg,
3186 		};
3187 		actions[2] = (struct rte_flow_action){
3188 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3189 			.conf = &jump,
3190 		};
3191 		actions[3] = (struct rte_flow_action){
3192 			.type = RTE_FLOW_ACTION_TYPE_END,
3193 		};
3194 	} else {
3195 		/* Default rule, wildcard match. */
3196 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3197 		items[0] = (struct rte_flow_item){
3198 			.type = RTE_FLOW_ITEM_TYPE_END,
3199 		};
3200 		actions[0] = (struct rte_flow_action){
3201 			.type = (enum rte_flow_action_type)
3202 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3203 			.conf = &cp_mreg,
3204 		};
3205 		actions[1] = (struct rte_flow_action){
3206 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3207 			.conf = &jump,
3208 		};
3209 		actions[2] = (struct rte_flow_action){
3210 			.type = RTE_FLOW_ACTION_TYPE_END,
3211 		};
3212 	}
3213 	/* Build a new entry. */
3214 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3215 	if (!mcp_res) {
3216 		rte_errno = ENOMEM;
3217 		return NULL;
3218 	}
3219 	mcp_res->idx = idx;
3220 	/*
3221 	 * The copy Flows are not included in any list. There
3222 	 * ones are referenced from other Flows and can not
3223 	 * be applied, removed, deleted in ardbitrary order
3224 	 * by list traversing.
3225 	 */
3226 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3227 					 actions, false, error);
3228 	if (!mcp_res->rix_flow)
3229 		goto error;
3230 	mcp_res->refcnt++;
3231 	mcp_res->hlist_ent.key = mark_id;
3232 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3233 				&mcp_res->hlist_ent);
3234 	MLX5_ASSERT(!ret);
3235 	if (ret)
3236 		goto error;
3237 	return mcp_res;
3238 error:
3239 	if (mcp_res->rix_flow)
3240 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3241 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3242 	return NULL;
3243 }
3244 
3245 /**
3246  * Release flow in RX_CP_TBL.
3247  *
3248  * @param dev
3249  *   Pointer to Ethernet device.
3250  * @flow
3251  *   Parent flow for wich copying is provided.
3252  */
3253 static void
3254 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3255 			  struct rte_flow *flow)
3256 {
3257 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3258 	struct mlx5_priv *priv = dev->data->dev_private;
3259 
3260 	if (!flow->rix_mreg_copy)
3261 		return;
3262 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3263 				 flow->rix_mreg_copy);
3264 	if (!mcp_res || !priv->mreg_cp_tbl)
3265 		return;
3266 	if (flow->copy_applied) {
3267 		MLX5_ASSERT(mcp_res->appcnt);
3268 		flow->copy_applied = 0;
3269 		--mcp_res->appcnt;
3270 		if (!mcp_res->appcnt) {
3271 			struct rte_flow *mcp_flow = mlx5_ipool_get
3272 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3273 					mcp_res->rix_flow);
3274 
3275 			if (mcp_flow)
3276 				flow_drv_remove(dev, mcp_flow);
3277 		}
3278 	}
3279 	/*
3280 	 * We do not check availability of metadata registers here,
3281 	 * because copy resources are not allocated in this case.
3282 	 */
3283 	if (--mcp_res->refcnt)
3284 		return;
3285 	MLX5_ASSERT(mcp_res->rix_flow);
3286 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3287 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3288 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3289 	flow->rix_mreg_copy = 0;
3290 }
3291 
3292 /**
3293  * Start flow in RX_CP_TBL.
3294  *
3295  * @param dev
3296  *   Pointer to Ethernet device.
3297  * @flow
3298  *   Parent flow for wich copying is provided.
3299  *
3300  * @return
3301  *   0 on success, a negative errno value otherwise and rte_errno is set.
3302  */
3303 static int
3304 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3305 			    struct rte_flow *flow)
3306 {
3307 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3308 	struct mlx5_priv *priv = dev->data->dev_private;
3309 	int ret;
3310 
3311 	if (!flow->rix_mreg_copy || flow->copy_applied)
3312 		return 0;
3313 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3314 				 flow->rix_mreg_copy);
3315 	if (!mcp_res)
3316 		return 0;
3317 	if (!mcp_res->appcnt) {
3318 		struct rte_flow *mcp_flow = mlx5_ipool_get
3319 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3320 				mcp_res->rix_flow);
3321 
3322 		if (mcp_flow) {
3323 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3324 			if (ret)
3325 				return ret;
3326 		}
3327 	}
3328 	++mcp_res->appcnt;
3329 	flow->copy_applied = 1;
3330 	return 0;
3331 }
3332 
3333 /**
3334  * Stop flow in RX_CP_TBL.
3335  *
3336  * @param dev
3337  *   Pointer to Ethernet device.
3338  * @flow
3339  *   Parent flow for wich copying is provided.
3340  */
3341 static void
3342 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3343 			   struct rte_flow *flow)
3344 {
3345 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3346 	struct mlx5_priv *priv = dev->data->dev_private;
3347 
3348 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3349 		return;
3350 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3351 				 flow->rix_mreg_copy);
3352 	if (!mcp_res)
3353 		return;
3354 	MLX5_ASSERT(mcp_res->appcnt);
3355 	--mcp_res->appcnt;
3356 	flow->copy_applied = 0;
3357 	if (!mcp_res->appcnt) {
3358 		struct rte_flow *mcp_flow = mlx5_ipool_get
3359 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3360 				mcp_res->rix_flow);
3361 
3362 		if (mcp_flow)
3363 			flow_drv_remove(dev, mcp_flow);
3364 	}
3365 }
3366 
3367 /**
3368  * Remove the default copy action from RX_CP_TBL.
3369  *
3370  * @param dev
3371  *   Pointer to Ethernet device.
3372  */
3373 static void
3374 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3375 {
3376 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3377 	struct mlx5_priv *priv = dev->data->dev_private;
3378 
3379 	/* Check if default flow is registered. */
3380 	if (!priv->mreg_cp_tbl)
3381 		return;
3382 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3383 					    MLX5_DEFAULT_COPY_ID);
3384 	if (!mcp_res)
3385 		return;
3386 	MLX5_ASSERT(mcp_res->rix_flow);
3387 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3388 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3389 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3390 }
3391 
3392 /**
3393  * Add the default copy action in in RX_CP_TBL.
3394  *
3395  * @param dev
3396  *   Pointer to Ethernet device.
3397  * @param[out] error
3398  *   Perform verbose error reporting if not NULL.
3399  *
3400  * @return
3401  *   0 for success, negative value otherwise and rte_errno is set.
3402  */
3403 static int
3404 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3405 				  struct rte_flow_error *error)
3406 {
3407 	struct mlx5_priv *priv = dev->data->dev_private;
3408 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3409 
3410 	/* Check whether extensive metadata feature is engaged. */
3411 	if (!priv->config.dv_flow_en ||
3412 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3413 	    !mlx5_flow_ext_mreg_supported(dev) ||
3414 	    !priv->sh->dv_regc0_mask)
3415 		return 0;
3416 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3417 	if (!mcp_res)
3418 		return -rte_errno;
3419 	return 0;
3420 }
3421 
3422 /**
3423  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3424  *
3425  * All the flow having Q/RSS action should be split by
3426  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3427  * performs the following,
3428  *   - CQE->flow_tag := reg_c[1] (MARK)
3429  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3430  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3431  * but there should be a flow per each MARK ID set by MARK action.
3432  *
3433  * For the aforementioned reason, if there's a MARK action in flow's action
3434  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3435  * the MARK ID to CQE's flow_tag like,
3436  *   - If reg_c[1] is mark_id,
3437  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3438  *
3439  * For SET_META action which stores value in reg_c[0], as the destination is
3440  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3441  * MARK ID means the default flow. The default flow looks like,
3442  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3443  *
3444  * @param dev
3445  *   Pointer to Ethernet device.
3446  * @param flow
3447  *   Pointer to flow structure.
3448  * @param[in] actions
3449  *   Pointer to the list of actions.
3450  * @param[out] error
3451  *   Perform verbose error reporting if not NULL.
3452  *
3453  * @return
3454  *   0 on success, negative value otherwise and rte_errno is set.
3455  */
3456 static int
3457 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3458 			    struct rte_flow *flow,
3459 			    const struct rte_flow_action *actions,
3460 			    struct rte_flow_error *error)
3461 {
3462 	struct mlx5_priv *priv = dev->data->dev_private;
3463 	struct mlx5_dev_config *config = &priv->config;
3464 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3465 	const struct rte_flow_action_mark *mark;
3466 
3467 	/* Check whether extensive metadata feature is engaged. */
3468 	if (!config->dv_flow_en ||
3469 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3470 	    !mlx5_flow_ext_mreg_supported(dev) ||
3471 	    !priv->sh->dv_regc0_mask)
3472 		return 0;
3473 	/* Find MARK action. */
3474 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3475 		switch (actions->type) {
3476 		case RTE_FLOW_ACTION_TYPE_FLAG:
3477 			mcp_res = flow_mreg_add_copy_action
3478 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3479 			if (!mcp_res)
3480 				return -rte_errno;
3481 			flow->rix_mreg_copy = mcp_res->idx;
3482 			if (dev->data->dev_started) {
3483 				mcp_res->appcnt++;
3484 				flow->copy_applied = 1;
3485 			}
3486 			return 0;
3487 		case RTE_FLOW_ACTION_TYPE_MARK:
3488 			mark = (const struct rte_flow_action_mark *)
3489 				actions->conf;
3490 			mcp_res =
3491 				flow_mreg_add_copy_action(dev, mark->id, error);
3492 			if (!mcp_res)
3493 				return -rte_errno;
3494 			flow->rix_mreg_copy = mcp_res->idx;
3495 			if (dev->data->dev_started) {
3496 				mcp_res->appcnt++;
3497 				flow->copy_applied = 1;
3498 			}
3499 			return 0;
3500 		default:
3501 			break;
3502 		}
3503 	}
3504 	return 0;
3505 }
3506 
3507 #define MLX5_MAX_SPLIT_ACTIONS 24
3508 #define MLX5_MAX_SPLIT_ITEMS 24
3509 
3510 /**
3511  * Split the hairpin flow.
3512  * Since HW can't support encap on Rx we move the encap to Tx.
3513  * If the count action is after the encap then we also
3514  * move the count action. in this case the count will also measure
3515  * the outer bytes.
3516  *
3517  * @param dev
3518  *   Pointer to Ethernet device.
3519  * @param[in] actions
3520  *   Associated actions (list terminated by the END action).
3521  * @param[out] actions_rx
3522  *   Rx flow actions.
3523  * @param[out] actions_tx
3524  *   Tx flow actions..
3525  * @param[out] pattern_tx
3526  *   The pattern items for the Tx flow.
3527  * @param[out] flow_id
3528  *   The flow ID connected to this flow.
3529  *
3530  * @return
3531  *   0 on success.
3532  */
3533 static int
3534 flow_hairpin_split(struct rte_eth_dev *dev,
3535 		   const struct rte_flow_action actions[],
3536 		   struct rte_flow_action actions_rx[],
3537 		   struct rte_flow_action actions_tx[],
3538 		   struct rte_flow_item pattern_tx[],
3539 		   uint32_t *flow_id)
3540 {
3541 	struct mlx5_priv *priv = dev->data->dev_private;
3542 	const struct rte_flow_action_raw_encap *raw_encap;
3543 	const struct rte_flow_action_raw_decap *raw_decap;
3544 	struct mlx5_rte_flow_action_set_tag *set_tag;
3545 	struct rte_flow_action *tag_action;
3546 	struct mlx5_rte_flow_item_tag *tag_item;
3547 	struct rte_flow_item *item;
3548 	char *addr;
3549 	int encap = 0;
3550 
3551 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3552 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3553 		switch (actions->type) {
3554 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3555 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3556 			rte_memcpy(actions_tx, actions,
3557 			       sizeof(struct rte_flow_action));
3558 			actions_tx++;
3559 			break;
3560 		case RTE_FLOW_ACTION_TYPE_COUNT:
3561 			if (encap) {
3562 				rte_memcpy(actions_tx, actions,
3563 					   sizeof(struct rte_flow_action));
3564 				actions_tx++;
3565 			} else {
3566 				rte_memcpy(actions_rx, actions,
3567 					   sizeof(struct rte_flow_action));
3568 				actions_rx++;
3569 			}
3570 			break;
3571 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3572 			raw_encap = actions->conf;
3573 			if (raw_encap->size >
3574 			    (sizeof(struct rte_flow_item_eth) +
3575 			     sizeof(struct rte_flow_item_ipv4))) {
3576 				memcpy(actions_tx, actions,
3577 				       sizeof(struct rte_flow_action));
3578 				actions_tx++;
3579 				encap = 1;
3580 			} else {
3581 				rte_memcpy(actions_rx, actions,
3582 					   sizeof(struct rte_flow_action));
3583 				actions_rx++;
3584 			}
3585 			break;
3586 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3587 			raw_decap = actions->conf;
3588 			if (raw_decap->size <
3589 			    (sizeof(struct rte_flow_item_eth) +
3590 			     sizeof(struct rte_flow_item_ipv4))) {
3591 				memcpy(actions_tx, actions,
3592 				       sizeof(struct rte_flow_action));
3593 				actions_tx++;
3594 			} else {
3595 				rte_memcpy(actions_rx, actions,
3596 					   sizeof(struct rte_flow_action));
3597 				actions_rx++;
3598 			}
3599 			break;
3600 		default:
3601 			rte_memcpy(actions_rx, actions,
3602 				   sizeof(struct rte_flow_action));
3603 			actions_rx++;
3604 			break;
3605 		}
3606 	}
3607 	/* Add set meta action and end action for the Rx flow. */
3608 	tag_action = actions_rx;
3609 	tag_action->type = (enum rte_flow_action_type)
3610 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3611 	actions_rx++;
3612 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3613 	actions_rx++;
3614 	set_tag = (void *)actions_rx;
3615 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3616 	MLX5_ASSERT(set_tag->id > REG_NONE);
3617 	set_tag->data = *flow_id;
3618 	tag_action->conf = set_tag;
3619 	/* Create Tx item list. */
3620 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3621 	addr = (void *)&pattern_tx[2];
3622 	item = pattern_tx;
3623 	item->type = (enum rte_flow_item_type)
3624 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3625 	tag_item = (void *)addr;
3626 	tag_item->data = *flow_id;
3627 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3628 	MLX5_ASSERT(set_tag->id > REG_NONE);
3629 	item->spec = tag_item;
3630 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3631 	tag_item = (void *)addr;
3632 	tag_item->data = UINT32_MAX;
3633 	tag_item->id = UINT16_MAX;
3634 	item->mask = tag_item;
3635 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3636 	item->last = NULL;
3637 	item++;
3638 	item->type = RTE_FLOW_ITEM_TYPE_END;
3639 	return 0;
3640 }
3641 
3642 /**
3643  * The last stage of splitting chain, just creates the subflow
3644  * without any modification.
3645  *
3646  * @param[in] dev
3647  *   Pointer to Ethernet device.
3648  * @param[in] flow
3649  *   Parent flow structure pointer.
3650  * @param[in, out] sub_flow
3651  *   Pointer to return the created subflow, may be NULL.
3652  * @param[in] prefix_layers
3653  *   Prefix subflow layers, may be 0.
3654  * @param[in] attr
3655  *   Flow rule attributes.
3656  * @param[in] items
3657  *   Pattern specification (list terminated by the END pattern item).
3658  * @param[in] actions
3659  *   Associated actions (list terminated by the END action).
3660  * @param[in] external
3661  *   This flow rule is created by request external to PMD.
3662  * @param[in] flow_idx
3663  *   This memory pool index to the flow.
3664  * @param[out] error
3665  *   Perform verbose error reporting if not NULL.
3666  * @return
3667  *   0 on success, negative value otherwise
3668  */
3669 static int
3670 flow_create_split_inner(struct rte_eth_dev *dev,
3671 			struct rte_flow *flow,
3672 			struct mlx5_flow **sub_flow,
3673 			uint64_t prefix_layers,
3674 			const struct rte_flow_attr *attr,
3675 			const struct rte_flow_item items[],
3676 			const struct rte_flow_action actions[],
3677 			bool external, uint32_t flow_idx,
3678 			struct rte_flow_error *error)
3679 {
3680 	struct mlx5_flow *dev_flow;
3681 
3682 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3683 		flow_idx, error);
3684 	if (!dev_flow)
3685 		return -rte_errno;
3686 	dev_flow->flow = flow;
3687 	dev_flow->external = external;
3688 	/* Subflow object was created, we must include one in the list. */
3689 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3690 		      dev_flow->handle, next);
3691 	/*
3692 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3693 	 * flow may need some user defined item layer flags.
3694 	 */
3695 	if (prefix_layers)
3696 		dev_flow->handle->layers = prefix_layers;
3697 	if (sub_flow)
3698 		*sub_flow = dev_flow;
3699 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3700 }
3701 
3702 /**
3703  * Split the meter flow.
3704  *
3705  * As meter flow will split to three sub flow, other than meter
3706  * action, the other actions make sense to only meter accepts
3707  * the packet. If it need to be dropped, no other additional
3708  * actions should be take.
3709  *
3710  * One kind of special action which decapsulates the L3 tunnel
3711  * header will be in the prefix sub flow, as not to take the
3712  * L3 tunnel header into account.
3713  *
3714  * @param dev
3715  *   Pointer to Ethernet device.
3716  * @param[in] items
3717  *   Pattern specification (list terminated by the END pattern item).
3718  * @param[out] sfx_items
3719  *   Suffix flow match items (list terminated by the END pattern item).
3720  * @param[in] actions
3721  *   Associated actions (list terminated by the END action).
3722  * @param[out] actions_sfx
3723  *   Suffix flow actions.
3724  * @param[out] actions_pre
3725  *   Prefix flow actions.
3726  * @param[out] pattern_sfx
3727  *   The pattern items for the suffix flow.
3728  * @param[out] tag_sfx
3729  *   Pointer to suffix flow tag.
3730  *
3731  * @return
3732  *   0 on success.
3733  */
3734 static int
3735 flow_meter_split_prep(struct rte_eth_dev *dev,
3736 		 const struct rte_flow_item items[],
3737 		 struct rte_flow_item sfx_items[],
3738 		 const struct rte_flow_action actions[],
3739 		 struct rte_flow_action actions_sfx[],
3740 		 struct rte_flow_action actions_pre[])
3741 {
3742 	struct rte_flow_action *tag_action = NULL;
3743 	struct rte_flow_item *tag_item;
3744 	struct mlx5_rte_flow_action_set_tag *set_tag;
3745 	struct rte_flow_error error;
3746 	const struct rte_flow_action_raw_encap *raw_encap;
3747 	const struct rte_flow_action_raw_decap *raw_decap;
3748 	struct mlx5_rte_flow_item_tag *tag_spec;
3749 	struct mlx5_rte_flow_item_tag *tag_mask;
3750 	uint32_t tag_id;
3751 	bool copy_vlan = false;
3752 
3753 	/* Prepare the actions for prefix and suffix flow. */
3754 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3755 		struct rte_flow_action **action_cur = NULL;
3756 
3757 		switch (actions->type) {
3758 		case RTE_FLOW_ACTION_TYPE_METER:
3759 			/* Add the extra tag action first. */
3760 			tag_action = actions_pre;
3761 			tag_action->type = (enum rte_flow_action_type)
3762 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3763 			actions_pre++;
3764 			action_cur = &actions_pre;
3765 			break;
3766 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3767 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3768 			action_cur = &actions_pre;
3769 			break;
3770 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3771 			raw_encap = actions->conf;
3772 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3773 				action_cur = &actions_pre;
3774 			break;
3775 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3776 			raw_decap = actions->conf;
3777 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3778 				action_cur = &actions_pre;
3779 			break;
3780 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3781 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3782 			copy_vlan = true;
3783 			break;
3784 		default:
3785 			break;
3786 		}
3787 		if (!action_cur)
3788 			action_cur = &actions_sfx;
3789 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3790 		(*action_cur)++;
3791 	}
3792 	/* Add end action to the actions. */
3793 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3794 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3795 	actions_pre++;
3796 	/* Set the tag. */
3797 	set_tag = (void *)actions_pre;
3798 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3799 	/*
3800 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3801 	 */
3802 	tag_id = flow_qrss_get_id(dev);
3803 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3804 	assert(tag_action);
3805 	tag_action->conf = set_tag;
3806 	/* Prepare the suffix subflow items. */
3807 	tag_item = sfx_items++;
3808 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3809 		int item_type = items->type;
3810 
3811 		switch (item_type) {
3812 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3813 			memcpy(sfx_items, items, sizeof(*sfx_items));
3814 			sfx_items++;
3815 			break;
3816 		case RTE_FLOW_ITEM_TYPE_VLAN:
3817 			if (copy_vlan) {
3818 				memcpy(sfx_items, items, sizeof(*sfx_items));
3819 				/*
3820 				 * Convert to internal match item, it is used
3821 				 * for vlan push and set vid.
3822 				 */
3823 				sfx_items->type = (enum rte_flow_item_type)
3824 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3825 				sfx_items++;
3826 			}
3827 			break;
3828 		default:
3829 			break;
3830 		}
3831 	}
3832 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3833 	sfx_items++;
3834 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3835 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3836 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3837 	tag_mask = tag_spec + 1;
3838 	tag_mask->data = 0xffffff00;
3839 	tag_item->type = (enum rte_flow_item_type)
3840 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3841 	tag_item->spec = tag_spec;
3842 	tag_item->last = NULL;
3843 	tag_item->mask = tag_mask;
3844 	return tag_id;
3845 }
3846 
3847 /**
3848  * Split action list having QUEUE/RSS for metadata register copy.
3849  *
3850  * Once Q/RSS action is detected in user's action list, the flow action
3851  * should be split in order to copy metadata registers, which will happen in
3852  * RX_CP_TBL like,
3853  *   - CQE->flow_tag := reg_c[1] (MARK)
3854  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3855  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3856  * This is because the last action of each flow must be a terminal action
3857  * (QUEUE, RSS or DROP).
3858  *
3859  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3860  * stored and kept in the mlx5_flow structure per each sub_flow.
3861  *
3862  * The Q/RSS action is replaced with,
3863  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3864  * And the following JUMP action is added at the end,
3865  *   - JUMP, to RX_CP_TBL.
3866  *
3867  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3868  * flow_create_split_metadata() routine. The flow will look like,
3869  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3870  *
3871  * @param dev
3872  *   Pointer to Ethernet device.
3873  * @param[out] split_actions
3874  *   Pointer to store split actions to jump to CP_TBL.
3875  * @param[in] actions
3876  *   Pointer to the list of original flow actions.
3877  * @param[in] qrss
3878  *   Pointer to the Q/RSS action.
3879  * @param[in] actions_n
3880  *   Number of original actions.
3881  * @param[out] error
3882  *   Perform verbose error reporting if not NULL.
3883  *
3884  * @return
3885  *   non-zero unique flow_id on success, otherwise 0 and
3886  *   error/rte_error are set.
3887  */
3888 static uint32_t
3889 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3890 			  struct rte_flow_action *split_actions,
3891 			  const struct rte_flow_action *actions,
3892 			  const struct rte_flow_action *qrss,
3893 			  int actions_n, struct rte_flow_error *error)
3894 {
3895 	struct mlx5_rte_flow_action_set_tag *set_tag;
3896 	struct rte_flow_action_jump *jump;
3897 	const int qrss_idx = qrss - actions;
3898 	uint32_t flow_id = 0;
3899 	int ret = 0;
3900 
3901 	/*
3902 	 * Given actions will be split
3903 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3904 	 * - Add jump to mreg CP_TBL.
3905 	 * As a result, there will be one more action.
3906 	 */
3907 	++actions_n;
3908 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3909 	set_tag = (void *)(split_actions + actions_n);
3910 	/*
3911 	 * If tag action is not set to void(it means we are not the meter
3912 	 * suffix flow), add the tag action. Since meter suffix flow already
3913 	 * has the tag added.
3914 	 */
3915 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3916 		/*
3917 		 * Allocate the new subflow ID. This one is unique within
3918 		 * device and not shared with representors. Otherwise,
3919 		 * we would have to resolve multi-thread access synch
3920 		 * issue. Each flow on the shared device is appended
3921 		 * with source vport identifier, so the resulting
3922 		 * flows will be unique in the shared (by master and
3923 		 * representors) domain even if they have coinciding
3924 		 * IDs.
3925 		 */
3926 		flow_id = flow_qrss_get_id(dev);
3927 		if (!flow_id)
3928 			return rte_flow_error_set(error, ENOMEM,
3929 						  RTE_FLOW_ERROR_TYPE_ACTION,
3930 						  NULL, "can't allocate id "
3931 						  "for split Q/RSS subflow");
3932 		/* Internal SET_TAG action to set flow ID. */
3933 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3934 			.data = flow_id,
3935 		};
3936 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3937 		if (ret < 0)
3938 			return ret;
3939 		set_tag->id = ret;
3940 		/* Construct new actions array. */
3941 		/* Replace QUEUE/RSS action. */
3942 		split_actions[qrss_idx] = (struct rte_flow_action){
3943 			.type = (enum rte_flow_action_type)
3944 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3945 			.conf = set_tag,
3946 		};
3947 	}
3948 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3949 	jump = (void *)(set_tag + 1);
3950 	*jump = (struct rte_flow_action_jump){
3951 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3952 	};
3953 	split_actions[actions_n - 2] = (struct rte_flow_action){
3954 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3955 		.conf = jump,
3956 	};
3957 	split_actions[actions_n - 1] = (struct rte_flow_action){
3958 		.type = RTE_FLOW_ACTION_TYPE_END,
3959 	};
3960 	return flow_id;
3961 }
3962 
3963 /**
3964  * Extend the given action list for Tx metadata copy.
3965  *
3966  * Copy the given action list to the ext_actions and add flow metadata register
3967  * copy action in order to copy reg_a set by WQE to reg_c[0].
3968  *
3969  * @param[out] ext_actions
3970  *   Pointer to the extended action list.
3971  * @param[in] actions
3972  *   Pointer to the list of actions.
3973  * @param[in] actions_n
3974  *   Number of actions in the list.
3975  * @param[out] error
3976  *   Perform verbose error reporting if not NULL.
3977  * @param[in] encap_idx
3978  *   The encap action inndex.
3979  *
3980  * @return
3981  *   0 on success, negative value otherwise
3982  */
3983 static int
3984 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3985 		       struct rte_flow_action *ext_actions,
3986 		       const struct rte_flow_action *actions,
3987 		       int actions_n, struct rte_flow_error *error,
3988 		       int encap_idx)
3989 {
3990 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3991 		(struct mlx5_flow_action_copy_mreg *)
3992 			(ext_actions + actions_n + 1);
3993 	int ret;
3994 
3995 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3996 	if (ret < 0)
3997 		return ret;
3998 	cp_mreg->dst = ret;
3999 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4000 	if (ret < 0)
4001 		return ret;
4002 	cp_mreg->src = ret;
4003 	if (encap_idx != 0)
4004 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4005 	if (encap_idx == actions_n - 1) {
4006 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4007 			.type = (enum rte_flow_action_type)
4008 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4009 			.conf = cp_mreg,
4010 		};
4011 		ext_actions[actions_n] = (struct rte_flow_action){
4012 			.type = RTE_FLOW_ACTION_TYPE_END,
4013 		};
4014 	} else {
4015 		ext_actions[encap_idx] = (struct rte_flow_action){
4016 			.type = (enum rte_flow_action_type)
4017 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4018 			.conf = cp_mreg,
4019 		};
4020 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4021 				sizeof(*ext_actions) * (actions_n - encap_idx));
4022 	}
4023 	return 0;
4024 }
4025 
4026 /**
4027  * The splitting for metadata feature.
4028  *
4029  * - Q/RSS action on NIC Rx should be split in order to pass by
4030  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4031  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4032  *
4033  * - All the actions on NIC Tx should have a mreg copy action to
4034  *   copy reg_a from WQE to reg_c[0].
4035  *
4036  * @param dev
4037  *   Pointer to Ethernet device.
4038  * @param[in] flow
4039  *   Parent flow structure pointer.
4040  * @param[in] prefix_layers
4041  *   Prefix flow layer flags.
4042  * @param[in] attr
4043  *   Flow rule attributes.
4044  * @param[in] items
4045  *   Pattern specification (list terminated by the END pattern item).
4046  * @param[in] actions
4047  *   Associated actions (list terminated by the END action).
4048  * @param[in] external
4049  *   This flow rule is created by request external to PMD.
4050  * @param[in] flow_idx
4051  *   This memory pool index to the flow.
4052  * @param[out] error
4053  *   Perform verbose error reporting if not NULL.
4054  * @return
4055  *   0 on success, negative value otherwise
4056  */
4057 static int
4058 flow_create_split_metadata(struct rte_eth_dev *dev,
4059 			   struct rte_flow *flow,
4060 			   uint64_t prefix_layers,
4061 			   const struct rte_flow_attr *attr,
4062 			   const struct rte_flow_item items[],
4063 			   const struct rte_flow_action actions[],
4064 			   bool external, uint32_t flow_idx,
4065 			   struct rte_flow_error *error)
4066 {
4067 	struct mlx5_priv *priv = dev->data->dev_private;
4068 	struct mlx5_dev_config *config = &priv->config;
4069 	const struct rte_flow_action *qrss = NULL;
4070 	struct rte_flow_action *ext_actions = NULL;
4071 	struct mlx5_flow *dev_flow = NULL;
4072 	uint32_t qrss_id = 0;
4073 	int mtr_sfx = 0;
4074 	size_t act_size;
4075 	int actions_n;
4076 	int encap_idx;
4077 	int ret;
4078 
4079 	/* Check whether extensive metadata feature is engaged. */
4080 	if (!config->dv_flow_en ||
4081 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4082 	    !mlx5_flow_ext_mreg_supported(dev))
4083 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
4084 					       attr, items, actions, external,
4085 					       flow_idx, error);
4086 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4087 							   &encap_idx);
4088 	if (qrss) {
4089 		/* Exclude hairpin flows from splitting. */
4090 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4091 			const struct rte_flow_action_queue *queue;
4092 
4093 			queue = qrss->conf;
4094 			if (mlx5_rxq_get_type(dev, queue->index) ==
4095 			    MLX5_RXQ_TYPE_HAIRPIN)
4096 				qrss = NULL;
4097 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4098 			const struct rte_flow_action_rss *rss;
4099 
4100 			rss = qrss->conf;
4101 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4102 			    MLX5_RXQ_TYPE_HAIRPIN)
4103 				qrss = NULL;
4104 		}
4105 	}
4106 	if (qrss) {
4107 		/* Check if it is in meter suffix table. */
4108 		mtr_sfx = attr->group == (attr->transfer ?
4109 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4110 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4111 		/*
4112 		 * Q/RSS action on NIC Rx should be split in order to pass by
4113 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4114 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4115 		 */
4116 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4117 			   sizeof(struct rte_flow_action_set_tag) +
4118 			   sizeof(struct rte_flow_action_jump);
4119 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4120 					  SOCKET_ID_ANY);
4121 		if (!ext_actions)
4122 			return rte_flow_error_set(error, ENOMEM,
4123 						  RTE_FLOW_ERROR_TYPE_ACTION,
4124 						  NULL, "no memory to split "
4125 						  "metadata flow");
4126 		/*
4127 		 * If we are the suffix flow of meter, tag already exist.
4128 		 * Set the tag action to void.
4129 		 */
4130 		if (mtr_sfx)
4131 			ext_actions[qrss - actions].type =
4132 						RTE_FLOW_ACTION_TYPE_VOID;
4133 		else
4134 			ext_actions[qrss - actions].type =
4135 						(enum rte_flow_action_type)
4136 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4137 		/*
4138 		 * Create the new actions list with removed Q/RSS action
4139 		 * and appended set tag and jump to register copy table
4140 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4141 		 * in advance, because it is needed for set tag action.
4142 		 */
4143 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4144 						    qrss, actions_n, error);
4145 		if (!mtr_sfx && !qrss_id) {
4146 			ret = -rte_errno;
4147 			goto exit;
4148 		}
4149 	} else if (attr->egress && !attr->transfer) {
4150 		/*
4151 		 * All the actions on NIC Tx should have a metadata register
4152 		 * copy action to copy reg_a from WQE to reg_c[meta]
4153 		 */
4154 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4155 			   sizeof(struct mlx5_flow_action_copy_mreg);
4156 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4157 					  SOCKET_ID_ANY);
4158 		if (!ext_actions)
4159 			return rte_flow_error_set(error, ENOMEM,
4160 						  RTE_FLOW_ERROR_TYPE_ACTION,
4161 						  NULL, "no memory to split "
4162 						  "metadata flow");
4163 		/* Create the action list appended with copy register. */
4164 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4165 					     actions_n, error, encap_idx);
4166 		if (ret < 0)
4167 			goto exit;
4168 	}
4169 	/* Add the unmodified original or prefix subflow. */
4170 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4171 				      items, ext_actions ? ext_actions :
4172 				      actions, external, flow_idx, error);
4173 	if (ret < 0)
4174 		goto exit;
4175 	MLX5_ASSERT(dev_flow);
4176 	if (qrss) {
4177 		const struct rte_flow_attr q_attr = {
4178 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4179 			.ingress = 1,
4180 		};
4181 		/* Internal PMD action to set register. */
4182 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4183 			.data = qrss_id,
4184 			.id = 0,
4185 		};
4186 		struct rte_flow_item q_items[] = {
4187 			{
4188 				.type = (enum rte_flow_item_type)
4189 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4190 				.spec = &q_tag_spec,
4191 				.last = NULL,
4192 				.mask = NULL,
4193 			},
4194 			{
4195 				.type = RTE_FLOW_ITEM_TYPE_END,
4196 			},
4197 		};
4198 		struct rte_flow_action q_actions[] = {
4199 			{
4200 				.type = qrss->type,
4201 				.conf = qrss->conf,
4202 			},
4203 			{
4204 				.type = RTE_FLOW_ACTION_TYPE_END,
4205 			},
4206 		};
4207 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4208 
4209 		/*
4210 		 * Configure the tag item only if there is no meter subflow.
4211 		 * Since tag is already marked in the meter suffix subflow
4212 		 * we can just use the meter suffix items as is.
4213 		 */
4214 		if (qrss_id) {
4215 			/* Not meter subflow. */
4216 			MLX5_ASSERT(!mtr_sfx);
4217 			/*
4218 			 * Put unique id in prefix flow due to it is destroyed
4219 			 * after suffix flow and id will be freed after there
4220 			 * is no actual flows with this id and identifier
4221 			 * reallocation becomes possible (for example, for
4222 			 * other flows in other threads).
4223 			 */
4224 			dev_flow->handle->split_flow_id = qrss_id;
4225 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4226 						   error);
4227 			if (ret < 0)
4228 				goto exit;
4229 			q_tag_spec.id = ret;
4230 		}
4231 		dev_flow = NULL;
4232 		/* Add suffix subflow to execute Q/RSS. */
4233 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4234 					      &q_attr, mtr_sfx ? items :
4235 					      q_items, q_actions,
4236 					      external, flow_idx, error);
4237 		if (ret < 0)
4238 			goto exit;
4239 		/* qrss ID should be freed if failed. */
4240 		qrss_id = 0;
4241 		MLX5_ASSERT(dev_flow);
4242 	}
4243 
4244 exit:
4245 	/*
4246 	 * We do not destroy the partially created sub_flows in case of error.
4247 	 * These ones are included into parent flow list and will be destroyed
4248 	 * by flow_drv_destroy.
4249 	 */
4250 	flow_qrss_free_id(dev, qrss_id);
4251 	mlx5_free(ext_actions);
4252 	return ret;
4253 }
4254 
4255 /**
4256  * The splitting for meter feature.
4257  *
4258  * - The meter flow will be split to two flows as prefix and
4259  *   suffix flow. The packets make sense only it pass the prefix
4260  *   meter action.
4261  *
4262  * - Reg_C_5 is used for the packet to match betweend prefix and
4263  *   suffix flow.
4264  *
4265  * @param dev
4266  *   Pointer to Ethernet device.
4267  * @param[in] flow
4268  *   Parent flow structure pointer.
4269  * @param[in] attr
4270  *   Flow rule attributes.
4271  * @param[in] items
4272  *   Pattern specification (list terminated by the END pattern item).
4273  * @param[in] actions
4274  *   Associated actions (list terminated by the END action).
4275  * @param[in] external
4276  *   This flow rule is created by request external to PMD.
4277  * @param[in] flow_idx
4278  *   This memory pool index to the flow.
4279  * @param[out] error
4280  *   Perform verbose error reporting if not NULL.
4281  * @return
4282  *   0 on success, negative value otherwise
4283  */
4284 static int
4285 flow_create_split_meter(struct rte_eth_dev *dev,
4286 			   struct rte_flow *flow,
4287 			   const struct rte_flow_attr *attr,
4288 			   const struct rte_flow_item items[],
4289 			   const struct rte_flow_action actions[],
4290 			   bool external, uint32_t flow_idx,
4291 			   struct rte_flow_error *error)
4292 {
4293 	struct mlx5_priv *priv = dev->data->dev_private;
4294 	struct rte_flow_action *sfx_actions = NULL;
4295 	struct rte_flow_action *pre_actions = NULL;
4296 	struct rte_flow_item *sfx_items = NULL;
4297 	struct mlx5_flow *dev_flow = NULL;
4298 	struct rte_flow_attr sfx_attr = *attr;
4299 	uint32_t mtr = 0;
4300 	uint32_t mtr_tag_id = 0;
4301 	size_t act_size;
4302 	size_t item_size;
4303 	int actions_n = 0;
4304 	int ret;
4305 
4306 	if (priv->mtr_en)
4307 		actions_n = flow_check_meter_action(actions, &mtr);
4308 	if (mtr) {
4309 		/* The five prefix actions: meter, decap, encap, tag, end. */
4310 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4311 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4312 		/* tag, vlan, port id, end. */
4313 #define METER_SUFFIX_ITEM 4
4314 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4315 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4316 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4317 					  0, SOCKET_ID_ANY);
4318 		if (!sfx_actions)
4319 			return rte_flow_error_set(error, ENOMEM,
4320 						  RTE_FLOW_ERROR_TYPE_ACTION,
4321 						  NULL, "no memory to split "
4322 						  "meter flow");
4323 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4324 			     act_size);
4325 		pre_actions = sfx_actions + actions_n;
4326 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4327 						   actions, sfx_actions,
4328 						   pre_actions);
4329 		if (!mtr_tag_id) {
4330 			ret = -rte_errno;
4331 			goto exit;
4332 		}
4333 		/* Add the prefix subflow. */
4334 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4335 					      items, pre_actions, external,
4336 					      flow_idx, error);
4337 		if (ret) {
4338 			ret = -rte_errno;
4339 			goto exit;
4340 		}
4341 		dev_flow->handle->split_flow_id = mtr_tag_id;
4342 		/* Setting the sfx group atrr. */
4343 		sfx_attr.group = sfx_attr.transfer ?
4344 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4345 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4346 	}
4347 	/* Add the prefix subflow. */
4348 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4349 					 flow_get_prefix_layer_flags(dev_flow) :
4350 					 0, &sfx_attr,
4351 					 sfx_items ? sfx_items : items,
4352 					 sfx_actions ? sfx_actions : actions,
4353 					 external, flow_idx, error);
4354 exit:
4355 	if (sfx_actions)
4356 		mlx5_free(sfx_actions);
4357 	return ret;
4358 }
4359 
4360 /**
4361  * Split the flow to subflow set. The splitters might be linked
4362  * in the chain, like this:
4363  * flow_create_split_outer() calls:
4364  *   flow_create_split_meter() calls:
4365  *     flow_create_split_metadata(meter_subflow_0) calls:
4366  *       flow_create_split_inner(metadata_subflow_0)
4367  *       flow_create_split_inner(metadata_subflow_1)
4368  *       flow_create_split_inner(metadata_subflow_2)
4369  *     flow_create_split_metadata(meter_subflow_1) calls:
4370  *       flow_create_split_inner(metadata_subflow_0)
4371  *       flow_create_split_inner(metadata_subflow_1)
4372  *       flow_create_split_inner(metadata_subflow_2)
4373  *
4374  * This provide flexible way to add new levels of flow splitting.
4375  * The all of successfully created subflows are included to the
4376  * parent flow dev_flow list.
4377  *
4378  * @param dev
4379  *   Pointer to Ethernet device.
4380  * @param[in] flow
4381  *   Parent flow structure pointer.
4382  * @param[in] attr
4383  *   Flow rule attributes.
4384  * @param[in] items
4385  *   Pattern specification (list terminated by the END pattern item).
4386  * @param[in] actions
4387  *   Associated actions (list terminated by the END action).
4388  * @param[in] external
4389  *   This flow rule is created by request external to PMD.
4390  * @param[in] flow_idx
4391  *   This memory pool index to the flow.
4392  * @param[out] error
4393  *   Perform verbose error reporting if not NULL.
4394  * @return
4395  *   0 on success, negative value otherwise
4396  */
4397 static int
4398 flow_create_split_outer(struct rte_eth_dev *dev,
4399 			struct rte_flow *flow,
4400 			const struct rte_flow_attr *attr,
4401 			const struct rte_flow_item items[],
4402 			const struct rte_flow_action actions[],
4403 			bool external, uint32_t flow_idx,
4404 			struct rte_flow_error *error)
4405 {
4406 	int ret;
4407 
4408 	ret = flow_create_split_meter(dev, flow, attr, items,
4409 					 actions, external, flow_idx, error);
4410 	MLX5_ASSERT(ret <= 0);
4411 	return ret;
4412 }
4413 
4414 /**
4415  * Create a flow and add it to @p list.
4416  *
4417  * @param dev
4418  *   Pointer to Ethernet device.
4419  * @param list
4420  *   Pointer to a TAILQ flow list. If this parameter NULL,
4421  *   no list insertion occurred, flow is just created,
4422  *   this is caller's responsibility to track the
4423  *   created flow.
4424  * @param[in] attr
4425  *   Flow rule attributes.
4426  * @param[in] items
4427  *   Pattern specification (list terminated by the END pattern item).
4428  * @param[in] actions
4429  *   Associated actions (list terminated by the END action).
4430  * @param[in] external
4431  *   This flow rule is created by request external to PMD.
4432  * @param[out] error
4433  *   Perform verbose error reporting if not NULL.
4434  *
4435  * @return
4436  *   A flow index on success, 0 otherwise and rte_errno is set.
4437  */
4438 static uint32_t
4439 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4440 		 const struct rte_flow_attr *attr,
4441 		 const struct rte_flow_item items[],
4442 		 const struct rte_flow_action actions[],
4443 		 bool external, struct rte_flow_error *error)
4444 {
4445 	struct mlx5_priv *priv = dev->data->dev_private;
4446 	struct rte_flow *flow = NULL;
4447 	struct mlx5_flow *dev_flow;
4448 	const struct rte_flow_action_rss *rss;
4449 	union {
4450 		struct rte_flow_expand_rss buf;
4451 		uint8_t buffer[2048];
4452 	} expand_buffer;
4453 	union {
4454 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4455 		uint8_t buffer[2048];
4456 	} actions_rx;
4457 	union {
4458 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4459 		uint8_t buffer[2048];
4460 	} actions_hairpin_tx;
4461 	union {
4462 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4463 		uint8_t buffer[2048];
4464 	} items_tx;
4465 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4466 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4467 					      priv->rss_desc)[!!priv->flow_idx];
4468 	const struct rte_flow_action *p_actions_rx = actions;
4469 	uint32_t i;
4470 	uint32_t idx = 0;
4471 	int hairpin_flow;
4472 	uint32_t hairpin_id = 0;
4473 	struct rte_flow_attr attr_tx = { .priority = 0 };
4474 	int ret;
4475 
4476 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4477 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4478 				external, hairpin_flow, error);
4479 	if (ret < 0)
4480 		return 0;
4481 	if (hairpin_flow > 0) {
4482 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4483 			rte_errno = EINVAL;
4484 			return 0;
4485 		}
4486 		flow_hairpin_split(dev, actions, actions_rx.actions,
4487 				   actions_hairpin_tx.actions, items_tx.items,
4488 				   &hairpin_id);
4489 		p_actions_rx = actions_rx.actions;
4490 	}
4491 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4492 	if (!flow) {
4493 		rte_errno = ENOMEM;
4494 		goto error_before_flow;
4495 	}
4496 	flow->drv_type = flow_get_drv_type(dev, attr);
4497 	if (hairpin_id != 0)
4498 		flow->hairpin_flow_id = hairpin_id;
4499 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4500 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4501 	memset(rss_desc, 0, sizeof(*rss_desc));
4502 	rss = flow_get_rss_action(p_actions_rx);
4503 	if (rss) {
4504 		/*
4505 		 * The following information is required by
4506 		 * mlx5_flow_hashfields_adjust() in advance.
4507 		 */
4508 		rss_desc->level = rss->level;
4509 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4510 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4511 	}
4512 	flow->dev_handles = 0;
4513 	if (rss && rss->types) {
4514 		unsigned int graph_root;
4515 
4516 		graph_root = find_graph_root(items, rss->level);
4517 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4518 					  items, rss->types,
4519 					  mlx5_support_expansion,
4520 					  graph_root);
4521 		MLX5_ASSERT(ret > 0 &&
4522 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4523 	} else {
4524 		buf->entries = 1;
4525 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4526 	}
4527 	/*
4528 	 * Record the start index when there is a nested call. All sub-flows
4529 	 * need to be translated before another calling.
4530 	 * No need to use ping-pong buffer to save memory here.
4531 	 */
4532 	if (priv->flow_idx) {
4533 		MLX5_ASSERT(!priv->flow_nested_idx);
4534 		priv->flow_nested_idx = priv->flow_idx;
4535 	}
4536 	for (i = 0; i < buf->entries; ++i) {
4537 		/*
4538 		 * The splitter may create multiple dev_flows,
4539 		 * depending on configuration. In the simplest
4540 		 * case it just creates unmodified original flow.
4541 		 */
4542 		ret = flow_create_split_outer(dev, flow, attr,
4543 					      buf->entry[i].pattern,
4544 					      p_actions_rx, external, idx,
4545 					      error);
4546 		if (ret < 0)
4547 			goto error;
4548 	}
4549 	/* Create the tx flow. */
4550 	if (hairpin_flow) {
4551 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4552 		attr_tx.ingress = 0;
4553 		attr_tx.egress = 1;
4554 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4555 					 actions_hairpin_tx.actions,
4556 					 idx, error);
4557 		if (!dev_flow)
4558 			goto error;
4559 		dev_flow->flow = flow;
4560 		dev_flow->external = 0;
4561 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4562 			      dev_flow->handle, next);
4563 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4564 					 items_tx.items,
4565 					 actions_hairpin_tx.actions, error);
4566 		if (ret < 0)
4567 			goto error;
4568 	}
4569 	/*
4570 	 * Update the metadata register copy table. If extensive
4571 	 * metadata feature is enabled and registers are supported
4572 	 * we might create the extra rte_flow for each unique
4573 	 * MARK/FLAG action ID.
4574 	 *
4575 	 * The table is updated for ingress Flows only, because
4576 	 * the egress Flows belong to the different device and
4577 	 * copy table should be updated in peer NIC Rx domain.
4578 	 */
4579 	if (attr->ingress &&
4580 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4581 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4582 		if (ret)
4583 			goto error;
4584 	}
4585 	/*
4586 	 * If the flow is external (from application) OR device is started, then
4587 	 * the flow will be applied immediately.
4588 	 */
4589 	if (external || dev->data->dev_started) {
4590 		ret = flow_drv_apply(dev, flow, error);
4591 		if (ret < 0)
4592 			goto error;
4593 	}
4594 	if (list)
4595 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4596 			     flow, next);
4597 	flow_rxq_flags_set(dev, flow);
4598 	/* Nested flow creation index recovery. */
4599 	priv->flow_idx = priv->flow_nested_idx;
4600 	if (priv->flow_nested_idx)
4601 		priv->flow_nested_idx = 0;
4602 	return idx;
4603 error:
4604 	MLX5_ASSERT(flow);
4605 	ret = rte_errno; /* Save rte_errno before cleanup. */
4606 	flow_mreg_del_copy_action(dev, flow);
4607 	flow_drv_destroy(dev, flow);
4608 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4609 	rte_errno = ret; /* Restore rte_errno. */
4610 error_before_flow:
4611 	ret = rte_errno;
4612 	if (hairpin_id)
4613 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4614 				     hairpin_id);
4615 	rte_errno = ret;
4616 	priv->flow_idx = priv->flow_nested_idx;
4617 	if (priv->flow_nested_idx)
4618 		priv->flow_nested_idx = 0;
4619 	return 0;
4620 }
4621 
4622 /**
4623  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4624  * incoming packets to table 1.
4625  *
4626  * Other flow rules, requested for group n, will be created in
4627  * e-switch table n+1.
4628  * Jump action to e-switch group n will be created to group n+1.
4629  *
4630  * Used when working in switchdev mode, to utilise advantages of table 1
4631  * and above.
4632  *
4633  * @param dev
4634  *   Pointer to Ethernet device.
4635  *
4636  * @return
4637  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4638  */
4639 struct rte_flow *
4640 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4641 {
4642 	const struct rte_flow_attr attr = {
4643 		.group = 0,
4644 		.priority = 0,
4645 		.ingress = 1,
4646 		.egress = 0,
4647 		.transfer = 1,
4648 	};
4649 	const struct rte_flow_item pattern = {
4650 		.type = RTE_FLOW_ITEM_TYPE_END,
4651 	};
4652 	struct rte_flow_action_jump jump = {
4653 		.group = 1,
4654 	};
4655 	const struct rte_flow_action actions[] = {
4656 		{
4657 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4658 			.conf = &jump,
4659 		},
4660 		{
4661 			.type = RTE_FLOW_ACTION_TYPE_END,
4662 		},
4663 	};
4664 	struct mlx5_priv *priv = dev->data->dev_private;
4665 	struct rte_flow_error error;
4666 
4667 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4668 						   &attr, &pattern,
4669 						   actions, false, &error);
4670 }
4671 
4672 /**
4673  * Validate a flow supported by the NIC.
4674  *
4675  * @see rte_flow_validate()
4676  * @see rte_flow_ops
4677  */
4678 int
4679 mlx5_flow_validate(struct rte_eth_dev *dev,
4680 		   const struct rte_flow_attr *attr,
4681 		   const struct rte_flow_item items[],
4682 		   const struct rte_flow_action actions[],
4683 		   struct rte_flow_error *error)
4684 {
4685 	int hairpin_flow;
4686 
4687 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4688 	return flow_drv_validate(dev, attr, items, actions,
4689 				true, hairpin_flow, error);
4690 }
4691 
4692 /**
4693  * Create a flow.
4694  *
4695  * @see rte_flow_create()
4696  * @see rte_flow_ops
4697  */
4698 struct rte_flow *
4699 mlx5_flow_create(struct rte_eth_dev *dev,
4700 		 const struct rte_flow_attr *attr,
4701 		 const struct rte_flow_item items[],
4702 		 const struct rte_flow_action actions[],
4703 		 struct rte_flow_error *error)
4704 {
4705 	struct mlx5_priv *priv = dev->data->dev_private;
4706 
4707 	/*
4708 	 * If the device is not started yet, it is not allowed to created a
4709 	 * flow from application. PMD default flows and traffic control flows
4710 	 * are not affected.
4711 	 */
4712 	if (unlikely(!dev->data->dev_started)) {
4713 		DRV_LOG(DEBUG, "port %u is not started when "
4714 			"inserting a flow", dev->data->port_id);
4715 		rte_flow_error_set(error, ENODEV,
4716 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4717 				   NULL,
4718 				   "port not started");
4719 		return NULL;
4720 	}
4721 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4722 				  attr, items, actions, true, error);
4723 }
4724 
4725 /**
4726  * Destroy a flow in a list.
4727  *
4728  * @param dev
4729  *   Pointer to Ethernet device.
4730  * @param list
4731  *   Pointer to the Indexed flow list. If this parameter NULL,
4732  *   there is no flow removal from the list. Be noted that as
4733  *   flow is add to the indexed list, memory of the indexed
4734  *   list points to maybe changed as flow destroyed.
4735  * @param[in] flow_idx
4736  *   Index of flow to destroy.
4737  */
4738 static void
4739 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4740 		  uint32_t flow_idx)
4741 {
4742 	struct mlx5_priv *priv = dev->data->dev_private;
4743 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4744 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4745 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4746 
4747 	if (!flow)
4748 		return;
4749 	/*
4750 	 * Update RX queue flags only if port is started, otherwise it is
4751 	 * already clean.
4752 	 */
4753 	if (dev->data->dev_started)
4754 		flow_rxq_flags_trim(dev, flow);
4755 	if (flow->hairpin_flow_id)
4756 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4757 				     flow->hairpin_flow_id);
4758 	flow_drv_destroy(dev, flow);
4759 	if (list)
4760 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4761 			     flow_idx, flow, next);
4762 	flow_mreg_del_copy_action(dev, flow);
4763 	if (flow->fdir) {
4764 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4765 			if (priv_fdir_flow->rix_flow == flow_idx)
4766 				break;
4767 		}
4768 		if (priv_fdir_flow) {
4769 			LIST_REMOVE(priv_fdir_flow, next);
4770 			mlx5_free(priv_fdir_flow->fdir);
4771 			mlx5_free(priv_fdir_flow);
4772 		}
4773 	}
4774 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4775 }
4776 
4777 /**
4778  * Destroy all flows.
4779  *
4780  * @param dev
4781  *   Pointer to Ethernet device.
4782  * @param list
4783  *   Pointer to the Indexed flow list.
4784  * @param active
4785  *   If flushing is called avtively.
4786  */
4787 void
4788 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4789 {
4790 	uint32_t num_flushed = 0;
4791 
4792 	while (*list) {
4793 		flow_list_destroy(dev, list, *list);
4794 		num_flushed++;
4795 	}
4796 	if (active) {
4797 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4798 			dev->data->port_id, num_flushed);
4799 	}
4800 }
4801 
4802 /**
4803  * Remove all flows.
4804  *
4805  * @param dev
4806  *   Pointer to Ethernet device.
4807  * @param list
4808  *   Pointer to the Indexed flow list.
4809  */
4810 void
4811 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4812 {
4813 	struct mlx5_priv *priv = dev->data->dev_private;
4814 	struct rte_flow *flow = NULL;
4815 	uint32_t idx;
4816 
4817 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4818 		      flow, next) {
4819 		flow_drv_remove(dev, flow);
4820 		flow_mreg_stop_copy_action(dev, flow);
4821 	}
4822 	flow_mreg_del_default_copy_action(dev);
4823 	flow_rxq_flags_clear(dev);
4824 }
4825 
4826 /**
4827  * Add all flows.
4828  *
4829  * @param dev
4830  *   Pointer to Ethernet device.
4831  * @param list
4832  *   Pointer to the Indexed flow list.
4833  *
4834  * @return
4835  *   0 on success, a negative errno value otherwise and rte_errno is set.
4836  */
4837 int
4838 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4839 {
4840 	struct mlx5_priv *priv = dev->data->dev_private;
4841 	struct rte_flow *flow = NULL;
4842 	struct rte_flow_error error;
4843 	uint32_t idx;
4844 	int ret = 0;
4845 
4846 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4847 	ret = flow_mreg_add_default_copy_action(dev, &error);
4848 	if (ret < 0)
4849 		return -rte_errno;
4850 	/* Apply Flows created by application. */
4851 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4852 		      flow, next) {
4853 		ret = flow_mreg_start_copy_action(dev, flow);
4854 		if (ret < 0)
4855 			goto error;
4856 		ret = flow_drv_apply(dev, flow, &error);
4857 		if (ret < 0)
4858 			goto error;
4859 		flow_rxq_flags_set(dev, flow);
4860 	}
4861 	return 0;
4862 error:
4863 	ret = rte_errno; /* Save rte_errno before cleanup. */
4864 	mlx5_flow_stop(dev, list);
4865 	rte_errno = ret; /* Restore rte_errno. */
4866 	return -rte_errno;
4867 }
4868 
4869 /**
4870  * Stop all default actions for flows.
4871  *
4872  * @param dev
4873  *   Pointer to Ethernet device.
4874  */
4875 void
4876 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4877 {
4878 	flow_mreg_del_default_copy_action(dev);
4879 	flow_rxq_flags_clear(dev);
4880 }
4881 
4882 /**
4883  * Start all default actions for flows.
4884  *
4885  * @param dev
4886  *   Pointer to Ethernet device.
4887  * @return
4888  *   0 on success, a negative errno value otherwise and rte_errno is set.
4889  */
4890 int
4891 mlx5_flow_start_default(struct rte_eth_dev *dev)
4892 {
4893 	struct rte_flow_error error;
4894 
4895 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4896 	return flow_mreg_add_default_copy_action(dev, &error);
4897 }
4898 
4899 /**
4900  * Allocate intermediate resources for flow creation.
4901  *
4902  * @param dev
4903  *   Pointer to Ethernet device.
4904  */
4905 void
4906 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4907 {
4908 	struct mlx5_priv *priv = dev->data->dev_private;
4909 
4910 	if (!priv->inter_flows) {
4911 		priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
4912 				    MLX5_NUM_MAX_DEV_FLOWS *
4913 				    sizeof(struct mlx5_flow) +
4914 				    (sizeof(struct mlx5_flow_rss_desc) +
4915 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0,
4916 				    SOCKET_ID_ANY);
4917 		if (!priv->inter_flows) {
4918 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4919 			return;
4920 		}
4921 	}
4922 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4923 			 [MLX5_NUM_MAX_DEV_FLOWS];
4924 	/* Reset the index. */
4925 	priv->flow_idx = 0;
4926 	priv->flow_nested_idx = 0;
4927 }
4928 
4929 /**
4930  * Free intermediate resources for flows.
4931  *
4932  * @param dev
4933  *   Pointer to Ethernet device.
4934  */
4935 void
4936 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4937 {
4938 	struct mlx5_priv *priv = dev->data->dev_private;
4939 
4940 	mlx5_free(priv->inter_flows);
4941 	priv->inter_flows = NULL;
4942 }
4943 
4944 /**
4945  * Verify the flow list is empty
4946  *
4947  * @param dev
4948  *  Pointer to Ethernet device.
4949  *
4950  * @return the number of flows not released.
4951  */
4952 int
4953 mlx5_flow_verify(struct rte_eth_dev *dev)
4954 {
4955 	struct mlx5_priv *priv = dev->data->dev_private;
4956 	struct rte_flow *flow;
4957 	uint32_t idx;
4958 	int ret = 0;
4959 
4960 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4961 		      flow, next) {
4962 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4963 			dev->data->port_id, (void *)flow);
4964 		++ret;
4965 	}
4966 	return ret;
4967 }
4968 
4969 /**
4970  * Enable default hairpin egress flow.
4971  *
4972  * @param dev
4973  *   Pointer to Ethernet device.
4974  * @param queue
4975  *   The queue index.
4976  *
4977  * @return
4978  *   0 on success, a negative errno value otherwise and rte_errno is set.
4979  */
4980 int
4981 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4982 			    uint32_t queue)
4983 {
4984 	struct mlx5_priv *priv = dev->data->dev_private;
4985 	const struct rte_flow_attr attr = {
4986 		.egress = 1,
4987 		.priority = 0,
4988 	};
4989 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4990 		.queue = queue,
4991 	};
4992 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4993 		.queue = UINT32_MAX,
4994 	};
4995 	struct rte_flow_item items[] = {
4996 		{
4997 			.type = (enum rte_flow_item_type)
4998 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4999 			.spec = &queue_spec,
5000 			.last = NULL,
5001 			.mask = &queue_mask,
5002 		},
5003 		{
5004 			.type = RTE_FLOW_ITEM_TYPE_END,
5005 		},
5006 	};
5007 	struct rte_flow_action_jump jump = {
5008 		.group = MLX5_HAIRPIN_TX_TABLE,
5009 	};
5010 	struct rte_flow_action actions[2];
5011 	uint32_t flow_idx;
5012 	struct rte_flow_error error;
5013 
5014 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
5015 	actions[0].conf = &jump;
5016 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
5017 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5018 				&attr, items, actions, false, &error);
5019 	if (!flow_idx) {
5020 		DRV_LOG(DEBUG,
5021 			"Failed to create ctrl flow: rte_errno(%d),"
5022 			" type(%d), message(%s)",
5023 			rte_errno, error.type,
5024 			error.message ? error.message : " (no stated reason)");
5025 		return -rte_errno;
5026 	}
5027 	return 0;
5028 }
5029 
5030 /**
5031  * Enable a control flow configured from the control plane.
5032  *
5033  * @param dev
5034  *   Pointer to Ethernet device.
5035  * @param eth_spec
5036  *   An Ethernet flow spec to apply.
5037  * @param eth_mask
5038  *   An Ethernet flow mask to apply.
5039  * @param vlan_spec
5040  *   A VLAN flow spec to apply.
5041  * @param vlan_mask
5042  *   A VLAN flow mask to apply.
5043  *
5044  * @return
5045  *   0 on success, a negative errno value otherwise and rte_errno is set.
5046  */
5047 int
5048 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
5049 		    struct rte_flow_item_eth *eth_spec,
5050 		    struct rte_flow_item_eth *eth_mask,
5051 		    struct rte_flow_item_vlan *vlan_spec,
5052 		    struct rte_flow_item_vlan *vlan_mask)
5053 {
5054 	struct mlx5_priv *priv = dev->data->dev_private;
5055 	const struct rte_flow_attr attr = {
5056 		.ingress = 1,
5057 		.priority = MLX5_FLOW_PRIO_RSVD,
5058 	};
5059 	struct rte_flow_item items[] = {
5060 		{
5061 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5062 			.spec = eth_spec,
5063 			.last = NULL,
5064 			.mask = eth_mask,
5065 		},
5066 		{
5067 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
5068 					      RTE_FLOW_ITEM_TYPE_END,
5069 			.spec = vlan_spec,
5070 			.last = NULL,
5071 			.mask = vlan_mask,
5072 		},
5073 		{
5074 			.type = RTE_FLOW_ITEM_TYPE_END,
5075 		},
5076 	};
5077 	uint16_t queue[priv->reta_idx_n];
5078 	struct rte_flow_action_rss action_rss = {
5079 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
5080 		.level = 0,
5081 		.types = priv->rss_conf.rss_hf,
5082 		.key_len = priv->rss_conf.rss_key_len,
5083 		.queue_num = priv->reta_idx_n,
5084 		.key = priv->rss_conf.rss_key,
5085 		.queue = queue,
5086 	};
5087 	struct rte_flow_action actions[] = {
5088 		{
5089 			.type = RTE_FLOW_ACTION_TYPE_RSS,
5090 			.conf = &action_rss,
5091 		},
5092 		{
5093 			.type = RTE_FLOW_ACTION_TYPE_END,
5094 		},
5095 	};
5096 	uint32_t flow_idx;
5097 	struct rte_flow_error error;
5098 	unsigned int i;
5099 
5100 	if (!priv->reta_idx_n || !priv->rxqs_n) {
5101 		return 0;
5102 	}
5103 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
5104 		action_rss.types = 0;
5105 	for (i = 0; i != priv->reta_idx_n; ++i)
5106 		queue[i] = (*priv->reta_idx)[i];
5107 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5108 				&attr, items, actions, false, &error);
5109 	if (!flow_idx)
5110 		return -rte_errno;
5111 	return 0;
5112 }
5113 
5114 /**
5115  * Enable a flow control configured from the control plane.
5116  *
5117  * @param dev
5118  *   Pointer to Ethernet device.
5119  * @param eth_spec
5120  *   An Ethernet flow spec to apply.
5121  * @param eth_mask
5122  *   An Ethernet flow mask to apply.
5123  *
5124  * @return
5125  *   0 on success, a negative errno value otherwise and rte_errno is set.
5126  */
5127 int
5128 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5129 	       struct rte_flow_item_eth *eth_spec,
5130 	       struct rte_flow_item_eth *eth_mask)
5131 {
5132 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5133 }
5134 
5135 /**
5136  * Create default miss flow rule matching lacp traffic
5137  *
5138  * @param dev
5139  *   Pointer to Ethernet device.
5140  * @param eth_spec
5141  *   An Ethernet flow spec to apply.
5142  *
5143  * @return
5144  *   0 on success, a negative errno value otherwise and rte_errno is set.
5145  */
5146 int
5147 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5148 {
5149 	struct mlx5_priv *priv = dev->data->dev_private;
5150 	/*
5151 	 * The LACP matching is done by only using ether type since using
5152 	 * a multicast dst mac causes kernel to give low priority to this flow.
5153 	 */
5154 	static const struct rte_flow_item_eth lacp_spec = {
5155 		.type = RTE_BE16(0x8809),
5156 	};
5157 	static const struct rte_flow_item_eth lacp_mask = {
5158 		.type = 0xffff,
5159 	};
5160 	const struct rte_flow_attr attr = {
5161 		.ingress = 1,
5162 	};
5163 	struct rte_flow_item items[] = {
5164 		{
5165 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5166 			.spec = &lacp_spec,
5167 			.mask = &lacp_mask,
5168 		},
5169 		{
5170 			.type = RTE_FLOW_ITEM_TYPE_END,
5171 		},
5172 	};
5173 	struct rte_flow_action actions[] = {
5174 		{
5175 			.type = (enum rte_flow_action_type)
5176 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5177 		},
5178 		{
5179 			.type = RTE_FLOW_ACTION_TYPE_END,
5180 		},
5181 	};
5182 	struct rte_flow_error error;
5183 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5184 				&attr, items, actions, false, &error);
5185 
5186 	if (!flow_idx)
5187 		return -rte_errno;
5188 	return 0;
5189 }
5190 
5191 /**
5192  * Destroy a flow.
5193  *
5194  * @see rte_flow_destroy()
5195  * @see rte_flow_ops
5196  */
5197 int
5198 mlx5_flow_destroy(struct rte_eth_dev *dev,
5199 		  struct rte_flow *flow,
5200 		  struct rte_flow_error *error __rte_unused)
5201 {
5202 	struct mlx5_priv *priv = dev->data->dev_private;
5203 
5204 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5205 	return 0;
5206 }
5207 
5208 /**
5209  * Destroy all flows.
5210  *
5211  * @see rte_flow_flush()
5212  * @see rte_flow_ops
5213  */
5214 int
5215 mlx5_flow_flush(struct rte_eth_dev *dev,
5216 		struct rte_flow_error *error __rte_unused)
5217 {
5218 	struct mlx5_priv *priv = dev->data->dev_private;
5219 
5220 	mlx5_flow_list_flush(dev, &priv->flows, false);
5221 	return 0;
5222 }
5223 
5224 /**
5225  * Isolated mode.
5226  *
5227  * @see rte_flow_isolate()
5228  * @see rte_flow_ops
5229  */
5230 int
5231 mlx5_flow_isolate(struct rte_eth_dev *dev,
5232 		  int enable,
5233 		  struct rte_flow_error *error)
5234 {
5235 	struct mlx5_priv *priv = dev->data->dev_private;
5236 
5237 	if (dev->data->dev_started) {
5238 		rte_flow_error_set(error, EBUSY,
5239 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5240 				   NULL,
5241 				   "port must be stopped first");
5242 		return -rte_errno;
5243 	}
5244 	priv->isolated = !!enable;
5245 	if (enable)
5246 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
5247 	else
5248 		dev->dev_ops = &mlx5_os_dev_ops;
5249 	return 0;
5250 }
5251 
5252 /**
5253  * Query a flow.
5254  *
5255  * @see rte_flow_query()
5256  * @see rte_flow_ops
5257  */
5258 static int
5259 flow_drv_query(struct rte_eth_dev *dev,
5260 	       uint32_t flow_idx,
5261 	       const struct rte_flow_action *actions,
5262 	       void *data,
5263 	       struct rte_flow_error *error)
5264 {
5265 	struct mlx5_priv *priv = dev->data->dev_private;
5266 	const struct mlx5_flow_driver_ops *fops;
5267 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5268 					       [MLX5_IPOOL_RTE_FLOW],
5269 					       flow_idx);
5270 	enum mlx5_flow_drv_type ftype;
5271 
5272 	if (!flow) {
5273 		return rte_flow_error_set(error, ENOENT,
5274 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5275 			  NULL,
5276 			  "invalid flow handle");
5277 	}
5278 	ftype = flow->drv_type;
5279 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5280 	fops = flow_get_drv_ops(ftype);
5281 
5282 	return fops->query(dev, flow, actions, data, error);
5283 }
5284 
5285 /**
5286  * Query a flow.
5287  *
5288  * @see rte_flow_query()
5289  * @see rte_flow_ops
5290  */
5291 int
5292 mlx5_flow_query(struct rte_eth_dev *dev,
5293 		struct rte_flow *flow,
5294 		const struct rte_flow_action *actions,
5295 		void *data,
5296 		struct rte_flow_error *error)
5297 {
5298 	int ret;
5299 
5300 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5301 			     error);
5302 	if (ret < 0)
5303 		return ret;
5304 	return 0;
5305 }
5306 
5307 /**
5308  * Convert a flow director filter to a generic flow.
5309  *
5310  * @param dev
5311  *   Pointer to Ethernet device.
5312  * @param fdir_filter
5313  *   Flow director filter to add.
5314  * @param attributes
5315  *   Generic flow parameters structure.
5316  *
5317  * @return
5318  *   0 on success, a negative errno value otherwise and rte_errno is set.
5319  */
5320 static int
5321 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5322 			 const struct rte_eth_fdir_filter *fdir_filter,
5323 			 struct mlx5_fdir *attributes)
5324 {
5325 	struct mlx5_priv *priv = dev->data->dev_private;
5326 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5327 	const struct rte_eth_fdir_masks *mask =
5328 		&dev->data->dev_conf.fdir_conf.mask;
5329 
5330 	/* Validate queue number. */
5331 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5332 		DRV_LOG(ERR, "port %u invalid queue number %d",
5333 			dev->data->port_id, fdir_filter->action.rx_queue);
5334 		rte_errno = EINVAL;
5335 		return -rte_errno;
5336 	}
5337 	attributes->attr.ingress = 1;
5338 	attributes->items[0] = (struct rte_flow_item) {
5339 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5340 		.spec = &attributes->l2,
5341 		.mask = &attributes->l2_mask,
5342 	};
5343 	switch (fdir_filter->action.behavior) {
5344 	case RTE_ETH_FDIR_ACCEPT:
5345 		attributes->actions[0] = (struct rte_flow_action){
5346 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5347 			.conf = &attributes->queue,
5348 		};
5349 		break;
5350 	case RTE_ETH_FDIR_REJECT:
5351 		attributes->actions[0] = (struct rte_flow_action){
5352 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5353 		};
5354 		break;
5355 	default:
5356 		DRV_LOG(ERR, "port %u invalid behavior %d",
5357 			dev->data->port_id,
5358 			fdir_filter->action.behavior);
5359 		rte_errno = ENOTSUP;
5360 		return -rte_errno;
5361 	}
5362 	attributes->queue.index = fdir_filter->action.rx_queue;
5363 	/* Handle L3. */
5364 	switch (fdir_filter->input.flow_type) {
5365 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5366 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5367 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5368 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5369 			.src_addr = input->flow.ip4_flow.src_ip,
5370 			.dst_addr = input->flow.ip4_flow.dst_ip,
5371 			.time_to_live = input->flow.ip4_flow.ttl,
5372 			.type_of_service = input->flow.ip4_flow.tos,
5373 		};
5374 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5375 			.src_addr = mask->ipv4_mask.src_ip,
5376 			.dst_addr = mask->ipv4_mask.dst_ip,
5377 			.time_to_live = mask->ipv4_mask.ttl,
5378 			.type_of_service = mask->ipv4_mask.tos,
5379 			.next_proto_id = mask->ipv4_mask.proto,
5380 		};
5381 		attributes->items[1] = (struct rte_flow_item){
5382 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5383 			.spec = &attributes->l3,
5384 			.mask = &attributes->l3_mask,
5385 		};
5386 		break;
5387 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5388 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5389 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5390 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5391 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5392 			.proto = input->flow.ipv6_flow.proto,
5393 		};
5394 
5395 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5396 		       input->flow.ipv6_flow.src_ip,
5397 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5398 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5399 		       input->flow.ipv6_flow.dst_ip,
5400 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5401 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5402 		       mask->ipv6_mask.src_ip,
5403 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5404 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5405 		       mask->ipv6_mask.dst_ip,
5406 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5407 		attributes->items[1] = (struct rte_flow_item){
5408 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5409 			.spec = &attributes->l3,
5410 			.mask = &attributes->l3_mask,
5411 		};
5412 		break;
5413 	default:
5414 		DRV_LOG(ERR, "port %u invalid flow type%d",
5415 			dev->data->port_id, fdir_filter->input.flow_type);
5416 		rte_errno = ENOTSUP;
5417 		return -rte_errno;
5418 	}
5419 	/* Handle L4. */
5420 	switch (fdir_filter->input.flow_type) {
5421 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5422 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5423 			.src_port = input->flow.udp4_flow.src_port,
5424 			.dst_port = input->flow.udp4_flow.dst_port,
5425 		};
5426 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5427 			.src_port = mask->src_port_mask,
5428 			.dst_port = mask->dst_port_mask,
5429 		};
5430 		attributes->items[2] = (struct rte_flow_item){
5431 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5432 			.spec = &attributes->l4,
5433 			.mask = &attributes->l4_mask,
5434 		};
5435 		break;
5436 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5437 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5438 			.src_port = input->flow.tcp4_flow.src_port,
5439 			.dst_port = input->flow.tcp4_flow.dst_port,
5440 		};
5441 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5442 			.src_port = mask->src_port_mask,
5443 			.dst_port = mask->dst_port_mask,
5444 		};
5445 		attributes->items[2] = (struct rte_flow_item){
5446 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5447 			.spec = &attributes->l4,
5448 			.mask = &attributes->l4_mask,
5449 		};
5450 		break;
5451 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5452 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5453 			.src_port = input->flow.udp6_flow.src_port,
5454 			.dst_port = input->flow.udp6_flow.dst_port,
5455 		};
5456 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5457 			.src_port = mask->src_port_mask,
5458 			.dst_port = mask->dst_port_mask,
5459 		};
5460 		attributes->items[2] = (struct rte_flow_item){
5461 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5462 			.spec = &attributes->l4,
5463 			.mask = &attributes->l4_mask,
5464 		};
5465 		break;
5466 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5467 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5468 			.src_port = input->flow.tcp6_flow.src_port,
5469 			.dst_port = input->flow.tcp6_flow.dst_port,
5470 		};
5471 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5472 			.src_port = mask->src_port_mask,
5473 			.dst_port = mask->dst_port_mask,
5474 		};
5475 		attributes->items[2] = (struct rte_flow_item){
5476 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5477 			.spec = &attributes->l4,
5478 			.mask = &attributes->l4_mask,
5479 		};
5480 		break;
5481 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5482 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5483 		break;
5484 	default:
5485 		DRV_LOG(ERR, "port %u invalid flow type%d",
5486 			dev->data->port_id, fdir_filter->input.flow_type);
5487 		rte_errno = ENOTSUP;
5488 		return -rte_errno;
5489 	}
5490 	return 0;
5491 }
5492 
5493 #define FLOW_FDIR_CMP(f1, f2, fld) \
5494 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5495 
5496 /**
5497  * Compare two FDIR flows. If items and actions are identical, the two flows are
5498  * regarded as same.
5499  *
5500  * @param dev
5501  *   Pointer to Ethernet device.
5502  * @param f1
5503  *   FDIR flow to compare.
5504  * @param f2
5505  *   FDIR flow to compare.
5506  *
5507  * @return
5508  *   Zero on match, 1 otherwise.
5509  */
5510 static int
5511 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5512 {
5513 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5514 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5515 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5516 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5517 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5518 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5519 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5520 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5521 		return 1;
5522 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5523 	    FLOW_FDIR_CMP(f1, f2, queue))
5524 		return 1;
5525 	return 0;
5526 }
5527 
5528 /**
5529  * Search device flow list to find out a matched FDIR flow.
5530  *
5531  * @param dev
5532  *   Pointer to Ethernet device.
5533  * @param fdir_flow
5534  *   FDIR flow to lookup.
5535  *
5536  * @return
5537  *   Index of flow if found, 0 otherwise.
5538  */
5539 static uint32_t
5540 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5541 {
5542 	struct mlx5_priv *priv = dev->data->dev_private;
5543 	uint32_t flow_idx = 0;
5544 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5545 
5546 	MLX5_ASSERT(fdir_flow);
5547 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5548 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5549 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5550 				dev->data->port_id, flow_idx);
5551 			flow_idx = priv_fdir_flow->rix_flow;
5552 			break;
5553 		}
5554 	}
5555 	return flow_idx;
5556 }
5557 
5558 /**
5559  * Add new flow director filter and store it in list.
5560  *
5561  * @param dev
5562  *   Pointer to Ethernet device.
5563  * @param fdir_filter
5564  *   Flow director filter to add.
5565  *
5566  * @return
5567  *   0 on success, a negative errno value otherwise and rte_errno is set.
5568  */
5569 static int
5570 flow_fdir_filter_add(struct rte_eth_dev *dev,
5571 		     const struct rte_eth_fdir_filter *fdir_filter)
5572 {
5573 	struct mlx5_priv *priv = dev->data->dev_private;
5574 	struct mlx5_fdir *fdir_flow;
5575 	struct rte_flow *flow;
5576 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5577 	uint32_t flow_idx;
5578 	int ret;
5579 
5580 	fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
5581 				SOCKET_ID_ANY);
5582 	if (!fdir_flow) {
5583 		rte_errno = ENOMEM;
5584 		return -rte_errno;
5585 	}
5586 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5587 	if (ret)
5588 		goto error;
5589 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5590 	if (flow_idx) {
5591 		rte_errno = EEXIST;
5592 		goto error;
5593 	}
5594 	priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
5595 				     sizeof(struct mlx5_fdir_flow),
5596 				     0, SOCKET_ID_ANY);
5597 	if (!priv_fdir_flow) {
5598 		rte_errno = ENOMEM;
5599 		goto error;
5600 	}
5601 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5602 				    fdir_flow->items, fdir_flow->actions, true,
5603 				    NULL);
5604 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5605 	if (!flow)
5606 		goto error;
5607 	flow->fdir = 1;
5608 	priv_fdir_flow->fdir = fdir_flow;
5609 	priv_fdir_flow->rix_flow = flow_idx;
5610 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5611 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5612 		dev->data->port_id, (void *)flow);
5613 	return 0;
5614 error:
5615 	mlx5_free(priv_fdir_flow);
5616 	mlx5_free(fdir_flow);
5617 	return -rte_errno;
5618 }
5619 
5620 /**
5621  * Delete specific filter.
5622  *
5623  * @param dev
5624  *   Pointer to Ethernet device.
5625  * @param fdir_filter
5626  *   Filter to be deleted.
5627  *
5628  * @return
5629  *   0 on success, a negative errno value otherwise and rte_errno is set.
5630  */
5631 static int
5632 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5633 			const struct rte_eth_fdir_filter *fdir_filter)
5634 {
5635 	struct mlx5_priv *priv = dev->data->dev_private;
5636 	uint32_t flow_idx;
5637 	struct mlx5_fdir fdir_flow = {
5638 		.attr.group = 0,
5639 	};
5640 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5641 	int ret;
5642 
5643 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5644 	if (ret)
5645 		return -rte_errno;
5646 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5647 		/* Find the fdir in priv list */
5648 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5649 			break;
5650 	}
5651 	if (!priv_fdir_flow)
5652 		return 0;
5653 	LIST_REMOVE(priv_fdir_flow, next);
5654 	flow_idx = priv_fdir_flow->rix_flow;
5655 	flow_list_destroy(dev, &priv->flows, flow_idx);
5656 	mlx5_free(priv_fdir_flow->fdir);
5657 	mlx5_free(priv_fdir_flow);
5658 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5659 		dev->data->port_id, flow_idx);
5660 	return 0;
5661 }
5662 
5663 /**
5664  * Update queue for specific filter.
5665  *
5666  * @param dev
5667  *   Pointer to Ethernet device.
5668  * @param fdir_filter
5669  *   Filter to be updated.
5670  *
5671  * @return
5672  *   0 on success, a negative errno value otherwise and rte_errno is set.
5673  */
5674 static int
5675 flow_fdir_filter_update(struct rte_eth_dev *dev,
5676 			const struct rte_eth_fdir_filter *fdir_filter)
5677 {
5678 	int ret;
5679 
5680 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5681 	if (ret)
5682 		return ret;
5683 	return flow_fdir_filter_add(dev, fdir_filter);
5684 }
5685 
5686 /**
5687  * Flush all filters.
5688  *
5689  * @param dev
5690  *   Pointer to Ethernet device.
5691  */
5692 static void
5693 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5694 {
5695 	struct mlx5_priv *priv = dev->data->dev_private;
5696 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5697 
5698 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5699 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5700 		LIST_REMOVE(priv_fdir_flow, next);
5701 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5702 		mlx5_free(priv_fdir_flow->fdir);
5703 		mlx5_free(priv_fdir_flow);
5704 	}
5705 }
5706 
5707 /**
5708  * Get flow director information.
5709  *
5710  * @param dev
5711  *   Pointer to Ethernet device.
5712  * @param[out] fdir_info
5713  *   Resulting flow director information.
5714  */
5715 static void
5716 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5717 {
5718 	struct rte_eth_fdir_masks *mask =
5719 		&dev->data->dev_conf.fdir_conf.mask;
5720 
5721 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5722 	fdir_info->guarant_spc = 0;
5723 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5724 	fdir_info->max_flexpayload = 0;
5725 	fdir_info->flow_types_mask[0] = 0;
5726 	fdir_info->flex_payload_unit = 0;
5727 	fdir_info->max_flex_payload_segment_num = 0;
5728 	fdir_info->flex_payload_limit = 0;
5729 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5730 }
5731 
5732 /**
5733  * Deal with flow director operations.
5734  *
5735  * @param dev
5736  *   Pointer to Ethernet device.
5737  * @param filter_op
5738  *   Operation to perform.
5739  * @param arg
5740  *   Pointer to operation-specific structure.
5741  *
5742  * @return
5743  *   0 on success, a negative errno value otherwise and rte_errno is set.
5744  */
5745 static int
5746 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5747 		    void *arg)
5748 {
5749 	enum rte_fdir_mode fdir_mode =
5750 		dev->data->dev_conf.fdir_conf.mode;
5751 
5752 	if (filter_op == RTE_ETH_FILTER_NOP)
5753 		return 0;
5754 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5755 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5756 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5757 			dev->data->port_id, fdir_mode);
5758 		rte_errno = EINVAL;
5759 		return -rte_errno;
5760 	}
5761 	switch (filter_op) {
5762 	case RTE_ETH_FILTER_ADD:
5763 		return flow_fdir_filter_add(dev, arg);
5764 	case RTE_ETH_FILTER_UPDATE:
5765 		return flow_fdir_filter_update(dev, arg);
5766 	case RTE_ETH_FILTER_DELETE:
5767 		return flow_fdir_filter_delete(dev, arg);
5768 	case RTE_ETH_FILTER_FLUSH:
5769 		flow_fdir_filter_flush(dev);
5770 		break;
5771 	case RTE_ETH_FILTER_INFO:
5772 		flow_fdir_info_get(dev, arg);
5773 		break;
5774 	default:
5775 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5776 			dev->data->port_id, filter_op);
5777 		rte_errno = EINVAL;
5778 		return -rte_errno;
5779 	}
5780 	return 0;
5781 }
5782 
5783 /**
5784  * Manage filter operations.
5785  *
5786  * @param dev
5787  *   Pointer to Ethernet device structure.
5788  * @param filter_type
5789  *   Filter type.
5790  * @param filter_op
5791  *   Operation to perform.
5792  * @param arg
5793  *   Pointer to operation-specific structure.
5794  *
5795  * @return
5796  *   0 on success, a negative errno value otherwise and rte_errno is set.
5797  */
5798 int
5799 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5800 		     enum rte_filter_type filter_type,
5801 		     enum rte_filter_op filter_op,
5802 		     void *arg)
5803 {
5804 	switch (filter_type) {
5805 	case RTE_ETH_FILTER_GENERIC:
5806 		if (filter_op != RTE_ETH_FILTER_GET) {
5807 			rte_errno = EINVAL;
5808 			return -rte_errno;
5809 		}
5810 		*(const void **)arg = &mlx5_flow_ops;
5811 		return 0;
5812 	case RTE_ETH_FILTER_FDIR:
5813 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5814 	default:
5815 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5816 			dev->data->port_id, filter_type);
5817 		rte_errno = ENOTSUP;
5818 		return -rte_errno;
5819 	}
5820 	return 0;
5821 }
5822 
5823 /**
5824  * Create the needed meter and suffix tables.
5825  *
5826  * @param[in] dev
5827  *   Pointer to Ethernet device.
5828  * @param[in] fm
5829  *   Pointer to the flow meter.
5830  *
5831  * @return
5832  *   Pointer to table set on success, NULL otherwise.
5833  */
5834 struct mlx5_meter_domains_infos *
5835 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5836 			  const struct mlx5_flow_meter *fm)
5837 {
5838 	const struct mlx5_flow_driver_ops *fops;
5839 
5840 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5841 	return fops->create_mtr_tbls(dev, fm);
5842 }
5843 
5844 /**
5845  * Destroy the meter table set.
5846  *
5847  * @param[in] dev
5848  *   Pointer to Ethernet device.
5849  * @param[in] tbl
5850  *   Pointer to the meter table set.
5851  *
5852  * @return
5853  *   0 on success.
5854  */
5855 int
5856 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5857 			   struct mlx5_meter_domains_infos *tbls)
5858 {
5859 	const struct mlx5_flow_driver_ops *fops;
5860 
5861 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5862 	return fops->destroy_mtr_tbls(dev, tbls);
5863 }
5864 
5865 /**
5866  * Create policer rules.
5867  *
5868  * @param[in] dev
5869  *   Pointer to Ethernet device.
5870  * @param[in] fm
5871  *   Pointer to flow meter structure.
5872  * @param[in] attr
5873  *   Pointer to flow attributes.
5874  *
5875  * @return
5876  *   0 on success, -1 otherwise.
5877  */
5878 int
5879 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5880 			       struct mlx5_flow_meter *fm,
5881 			       const struct rte_flow_attr *attr)
5882 {
5883 	const struct mlx5_flow_driver_ops *fops;
5884 
5885 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5886 	return fops->create_policer_rules(dev, fm, attr);
5887 }
5888 
5889 /**
5890  * Destroy policer rules.
5891  *
5892  * @param[in] fm
5893  *   Pointer to flow meter structure.
5894  * @param[in] attr
5895  *   Pointer to flow attributes.
5896  *
5897  * @return
5898  *   0 on success, -1 otherwise.
5899  */
5900 int
5901 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5902 				struct mlx5_flow_meter *fm,
5903 				const struct rte_flow_attr *attr)
5904 {
5905 	const struct mlx5_flow_driver_ops *fops;
5906 
5907 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5908 	return fops->destroy_policer_rules(dev, fm, attr);
5909 }
5910 
5911 /**
5912  * Allocate a counter.
5913  *
5914  * @param[in] dev
5915  *   Pointer to Ethernet device structure.
5916  *
5917  * @return
5918  *   Index to allocated counter  on success, 0 otherwise.
5919  */
5920 uint32_t
5921 mlx5_counter_alloc(struct rte_eth_dev *dev)
5922 {
5923 	const struct mlx5_flow_driver_ops *fops;
5924 	struct rte_flow_attr attr = { .transfer = 0 };
5925 
5926 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5927 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5928 		return fops->counter_alloc(dev);
5929 	}
5930 	DRV_LOG(ERR,
5931 		"port %u counter allocate is not supported.",
5932 		 dev->data->port_id);
5933 	return 0;
5934 }
5935 
5936 /**
5937  * Free a counter.
5938  *
5939  * @param[in] dev
5940  *   Pointer to Ethernet device structure.
5941  * @param[in] cnt
5942  *   Index to counter to be free.
5943  */
5944 void
5945 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5946 {
5947 	const struct mlx5_flow_driver_ops *fops;
5948 	struct rte_flow_attr attr = { .transfer = 0 };
5949 
5950 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5951 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5952 		fops->counter_free(dev, cnt);
5953 		return;
5954 	}
5955 	DRV_LOG(ERR,
5956 		"port %u counter free is not supported.",
5957 		 dev->data->port_id);
5958 }
5959 
5960 /**
5961  * Query counter statistics.
5962  *
5963  * @param[in] dev
5964  *   Pointer to Ethernet device structure.
5965  * @param[in] cnt
5966  *   Index to counter to query.
5967  * @param[in] clear
5968  *   Set to clear counter statistics.
5969  * @param[out] pkts
5970  *   The counter hits packets number to save.
5971  * @param[out] bytes
5972  *   The counter hits bytes number to save.
5973  *
5974  * @return
5975  *   0 on success, a negative errno value otherwise.
5976  */
5977 int
5978 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5979 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5980 {
5981 	const struct mlx5_flow_driver_ops *fops;
5982 	struct rte_flow_attr attr = { .transfer = 0 };
5983 
5984 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5985 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5986 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5987 	}
5988 	DRV_LOG(ERR,
5989 		"port %u counter query is not supported.",
5990 		 dev->data->port_id);
5991 	return -ENOTSUP;
5992 }
5993 
5994 #define MLX5_POOL_QUERY_FREQ_US 1000000
5995 
5996 /**
5997  * Get number of all validate pools.
5998  *
5999  * @param[in] sh
6000  *   Pointer to mlx5_dev_ctx_shared object.
6001  *
6002  * @return
6003  *   The number of all validate pools.
6004  */
6005 static uint32_t
6006 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
6007 {
6008 	int i;
6009 	uint32_t pools_n = 0;
6010 
6011 	for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
6012 		pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
6013 	return pools_n;
6014 }
6015 
6016 /**
6017  * Set the periodic procedure for triggering asynchronous batch queries for all
6018  * the counter pools.
6019  *
6020  * @param[in] sh
6021  *   Pointer to mlx5_dev_ctx_shared object.
6022  */
6023 void
6024 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6025 {
6026 	uint32_t pools_n, us;
6027 
6028 	pools_n = mlx5_get_all_valid_pool_count(sh);
6029 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6030 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6031 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6032 		sh->cmng.query_thread_on = 0;
6033 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6034 	} else {
6035 		sh->cmng.query_thread_on = 1;
6036 	}
6037 }
6038 
6039 /**
6040  * The periodic procedure for triggering asynchronous batch queries for all the
6041  * counter pools. This function is probably called by the host thread.
6042  *
6043  * @param[in] arg
6044  *   The parameter for the alarm process.
6045  */
6046 void
6047 mlx5_flow_query_alarm(void *arg)
6048 {
6049 	struct mlx5_dev_ctx_shared *sh = arg;
6050 	struct mlx5_devx_obj *dcs;
6051 	uint16_t offset;
6052 	int ret;
6053 	uint8_t batch = sh->cmng.batch;
6054 	uint8_t age = sh->cmng.age;
6055 	uint16_t pool_index = sh->cmng.pool_index;
6056 	struct mlx5_pools_container *cont;
6057 	struct mlx5_flow_counter_pool *pool;
6058 	int cont_loop = MLX5_CCONT_TYPE_MAX;
6059 
6060 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6061 		goto set_alarm;
6062 next_container:
6063 	cont = MLX5_CNT_CONTAINER(sh, batch, age);
6064 	rte_spinlock_lock(&cont->resize_sl);
6065 	if (!cont->pools) {
6066 		rte_spinlock_unlock(&cont->resize_sl);
6067 		/* Check if all the containers are empty. */
6068 		if (unlikely(--cont_loop == 0))
6069 			goto set_alarm;
6070 		batch ^= 0x1;
6071 		pool_index = 0;
6072 		if (batch == 0 && pool_index == 0) {
6073 			age ^= 0x1;
6074 			sh->cmng.batch = batch;
6075 			sh->cmng.age = age;
6076 		}
6077 		goto next_container;
6078 	}
6079 	pool = cont->pools[pool_index];
6080 	rte_spinlock_unlock(&cont->resize_sl);
6081 	if (pool->raw_hw)
6082 		/* There is a pool query in progress. */
6083 		goto set_alarm;
6084 	pool->raw_hw =
6085 		LIST_FIRST(&sh->cmng.free_stat_raws);
6086 	if (!pool->raw_hw)
6087 		/* No free counter statistics raw memory. */
6088 		goto set_alarm;
6089 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
6090 							      (&pool->a64_dcs);
6091 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
6092 	/*
6093 	 * Identify the counters released between query trigger and query
6094 	 * handle more effiecntly. The counter released in this gap period
6095 	 * should wait for a new round of query as the new arrived packets
6096 	 * will not be taken into account.
6097 	 */
6098 	pool->query_gen++;
6099 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
6100 					       offset, NULL, NULL,
6101 					       pool->raw_hw->mem_mng->dm->id,
6102 					       (void *)(uintptr_t)
6103 					       (pool->raw_hw->data + offset),
6104 					       sh->devx_comp,
6105 					       (uint64_t)(uintptr_t)pool);
6106 	if (ret) {
6107 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6108 			" %d", pool->min_dcs->id);
6109 		pool->raw_hw = NULL;
6110 		goto set_alarm;
6111 	}
6112 	pool->raw_hw->min_dcs_id = dcs->id;
6113 	LIST_REMOVE(pool->raw_hw, next);
6114 	sh->cmng.pending_queries++;
6115 	pool_index++;
6116 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6117 		batch ^= 0x1;
6118 		pool_index = 0;
6119 		if (batch == 0 && pool_index == 0)
6120 			age ^= 0x1;
6121 	}
6122 set_alarm:
6123 	sh->cmng.batch = batch;
6124 	sh->cmng.pool_index = pool_index;
6125 	sh->cmng.age = age;
6126 	mlx5_set_query_alarm(sh);
6127 }
6128 
6129 /**
6130  * Check and callback event for new aged flow in the counter pool
6131  *
6132  * @param[in] sh
6133  *   Pointer to mlx5_dev_ctx_shared object.
6134  * @param[in] pool
6135  *   Pointer to Current counter pool.
6136  */
6137 static void
6138 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6139 		   struct mlx5_flow_counter_pool *pool)
6140 {
6141 	struct mlx5_priv *priv;
6142 	struct mlx5_flow_counter *cnt;
6143 	struct mlx5_age_info *age_info;
6144 	struct mlx5_age_param *age_param;
6145 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6146 	struct mlx5_counter_stats_raw *prev = pool->raw;
6147 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6148 	uint32_t i;
6149 
6150 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6151 		cnt = MLX5_POOL_GET_CNT(pool, i);
6152 		age_param = MLX5_CNT_TO_AGE(cnt);
6153 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6154 			continue;
6155 		if (cur->data[i].hits != prev->data[i].hits) {
6156 			age_param->expire = curr + age_param->timeout;
6157 			continue;
6158 		}
6159 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6160 			continue;
6161 		/**
6162 		 * Hold the lock first, or if between the
6163 		 * state AGE_TMOUT and tailq operation the
6164 		 * release happened, the release procedure
6165 		 * may delete a non-existent tailq node.
6166 		 */
6167 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6168 		age_info = GET_PORT_AGE_INFO(priv);
6169 		rte_spinlock_lock(&age_info->aged_sl);
6170 		/* If the cpmset fails, release happens. */
6171 		if (rte_atomic16_cmpset((volatile uint16_t *)
6172 					&age_param->state,
6173 					AGE_CANDIDATE,
6174 					AGE_TMOUT) ==
6175 					AGE_CANDIDATE) {
6176 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6177 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6178 		}
6179 		rte_spinlock_unlock(&age_info->aged_sl);
6180 	}
6181 	for (i = 0; i < sh->max_port; i++) {
6182 		age_info = &sh->port[i].age_info;
6183 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6184 			continue;
6185 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6186 			_rte_eth_dev_callback_process
6187 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
6188 				RTE_ETH_EVENT_FLOW_AGED, NULL);
6189 		age_info->flags = 0;
6190 	}
6191 }
6192 
6193 /**
6194  * Handler for the HW respond about ready values from an asynchronous batch
6195  * query. This function is probably called by the host thread.
6196  *
6197  * @param[in] sh
6198  *   The pointer to the shared device context.
6199  * @param[in] async_id
6200  *   The Devx async ID.
6201  * @param[in] status
6202  *   The status of the completion.
6203  */
6204 void
6205 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6206 				  uint64_t async_id, int status)
6207 {
6208 	struct mlx5_flow_counter_pool *pool =
6209 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6210 	struct mlx5_counter_stats_raw *raw_to_free;
6211 	uint8_t age = !!IS_AGE_POOL(pool);
6212 	uint8_t query_gen = pool->query_gen ^ 1;
6213 	struct mlx5_pools_container *cont =
6214 		MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6215 
6216 	if (unlikely(status)) {
6217 		raw_to_free = pool->raw_hw;
6218 	} else {
6219 		raw_to_free = pool->raw;
6220 		if (IS_AGE_POOL(pool))
6221 			mlx5_flow_aging_check(sh, pool);
6222 		rte_spinlock_lock(&pool->sl);
6223 		pool->raw = pool->raw_hw;
6224 		rte_spinlock_unlock(&pool->sl);
6225 		/* Be sure the new raw counters data is updated in memory. */
6226 		rte_cio_wmb();
6227 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6228 			rte_spinlock_lock(&cont->csl);
6229 			TAILQ_CONCAT(&cont->counters,
6230 				     &pool->counters[query_gen], next);
6231 			rte_spinlock_unlock(&cont->csl);
6232 		}
6233 	}
6234 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6235 	pool->raw_hw = NULL;
6236 	sh->cmng.pending_queries--;
6237 }
6238 
6239 /**
6240  * Translate the rte_flow group index to HW table value.
6241  *
6242  * @param[in] attributes
6243  *   Pointer to flow attributes
6244  * @param[in] external
6245  *   Value is part of flow rule created by request external to PMD.
6246  * @param[in] group
6247  *   rte_flow group index value.
6248  * @param[out] fdb_def_rule
6249  *   Whether fdb jump to table 1 is configured.
6250  * @param[out] table
6251  *   HW table value.
6252  * @param[out] error
6253  *   Pointer to error structure.
6254  *
6255  * @return
6256  *   0 on success, a negative errno value otherwise and rte_errno is set.
6257  */
6258 int
6259 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6260 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6261 			 struct rte_flow_error *error)
6262 {
6263 	if (attributes->transfer && external && fdb_def_rule) {
6264 		if (group == UINT32_MAX)
6265 			return rte_flow_error_set
6266 						(error, EINVAL,
6267 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6268 						 NULL,
6269 						 "group index not supported");
6270 		*table = group + 1;
6271 	} else {
6272 		*table = group;
6273 	}
6274 	return 0;
6275 }
6276 
6277 /**
6278  * Discover availability of metadata reg_c's.
6279  *
6280  * Iteratively use test flows to check availability.
6281  *
6282  * @param[in] dev
6283  *   Pointer to the Ethernet device structure.
6284  *
6285  * @return
6286  *   0 on success, a negative errno value otherwise and rte_errno is set.
6287  */
6288 int
6289 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6290 {
6291 	struct mlx5_priv *priv = dev->data->dev_private;
6292 	struct mlx5_dev_config *config = &priv->config;
6293 	enum modify_reg idx;
6294 	int n = 0;
6295 
6296 	/* reg_c[0] and reg_c[1] are reserved. */
6297 	config->flow_mreg_c[n++] = REG_C_0;
6298 	config->flow_mreg_c[n++] = REG_C_1;
6299 	/* Discover availability of other reg_c's. */
6300 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6301 		struct rte_flow_attr attr = {
6302 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6303 			.priority = MLX5_FLOW_PRIO_RSVD,
6304 			.ingress = 1,
6305 		};
6306 		struct rte_flow_item items[] = {
6307 			[0] = {
6308 				.type = RTE_FLOW_ITEM_TYPE_END,
6309 			},
6310 		};
6311 		struct rte_flow_action actions[] = {
6312 			[0] = {
6313 				.type = (enum rte_flow_action_type)
6314 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6315 				.conf = &(struct mlx5_flow_action_copy_mreg){
6316 					.src = REG_C_1,
6317 					.dst = idx,
6318 				},
6319 			},
6320 			[1] = {
6321 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6322 				.conf = &(struct rte_flow_action_jump){
6323 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6324 				},
6325 			},
6326 			[2] = {
6327 				.type = RTE_FLOW_ACTION_TYPE_END,
6328 			},
6329 		};
6330 		uint32_t flow_idx;
6331 		struct rte_flow *flow;
6332 		struct rte_flow_error error;
6333 
6334 		if (!config->dv_flow_en)
6335 			break;
6336 		/* Create internal flow, validation skips copy action. */
6337 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6338 					    actions, false, &error);
6339 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6340 				      flow_idx);
6341 		if (!flow)
6342 			continue;
6343 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6344 			config->flow_mreg_c[n++] = idx;
6345 		flow_list_destroy(dev, NULL, flow_idx);
6346 	}
6347 	for (; n < MLX5_MREG_C_NUM; ++n)
6348 		config->flow_mreg_c[n] = REG_NONE;
6349 	return 0;
6350 }
6351 
6352 /**
6353  * Dump flow raw hw data to file
6354  *
6355  * @param[in] dev
6356  *    The pointer to Ethernet device.
6357  * @param[in] file
6358  *   A pointer to a file for output.
6359  * @param[out] error
6360  *   Perform verbose error reporting if not NULL. PMDs initialize this
6361  *   structure in case of error only.
6362  * @return
6363  *   0 on success, a nagative value otherwise.
6364  */
6365 int
6366 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6367 		   FILE *file,
6368 		   struct rte_flow_error *error __rte_unused)
6369 {
6370 	struct mlx5_priv *priv = dev->data->dev_private;
6371 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6372 
6373 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6374 				       sh->tx_domain, file);
6375 }
6376 
6377 /**
6378  * Get aged-out flows.
6379  *
6380  * @param[in] dev
6381  *   Pointer to the Ethernet device structure.
6382  * @param[in] context
6383  *   The address of an array of pointers to the aged-out flows contexts.
6384  * @param[in] nb_countexts
6385  *   The length of context array pointers.
6386  * @param[out] error
6387  *   Perform verbose error reporting if not NULL. Initialized in case of
6388  *   error only.
6389  *
6390  * @return
6391  *   how many contexts get in success, otherwise negative errno value.
6392  *   if nb_contexts is 0, return the amount of all aged contexts.
6393  *   if nb_contexts is not 0 , return the amount of aged flows reported
6394  *   in the context array.
6395  */
6396 int
6397 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6398 			uint32_t nb_contexts, struct rte_flow_error *error)
6399 {
6400 	const struct mlx5_flow_driver_ops *fops;
6401 	struct rte_flow_attr attr = { .transfer = 0 };
6402 
6403 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6404 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6405 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6406 						    error);
6407 	}
6408 	DRV_LOG(ERR,
6409 		"port %u get aged flows is not supported.",
6410 		 dev->data->port_id);
6411 	return -ENOTSUP;
6412 }
6413