xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 2df00d562d203fb1450261c26d0ffa47b8007089)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21 
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
25 #include <rte_flow.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
28 #include <rte_ip.h>
29 
30 #include <mlx5_glue.h>
31 #include <mlx5_devx_cmds.h>
32 #include <mlx5_prm.h>
33 
34 #include "mlx5_defs.h"
35 #include "mlx5.h"
36 #include "mlx5_flow.h"
37 #include "mlx5_rxtx.h"
38 
39 /* Dev ops structure defined in mlx5.c */
40 extern const struct eth_dev_ops mlx5_dev_ops;
41 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
42 
43 /** Device flow drivers. */
44 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
45 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
46 #endif
47 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
48 
49 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
50 
51 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
52 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
53 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
54 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
55 #endif
56 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 };
59 
60 enum mlx5_expansion {
61 	MLX5_EXPANSION_ROOT,
62 	MLX5_EXPANSION_ROOT_OUTER,
63 	MLX5_EXPANSION_ROOT_ETH_VLAN,
64 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_ETH,
66 	MLX5_EXPANSION_OUTER_ETH_VLAN,
67 	MLX5_EXPANSION_OUTER_VLAN,
68 	MLX5_EXPANSION_OUTER_IPV4,
69 	MLX5_EXPANSION_OUTER_IPV4_UDP,
70 	MLX5_EXPANSION_OUTER_IPV4_TCP,
71 	MLX5_EXPANSION_OUTER_IPV6,
72 	MLX5_EXPANSION_OUTER_IPV6_UDP,
73 	MLX5_EXPANSION_OUTER_IPV6_TCP,
74 	MLX5_EXPANSION_VXLAN,
75 	MLX5_EXPANSION_VXLAN_GPE,
76 	MLX5_EXPANSION_GRE,
77 	MLX5_EXPANSION_MPLS,
78 	MLX5_EXPANSION_ETH,
79 	MLX5_EXPANSION_ETH_VLAN,
80 	MLX5_EXPANSION_VLAN,
81 	MLX5_EXPANSION_IPV4,
82 	MLX5_EXPANSION_IPV4_UDP,
83 	MLX5_EXPANSION_IPV4_TCP,
84 	MLX5_EXPANSION_IPV6,
85 	MLX5_EXPANSION_IPV6_UDP,
86 	MLX5_EXPANSION_IPV6_TCP,
87 };
88 
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 	[MLX5_EXPANSION_ROOT] = {
92 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 						 MLX5_EXPANSION_IPV4,
94 						 MLX5_EXPANSION_IPV6),
95 		.type = RTE_FLOW_ITEM_TYPE_END,
96 	},
97 	[MLX5_EXPANSION_ROOT_OUTER] = {
98 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 						 MLX5_EXPANSION_OUTER_IPV4,
100 						 MLX5_EXPANSION_OUTER_IPV6),
101 		.type = RTE_FLOW_ITEM_TYPE_END,
102 	},
103 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 		.type = RTE_FLOW_ITEM_TYPE_END,
106 	},
107 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 		.type = RTE_FLOW_ITEM_TYPE_END,
110 	},
111 	[MLX5_EXPANSION_OUTER_ETH] = {
112 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 						 MLX5_EXPANSION_OUTER_IPV6,
114 						 MLX5_EXPANSION_MPLS),
115 		.type = RTE_FLOW_ITEM_TYPE_ETH,
116 		.rss_types = 0,
117 	},
118 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 		.type = RTE_FLOW_ITEM_TYPE_ETH,
121 		.rss_types = 0,
122 	},
123 	[MLX5_EXPANSION_OUTER_VLAN] = {
124 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 						 MLX5_EXPANSION_OUTER_IPV6),
126 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
127 	},
128 	[MLX5_EXPANSION_OUTER_IPV4] = {
129 		.next = RTE_FLOW_EXPAND_RSS_NEXT
130 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
131 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 			 MLX5_EXPANSION_GRE,
133 			 MLX5_EXPANSION_IPV4,
134 			 MLX5_EXPANSION_IPV6),
135 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
136 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
137 			ETH_RSS_NONFRAG_IPV4_OTHER,
138 	},
139 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
140 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
141 						 MLX5_EXPANSION_VXLAN_GPE),
142 		.type = RTE_FLOW_ITEM_TYPE_UDP,
143 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
144 	},
145 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
146 		.type = RTE_FLOW_ITEM_TYPE_TCP,
147 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
148 	},
149 	[MLX5_EXPANSION_OUTER_IPV6] = {
150 		.next = RTE_FLOW_EXPAND_RSS_NEXT
151 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
152 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
153 			 MLX5_EXPANSION_IPV4,
154 			 MLX5_EXPANSION_IPV6),
155 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
156 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
157 			ETH_RSS_NONFRAG_IPV6_OTHER,
158 	},
159 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
160 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
161 						 MLX5_EXPANSION_VXLAN_GPE),
162 		.type = RTE_FLOW_ITEM_TYPE_UDP,
163 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
164 	},
165 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
166 		.type = RTE_FLOW_ITEM_TYPE_TCP,
167 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
168 	},
169 	[MLX5_EXPANSION_VXLAN] = {
170 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
171 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
172 	},
173 	[MLX5_EXPANSION_VXLAN_GPE] = {
174 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
175 						 MLX5_EXPANSION_IPV4,
176 						 MLX5_EXPANSION_IPV6),
177 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
178 	},
179 	[MLX5_EXPANSION_GRE] = {
180 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
181 		.type = RTE_FLOW_ITEM_TYPE_GRE,
182 	},
183 	[MLX5_EXPANSION_MPLS] = {
184 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
185 						 MLX5_EXPANSION_IPV6),
186 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
187 	},
188 	[MLX5_EXPANSION_ETH] = {
189 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
190 						 MLX5_EXPANSION_IPV6),
191 		.type = RTE_FLOW_ITEM_TYPE_ETH,
192 	},
193 	[MLX5_EXPANSION_ETH_VLAN] = {
194 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
195 		.type = RTE_FLOW_ITEM_TYPE_ETH,
196 	},
197 	[MLX5_EXPANSION_VLAN] = {
198 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
199 						 MLX5_EXPANSION_IPV6),
200 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
201 	},
202 	[MLX5_EXPANSION_IPV4] = {
203 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
204 						 MLX5_EXPANSION_IPV4_TCP),
205 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
206 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
207 			ETH_RSS_NONFRAG_IPV4_OTHER,
208 	},
209 	[MLX5_EXPANSION_IPV4_UDP] = {
210 		.type = RTE_FLOW_ITEM_TYPE_UDP,
211 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
212 	},
213 	[MLX5_EXPANSION_IPV4_TCP] = {
214 		.type = RTE_FLOW_ITEM_TYPE_TCP,
215 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
216 	},
217 	[MLX5_EXPANSION_IPV6] = {
218 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
219 						 MLX5_EXPANSION_IPV6_TCP),
220 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
221 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
222 			ETH_RSS_NONFRAG_IPV6_OTHER,
223 	},
224 	[MLX5_EXPANSION_IPV6_UDP] = {
225 		.type = RTE_FLOW_ITEM_TYPE_UDP,
226 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
227 	},
228 	[MLX5_EXPANSION_IPV6_TCP] = {
229 		.type = RTE_FLOW_ITEM_TYPE_TCP,
230 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
231 	},
232 };
233 
234 static const struct rte_flow_ops mlx5_flow_ops = {
235 	.validate = mlx5_flow_validate,
236 	.create = mlx5_flow_create,
237 	.destroy = mlx5_flow_destroy,
238 	.flush = mlx5_flow_flush,
239 	.isolate = mlx5_flow_isolate,
240 	.query = mlx5_flow_query,
241 	.dev_dump = mlx5_flow_dev_dump,
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 (config->flow_mreg_c
441 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
442 				return config->flow_mreg_c
443 					       [id + 1 + start_reg - REG_C_0];
444 			return rte_flow_error_set(error, ENOTSUP,
445 						  RTE_FLOW_ERROR_TYPE_ITEM,
446 						  NULL, "unsupported tag id");
447 		}
448 		return config->flow_mreg_c[id + start_reg - REG_C_0];
449 	}
450 	MLX5_ASSERT(false);
451 	return rte_flow_error_set(error, EINVAL,
452 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
453 				  NULL, "invalid feature name");
454 }
455 
456 /**
457  * Check extensive flow metadata register support.
458  *
459  * @param dev
460  *   Pointer to rte_eth_dev structure.
461  *
462  * @return
463  *   True if device supports extensive flow metadata register, otherwise false.
464  */
465 bool
466 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
467 {
468 	struct mlx5_priv *priv = dev->data->dev_private;
469 	struct mlx5_dev_config *config = &priv->config;
470 
471 	/*
472 	 * Having available reg_c can be regarded inclusively as supporting
473 	 * extensive flow metadata register, which could mean,
474 	 * - metadata register copy action by modify header.
475 	 * - 16 modify header actions is supported.
476 	 * - reg_c's are preserved across different domain (FDB and NIC) on
477 	 *   packet loopback by flow lookup miss.
478 	 */
479 	return config->flow_mreg_c[2] != REG_NONE;
480 }
481 
482 /**
483  * Discover the maximum number of priority available.
484  *
485  * @param[in] dev
486  *   Pointer to the Ethernet device structure.
487  *
488  * @return
489  *   number of supported flow priority on success, a negative errno
490  *   value otherwise and rte_errno is set.
491  */
492 int
493 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
494 {
495 	struct mlx5_priv *priv = dev->data->dev_private;
496 	struct {
497 		struct ibv_flow_attr attr;
498 		struct ibv_flow_spec_eth eth;
499 		struct ibv_flow_spec_action_drop drop;
500 	} flow_attr = {
501 		.attr = {
502 			.num_of_specs = 2,
503 			.port = (uint8_t)priv->ibv_port,
504 		},
505 		.eth = {
506 			.type = IBV_FLOW_SPEC_ETH,
507 			.size = sizeof(struct ibv_flow_spec_eth),
508 		},
509 		.drop = {
510 			.size = sizeof(struct ibv_flow_spec_action_drop),
511 			.type = IBV_FLOW_SPEC_ACTION_DROP,
512 		},
513 	};
514 	struct ibv_flow *flow;
515 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
516 	uint16_t vprio[] = { 8, 16 };
517 	int i;
518 	int priority = 0;
519 
520 	if (!drop) {
521 		rte_errno = ENOTSUP;
522 		return -rte_errno;
523 	}
524 	for (i = 0; i != RTE_DIM(vprio); i++) {
525 		flow_attr.attr.priority = vprio[i] - 1;
526 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
527 		if (!flow)
528 			break;
529 		claim_zero(mlx5_glue->destroy_flow(flow));
530 		priority = vprio[i];
531 	}
532 	mlx5_hrxq_drop_release(dev);
533 	switch (priority) {
534 	case 8:
535 		priority = RTE_DIM(priority_map_3);
536 		break;
537 	case 16:
538 		priority = RTE_DIM(priority_map_5);
539 		break;
540 	default:
541 		rte_errno = ENOTSUP;
542 		DRV_LOG(ERR,
543 			"port %u verbs maximum priority: %d expected 8/16",
544 			dev->data->port_id, priority);
545 		return -rte_errno;
546 	}
547 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
548 		dev->data->port_id, priority);
549 	return priority;
550 }
551 
552 /**
553  * Adjust flow priority based on the highest layer and the request priority.
554  *
555  * @param[in] dev
556  *   Pointer to the Ethernet device structure.
557  * @param[in] priority
558  *   The rule base priority.
559  * @param[in] subpriority
560  *   The priority based on the items.
561  *
562  * @return
563  *   The new priority.
564  */
565 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
566 				   uint32_t subpriority)
567 {
568 	uint32_t res = 0;
569 	struct mlx5_priv *priv = dev->data->dev_private;
570 
571 	switch (priv->config.flow_prio) {
572 	case RTE_DIM(priority_map_3):
573 		res = priority_map_3[priority][subpriority];
574 		break;
575 	case RTE_DIM(priority_map_5):
576 		res = priority_map_5[priority][subpriority];
577 		break;
578 	}
579 	return  res;
580 }
581 
582 /**
583  * Verify the @p item specifications (spec, last, mask) are compatible with the
584  * NIC capabilities.
585  *
586  * @param[in] item
587  *   Item specification.
588  * @param[in] mask
589  *   @p item->mask or flow default bit-masks.
590  * @param[in] nic_mask
591  *   Bit-masks covering supported fields by the NIC to compare with user mask.
592  * @param[in] size
593  *   Bit-masks size in bytes.
594  * @param[out] error
595  *   Pointer to error structure.
596  *
597  * @return
598  *   0 on success, a negative errno value otherwise and rte_errno is set.
599  */
600 int
601 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
602 			  const uint8_t *mask,
603 			  const uint8_t *nic_mask,
604 			  unsigned int size,
605 			  struct rte_flow_error *error)
606 {
607 	unsigned int i;
608 
609 	MLX5_ASSERT(nic_mask);
610 	for (i = 0; i < size; ++i)
611 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
612 			return rte_flow_error_set(error, ENOTSUP,
613 						  RTE_FLOW_ERROR_TYPE_ITEM,
614 						  item,
615 						  "mask enables non supported"
616 						  " bits");
617 	if (!item->spec && (item->mask || item->last))
618 		return rte_flow_error_set(error, EINVAL,
619 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
620 					  "mask/last without a spec is not"
621 					  " supported");
622 	if (item->spec && item->last) {
623 		uint8_t spec[size];
624 		uint8_t last[size];
625 		unsigned int i;
626 		int ret;
627 
628 		for (i = 0; i < size; ++i) {
629 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
630 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
631 		}
632 		ret = memcmp(spec, last, size);
633 		if (ret != 0)
634 			return rte_flow_error_set(error, EINVAL,
635 						  RTE_FLOW_ERROR_TYPE_ITEM,
636 						  item,
637 						  "range is not valid");
638 	}
639 	return 0;
640 }
641 
642 /**
643  * Adjust the hash fields according to the @p flow information.
644  *
645  * @param[in] dev_flow.
646  *   Pointer to the mlx5_flow.
647  * @param[in] tunnel
648  *   1 when the hash field is for a tunnel item.
649  * @param[in] layer_types
650  *   ETH_RSS_* types.
651  * @param[in] hash_fields
652  *   Item hash fields.
653  *
654  * @return
655  *   The hash fields that should be used.
656  */
657 uint64_t
658 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
659 			    int tunnel __rte_unused, uint64_t layer_types,
660 			    uint64_t hash_fields)
661 {
662 	struct rte_flow *flow = dev_flow->flow;
663 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
664 	int rss_request_inner = flow->rss.level >= 2;
665 
666 	/* Check RSS hash level for tunnel. */
667 	if (tunnel && rss_request_inner)
668 		hash_fields |= IBV_RX_HASH_INNER;
669 	else if (tunnel || rss_request_inner)
670 		return 0;
671 #endif
672 	/* Check if requested layer matches RSS hash fields. */
673 	if (!(flow->rss.types & layer_types))
674 		return 0;
675 	return hash_fields;
676 }
677 
678 /**
679  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
680  * if several tunnel rules are used on this queue, the tunnel ptype will be
681  * cleared.
682  *
683  * @param rxq_ctrl
684  *   Rx queue to update.
685  */
686 static void
687 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
688 {
689 	unsigned int i;
690 	uint32_t tunnel_ptype = 0;
691 
692 	/* Look up for the ptype to use. */
693 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
694 		if (!rxq_ctrl->flow_tunnels_n[i])
695 			continue;
696 		if (!tunnel_ptype) {
697 			tunnel_ptype = tunnels_info[i].ptype;
698 		} else {
699 			tunnel_ptype = 0;
700 			break;
701 		}
702 	}
703 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
704 }
705 
706 /**
707  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
708  * flow.
709  *
710  * @param[in] dev
711  *   Pointer to the Ethernet device structure.
712  * @param[in] dev_flow
713  *   Pointer to device flow structure.
714  */
715 static void
716 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
717 {
718 	struct mlx5_priv *priv = dev->data->dev_private;
719 	struct rte_flow *flow = dev_flow->flow;
720 	const int mark = !!(dev_flow->actions &
721 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
722 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
723 	unsigned int i;
724 
725 	for (i = 0; i != flow->rss.queue_num; ++i) {
726 		int idx = (*flow->rss.queue)[i];
727 		struct mlx5_rxq_ctrl *rxq_ctrl =
728 			container_of((*priv->rxqs)[idx],
729 				     struct mlx5_rxq_ctrl, rxq);
730 
731 		/*
732 		 * To support metadata register copy on Tx loopback,
733 		 * this must be always enabled (metadata may arive
734 		 * from other port - not from local flows only.
735 		 */
736 		if (priv->config.dv_flow_en &&
737 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
738 		    mlx5_flow_ext_mreg_supported(dev)) {
739 			rxq_ctrl->rxq.mark = 1;
740 			rxq_ctrl->flow_mark_n = 1;
741 		} else if (mark) {
742 			rxq_ctrl->rxq.mark = 1;
743 			rxq_ctrl->flow_mark_n++;
744 		}
745 		if (tunnel) {
746 			unsigned int j;
747 
748 			/* Increase the counter matching the flow. */
749 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
750 				if ((tunnels_info[j].tunnel &
751 				     dev_flow->layers) ==
752 				    tunnels_info[j].tunnel) {
753 					rxq_ctrl->flow_tunnels_n[j]++;
754 					break;
755 				}
756 			}
757 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
758 		}
759 	}
760 }
761 
762 /**
763  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
764  *
765  * @param[in] dev
766  *   Pointer to the Ethernet device structure.
767  * @param[in] flow
768  *   Pointer to flow structure.
769  */
770 static void
771 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
772 {
773 	struct mlx5_flow *dev_flow;
774 
775 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
776 		flow_drv_rxq_flags_set(dev, dev_flow);
777 }
778 
779 /**
780  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
781  * device flow if no other flow uses it with the same kind of request.
782  *
783  * @param dev
784  *   Pointer to Ethernet device.
785  * @param[in] dev_flow
786  *   Pointer to the device flow.
787  */
788 static void
789 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
790 {
791 	struct mlx5_priv *priv = dev->data->dev_private;
792 	struct rte_flow *flow = dev_flow->flow;
793 	const int mark = !!(dev_flow->actions &
794 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
795 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
796 	unsigned int i;
797 
798 	MLX5_ASSERT(dev->data->dev_started);
799 	for (i = 0; i != flow->rss.queue_num; ++i) {
800 		int idx = (*flow->rss.queue)[i];
801 		struct mlx5_rxq_ctrl *rxq_ctrl =
802 			container_of((*priv->rxqs)[idx],
803 				     struct mlx5_rxq_ctrl, rxq);
804 
805 		if (priv->config.dv_flow_en &&
806 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
807 		    mlx5_flow_ext_mreg_supported(dev)) {
808 			rxq_ctrl->rxq.mark = 1;
809 			rxq_ctrl->flow_mark_n = 1;
810 		} else if (mark) {
811 			rxq_ctrl->flow_mark_n--;
812 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
813 		}
814 		if (tunnel) {
815 			unsigned int j;
816 
817 			/* Decrease the counter matching the flow. */
818 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
819 				if ((tunnels_info[j].tunnel &
820 				     dev_flow->layers) ==
821 				    tunnels_info[j].tunnel) {
822 					rxq_ctrl->flow_tunnels_n[j]--;
823 					break;
824 				}
825 			}
826 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
827 		}
828 	}
829 }
830 
831 /**
832  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
833  * @p flow if no other flow uses it with the same kind of request.
834  *
835  * @param dev
836  *   Pointer to Ethernet device.
837  * @param[in] flow
838  *   Pointer to the flow.
839  */
840 static void
841 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
842 {
843 	struct mlx5_flow *dev_flow;
844 
845 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
846 		flow_drv_rxq_flags_trim(dev, dev_flow);
847 }
848 
849 /**
850  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
851  *
852  * @param dev
853  *   Pointer to Ethernet device.
854  */
855 static void
856 flow_rxq_flags_clear(struct rte_eth_dev *dev)
857 {
858 	struct mlx5_priv *priv = dev->data->dev_private;
859 	unsigned int i;
860 
861 	for (i = 0; i != priv->rxqs_n; ++i) {
862 		struct mlx5_rxq_ctrl *rxq_ctrl;
863 		unsigned int j;
864 
865 		if (!(*priv->rxqs)[i])
866 			continue;
867 		rxq_ctrl = container_of((*priv->rxqs)[i],
868 					struct mlx5_rxq_ctrl, rxq);
869 		rxq_ctrl->flow_mark_n = 0;
870 		rxq_ctrl->rxq.mark = 0;
871 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
872 			rxq_ctrl->flow_tunnels_n[j] = 0;
873 		rxq_ctrl->rxq.tunnel = 0;
874 	}
875 }
876 
877 /*
878  * return a pointer to the desired action in the list of actions.
879  *
880  * @param[in] actions
881  *   The list of actions to search the action in.
882  * @param[in] action
883  *   The action to find.
884  *
885  * @return
886  *   Pointer to the action in the list, if found. NULL otherwise.
887  */
888 const struct rte_flow_action *
889 mlx5_flow_find_action(const struct rte_flow_action *actions,
890 		      enum rte_flow_action_type action)
891 {
892 	if (actions == NULL)
893 		return NULL;
894 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
895 		if (actions->type == action)
896 			return actions;
897 	return NULL;
898 }
899 
900 /*
901  * Validate the flag action.
902  *
903  * @param[in] action_flags
904  *   Bit-fields that holds the actions detected until now.
905  * @param[in] attr
906  *   Attributes of flow that includes this action.
907  * @param[out] error
908  *   Pointer to error structure.
909  *
910  * @return
911  *   0 on success, a negative errno value otherwise and rte_errno is set.
912  */
913 int
914 mlx5_flow_validate_action_flag(uint64_t action_flags,
915 			       const struct rte_flow_attr *attr,
916 			       struct rte_flow_error *error)
917 {
918 	if (action_flags & MLX5_FLOW_ACTION_MARK)
919 		return rte_flow_error_set(error, EINVAL,
920 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
921 					  "can't mark and flag in same flow");
922 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
923 		return rte_flow_error_set(error, EINVAL,
924 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
925 					  "can't have 2 flag"
926 					  " actions in same flow");
927 	if (attr->egress)
928 		return rte_flow_error_set(error, ENOTSUP,
929 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
930 					  "flag action not supported for "
931 					  "egress");
932 	return 0;
933 }
934 
935 /*
936  * Validate the mark action.
937  *
938  * @param[in] action
939  *   Pointer to the queue action.
940  * @param[in] action_flags
941  *   Bit-fields that holds the actions detected until now.
942  * @param[in] attr
943  *   Attributes of flow that includes this action.
944  * @param[out] error
945  *   Pointer to error structure.
946  *
947  * @return
948  *   0 on success, a negative errno value otherwise and rte_errno is set.
949  */
950 int
951 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
952 			       uint64_t action_flags,
953 			       const struct rte_flow_attr *attr,
954 			       struct rte_flow_error *error)
955 {
956 	const struct rte_flow_action_mark *mark = action->conf;
957 
958 	if (!mark)
959 		return rte_flow_error_set(error, EINVAL,
960 					  RTE_FLOW_ERROR_TYPE_ACTION,
961 					  action,
962 					  "configuration cannot be null");
963 	if (mark->id >= MLX5_FLOW_MARK_MAX)
964 		return rte_flow_error_set(error, EINVAL,
965 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
966 					  &mark->id,
967 					  "mark id must in 0 <= id < "
968 					  RTE_STR(MLX5_FLOW_MARK_MAX));
969 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
970 		return rte_flow_error_set(error, EINVAL,
971 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
972 					  "can't flag and mark in same flow");
973 	if (action_flags & MLX5_FLOW_ACTION_MARK)
974 		return rte_flow_error_set(error, EINVAL,
975 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 					  "can't have 2 mark actions in same"
977 					  " flow");
978 	if (attr->egress)
979 		return rte_flow_error_set(error, ENOTSUP,
980 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
981 					  "mark action not supported for "
982 					  "egress");
983 	return 0;
984 }
985 
986 /*
987  * Validate the drop action.
988  *
989  * @param[in] action_flags
990  *   Bit-fields that holds the actions detected until now.
991  * @param[in] attr
992  *   Attributes of flow that includes this action.
993  * @param[out] error
994  *   Pointer to error structure.
995  *
996  * @return
997  *   0 on success, a negative errno value otherwise and rte_errno is set.
998  */
999 int
1000 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1001 			       const struct rte_flow_attr *attr,
1002 			       struct rte_flow_error *error)
1003 {
1004 	if (attr->egress)
1005 		return rte_flow_error_set(error, ENOTSUP,
1006 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007 					  "drop action not supported for "
1008 					  "egress");
1009 	return 0;
1010 }
1011 
1012 /*
1013  * Validate the queue action.
1014  *
1015  * @param[in] action
1016  *   Pointer to the queue action.
1017  * @param[in] action_flags
1018  *   Bit-fields that holds the actions detected until now.
1019  * @param[in] dev
1020  *   Pointer to the Ethernet device structure.
1021  * @param[in] attr
1022  *   Attributes of flow that includes this action.
1023  * @param[out] error
1024  *   Pointer to error structure.
1025  *
1026  * @return
1027  *   0 on success, a negative errno value otherwise and rte_errno is set.
1028  */
1029 int
1030 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1031 				uint64_t action_flags,
1032 				struct rte_eth_dev *dev,
1033 				const struct rte_flow_attr *attr,
1034 				struct rte_flow_error *error)
1035 {
1036 	struct mlx5_priv *priv = dev->data->dev_private;
1037 	const struct rte_flow_action_queue *queue = action->conf;
1038 
1039 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1040 		return rte_flow_error_set(error, EINVAL,
1041 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1042 					  "can't have 2 fate actions in"
1043 					  " same flow");
1044 	if (!priv->rxqs_n)
1045 		return rte_flow_error_set(error, EINVAL,
1046 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1047 					  NULL, "No Rx queues configured");
1048 	if (queue->index >= priv->rxqs_n)
1049 		return rte_flow_error_set(error, EINVAL,
1050 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1051 					  &queue->index,
1052 					  "queue index out of range");
1053 	if (!(*priv->rxqs)[queue->index])
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056 					  &queue->index,
1057 					  "queue is not configured");
1058 	if (attr->egress)
1059 		return rte_flow_error_set(error, ENOTSUP,
1060 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1061 					  "queue action not supported for "
1062 					  "egress");
1063 	return 0;
1064 }
1065 
1066 /*
1067  * Validate the rss action.
1068  *
1069  * @param[in] action
1070  *   Pointer to the queue action.
1071  * @param[in] action_flags
1072  *   Bit-fields that holds the actions detected until now.
1073  * @param[in] dev
1074  *   Pointer to the Ethernet device structure.
1075  * @param[in] attr
1076  *   Attributes of flow that includes this action.
1077  * @param[in] item_flags
1078  *   Items that were detected.
1079  * @param[out] error
1080  *   Pointer to error structure.
1081  *
1082  * @return
1083  *   0 on success, a negative errno value otherwise and rte_errno is set.
1084  */
1085 int
1086 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1087 			      uint64_t action_flags,
1088 			      struct rte_eth_dev *dev,
1089 			      const struct rte_flow_attr *attr,
1090 			      uint64_t item_flags,
1091 			      struct rte_flow_error *error)
1092 {
1093 	struct mlx5_priv *priv = dev->data->dev_private;
1094 	const struct rte_flow_action_rss *rss = action->conf;
1095 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1096 	unsigned int i;
1097 
1098 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1099 		return rte_flow_error_set(error, EINVAL,
1100 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1101 					  "can't have 2 fate actions"
1102 					  " in same flow");
1103 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1104 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1105 		return rte_flow_error_set(error, ENOTSUP,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &rss->func,
1108 					  "RSS hash function not supported");
1109 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1110 	if (rss->level > 2)
1111 #else
1112 	if (rss->level > 1)
1113 #endif
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 					  &rss->level,
1117 					  "tunnel RSS is not supported");
1118 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1119 	if (rss->key_len == 0 && rss->key != NULL)
1120 		return rte_flow_error_set(error, ENOTSUP,
1121 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1122 					  &rss->key_len,
1123 					  "RSS hash key length 0");
1124 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127 					  &rss->key_len,
1128 					  "RSS hash key too small");
1129 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1130 		return rte_flow_error_set(error, ENOTSUP,
1131 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1132 					  &rss->key_len,
1133 					  "RSS hash key too large");
1134 	if (rss->queue_num > priv->config.ind_table_max_size)
1135 		return rte_flow_error_set(error, ENOTSUP,
1136 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1137 					  &rss->queue_num,
1138 					  "number of queues too large");
1139 	if (rss->types & MLX5_RSS_HF_MASK)
1140 		return rte_flow_error_set(error, ENOTSUP,
1141 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142 					  &rss->types,
1143 					  "some RSS protocols are not"
1144 					  " supported");
1145 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1146 	    !(rss->types & ETH_RSS_IP))
1147 		return rte_flow_error_set(error, EINVAL,
1148 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1149 					  "L3 partial RSS requested but L3 RSS"
1150 					  " type not specified");
1151 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1152 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1153 		return rte_flow_error_set(error, EINVAL,
1154 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1155 					  "L4 partial RSS requested but L4 RSS"
1156 					  " type not specified");
1157 	if (!priv->rxqs_n)
1158 		return rte_flow_error_set(error, EINVAL,
1159 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1160 					  NULL, "No Rx queues configured");
1161 	if (!rss->queue_num)
1162 		return rte_flow_error_set(error, EINVAL,
1163 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1164 					  NULL, "No queues configured");
1165 	for (i = 0; i != rss->queue_num; ++i) {
1166 		if (rss->queue[i] >= priv->rxqs_n)
1167 			return rte_flow_error_set
1168 				(error, EINVAL,
1169 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 				 &rss->queue[i], "queue index out of range");
1171 		if (!(*priv->rxqs)[rss->queue[i]])
1172 			return rte_flow_error_set
1173 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1174 				 &rss->queue[i], "queue is not configured");
1175 	}
1176 	if (attr->egress)
1177 		return rte_flow_error_set(error, ENOTSUP,
1178 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1179 					  "rss action not supported for "
1180 					  "egress");
1181 	if (rss->level > 1 &&  !tunnel)
1182 		return rte_flow_error_set(error, EINVAL,
1183 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1184 					  "inner RSS is not supported for "
1185 					  "non-tunnel flows");
1186 	return 0;
1187 }
1188 
1189 /*
1190  * Validate the count action.
1191  *
1192  * @param[in] dev
1193  *   Pointer to the Ethernet device structure.
1194  * @param[in] attr
1195  *   Attributes of flow that includes this action.
1196  * @param[out] error
1197  *   Pointer to error structure.
1198  *
1199  * @return
1200  *   0 on success, a negative errno value otherwise and rte_errno is set.
1201  */
1202 int
1203 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1204 				const struct rte_flow_attr *attr,
1205 				struct rte_flow_error *error)
1206 {
1207 	if (attr->egress)
1208 		return rte_flow_error_set(error, ENOTSUP,
1209 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1210 					  "count action not supported for "
1211 					  "egress");
1212 	return 0;
1213 }
1214 
1215 /**
1216  * Verify the @p attributes will be correctly understood by the NIC and store
1217  * them in the @p flow if everything is correct.
1218  *
1219  * @param[in] dev
1220  *   Pointer to the Ethernet device structure.
1221  * @param[in] attributes
1222  *   Pointer to flow attributes
1223  * @param[out] error
1224  *   Pointer to error structure.
1225  *
1226  * @return
1227  *   0 on success, a negative errno value otherwise and rte_errno is set.
1228  */
1229 int
1230 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1231 			      const struct rte_flow_attr *attributes,
1232 			      struct rte_flow_error *error)
1233 {
1234 	struct mlx5_priv *priv = dev->data->dev_private;
1235 	uint32_t priority_max = priv->config.flow_prio - 1;
1236 
1237 	if (attributes->group)
1238 		return rte_flow_error_set(error, ENOTSUP,
1239 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1240 					  NULL, "groups is not supported");
1241 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1242 	    attributes->priority >= priority_max)
1243 		return rte_flow_error_set(error, ENOTSUP,
1244 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1245 					  NULL, "priority out of range");
1246 	if (attributes->egress)
1247 		return rte_flow_error_set(error, ENOTSUP,
1248 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1249 					  "egress is not supported");
1250 	if (attributes->transfer && !priv->config.dv_esw_en)
1251 		return rte_flow_error_set(error, ENOTSUP,
1252 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1253 					  NULL, "transfer is not supported");
1254 	if (!attributes->ingress)
1255 		return rte_flow_error_set(error, EINVAL,
1256 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1257 					  NULL,
1258 					  "ingress attribute is mandatory");
1259 	return 0;
1260 }
1261 
1262 /**
1263  * Validate ICMP6 item.
1264  *
1265  * @param[in] item
1266  *   Item specification.
1267  * @param[in] item_flags
1268  *   Bit-fields that holds the items detected until now.
1269  * @param[out] error
1270  *   Pointer to error structure.
1271  *
1272  * @return
1273  *   0 on success, a negative errno value otherwise and rte_errno is set.
1274  */
1275 int
1276 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1277 			       uint64_t item_flags,
1278 			       uint8_t target_protocol,
1279 			       struct rte_flow_error *error)
1280 {
1281 	const struct rte_flow_item_icmp6 *mask = item->mask;
1282 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1283 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1284 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1285 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1286 				      MLX5_FLOW_LAYER_OUTER_L4;
1287 	int ret;
1288 
1289 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1290 		return rte_flow_error_set(error, EINVAL,
1291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 					  "protocol filtering not compatible"
1293 					  " with ICMP6 layer");
1294 	if (!(item_flags & l3m))
1295 		return rte_flow_error_set(error, EINVAL,
1296 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1297 					  "IPv6 is mandatory to filter on"
1298 					  " ICMP6");
1299 	if (item_flags & l4m)
1300 		return rte_flow_error_set(error, EINVAL,
1301 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1302 					  "multiple L4 layers not supported");
1303 	if (!mask)
1304 		mask = &rte_flow_item_icmp6_mask;
1305 	ret = mlx5_flow_item_acceptable
1306 		(item, (const uint8_t *)mask,
1307 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1308 		 sizeof(struct rte_flow_item_icmp6), error);
1309 	if (ret < 0)
1310 		return ret;
1311 	return 0;
1312 }
1313 
1314 /**
1315  * Validate ICMP item.
1316  *
1317  * @param[in] item
1318  *   Item specification.
1319  * @param[in] item_flags
1320  *   Bit-fields that holds the items detected until now.
1321  * @param[out] error
1322  *   Pointer to error structure.
1323  *
1324  * @return
1325  *   0 on success, a negative errno value otherwise and rte_errno is set.
1326  */
1327 int
1328 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1329 			     uint64_t item_flags,
1330 			     uint8_t target_protocol,
1331 			     struct rte_flow_error *error)
1332 {
1333 	const struct rte_flow_item_icmp *mask = item->mask;
1334 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1335 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1336 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1337 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1338 				      MLX5_FLOW_LAYER_OUTER_L4;
1339 	int ret;
1340 
1341 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1342 		return rte_flow_error_set(error, EINVAL,
1343 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1344 					  "protocol filtering not compatible"
1345 					  " with ICMP layer");
1346 	if (!(item_flags & l3m))
1347 		return rte_flow_error_set(error, EINVAL,
1348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1349 					  "IPv4 is mandatory to filter"
1350 					  " on ICMP");
1351 	if (item_flags & l4m)
1352 		return rte_flow_error_set(error, EINVAL,
1353 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1354 					  "multiple L4 layers not supported");
1355 	if (!mask)
1356 		mask = &rte_flow_item_icmp_mask;
1357 	ret = mlx5_flow_item_acceptable
1358 		(item, (const uint8_t *)mask,
1359 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1360 		 sizeof(struct rte_flow_item_icmp), error);
1361 	if (ret < 0)
1362 		return ret;
1363 	return 0;
1364 }
1365 
1366 /**
1367  * Validate Ethernet item.
1368  *
1369  * @param[in] item
1370  *   Item specification.
1371  * @param[in] item_flags
1372  *   Bit-fields that holds the items detected until now.
1373  * @param[out] error
1374  *   Pointer to error structure.
1375  *
1376  * @return
1377  *   0 on success, a negative errno value otherwise and rte_errno is set.
1378  */
1379 int
1380 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1381 			    uint64_t item_flags,
1382 			    struct rte_flow_error *error)
1383 {
1384 	const struct rte_flow_item_eth *mask = item->mask;
1385 	const struct rte_flow_item_eth nic_mask = {
1386 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1387 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1388 		.type = RTE_BE16(0xffff),
1389 	};
1390 	int ret;
1391 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1392 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1393 				       MLX5_FLOW_LAYER_OUTER_L2;
1394 
1395 	if (item_flags & ethm)
1396 		return rte_flow_error_set(error, ENOTSUP,
1397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 					  "multiple L2 layers not supported");
1399 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1400 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1401 		return rte_flow_error_set(error, EINVAL,
1402 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 					  "L2 layer should not follow "
1404 					  "L3 layers");
1405 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1406 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1407 		return rte_flow_error_set(error, EINVAL,
1408 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1409 					  "L2 layer should not follow VLAN");
1410 	if (!mask)
1411 		mask = &rte_flow_item_eth_mask;
1412 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1413 					(const uint8_t *)&nic_mask,
1414 					sizeof(struct rte_flow_item_eth),
1415 					error);
1416 	return ret;
1417 }
1418 
1419 /**
1420  * Validate VLAN item.
1421  *
1422  * @param[in] item
1423  *   Item specification.
1424  * @param[in] item_flags
1425  *   Bit-fields that holds the items detected until now.
1426  * @param[in] dev
1427  *   Ethernet device flow is being created on.
1428  * @param[out] error
1429  *   Pointer to error structure.
1430  *
1431  * @return
1432  *   0 on success, a negative errno value otherwise and rte_errno is set.
1433  */
1434 int
1435 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1436 			     uint64_t item_flags,
1437 			     struct rte_eth_dev *dev,
1438 			     struct rte_flow_error *error)
1439 {
1440 	const struct rte_flow_item_vlan *spec = item->spec;
1441 	const struct rte_flow_item_vlan *mask = item->mask;
1442 	const struct rte_flow_item_vlan nic_mask = {
1443 		.tci = RTE_BE16(UINT16_MAX),
1444 		.inner_type = RTE_BE16(UINT16_MAX),
1445 	};
1446 	uint16_t vlan_tag = 0;
1447 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1448 	int ret;
1449 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1450 					MLX5_FLOW_LAYER_INNER_L4) :
1451 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1452 					MLX5_FLOW_LAYER_OUTER_L4);
1453 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1454 					MLX5_FLOW_LAYER_OUTER_VLAN;
1455 
1456 	if (item_flags & vlanm)
1457 		return rte_flow_error_set(error, EINVAL,
1458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1459 					  "multiple VLAN layers not supported");
1460 	else if ((item_flags & l34m) != 0)
1461 		return rte_flow_error_set(error, EINVAL,
1462 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1463 					  "VLAN cannot follow L3/L4 layer");
1464 	if (!mask)
1465 		mask = &rte_flow_item_vlan_mask;
1466 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1467 					(const uint8_t *)&nic_mask,
1468 					sizeof(struct rte_flow_item_vlan),
1469 					error);
1470 	if (ret)
1471 		return ret;
1472 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1473 		struct mlx5_priv *priv = dev->data->dev_private;
1474 
1475 		if (priv->vmwa_context) {
1476 			/*
1477 			 * Non-NULL context means we have a virtual machine
1478 			 * and SR-IOV enabled, we have to create VLAN interface
1479 			 * to make hypervisor to setup E-Switch vport
1480 			 * context correctly. We avoid creating the multiple
1481 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1482 			 */
1483 			return rte_flow_error_set(error, EINVAL,
1484 						  RTE_FLOW_ERROR_TYPE_ITEM,
1485 						  item,
1486 						  "VLAN tag mask is not"
1487 						  " supported in virtual"
1488 						  " environment");
1489 		}
1490 	}
1491 	if (spec) {
1492 		vlan_tag = spec->tci;
1493 		vlan_tag &= mask->tci;
1494 	}
1495 	/*
1496 	 * From verbs perspective an empty VLAN is equivalent
1497 	 * to a packet without VLAN layer.
1498 	 */
1499 	if (!vlan_tag)
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1502 					  item->spec,
1503 					  "VLAN cannot be empty");
1504 	return 0;
1505 }
1506 
1507 /**
1508  * Validate IPV4 item.
1509  *
1510  * @param[in] item
1511  *   Item specification.
1512  * @param[in] item_flags
1513  *   Bit-fields that holds the items detected until now.
1514  * @param[in] acc_mask
1515  *   Acceptable mask, if NULL default internal default mask
1516  *   will be used to check whether item fields are supported.
1517  * @param[out] error
1518  *   Pointer to error structure.
1519  *
1520  * @return
1521  *   0 on success, a negative errno value otherwise and rte_errno is set.
1522  */
1523 int
1524 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1525 			     uint64_t item_flags,
1526 			     uint64_t last_item,
1527 			     uint16_t ether_type,
1528 			     const struct rte_flow_item_ipv4 *acc_mask,
1529 			     struct rte_flow_error *error)
1530 {
1531 	const struct rte_flow_item_ipv4 *mask = item->mask;
1532 	const struct rte_flow_item_ipv4 *spec = item->spec;
1533 	const struct rte_flow_item_ipv4 nic_mask = {
1534 		.hdr = {
1535 			.src_addr = RTE_BE32(0xffffffff),
1536 			.dst_addr = RTE_BE32(0xffffffff),
1537 			.type_of_service = 0xff,
1538 			.next_proto_id = 0xff,
1539 		},
1540 	};
1541 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1542 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1543 				      MLX5_FLOW_LAYER_OUTER_L3;
1544 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1545 				      MLX5_FLOW_LAYER_OUTER_L4;
1546 	int ret;
1547 	uint8_t next_proto = 0xFF;
1548 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1549 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1550 				  MLX5_FLOW_LAYER_INNER_VLAN);
1551 
1552 	if ((last_item & l2_vlan) && ether_type &&
1553 	    ether_type != RTE_ETHER_TYPE_IPV4)
1554 		return rte_flow_error_set(error, EINVAL,
1555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1556 					  "IPv4 cannot follow L2/VLAN layer "
1557 					  "which ether type is not IPv4");
1558 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1559 		if (mask && spec)
1560 			next_proto = mask->hdr.next_proto_id &
1561 				     spec->hdr.next_proto_id;
1562 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1563 			return rte_flow_error_set(error, EINVAL,
1564 						  RTE_FLOW_ERROR_TYPE_ITEM,
1565 						  item,
1566 						  "multiple tunnel "
1567 						  "not supported");
1568 	}
1569 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1570 		return rte_flow_error_set(error, EINVAL,
1571 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1572 					  "wrong tunnel type - IPv6 specified "
1573 					  "but IPv4 item provided");
1574 	if (item_flags & l3m)
1575 		return rte_flow_error_set(error, ENOTSUP,
1576 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1577 					  "multiple L3 layers not supported");
1578 	else if (item_flags & l4m)
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 					  "L3 cannot follow an L4 layer.");
1582 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1583 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1584 		return rte_flow_error_set(error, EINVAL,
1585 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1586 					  "L3 cannot follow an NVGRE layer.");
1587 	if (!mask)
1588 		mask = &rte_flow_item_ipv4_mask;
1589 	else if (mask->hdr.next_proto_id != 0 &&
1590 		 mask->hdr.next_proto_id != 0xff)
1591 		return rte_flow_error_set(error, EINVAL,
1592 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1593 					  "partial mask is not supported"
1594 					  " for protocol");
1595 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1596 					acc_mask ? (const uint8_t *)acc_mask
1597 						 : (const uint8_t *)&nic_mask,
1598 					sizeof(struct rte_flow_item_ipv4),
1599 					error);
1600 	if (ret < 0)
1601 		return ret;
1602 	return 0;
1603 }
1604 
1605 /**
1606  * Validate IPV6 item.
1607  *
1608  * @param[in] item
1609  *   Item specification.
1610  * @param[in] item_flags
1611  *   Bit-fields that holds the items detected until now.
1612  * @param[in] acc_mask
1613  *   Acceptable mask, if NULL default internal default mask
1614  *   will be used to check whether item fields are supported.
1615  * @param[out] error
1616  *   Pointer to error structure.
1617  *
1618  * @return
1619  *   0 on success, a negative errno value otherwise and rte_errno is set.
1620  */
1621 int
1622 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1623 			     uint64_t item_flags,
1624 			     uint64_t last_item,
1625 			     uint16_t ether_type,
1626 			     const struct rte_flow_item_ipv6 *acc_mask,
1627 			     struct rte_flow_error *error)
1628 {
1629 	const struct rte_flow_item_ipv6 *mask = item->mask;
1630 	const struct rte_flow_item_ipv6 *spec = item->spec;
1631 	const struct rte_flow_item_ipv6 nic_mask = {
1632 		.hdr = {
1633 			.src_addr =
1634 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1635 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1636 			.dst_addr =
1637 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1638 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1639 			.vtc_flow = RTE_BE32(0xffffffff),
1640 			.proto = 0xff,
1641 			.hop_limits = 0xff,
1642 		},
1643 	};
1644 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1645 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1646 				      MLX5_FLOW_LAYER_OUTER_L3;
1647 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1648 				      MLX5_FLOW_LAYER_OUTER_L4;
1649 	int ret;
1650 	uint8_t next_proto = 0xFF;
1651 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1652 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1653 				  MLX5_FLOW_LAYER_INNER_VLAN);
1654 
1655 	if ((last_item & l2_vlan) && ether_type &&
1656 	    ether_type != RTE_ETHER_TYPE_IPV6)
1657 		return rte_flow_error_set(error, EINVAL,
1658 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1659 					  "IPv6 cannot follow L2/VLAN layer "
1660 					  "which ether type is not IPv6");
1661 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1662 		if (mask && spec)
1663 			next_proto = mask->hdr.proto & spec->hdr.proto;
1664 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1665 			return rte_flow_error_set(error, EINVAL,
1666 						  RTE_FLOW_ERROR_TYPE_ITEM,
1667 						  item,
1668 						  "multiple tunnel "
1669 						  "not supported");
1670 	}
1671 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1672 		return rte_flow_error_set(error, EINVAL,
1673 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1674 					  "wrong tunnel type - IPv4 specified "
1675 					  "but IPv6 item provided");
1676 	if (item_flags & l3m)
1677 		return rte_flow_error_set(error, ENOTSUP,
1678 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 					  "multiple L3 layers not supported");
1680 	else if (item_flags & l4m)
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 					  "L3 cannot follow an L4 layer.");
1684 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1685 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1688 					  "L3 cannot follow an NVGRE layer.");
1689 	if (!mask)
1690 		mask = &rte_flow_item_ipv6_mask;
1691 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1692 					acc_mask ? (const uint8_t *)acc_mask
1693 						 : (const uint8_t *)&nic_mask,
1694 					sizeof(struct rte_flow_item_ipv6),
1695 					error);
1696 	if (ret < 0)
1697 		return ret;
1698 	return 0;
1699 }
1700 
1701 /**
1702  * Validate UDP item.
1703  *
1704  * @param[in] item
1705  *   Item specification.
1706  * @param[in] item_flags
1707  *   Bit-fields that holds the items detected until now.
1708  * @param[in] target_protocol
1709  *   The next protocol in the previous item.
1710  * @param[in] flow_mask
1711  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1712  * @param[out] error
1713  *   Pointer to error structure.
1714  *
1715  * @return
1716  *   0 on success, a negative errno value otherwise and rte_errno is set.
1717  */
1718 int
1719 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1720 			    uint64_t item_flags,
1721 			    uint8_t target_protocol,
1722 			    struct rte_flow_error *error)
1723 {
1724 	const struct rte_flow_item_udp *mask = item->mask;
1725 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1726 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1727 				      MLX5_FLOW_LAYER_OUTER_L3;
1728 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1729 				      MLX5_FLOW_LAYER_OUTER_L4;
1730 	int ret;
1731 
1732 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1733 		return rte_flow_error_set(error, EINVAL,
1734 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1735 					  "protocol filtering not compatible"
1736 					  " with UDP layer");
1737 	if (!(item_flags & l3m))
1738 		return rte_flow_error_set(error, EINVAL,
1739 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1740 					  "L3 is mandatory to filter on L4");
1741 	if (item_flags & l4m)
1742 		return rte_flow_error_set(error, EINVAL,
1743 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1744 					  "multiple L4 layers not supported");
1745 	if (!mask)
1746 		mask = &rte_flow_item_udp_mask;
1747 	ret = mlx5_flow_item_acceptable
1748 		(item, (const uint8_t *)mask,
1749 		 (const uint8_t *)&rte_flow_item_udp_mask,
1750 		 sizeof(struct rte_flow_item_udp), error);
1751 	if (ret < 0)
1752 		return ret;
1753 	return 0;
1754 }
1755 
1756 /**
1757  * Validate TCP item.
1758  *
1759  * @param[in] item
1760  *   Item specification.
1761  * @param[in] item_flags
1762  *   Bit-fields that holds the items detected until now.
1763  * @param[in] target_protocol
1764  *   The next protocol in the previous item.
1765  * @param[out] error
1766  *   Pointer to error structure.
1767  *
1768  * @return
1769  *   0 on success, a negative errno value otherwise and rte_errno is set.
1770  */
1771 int
1772 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1773 			    uint64_t item_flags,
1774 			    uint8_t target_protocol,
1775 			    const struct rte_flow_item_tcp *flow_mask,
1776 			    struct rte_flow_error *error)
1777 {
1778 	const struct rte_flow_item_tcp *mask = item->mask;
1779 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1780 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1781 				      MLX5_FLOW_LAYER_OUTER_L3;
1782 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1783 				      MLX5_FLOW_LAYER_OUTER_L4;
1784 	int ret;
1785 
1786 	MLX5_ASSERT(flow_mask);
1787 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1788 		return rte_flow_error_set(error, EINVAL,
1789 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1790 					  "protocol filtering not compatible"
1791 					  " with TCP layer");
1792 	if (!(item_flags & l3m))
1793 		return rte_flow_error_set(error, EINVAL,
1794 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1795 					  "L3 is mandatory to filter on L4");
1796 	if (item_flags & l4m)
1797 		return rte_flow_error_set(error, EINVAL,
1798 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1799 					  "multiple L4 layers not supported");
1800 	if (!mask)
1801 		mask = &rte_flow_item_tcp_mask;
1802 	ret = mlx5_flow_item_acceptable
1803 		(item, (const uint8_t *)mask,
1804 		 (const uint8_t *)flow_mask,
1805 		 sizeof(struct rte_flow_item_tcp), error);
1806 	if (ret < 0)
1807 		return ret;
1808 	return 0;
1809 }
1810 
1811 /**
1812  * Validate VXLAN item.
1813  *
1814  * @param[in] item
1815  *   Item specification.
1816  * @param[in] item_flags
1817  *   Bit-fields that holds the items detected until now.
1818  * @param[in] target_protocol
1819  *   The next protocol in the previous item.
1820  * @param[out] error
1821  *   Pointer to error structure.
1822  *
1823  * @return
1824  *   0 on success, a negative errno value otherwise and rte_errno is set.
1825  */
1826 int
1827 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1828 			      uint64_t item_flags,
1829 			      struct rte_flow_error *error)
1830 {
1831 	const struct rte_flow_item_vxlan *spec = item->spec;
1832 	const struct rte_flow_item_vxlan *mask = item->mask;
1833 	int ret;
1834 	union vni {
1835 		uint32_t vlan_id;
1836 		uint8_t vni[4];
1837 	} id = { .vlan_id = 0, };
1838 	uint32_t vlan_id = 0;
1839 
1840 
1841 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1842 		return rte_flow_error_set(error, ENOTSUP,
1843 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1844 					  "multiple tunnel layers not"
1845 					  " supported");
1846 	/*
1847 	 * Verify only UDPv4 is present as defined in
1848 	 * https://tools.ietf.org/html/rfc7348
1849 	 */
1850 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1851 		return rte_flow_error_set(error, EINVAL,
1852 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 					  "no outer UDP layer found");
1854 	if (!mask)
1855 		mask = &rte_flow_item_vxlan_mask;
1856 	ret = mlx5_flow_item_acceptable
1857 		(item, (const uint8_t *)mask,
1858 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1859 		 sizeof(struct rte_flow_item_vxlan),
1860 		 error);
1861 	if (ret < 0)
1862 		return ret;
1863 	if (spec) {
1864 		memcpy(&id.vni[1], spec->vni, 3);
1865 		vlan_id = id.vlan_id;
1866 		memcpy(&id.vni[1], mask->vni, 3);
1867 		vlan_id &= id.vlan_id;
1868 	}
1869 	/*
1870 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1871 	 * only this layer is defined in the Verbs specification it is
1872 	 * interpreted as wildcard and all packets will match this
1873 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1874 	 * udp), all packets matching the layers before will also
1875 	 * match this rule.  To avoid such situation, VNI 0 is
1876 	 * currently refused.
1877 	 */
1878 	if (!vlan_id)
1879 		return rte_flow_error_set(error, ENOTSUP,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "VXLAN vni cannot be 0");
1882 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1883 		return rte_flow_error_set(error, ENOTSUP,
1884 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1885 					  "VXLAN tunnel must be fully defined");
1886 	return 0;
1887 }
1888 
1889 /**
1890  * Validate VXLAN_GPE item.
1891  *
1892  * @param[in] item
1893  *   Item specification.
1894  * @param[in] item_flags
1895  *   Bit-fields that holds the items detected until now.
1896  * @param[in] priv
1897  *   Pointer to the private data structure.
1898  * @param[in] target_protocol
1899  *   The next protocol in the previous item.
1900  * @param[out] error
1901  *   Pointer to error structure.
1902  *
1903  * @return
1904  *   0 on success, a negative errno value otherwise and rte_errno is set.
1905  */
1906 int
1907 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1908 				  uint64_t item_flags,
1909 				  struct rte_eth_dev *dev,
1910 				  struct rte_flow_error *error)
1911 {
1912 	struct mlx5_priv *priv = dev->data->dev_private;
1913 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1914 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1915 	int ret;
1916 	union vni {
1917 		uint32_t vlan_id;
1918 		uint8_t vni[4];
1919 	} id = { .vlan_id = 0, };
1920 	uint32_t vlan_id = 0;
1921 
1922 	if (!priv->config.l3_vxlan_en)
1923 		return rte_flow_error_set(error, ENOTSUP,
1924 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1925 					  "L3 VXLAN is not enabled by device"
1926 					  " parameter and/or not configured in"
1927 					  " firmware");
1928 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1929 		return rte_flow_error_set(error, ENOTSUP,
1930 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1931 					  "multiple tunnel layers not"
1932 					  " supported");
1933 	/*
1934 	 * Verify only UDPv4 is present as defined in
1935 	 * https://tools.ietf.org/html/rfc7348
1936 	 */
1937 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1938 		return rte_flow_error_set(error, EINVAL,
1939 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1940 					  "no outer UDP layer found");
1941 	if (!mask)
1942 		mask = &rte_flow_item_vxlan_gpe_mask;
1943 	ret = mlx5_flow_item_acceptable
1944 		(item, (const uint8_t *)mask,
1945 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1946 		 sizeof(struct rte_flow_item_vxlan_gpe),
1947 		 error);
1948 	if (ret < 0)
1949 		return ret;
1950 	if (spec) {
1951 		if (spec->protocol)
1952 			return rte_flow_error_set(error, ENOTSUP,
1953 						  RTE_FLOW_ERROR_TYPE_ITEM,
1954 						  item,
1955 						  "VxLAN-GPE protocol"
1956 						  " not supported");
1957 		memcpy(&id.vni[1], spec->vni, 3);
1958 		vlan_id = id.vlan_id;
1959 		memcpy(&id.vni[1], mask->vni, 3);
1960 		vlan_id &= id.vlan_id;
1961 	}
1962 	/*
1963 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1964 	 * layer is defined in the Verbs specification it is interpreted as
1965 	 * wildcard and all packets will match this rule, if it follows a full
1966 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1967 	 * before will also match this rule.  To avoid such situation, VNI 0
1968 	 * is currently refused.
1969 	 */
1970 	if (!vlan_id)
1971 		return rte_flow_error_set(error, ENOTSUP,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "VXLAN-GPE vni cannot be 0");
1974 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1975 		return rte_flow_error_set(error, ENOTSUP,
1976 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 					  "VXLAN-GPE tunnel must be fully"
1978 					  " defined");
1979 	return 0;
1980 }
1981 /**
1982  * Validate GRE Key item.
1983  *
1984  * @param[in] item
1985  *   Item specification.
1986  * @param[in] item_flags
1987  *   Bit flags to mark detected items.
1988  * @param[in] gre_item
1989  *   Pointer to gre_item
1990  * @param[out] error
1991  *   Pointer to error structure.
1992  *
1993  * @return
1994  *   0 on success, a negative errno value otherwise and rte_errno is set.
1995  */
1996 int
1997 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1998 				uint64_t item_flags,
1999 				const struct rte_flow_item *gre_item,
2000 				struct rte_flow_error *error)
2001 {
2002 	const rte_be32_t *mask = item->mask;
2003 	int ret = 0;
2004 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2005 	const struct rte_flow_item_gre *gre_spec;
2006 	const struct rte_flow_item_gre *gre_mask;
2007 
2008 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2009 		return rte_flow_error_set(error, ENOTSUP,
2010 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2011 					  "Multiple GRE key not support");
2012 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2013 		return rte_flow_error_set(error, ENOTSUP,
2014 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2015 					  "No preceding GRE header");
2016 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2017 		return rte_flow_error_set(error, ENOTSUP,
2018 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2019 					  "GRE key following a wrong item");
2020 	gre_mask = gre_item->mask;
2021 	if (!gre_mask)
2022 		gre_mask = &rte_flow_item_gre_mask;
2023 	gre_spec = gre_item->spec;
2024 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2025 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2026 		return rte_flow_error_set(error, EINVAL,
2027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 					  "Key bit must be on");
2029 
2030 	if (!mask)
2031 		mask = &gre_key_default_mask;
2032 	ret = mlx5_flow_item_acceptable
2033 		(item, (const uint8_t *)mask,
2034 		 (const uint8_t *)&gre_key_default_mask,
2035 		 sizeof(rte_be32_t), error);
2036 	return ret;
2037 }
2038 
2039 /**
2040  * Validate GRE item.
2041  *
2042  * @param[in] item
2043  *   Item specification.
2044  * @param[in] item_flags
2045  *   Bit flags to mark detected items.
2046  * @param[in] target_protocol
2047  *   The next protocol in the previous item.
2048  * @param[out] error
2049  *   Pointer to error structure.
2050  *
2051  * @return
2052  *   0 on success, a negative errno value otherwise and rte_errno is set.
2053  */
2054 int
2055 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2056 			    uint64_t item_flags,
2057 			    uint8_t target_protocol,
2058 			    struct rte_flow_error *error)
2059 {
2060 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2061 	const struct rte_flow_item_gre *mask = item->mask;
2062 	int ret;
2063 	const struct rte_flow_item_gre nic_mask = {
2064 		.c_rsvd0_ver = RTE_BE16(0xB000),
2065 		.protocol = RTE_BE16(UINT16_MAX),
2066 	};
2067 
2068 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2069 		return rte_flow_error_set(error, EINVAL,
2070 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2071 					  "protocol filtering not compatible"
2072 					  " with this GRE layer");
2073 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2074 		return rte_flow_error_set(error, ENOTSUP,
2075 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 					  "multiple tunnel layers not"
2077 					  " supported");
2078 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2079 		return rte_flow_error_set(error, ENOTSUP,
2080 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2081 					  "L3 Layer is missing");
2082 	if (!mask)
2083 		mask = &rte_flow_item_gre_mask;
2084 	ret = mlx5_flow_item_acceptable
2085 		(item, (const uint8_t *)mask,
2086 		 (const uint8_t *)&nic_mask,
2087 		 sizeof(struct rte_flow_item_gre), error);
2088 	if (ret < 0)
2089 		return ret;
2090 #ifndef HAVE_MLX5DV_DR
2091 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2092 	if (spec && (spec->protocol & mask->protocol))
2093 		return rte_flow_error_set(error, ENOTSUP,
2094 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 					  "without MPLS support the"
2096 					  " specification cannot be used for"
2097 					  " filtering");
2098 #endif
2099 #endif
2100 	return 0;
2101 }
2102 
2103 /**
2104  * Validate Geneve item.
2105  *
2106  * @param[in] item
2107  *   Item specification.
2108  * @param[in] itemFlags
2109  *   Bit-fields that holds the items detected until now.
2110  * @param[in] enPriv
2111  *   Pointer to the private data structure.
2112  * @param[out] error
2113  *   Pointer to error structure.
2114  *
2115  * @return
2116  *   0 on success, a negative errno value otherwise and rte_errno is set.
2117  */
2118 
2119 int
2120 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2121 			       uint64_t item_flags,
2122 			       struct rte_eth_dev *dev,
2123 			       struct rte_flow_error *error)
2124 {
2125 	struct mlx5_priv *priv = dev->data->dev_private;
2126 	const struct rte_flow_item_geneve *spec = item->spec;
2127 	const struct rte_flow_item_geneve *mask = item->mask;
2128 	int ret;
2129 	uint16_t gbhdr;
2130 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2131 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2132 	const struct rte_flow_item_geneve nic_mask = {
2133 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2134 		.vni = "\xff\xff\xff",
2135 		.protocol = RTE_BE16(UINT16_MAX),
2136 	};
2137 
2138 	if (!(priv->config.hca_attr.flex_parser_protocols &
2139 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2140 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2141 		return rte_flow_error_set(error, ENOTSUP,
2142 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2143 					  "L3 Geneve is not enabled by device"
2144 					  " parameter and/or not configured in"
2145 					  " firmware");
2146 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2147 		return rte_flow_error_set(error, ENOTSUP,
2148 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2149 					  "multiple tunnel layers not"
2150 					  " supported");
2151 	/*
2152 	 * Verify only UDPv4 is present as defined in
2153 	 * https://tools.ietf.org/html/rfc7348
2154 	 */
2155 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2156 		return rte_flow_error_set(error, EINVAL,
2157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 					  "no outer UDP layer found");
2159 	if (!mask)
2160 		mask = &rte_flow_item_geneve_mask;
2161 	ret = mlx5_flow_item_acceptable
2162 				  (item, (const uint8_t *)mask,
2163 				   (const uint8_t *)&nic_mask,
2164 				   sizeof(struct rte_flow_item_geneve), error);
2165 	if (ret)
2166 		return ret;
2167 	if (spec) {
2168 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2169 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2170 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2171 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2172 			return rte_flow_error_set(error, ENOTSUP,
2173 						  RTE_FLOW_ERROR_TYPE_ITEM,
2174 						  item,
2175 						  "Geneve protocol unsupported"
2176 						  " fields are being used");
2177 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2178 			return rte_flow_error_set
2179 					(error, ENOTSUP,
2180 					 RTE_FLOW_ERROR_TYPE_ITEM,
2181 					 item,
2182 					 "Unsupported Geneve options length");
2183 	}
2184 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2185 		return rte_flow_error_set
2186 				    (error, ENOTSUP,
2187 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2188 				     "Geneve tunnel must be fully defined");
2189 	return 0;
2190 }
2191 
2192 /**
2193  * Validate MPLS item.
2194  *
2195  * @param[in] dev
2196  *   Pointer to the rte_eth_dev structure.
2197  * @param[in] item
2198  *   Item specification.
2199  * @param[in] item_flags
2200  *   Bit-fields that holds the items detected until now.
2201  * @param[in] prev_layer
2202  *   The protocol layer indicated in previous item.
2203  * @param[out] error
2204  *   Pointer to error structure.
2205  *
2206  * @return
2207  *   0 on success, a negative errno value otherwise and rte_errno is set.
2208  */
2209 int
2210 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2211 			     const struct rte_flow_item *item __rte_unused,
2212 			     uint64_t item_flags __rte_unused,
2213 			     uint64_t prev_layer __rte_unused,
2214 			     struct rte_flow_error *error)
2215 {
2216 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2217 	const struct rte_flow_item_mpls *mask = item->mask;
2218 	struct mlx5_priv *priv = dev->data->dev_private;
2219 	int ret;
2220 
2221 	if (!priv->config.mpls_en)
2222 		return rte_flow_error_set(error, ENOTSUP,
2223 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2224 					  "MPLS not supported or"
2225 					  " disabled in firmware"
2226 					  " configuration.");
2227 	/* MPLS over IP, UDP, GRE is allowed */
2228 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2229 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2230 			    MLX5_FLOW_LAYER_GRE)))
2231 		return rte_flow_error_set(error, EINVAL,
2232 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 					  "protocol filtering not compatible"
2234 					  " with MPLS layer");
2235 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2236 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2237 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2238 		return rte_flow_error_set(error, ENOTSUP,
2239 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 					  "multiple tunnel layers not"
2241 					  " supported");
2242 	if (!mask)
2243 		mask = &rte_flow_item_mpls_mask;
2244 	ret = mlx5_flow_item_acceptable
2245 		(item, (const uint8_t *)mask,
2246 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2247 		 sizeof(struct rte_flow_item_mpls), error);
2248 	if (ret < 0)
2249 		return ret;
2250 	return 0;
2251 #endif
2252 	return rte_flow_error_set(error, ENOTSUP,
2253 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 				  "MPLS is not supported by Verbs, please"
2255 				  " update.");
2256 }
2257 
2258 /**
2259  * Validate NVGRE item.
2260  *
2261  * @param[in] item
2262  *   Item specification.
2263  * @param[in] item_flags
2264  *   Bit flags to mark detected items.
2265  * @param[in] target_protocol
2266  *   The next protocol in the previous item.
2267  * @param[out] error
2268  *   Pointer to error structure.
2269  *
2270  * @return
2271  *   0 on success, a negative errno value otherwise and rte_errno is set.
2272  */
2273 int
2274 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2275 			      uint64_t item_flags,
2276 			      uint8_t target_protocol,
2277 			      struct rte_flow_error *error)
2278 {
2279 	const struct rte_flow_item_nvgre *mask = item->mask;
2280 	int ret;
2281 
2282 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2283 		return rte_flow_error_set(error, EINVAL,
2284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 					  "protocol filtering not compatible"
2286 					  " with this GRE layer");
2287 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2288 		return rte_flow_error_set(error, ENOTSUP,
2289 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2290 					  "multiple tunnel layers not"
2291 					  " supported");
2292 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2293 		return rte_flow_error_set(error, ENOTSUP,
2294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 					  "L3 Layer is missing");
2296 	if (!mask)
2297 		mask = &rte_flow_item_nvgre_mask;
2298 	ret = mlx5_flow_item_acceptable
2299 		(item, (const uint8_t *)mask,
2300 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2301 		 sizeof(struct rte_flow_item_nvgre), error);
2302 	if (ret < 0)
2303 		return ret;
2304 	return 0;
2305 }
2306 
2307 /* Allocate unique ID for the split Q/RSS subflows. */
2308 static uint32_t
2309 flow_qrss_get_id(struct rte_eth_dev *dev)
2310 {
2311 	struct mlx5_priv *priv = dev->data->dev_private;
2312 	uint32_t qrss_id, ret;
2313 
2314 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2315 	if (ret)
2316 		return 0;
2317 	MLX5_ASSERT(qrss_id);
2318 	return qrss_id;
2319 }
2320 
2321 /* Free unique ID for the split Q/RSS subflows. */
2322 static void
2323 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2324 {
2325 	struct mlx5_priv *priv = dev->data->dev_private;
2326 
2327 	if (qrss_id)
2328 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2329 }
2330 
2331 /**
2332  * Release resource related QUEUE/RSS action split.
2333  *
2334  * @param dev
2335  *   Pointer to Ethernet device.
2336  * @param flow
2337  *   Flow to release id's from.
2338  */
2339 static void
2340 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2341 			     struct rte_flow *flow)
2342 {
2343 	struct mlx5_flow *dev_flow;
2344 
2345 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2346 		if (dev_flow->qrss_id)
2347 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2348 }
2349 
2350 static int
2351 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2352 		   const struct rte_flow_attr *attr __rte_unused,
2353 		   const struct rte_flow_item items[] __rte_unused,
2354 		   const struct rte_flow_action actions[] __rte_unused,
2355 		   bool external __rte_unused,
2356 		   struct rte_flow_error *error)
2357 {
2358 	return rte_flow_error_set(error, ENOTSUP,
2359 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2360 }
2361 
2362 static struct mlx5_flow *
2363 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2364 		  const struct rte_flow_item items[] __rte_unused,
2365 		  const struct rte_flow_action actions[] __rte_unused,
2366 		  struct rte_flow_error *error)
2367 {
2368 	rte_flow_error_set(error, ENOTSUP,
2369 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2370 	return NULL;
2371 }
2372 
2373 static int
2374 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2375 		    struct mlx5_flow *dev_flow __rte_unused,
2376 		    const struct rte_flow_attr *attr __rte_unused,
2377 		    const struct rte_flow_item items[] __rte_unused,
2378 		    const struct rte_flow_action actions[] __rte_unused,
2379 		    struct rte_flow_error *error)
2380 {
2381 	return rte_flow_error_set(error, ENOTSUP,
2382 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2383 }
2384 
2385 static int
2386 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2387 		struct rte_flow *flow __rte_unused,
2388 		struct rte_flow_error *error)
2389 {
2390 	return rte_flow_error_set(error, ENOTSUP,
2391 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2392 }
2393 
2394 static void
2395 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2396 		 struct rte_flow *flow __rte_unused)
2397 {
2398 }
2399 
2400 static void
2401 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2402 		  struct rte_flow *flow __rte_unused)
2403 {
2404 }
2405 
2406 static int
2407 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2408 		struct rte_flow *flow __rte_unused,
2409 		const struct rte_flow_action *actions __rte_unused,
2410 		void *data __rte_unused,
2411 		struct rte_flow_error *error)
2412 {
2413 	return rte_flow_error_set(error, ENOTSUP,
2414 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2415 }
2416 
2417 /* Void driver to protect from null pointer reference. */
2418 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2419 	.validate = flow_null_validate,
2420 	.prepare = flow_null_prepare,
2421 	.translate = flow_null_translate,
2422 	.apply = flow_null_apply,
2423 	.remove = flow_null_remove,
2424 	.destroy = flow_null_destroy,
2425 	.query = flow_null_query,
2426 };
2427 
2428 /**
2429  * Select flow driver type according to flow attributes and device
2430  * configuration.
2431  *
2432  * @param[in] dev
2433  *   Pointer to the dev structure.
2434  * @param[in] attr
2435  *   Pointer to the flow attributes.
2436  *
2437  * @return
2438  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2439  */
2440 static enum mlx5_flow_drv_type
2441 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2442 {
2443 	struct mlx5_priv *priv = dev->data->dev_private;
2444 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2445 
2446 	if (attr->transfer && priv->config.dv_esw_en)
2447 		type = MLX5_FLOW_TYPE_DV;
2448 	if (!attr->transfer)
2449 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2450 						 MLX5_FLOW_TYPE_VERBS;
2451 	return type;
2452 }
2453 
2454 #define flow_get_drv_ops(type) flow_drv_ops[type]
2455 
2456 /**
2457  * Flow driver validation API. This abstracts calling driver specific functions.
2458  * The type of flow driver is determined according to flow attributes.
2459  *
2460  * @param[in] dev
2461  *   Pointer to the dev structure.
2462  * @param[in] attr
2463  *   Pointer to the flow attributes.
2464  * @param[in] items
2465  *   Pointer to the list of items.
2466  * @param[in] actions
2467  *   Pointer to the list of actions.
2468  * @param[in] external
2469  *   This flow rule is created by request external to PMD.
2470  * @param[out] error
2471  *   Pointer to the error structure.
2472  *
2473  * @return
2474  *   0 on success, a negative errno value otherwise and rte_errno is set.
2475  */
2476 static inline int
2477 flow_drv_validate(struct rte_eth_dev *dev,
2478 		  const struct rte_flow_attr *attr,
2479 		  const struct rte_flow_item items[],
2480 		  const struct rte_flow_action actions[],
2481 		  bool external, struct rte_flow_error *error)
2482 {
2483 	const struct mlx5_flow_driver_ops *fops;
2484 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2485 
2486 	fops = flow_get_drv_ops(type);
2487 	return fops->validate(dev, attr, items, actions, external, error);
2488 }
2489 
2490 /**
2491  * Flow driver preparation API. This abstracts calling driver specific
2492  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2493  * calculates the size of memory required for device flow, allocates the memory,
2494  * initializes the device flow and returns the pointer.
2495  *
2496  * @note
2497  *   This function initializes device flow structure such as dv or verbs in
2498  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2499  *   rest. For example, adding returning device flow to flow->dev_flow list and
2500  *   setting backward reference to the flow should be done out of this function.
2501  *   layers field is not filled either.
2502  *
2503  * @param[in] attr
2504  *   Pointer to the flow attributes.
2505  * @param[in] items
2506  *   Pointer to the list of items.
2507  * @param[in] actions
2508  *   Pointer to the list of actions.
2509  * @param[out] error
2510  *   Pointer to the error structure.
2511  *
2512  * @return
2513  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2514  */
2515 static inline struct mlx5_flow *
2516 flow_drv_prepare(const struct rte_flow *flow,
2517 		 const struct rte_flow_attr *attr,
2518 		 const struct rte_flow_item items[],
2519 		 const struct rte_flow_action actions[],
2520 		 struct rte_flow_error *error)
2521 {
2522 	const struct mlx5_flow_driver_ops *fops;
2523 	enum mlx5_flow_drv_type type = flow->drv_type;
2524 
2525 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2526 	fops = flow_get_drv_ops(type);
2527 	return fops->prepare(attr, items, actions, error);
2528 }
2529 
2530 /**
2531  * Flow driver translation API. This abstracts calling driver specific
2532  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2533  * translates a generic flow into a driver flow. flow_drv_prepare() must
2534  * precede.
2535  *
2536  * @note
2537  *   dev_flow->layers could be filled as a result of parsing during translation
2538  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2539  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2540  *   flow->actions could be overwritten even though all the expanded dev_flows
2541  *   have the same actions.
2542  *
2543  * @param[in] dev
2544  *   Pointer to the rte dev structure.
2545  * @param[in, out] dev_flow
2546  *   Pointer to the mlx5 flow.
2547  * @param[in] attr
2548  *   Pointer to the flow attributes.
2549  * @param[in] items
2550  *   Pointer to the list of items.
2551  * @param[in] actions
2552  *   Pointer to the list of actions.
2553  * @param[out] error
2554  *   Pointer to the error structure.
2555  *
2556  * @return
2557  *   0 on success, a negative errno value otherwise and rte_errno is set.
2558  */
2559 static inline int
2560 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2561 		   const struct rte_flow_attr *attr,
2562 		   const struct rte_flow_item items[],
2563 		   const struct rte_flow_action actions[],
2564 		   struct rte_flow_error *error)
2565 {
2566 	const struct mlx5_flow_driver_ops *fops;
2567 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2568 
2569 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2570 	fops = flow_get_drv_ops(type);
2571 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2572 }
2573 
2574 /**
2575  * Flow driver apply API. This abstracts calling driver specific functions.
2576  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2577  * translated driver flows on to device. flow_drv_translate() must precede.
2578  *
2579  * @param[in] dev
2580  *   Pointer to Ethernet device structure.
2581  * @param[in, out] flow
2582  *   Pointer to flow structure.
2583  * @param[out] error
2584  *   Pointer to error structure.
2585  *
2586  * @return
2587  *   0 on success, a negative errno value otherwise and rte_errno is set.
2588  */
2589 static inline int
2590 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2591 	       struct rte_flow_error *error)
2592 {
2593 	const struct mlx5_flow_driver_ops *fops;
2594 	enum mlx5_flow_drv_type type = flow->drv_type;
2595 
2596 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2597 	fops = flow_get_drv_ops(type);
2598 	return fops->apply(dev, flow, error);
2599 }
2600 
2601 /**
2602  * Flow driver remove API. This abstracts calling driver specific functions.
2603  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2604  * on device. All the resources of the flow should be freed by calling
2605  * flow_drv_destroy().
2606  *
2607  * @param[in] dev
2608  *   Pointer to Ethernet device.
2609  * @param[in, out] flow
2610  *   Pointer to flow structure.
2611  */
2612 static inline void
2613 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2614 {
2615 	const struct mlx5_flow_driver_ops *fops;
2616 	enum mlx5_flow_drv_type type = flow->drv_type;
2617 
2618 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2619 	fops = flow_get_drv_ops(type);
2620 	fops->remove(dev, flow);
2621 }
2622 
2623 /**
2624  * Flow driver destroy API. This abstracts calling driver specific functions.
2625  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2626  * on device and releases resources of the flow.
2627  *
2628  * @param[in] dev
2629  *   Pointer to Ethernet device.
2630  * @param[in, out] flow
2631  *   Pointer to flow structure.
2632  */
2633 static inline void
2634 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2635 {
2636 	const struct mlx5_flow_driver_ops *fops;
2637 	enum mlx5_flow_drv_type type = flow->drv_type;
2638 
2639 	flow_mreg_split_qrss_release(dev, flow);
2640 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2641 	fops = flow_get_drv_ops(type);
2642 	fops->destroy(dev, flow);
2643 }
2644 
2645 /**
2646  * Validate a flow supported by the NIC.
2647  *
2648  * @see rte_flow_validate()
2649  * @see rte_flow_ops
2650  */
2651 int
2652 mlx5_flow_validate(struct rte_eth_dev *dev,
2653 		   const struct rte_flow_attr *attr,
2654 		   const struct rte_flow_item items[],
2655 		   const struct rte_flow_action actions[],
2656 		   struct rte_flow_error *error)
2657 {
2658 	int ret;
2659 
2660 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2661 	if (ret < 0)
2662 		return ret;
2663 	return 0;
2664 }
2665 
2666 /**
2667  * Get port id item from the item list.
2668  *
2669  * @param[in] item
2670  *   Pointer to the list of items.
2671  *
2672  * @return
2673  *   Pointer to the port id item if exist, else return NULL.
2674  */
2675 static const struct rte_flow_item *
2676 find_port_id_item(const struct rte_flow_item *item)
2677 {
2678 	MLX5_ASSERT(item);
2679 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2680 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2681 			return item;
2682 	}
2683 	return NULL;
2684 }
2685 
2686 /**
2687  * Get RSS action from the action list.
2688  *
2689  * @param[in] actions
2690  *   Pointer to the list of actions.
2691  *
2692  * @return
2693  *   Pointer to the RSS action if exist, else return NULL.
2694  */
2695 static const struct rte_flow_action_rss*
2696 flow_get_rss_action(const struct rte_flow_action actions[])
2697 {
2698 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2699 		switch (actions->type) {
2700 		case RTE_FLOW_ACTION_TYPE_RSS:
2701 			return (const struct rte_flow_action_rss *)
2702 			       actions->conf;
2703 		default:
2704 			break;
2705 		}
2706 	}
2707 	return NULL;
2708 }
2709 
2710 static unsigned int
2711 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2712 {
2713 	const struct rte_flow_item *item;
2714 	unsigned int has_vlan = 0;
2715 
2716 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2717 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2718 			has_vlan = 1;
2719 			break;
2720 		}
2721 	}
2722 	if (has_vlan)
2723 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2724 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2725 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2726 			       MLX5_EXPANSION_ROOT_OUTER;
2727 }
2728 
2729 /**
2730  * Get QUEUE/RSS action from the action list.
2731  *
2732  * @param[in] actions
2733  *   Pointer to the list of actions.
2734  * @param[out] qrss
2735  *   Pointer to the return pointer.
2736  * @param[out] qrss_type
2737  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2738  *   if no QUEUE/RSS is found.
2739  *
2740  * @return
2741  *   Total number of actions.
2742  */
2743 static int
2744 flow_parse_qrss_action(const struct rte_flow_action actions[],
2745 		       const struct rte_flow_action **qrss)
2746 {
2747 	int actions_n = 0;
2748 
2749 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2750 		switch (actions->type) {
2751 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2752 		case RTE_FLOW_ACTION_TYPE_RSS:
2753 			*qrss = actions;
2754 			break;
2755 		default:
2756 			break;
2757 		}
2758 		actions_n++;
2759 	}
2760 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2761 	return actions_n + 1;
2762 }
2763 
2764 /**
2765  * Check meter action from the action list.
2766  *
2767  * @param[in] actions
2768  *   Pointer to the list of actions.
2769  * @param[out] mtr
2770  *   Pointer to the meter exist flag.
2771  *
2772  * @return
2773  *   Total number of actions.
2774  */
2775 static int
2776 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2777 {
2778 	int actions_n = 0;
2779 
2780 	MLX5_ASSERT(mtr);
2781 	*mtr = 0;
2782 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2783 		switch (actions->type) {
2784 		case RTE_FLOW_ACTION_TYPE_METER:
2785 			*mtr = 1;
2786 			break;
2787 		default:
2788 			break;
2789 		}
2790 		actions_n++;
2791 	}
2792 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2793 	return actions_n + 1;
2794 }
2795 
2796 /**
2797  * Check if the flow should be splited due to hairpin.
2798  * The reason for the split is that in current HW we can't
2799  * support encap on Rx, so if a flow have encap we move it
2800  * to Tx.
2801  *
2802  * @param dev
2803  *   Pointer to Ethernet device.
2804  * @param[in] attr
2805  *   Flow rule attributes.
2806  * @param[in] actions
2807  *   Associated actions (list terminated by the END action).
2808  *
2809  * @return
2810  *   > 0 the number of actions and the flow should be split,
2811  *   0 when no split required.
2812  */
2813 static int
2814 flow_check_hairpin_split(struct rte_eth_dev *dev,
2815 			 const struct rte_flow_attr *attr,
2816 			 const struct rte_flow_action actions[])
2817 {
2818 	int queue_action = 0;
2819 	int action_n = 0;
2820 	int encap = 0;
2821 	const struct rte_flow_action_queue *queue;
2822 	const struct rte_flow_action_rss *rss;
2823 	const struct rte_flow_action_raw_encap *raw_encap;
2824 
2825 	if (!attr->ingress)
2826 		return 0;
2827 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2828 		switch (actions->type) {
2829 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2830 			queue = actions->conf;
2831 			if (queue == NULL)
2832 				return 0;
2833 			if (mlx5_rxq_get_type(dev, queue->index) !=
2834 			    MLX5_RXQ_TYPE_HAIRPIN)
2835 				return 0;
2836 			queue_action = 1;
2837 			action_n++;
2838 			break;
2839 		case RTE_FLOW_ACTION_TYPE_RSS:
2840 			rss = actions->conf;
2841 			if (rss == NULL || rss->queue_num == 0)
2842 				return 0;
2843 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2844 			    MLX5_RXQ_TYPE_HAIRPIN)
2845 				return 0;
2846 			queue_action = 1;
2847 			action_n++;
2848 			break;
2849 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2850 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2851 			encap = 1;
2852 			action_n++;
2853 			break;
2854 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2855 			raw_encap = actions->conf;
2856 			if (raw_encap->size >
2857 			    (sizeof(struct rte_flow_item_eth) +
2858 			     sizeof(struct rte_flow_item_ipv4)))
2859 				encap = 1;
2860 			action_n++;
2861 			break;
2862 		default:
2863 			action_n++;
2864 			break;
2865 		}
2866 	}
2867 	if (encap == 1 && queue_action)
2868 		return action_n;
2869 	return 0;
2870 }
2871 
2872 /* Declare flow create/destroy prototype in advance. */
2873 static struct rte_flow *
2874 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2875 		 const struct rte_flow_attr *attr,
2876 		 const struct rte_flow_item items[],
2877 		 const struct rte_flow_action actions[],
2878 		 bool external, struct rte_flow_error *error);
2879 
2880 static void
2881 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2882 		  struct rte_flow *flow);
2883 
2884 /**
2885  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2886  *
2887  * As mark_id is unique, if there's already a registered flow for the mark_id,
2888  * return by increasing the reference counter of the resource. Otherwise, create
2889  * the resource (mcp_res) and flow.
2890  *
2891  * Flow looks like,
2892  *   - If ingress port is ANY and reg_c[1] is mark_id,
2893  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2894  *
2895  * For default flow (zero mark_id), flow is like,
2896  *   - If ingress port is ANY,
2897  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2898  *
2899  * @param dev
2900  *   Pointer to Ethernet device.
2901  * @param mark_id
2902  *   ID of MARK action, zero means default flow for META.
2903  * @param[out] error
2904  *   Perform verbose error reporting if not NULL.
2905  *
2906  * @return
2907  *   Associated resource on success, NULL otherwise and rte_errno is set.
2908  */
2909 static struct mlx5_flow_mreg_copy_resource *
2910 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2911 			  struct rte_flow_error *error)
2912 {
2913 	struct mlx5_priv *priv = dev->data->dev_private;
2914 	struct rte_flow_attr attr = {
2915 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2916 		.ingress = 1,
2917 	};
2918 	struct mlx5_rte_flow_item_tag tag_spec = {
2919 		.data = mark_id,
2920 	};
2921 	struct rte_flow_item items[] = {
2922 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2923 	};
2924 	struct rte_flow_action_mark ftag = {
2925 		.id = mark_id,
2926 	};
2927 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2928 		.dst = REG_B,
2929 		.src = 0,
2930 	};
2931 	struct rte_flow_action_jump jump = {
2932 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2933 	};
2934 	struct rte_flow_action actions[] = {
2935 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2936 	};
2937 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2938 	int ret;
2939 
2940 	/* Fill the register fileds in the flow. */
2941 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2942 	if (ret < 0)
2943 		return NULL;
2944 	tag_spec.id = ret;
2945 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2946 	if (ret < 0)
2947 		return NULL;
2948 	cp_mreg.src = ret;
2949 	/* Check if already registered. */
2950 	MLX5_ASSERT(priv->mreg_cp_tbl);
2951 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2952 	if (mcp_res) {
2953 		/* For non-default rule. */
2954 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2955 			mcp_res->refcnt++;
2956 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2957 			    mcp_res->refcnt == 1);
2958 		return mcp_res;
2959 	}
2960 	/* Provide the full width of FLAG specific value. */
2961 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2962 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2963 	/* Build a new flow. */
2964 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2965 		items[0] = (struct rte_flow_item){
2966 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2967 			.spec = &tag_spec,
2968 		};
2969 		items[1] = (struct rte_flow_item){
2970 			.type = RTE_FLOW_ITEM_TYPE_END,
2971 		};
2972 		actions[0] = (struct rte_flow_action){
2973 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2974 			.conf = &ftag,
2975 		};
2976 		actions[1] = (struct rte_flow_action){
2977 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2978 			.conf = &cp_mreg,
2979 		};
2980 		actions[2] = (struct rte_flow_action){
2981 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2982 			.conf = &jump,
2983 		};
2984 		actions[3] = (struct rte_flow_action){
2985 			.type = RTE_FLOW_ACTION_TYPE_END,
2986 		};
2987 	} else {
2988 		/* Default rule, wildcard match. */
2989 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2990 		items[0] = (struct rte_flow_item){
2991 			.type = RTE_FLOW_ITEM_TYPE_END,
2992 		};
2993 		actions[0] = (struct rte_flow_action){
2994 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2995 			.conf = &cp_mreg,
2996 		};
2997 		actions[1] = (struct rte_flow_action){
2998 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2999 			.conf = &jump,
3000 		};
3001 		actions[2] = (struct rte_flow_action){
3002 			.type = RTE_FLOW_ACTION_TYPE_END,
3003 		};
3004 	}
3005 	/* Build a new entry. */
3006 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3007 	if (!mcp_res) {
3008 		rte_errno = ENOMEM;
3009 		return NULL;
3010 	}
3011 	/*
3012 	 * The copy Flows are not included in any list. There
3013 	 * ones are referenced from other Flows and can not
3014 	 * be applied, removed, deleted in ardbitrary order
3015 	 * by list traversing.
3016 	 */
3017 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3018 					 actions, false, error);
3019 	if (!mcp_res->flow)
3020 		goto error;
3021 	mcp_res->refcnt++;
3022 	mcp_res->hlist_ent.key = mark_id;
3023 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3024 				&mcp_res->hlist_ent);
3025 	MLX5_ASSERT(!ret);
3026 	if (ret)
3027 		goto error;
3028 	return mcp_res;
3029 error:
3030 	if (mcp_res->flow)
3031 		flow_list_destroy(dev, NULL, mcp_res->flow);
3032 	rte_free(mcp_res);
3033 	return NULL;
3034 }
3035 
3036 /**
3037  * Release flow in RX_CP_TBL.
3038  *
3039  * @param dev
3040  *   Pointer to Ethernet device.
3041  * @flow
3042  *   Parent flow for wich copying is provided.
3043  */
3044 static void
3045 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3046 			  struct rte_flow *flow)
3047 {
3048 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3049 	struct mlx5_priv *priv = dev->data->dev_private;
3050 
3051 	if (!mcp_res || !priv->mreg_cp_tbl)
3052 		return;
3053 	if (flow->copy_applied) {
3054 		MLX5_ASSERT(mcp_res->appcnt);
3055 		flow->copy_applied = 0;
3056 		--mcp_res->appcnt;
3057 		if (!mcp_res->appcnt)
3058 			flow_drv_remove(dev, mcp_res->flow);
3059 	}
3060 	/*
3061 	 * We do not check availability of metadata registers here,
3062 	 * because copy resources are not allocated in this case.
3063 	 */
3064 	if (--mcp_res->refcnt)
3065 		return;
3066 	MLX5_ASSERT(mcp_res->flow);
3067 	flow_list_destroy(dev, NULL, mcp_res->flow);
3068 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3069 	rte_free(mcp_res);
3070 	flow->mreg_copy = NULL;
3071 }
3072 
3073 /**
3074  * Start flow in RX_CP_TBL.
3075  *
3076  * @param dev
3077  *   Pointer to Ethernet device.
3078  * @flow
3079  *   Parent flow for wich copying is provided.
3080  *
3081  * @return
3082  *   0 on success, a negative errno value otherwise and rte_errno is set.
3083  */
3084 static int
3085 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3086 			    struct rte_flow *flow)
3087 {
3088 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3089 	int ret;
3090 
3091 	if (!mcp_res || flow->copy_applied)
3092 		return 0;
3093 	if (!mcp_res->appcnt) {
3094 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3095 		if (ret)
3096 			return ret;
3097 	}
3098 	++mcp_res->appcnt;
3099 	flow->copy_applied = 1;
3100 	return 0;
3101 }
3102 
3103 /**
3104  * Stop flow in RX_CP_TBL.
3105  *
3106  * @param dev
3107  *   Pointer to Ethernet device.
3108  * @flow
3109  *   Parent flow for wich copying is provided.
3110  */
3111 static void
3112 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3113 			   struct rte_flow *flow)
3114 {
3115 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3116 
3117 	if (!mcp_res || !flow->copy_applied)
3118 		return;
3119 	MLX5_ASSERT(mcp_res->appcnt);
3120 	--mcp_res->appcnt;
3121 	flow->copy_applied = 0;
3122 	if (!mcp_res->appcnt)
3123 		flow_drv_remove(dev, mcp_res->flow);
3124 }
3125 
3126 /**
3127  * Remove the default copy action from RX_CP_TBL.
3128  *
3129  * @param dev
3130  *   Pointer to Ethernet device.
3131  */
3132 static void
3133 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3134 {
3135 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3136 	struct mlx5_priv *priv = dev->data->dev_private;
3137 
3138 	/* Check if default flow is registered. */
3139 	if (!priv->mreg_cp_tbl)
3140 		return;
3141 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3142 					    MLX5_DEFAULT_COPY_ID);
3143 	if (!mcp_res)
3144 		return;
3145 	MLX5_ASSERT(mcp_res->flow);
3146 	flow_list_destroy(dev, NULL, mcp_res->flow);
3147 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3148 	rte_free(mcp_res);
3149 }
3150 
3151 /**
3152  * Add the default copy action in in RX_CP_TBL.
3153  *
3154  * @param dev
3155  *   Pointer to Ethernet device.
3156  * @param[out] error
3157  *   Perform verbose error reporting if not NULL.
3158  *
3159  * @return
3160  *   0 for success, negative value otherwise and rte_errno is set.
3161  */
3162 static int
3163 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3164 				  struct rte_flow_error *error)
3165 {
3166 	struct mlx5_priv *priv = dev->data->dev_private;
3167 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3168 
3169 	/* Check whether extensive metadata feature is engaged. */
3170 	if (!priv->config.dv_flow_en ||
3171 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3172 	    !mlx5_flow_ext_mreg_supported(dev) ||
3173 	    !priv->sh->dv_regc0_mask)
3174 		return 0;
3175 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3176 	if (!mcp_res)
3177 		return -rte_errno;
3178 	return 0;
3179 }
3180 
3181 /**
3182  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3183  *
3184  * All the flow having Q/RSS action should be split by
3185  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3186  * performs the following,
3187  *   - CQE->flow_tag := reg_c[1] (MARK)
3188  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3189  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3190  * but there should be a flow per each MARK ID set by MARK action.
3191  *
3192  * For the aforementioned reason, if there's a MARK action in flow's action
3193  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3194  * the MARK ID to CQE's flow_tag like,
3195  *   - If reg_c[1] is mark_id,
3196  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3197  *
3198  * For SET_META action which stores value in reg_c[0], as the destination is
3199  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3200  * MARK ID means the default flow. The default flow looks like,
3201  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3202  *
3203  * @param dev
3204  *   Pointer to Ethernet device.
3205  * @param flow
3206  *   Pointer to flow structure.
3207  * @param[in] actions
3208  *   Pointer to the list of actions.
3209  * @param[out] error
3210  *   Perform verbose error reporting if not NULL.
3211  *
3212  * @return
3213  *   0 on success, negative value otherwise and rte_errno is set.
3214  */
3215 static int
3216 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3217 			    struct rte_flow *flow,
3218 			    const struct rte_flow_action *actions,
3219 			    struct rte_flow_error *error)
3220 {
3221 	struct mlx5_priv *priv = dev->data->dev_private;
3222 	struct mlx5_dev_config *config = &priv->config;
3223 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3224 	const struct rte_flow_action_mark *mark;
3225 
3226 	/* Check whether extensive metadata feature is engaged. */
3227 	if (!config->dv_flow_en ||
3228 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3229 	    !mlx5_flow_ext_mreg_supported(dev) ||
3230 	    !priv->sh->dv_regc0_mask)
3231 		return 0;
3232 	/* Find MARK action. */
3233 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3234 		switch (actions->type) {
3235 		case RTE_FLOW_ACTION_TYPE_FLAG:
3236 			mcp_res = flow_mreg_add_copy_action
3237 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3238 			if (!mcp_res)
3239 				return -rte_errno;
3240 			flow->mreg_copy = mcp_res;
3241 			if (dev->data->dev_started) {
3242 				mcp_res->appcnt++;
3243 				flow->copy_applied = 1;
3244 			}
3245 			return 0;
3246 		case RTE_FLOW_ACTION_TYPE_MARK:
3247 			mark = (const struct rte_flow_action_mark *)
3248 				actions->conf;
3249 			mcp_res =
3250 				flow_mreg_add_copy_action(dev, mark->id, error);
3251 			if (!mcp_res)
3252 				return -rte_errno;
3253 			flow->mreg_copy = mcp_res;
3254 			if (dev->data->dev_started) {
3255 				mcp_res->appcnt++;
3256 				flow->copy_applied = 1;
3257 			}
3258 			return 0;
3259 		default:
3260 			break;
3261 		}
3262 	}
3263 	return 0;
3264 }
3265 
3266 #define MLX5_MAX_SPLIT_ACTIONS 24
3267 #define MLX5_MAX_SPLIT_ITEMS 24
3268 
3269 /**
3270  * Split the hairpin flow.
3271  * Since HW can't support encap on Rx we move the encap to Tx.
3272  * If the count action is after the encap then we also
3273  * move the count action. in this case the count will also measure
3274  * the outer bytes.
3275  *
3276  * @param dev
3277  *   Pointer to Ethernet device.
3278  * @param[in] actions
3279  *   Associated actions (list terminated by the END action).
3280  * @param[out] actions_rx
3281  *   Rx flow actions.
3282  * @param[out] actions_tx
3283  *   Tx flow actions..
3284  * @param[out] pattern_tx
3285  *   The pattern items for the Tx flow.
3286  * @param[out] flow_id
3287  *   The flow ID connected to this flow.
3288  *
3289  * @return
3290  *   0 on success.
3291  */
3292 static int
3293 flow_hairpin_split(struct rte_eth_dev *dev,
3294 		   const struct rte_flow_action actions[],
3295 		   struct rte_flow_action actions_rx[],
3296 		   struct rte_flow_action actions_tx[],
3297 		   struct rte_flow_item pattern_tx[],
3298 		   uint32_t *flow_id)
3299 {
3300 	struct mlx5_priv *priv = dev->data->dev_private;
3301 	const struct rte_flow_action_raw_encap *raw_encap;
3302 	const struct rte_flow_action_raw_decap *raw_decap;
3303 	struct mlx5_rte_flow_action_set_tag *set_tag;
3304 	struct rte_flow_action *tag_action;
3305 	struct mlx5_rte_flow_item_tag *tag_item;
3306 	struct rte_flow_item *item;
3307 	char *addr;
3308 	int encap = 0;
3309 
3310 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3311 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3312 		switch (actions->type) {
3313 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3314 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3315 			rte_memcpy(actions_tx, actions,
3316 			       sizeof(struct rte_flow_action));
3317 			actions_tx++;
3318 			break;
3319 		case RTE_FLOW_ACTION_TYPE_COUNT:
3320 			if (encap) {
3321 				rte_memcpy(actions_tx, actions,
3322 					   sizeof(struct rte_flow_action));
3323 				actions_tx++;
3324 			} else {
3325 				rte_memcpy(actions_rx, actions,
3326 					   sizeof(struct rte_flow_action));
3327 				actions_rx++;
3328 			}
3329 			break;
3330 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3331 			raw_encap = actions->conf;
3332 			if (raw_encap->size >
3333 			    (sizeof(struct rte_flow_item_eth) +
3334 			     sizeof(struct rte_flow_item_ipv4))) {
3335 				memcpy(actions_tx, actions,
3336 				       sizeof(struct rte_flow_action));
3337 				actions_tx++;
3338 				encap = 1;
3339 			} else {
3340 				rte_memcpy(actions_rx, actions,
3341 					   sizeof(struct rte_flow_action));
3342 				actions_rx++;
3343 			}
3344 			break;
3345 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3346 			raw_decap = actions->conf;
3347 			if (raw_decap->size <
3348 			    (sizeof(struct rte_flow_item_eth) +
3349 			     sizeof(struct rte_flow_item_ipv4))) {
3350 				memcpy(actions_tx, actions,
3351 				       sizeof(struct rte_flow_action));
3352 				actions_tx++;
3353 			} else {
3354 				rte_memcpy(actions_rx, actions,
3355 					   sizeof(struct rte_flow_action));
3356 				actions_rx++;
3357 			}
3358 			break;
3359 		default:
3360 			rte_memcpy(actions_rx, actions,
3361 				   sizeof(struct rte_flow_action));
3362 			actions_rx++;
3363 			break;
3364 		}
3365 	}
3366 	/* Add set meta action and end action for the Rx flow. */
3367 	tag_action = actions_rx;
3368 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3369 	actions_rx++;
3370 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3371 	actions_rx++;
3372 	set_tag = (void *)actions_rx;
3373 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3374 	MLX5_ASSERT(set_tag->id > REG_NONE);
3375 	set_tag->data = *flow_id;
3376 	tag_action->conf = set_tag;
3377 	/* Create Tx item list. */
3378 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3379 	addr = (void *)&pattern_tx[2];
3380 	item = pattern_tx;
3381 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3382 	tag_item = (void *)addr;
3383 	tag_item->data = *flow_id;
3384 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3385 	MLX5_ASSERT(set_tag->id > REG_NONE);
3386 	item->spec = tag_item;
3387 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3388 	tag_item = (void *)addr;
3389 	tag_item->data = UINT32_MAX;
3390 	tag_item->id = UINT16_MAX;
3391 	item->mask = tag_item;
3392 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3393 	item->last = NULL;
3394 	item++;
3395 	item->type = RTE_FLOW_ITEM_TYPE_END;
3396 	return 0;
3397 }
3398 
3399 /**
3400  * The last stage of splitting chain, just creates the subflow
3401  * without any modification.
3402  *
3403  * @param dev
3404  *   Pointer to Ethernet device.
3405  * @param[in] flow
3406  *   Parent flow structure pointer.
3407  * @param[in, out] sub_flow
3408  *   Pointer to return the created subflow, may be NULL.
3409  * @param[in] attr
3410  *   Flow rule attributes.
3411  * @param[in] items
3412  *   Pattern specification (list terminated by the END pattern item).
3413  * @param[in] actions
3414  *   Associated actions (list terminated by the END action).
3415  * @param[in] external
3416  *   This flow rule is created by request external to PMD.
3417  * @param[out] error
3418  *   Perform verbose error reporting if not NULL.
3419  * @return
3420  *   0 on success, negative value otherwise
3421  */
3422 static int
3423 flow_create_split_inner(struct rte_eth_dev *dev,
3424 			struct rte_flow *flow,
3425 			struct mlx5_flow **sub_flow,
3426 			const struct rte_flow_attr *attr,
3427 			const struct rte_flow_item items[],
3428 			const struct rte_flow_action actions[],
3429 			bool external, struct rte_flow_error *error)
3430 {
3431 	struct mlx5_flow *dev_flow;
3432 
3433 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3434 	if (!dev_flow)
3435 		return -rte_errno;
3436 	dev_flow->flow = flow;
3437 	dev_flow->external = external;
3438 	/* Subflow object was created, we must include one in the list. */
3439 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3440 	if (sub_flow)
3441 		*sub_flow = dev_flow;
3442 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3443 }
3444 
3445 /**
3446  * Split the meter flow.
3447  *
3448  * As meter flow will split to three sub flow, other than meter
3449  * action, the other actions make sense to only meter accepts
3450  * the packet. If it need to be dropped, no other additional
3451  * actions should be take.
3452  *
3453  * One kind of special action which decapsulates the L3 tunnel
3454  * header will be in the prefix sub flow, as not to take the
3455  * L3 tunnel header into account.
3456  *
3457  * @param dev
3458  *   Pointer to Ethernet device.
3459  * @param[in] actions
3460  *   Associated actions (list terminated by the END action).
3461  * @param[out] actions_sfx
3462  *   Suffix flow actions.
3463  * @param[out] actions_pre
3464  *   Prefix flow actions.
3465  * @param[out] pattern_sfx
3466  *   The pattern items for the suffix flow.
3467  * @param[out] tag_sfx
3468  *   Pointer to suffix flow tag.
3469  *
3470  * @return
3471  *   0 on success.
3472  */
3473 static int
3474 flow_meter_split_prep(struct rte_eth_dev *dev,
3475 		 const struct rte_flow_action actions[],
3476 		 struct rte_flow_action actions_sfx[],
3477 		 struct rte_flow_action actions_pre[])
3478 {
3479 	struct rte_flow_action *tag_action = NULL;
3480 	struct mlx5_rte_flow_action_set_tag *set_tag;
3481 	struct rte_flow_error error;
3482 	const struct rte_flow_action_raw_encap *raw_encap;
3483 	const struct rte_flow_action_raw_decap *raw_decap;
3484 	uint32_t tag_id;
3485 
3486 	/* Prepare the actions for prefix and suffix flow. */
3487 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3488 		switch (actions->type) {
3489 		case RTE_FLOW_ACTION_TYPE_METER:
3490 			/* Add the extra tag action first. */
3491 			tag_action = actions_pre;
3492 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3493 			actions_pre++;
3494 			memcpy(actions_pre, actions,
3495 			       sizeof(struct rte_flow_action));
3496 			actions_pre++;
3497 			break;
3498 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3499 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3500 			memcpy(actions_pre, actions,
3501 			       sizeof(struct rte_flow_action));
3502 			actions_pre++;
3503 			break;
3504 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3505 			raw_encap = actions->conf;
3506 			if (raw_encap->size >
3507 			    (sizeof(struct rte_flow_item_eth) +
3508 			     sizeof(struct rte_flow_item_ipv4))) {
3509 				memcpy(actions_sfx, actions,
3510 				       sizeof(struct rte_flow_action));
3511 				actions_sfx++;
3512 			} else {
3513 				rte_memcpy(actions_pre, actions,
3514 					   sizeof(struct rte_flow_action));
3515 				actions_pre++;
3516 			}
3517 			break;
3518 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3519 			raw_decap = actions->conf;
3520 			/* Size 0 decap means 50 bytes as vxlan decap. */
3521 			if (raw_decap->size && (raw_decap->size <
3522 			    (sizeof(struct rte_flow_item_eth) +
3523 			     sizeof(struct rte_flow_item_ipv4)))) {
3524 				memcpy(actions_sfx, actions,
3525 				       sizeof(struct rte_flow_action));
3526 				actions_sfx++;
3527 			} else {
3528 				rte_memcpy(actions_pre, actions,
3529 					   sizeof(struct rte_flow_action));
3530 				actions_pre++;
3531 			}
3532 			break;
3533 		default:
3534 			memcpy(actions_sfx, actions,
3535 				sizeof(struct rte_flow_action));
3536 			actions_sfx++;
3537 			break;
3538 		}
3539 	}
3540 	/* Add end action to the actions. */
3541 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3542 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3543 	actions_pre++;
3544 	/* Set the tag. */
3545 	set_tag = (void *)actions_pre;
3546 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3547 	/*
3548 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3549 	 */
3550 	tag_id = flow_qrss_get_id(dev);
3551 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3552 	assert(tag_action);
3553 	tag_action->conf = set_tag;
3554 	return tag_id;
3555 }
3556 
3557 /**
3558  * Split action list having QUEUE/RSS for metadata register copy.
3559  *
3560  * Once Q/RSS action is detected in user's action list, the flow action
3561  * should be split in order to copy metadata registers, which will happen in
3562  * RX_CP_TBL like,
3563  *   - CQE->flow_tag := reg_c[1] (MARK)
3564  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3565  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3566  * This is because the last action of each flow must be a terminal action
3567  * (QUEUE, RSS or DROP).
3568  *
3569  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3570  * stored and kept in the mlx5_flow structure per each sub_flow.
3571  *
3572  * The Q/RSS action is replaced with,
3573  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3574  * And the following JUMP action is added at the end,
3575  *   - JUMP, to RX_CP_TBL.
3576  *
3577  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3578  * flow_create_split_metadata() routine. The flow will look like,
3579  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3580  *
3581  * @param dev
3582  *   Pointer to Ethernet device.
3583  * @param[out] split_actions
3584  *   Pointer to store split actions to jump to CP_TBL.
3585  * @param[in] actions
3586  *   Pointer to the list of original flow actions.
3587  * @param[in] qrss
3588  *   Pointer to the Q/RSS action.
3589  * @param[in] actions_n
3590  *   Number of original actions.
3591  * @param[out] error
3592  *   Perform verbose error reporting if not NULL.
3593  *
3594  * @return
3595  *   non-zero unique flow_id on success, otherwise 0 and
3596  *   error/rte_error are set.
3597  */
3598 static uint32_t
3599 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3600 			  struct rte_flow_action *split_actions,
3601 			  const struct rte_flow_action *actions,
3602 			  const struct rte_flow_action *qrss,
3603 			  int actions_n, struct rte_flow_error *error)
3604 {
3605 	struct mlx5_rte_flow_action_set_tag *set_tag;
3606 	struct rte_flow_action_jump *jump;
3607 	const int qrss_idx = qrss - actions;
3608 	uint32_t flow_id = 0;
3609 	int ret = 0;
3610 
3611 	/*
3612 	 * Given actions will be split
3613 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3614 	 * - Add jump to mreg CP_TBL.
3615 	 * As a result, there will be one more action.
3616 	 */
3617 	++actions_n;
3618 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3619 	set_tag = (void *)(split_actions + actions_n);
3620 	/*
3621 	 * If tag action is not set to void(it means we are not the meter
3622 	 * suffix flow), add the tag action. Since meter suffix flow already
3623 	 * has the tag added.
3624 	 */
3625 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3626 		/*
3627 		 * Allocate the new subflow ID. This one is unique within
3628 		 * device and not shared with representors. Otherwise,
3629 		 * we would have to resolve multi-thread access synch
3630 		 * issue. Each flow on the shared device is appended
3631 		 * with source vport identifier, so the resulting
3632 		 * flows will be unique in the shared (by master and
3633 		 * representors) domain even if they have coinciding
3634 		 * IDs.
3635 		 */
3636 		flow_id = flow_qrss_get_id(dev);
3637 		if (!flow_id)
3638 			return rte_flow_error_set(error, ENOMEM,
3639 						  RTE_FLOW_ERROR_TYPE_ACTION,
3640 						  NULL, "can't allocate id "
3641 						  "for split Q/RSS subflow");
3642 		/* Internal SET_TAG action to set flow ID. */
3643 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3644 			.data = flow_id,
3645 		};
3646 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3647 		if (ret < 0)
3648 			return ret;
3649 		set_tag->id = ret;
3650 		/* Construct new actions array. */
3651 		/* Replace QUEUE/RSS action. */
3652 		split_actions[qrss_idx] = (struct rte_flow_action){
3653 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3654 			.conf = set_tag,
3655 		};
3656 	}
3657 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3658 	jump = (void *)(set_tag + 1);
3659 	*jump = (struct rte_flow_action_jump){
3660 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3661 	};
3662 	split_actions[actions_n - 2] = (struct rte_flow_action){
3663 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3664 		.conf = jump,
3665 	};
3666 	split_actions[actions_n - 1] = (struct rte_flow_action){
3667 		.type = RTE_FLOW_ACTION_TYPE_END,
3668 	};
3669 	return flow_id;
3670 }
3671 
3672 /**
3673  * Extend the given action list for Tx metadata copy.
3674  *
3675  * Copy the given action list to the ext_actions and add flow metadata register
3676  * copy action in order to copy reg_a set by WQE to reg_c[0].
3677  *
3678  * @param[out] ext_actions
3679  *   Pointer to the extended action list.
3680  * @param[in] actions
3681  *   Pointer to the list of actions.
3682  * @param[in] actions_n
3683  *   Number of actions in the list.
3684  * @param[out] error
3685  *   Perform verbose error reporting if not NULL.
3686  *
3687  * @return
3688  *   0 on success, negative value otherwise
3689  */
3690 static int
3691 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3692 		       struct rte_flow_action *ext_actions,
3693 		       const struct rte_flow_action *actions,
3694 		       int actions_n, struct rte_flow_error *error)
3695 {
3696 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3697 		(struct mlx5_flow_action_copy_mreg *)
3698 			(ext_actions + actions_n + 1);
3699 	int ret;
3700 
3701 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3702 	if (ret < 0)
3703 		return ret;
3704 	cp_mreg->dst = ret;
3705 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3706 	if (ret < 0)
3707 		return ret;
3708 	cp_mreg->src = ret;
3709 	memcpy(ext_actions, actions,
3710 			sizeof(*ext_actions) * actions_n);
3711 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3712 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3713 		.conf = cp_mreg,
3714 	};
3715 	ext_actions[actions_n] = (struct rte_flow_action){
3716 		.type = RTE_FLOW_ACTION_TYPE_END,
3717 	};
3718 	return 0;
3719 }
3720 
3721 /**
3722  * The splitting for metadata feature.
3723  *
3724  * - Q/RSS action on NIC Rx should be split in order to pass by
3725  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3726  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3727  *
3728  * - All the actions on NIC Tx should have a mreg copy action to
3729  *   copy reg_a from WQE to reg_c[0].
3730  *
3731  * @param dev
3732  *   Pointer to Ethernet device.
3733  * @param[in] flow
3734  *   Parent flow structure pointer.
3735  * @param[in] attr
3736  *   Flow rule attributes.
3737  * @param[in] items
3738  *   Pattern specification (list terminated by the END pattern item).
3739  * @param[in] actions
3740  *   Associated actions (list terminated by the END action).
3741  * @param[in] external
3742  *   This flow rule is created by request external to PMD.
3743  * @param[out] error
3744  *   Perform verbose error reporting if not NULL.
3745  * @return
3746  *   0 on success, negative value otherwise
3747  */
3748 static int
3749 flow_create_split_metadata(struct rte_eth_dev *dev,
3750 			   struct rte_flow *flow,
3751 			   const struct rte_flow_attr *attr,
3752 			   const struct rte_flow_item items[],
3753 			   const struct rte_flow_action actions[],
3754 			   bool external, struct rte_flow_error *error)
3755 {
3756 	struct mlx5_priv *priv = dev->data->dev_private;
3757 	struct mlx5_dev_config *config = &priv->config;
3758 	const struct rte_flow_action *qrss = NULL;
3759 	struct rte_flow_action *ext_actions = NULL;
3760 	struct mlx5_flow *dev_flow = NULL;
3761 	uint32_t qrss_id = 0;
3762 	int mtr_sfx = 0;
3763 	size_t act_size;
3764 	int actions_n;
3765 	int ret;
3766 
3767 	/* Check whether extensive metadata feature is engaged. */
3768 	if (!config->dv_flow_en ||
3769 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3770 	    !mlx5_flow_ext_mreg_supported(dev))
3771 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3772 					       actions, external, error);
3773 	actions_n = flow_parse_qrss_action(actions, &qrss);
3774 	if (qrss) {
3775 		/* Exclude hairpin flows from splitting. */
3776 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3777 			const struct rte_flow_action_queue *queue;
3778 
3779 			queue = qrss->conf;
3780 			if (mlx5_rxq_get_type(dev, queue->index) ==
3781 			    MLX5_RXQ_TYPE_HAIRPIN)
3782 				qrss = NULL;
3783 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3784 			const struct rte_flow_action_rss *rss;
3785 
3786 			rss = qrss->conf;
3787 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3788 			    MLX5_RXQ_TYPE_HAIRPIN)
3789 				qrss = NULL;
3790 		}
3791 	}
3792 	if (qrss) {
3793 		/* Check if it is in meter suffix table. */
3794 		mtr_sfx = attr->group == (attr->transfer ?
3795 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3796 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3797 		/*
3798 		 * Q/RSS action on NIC Rx should be split in order to pass by
3799 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3800 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3801 		 */
3802 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3803 			   sizeof(struct rte_flow_action_set_tag) +
3804 			   sizeof(struct rte_flow_action_jump);
3805 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3806 		if (!ext_actions)
3807 			return rte_flow_error_set(error, ENOMEM,
3808 						  RTE_FLOW_ERROR_TYPE_ACTION,
3809 						  NULL, "no memory to split "
3810 						  "metadata flow");
3811 		/*
3812 		 * If we are the suffix flow of meter, tag already exist.
3813 		 * Set the tag action to void.
3814 		 */
3815 		if (mtr_sfx)
3816 			ext_actions[qrss - actions].type =
3817 						RTE_FLOW_ACTION_TYPE_VOID;
3818 		else
3819 			ext_actions[qrss - actions].type =
3820 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3821 		/*
3822 		 * Create the new actions list with removed Q/RSS action
3823 		 * and appended set tag and jump to register copy table
3824 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3825 		 * in advance, because it is needed for set tag action.
3826 		 */
3827 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3828 						    qrss, actions_n, error);
3829 		if (!mtr_sfx && !qrss_id) {
3830 			ret = -rte_errno;
3831 			goto exit;
3832 		}
3833 	} else if (attr->egress && !attr->transfer) {
3834 		/*
3835 		 * All the actions on NIC Tx should have a metadata register
3836 		 * copy action to copy reg_a from WQE to reg_c[meta]
3837 		 */
3838 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3839 			   sizeof(struct mlx5_flow_action_copy_mreg);
3840 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3841 		if (!ext_actions)
3842 			return rte_flow_error_set(error, ENOMEM,
3843 						  RTE_FLOW_ERROR_TYPE_ACTION,
3844 						  NULL, "no memory to split "
3845 						  "metadata flow");
3846 		/* Create the action list appended with copy register. */
3847 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3848 					     actions_n, error);
3849 		if (ret < 0)
3850 			goto exit;
3851 	}
3852 	/* Add the unmodified original or prefix subflow. */
3853 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3854 				      ext_actions ? ext_actions : actions,
3855 				      external, error);
3856 	if (ret < 0)
3857 		goto exit;
3858 	MLX5_ASSERT(dev_flow);
3859 	if (qrss) {
3860 		const struct rte_flow_attr q_attr = {
3861 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3862 			.ingress = 1,
3863 		};
3864 		/* Internal PMD action to set register. */
3865 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3866 			.data = qrss_id,
3867 			.id = 0,
3868 		};
3869 		struct rte_flow_item q_items[] = {
3870 			{
3871 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3872 				.spec = &q_tag_spec,
3873 				.last = NULL,
3874 				.mask = NULL,
3875 			},
3876 			{
3877 				.type = RTE_FLOW_ITEM_TYPE_END,
3878 			},
3879 		};
3880 		struct rte_flow_action q_actions[] = {
3881 			{
3882 				.type = qrss->type,
3883 				.conf = qrss->conf,
3884 			},
3885 			{
3886 				.type = RTE_FLOW_ACTION_TYPE_END,
3887 			},
3888 		};
3889 		uint64_t hash_fields = dev_flow->hash_fields;
3890 
3891 		/*
3892 		 * Configure the tag item only if there is no meter subflow.
3893 		 * Since tag is already marked in the meter suffix subflow
3894 		 * we can just use the meter suffix items as is.
3895 		 */
3896 		if (qrss_id) {
3897 			/* Not meter subflow. */
3898 			MLX5_ASSERT(!mtr_sfx);
3899 			/*
3900 			 * Put unique id in prefix flow due to it is destroyed
3901 			 * after suffix flow and id will be freed after there
3902 			 * is no actual flows with this id and identifier
3903 			 * reallocation becomes possible (for example, for
3904 			 * other flows in other threads).
3905 			 */
3906 			dev_flow->qrss_id = qrss_id;
3907 			qrss_id = 0;
3908 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3909 						   error);
3910 			if (ret < 0)
3911 				goto exit;
3912 			q_tag_spec.id = ret;
3913 		}
3914 		dev_flow = NULL;
3915 		/* Add suffix subflow to execute Q/RSS. */
3916 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3917 					      &q_attr, mtr_sfx ? items :
3918 					      q_items, q_actions,
3919 					      external, error);
3920 		if (ret < 0)
3921 			goto exit;
3922 		MLX5_ASSERT(dev_flow);
3923 		dev_flow->hash_fields = hash_fields;
3924 	}
3925 
3926 exit:
3927 	/*
3928 	 * We do not destroy the partially created sub_flows in case of error.
3929 	 * These ones are included into parent flow list and will be destroyed
3930 	 * by flow_drv_destroy.
3931 	 */
3932 	flow_qrss_free_id(dev, qrss_id);
3933 	rte_free(ext_actions);
3934 	return ret;
3935 }
3936 
3937 /**
3938  * The splitting for meter feature.
3939  *
3940  * - The meter flow will be split to two flows as prefix and
3941  *   suffix flow. The packets make sense only it pass the prefix
3942  *   meter action.
3943  *
3944  * - Reg_C_5 is used for the packet to match betweend prefix and
3945  *   suffix flow.
3946  *
3947  * @param dev
3948  *   Pointer to Ethernet device.
3949  * @param[in] flow
3950  *   Parent flow structure pointer.
3951  * @param[in] attr
3952  *   Flow rule attributes.
3953  * @param[in] items
3954  *   Pattern specification (list terminated by the END pattern item).
3955  * @param[in] actions
3956  *   Associated actions (list terminated by the END action).
3957  * @param[in] external
3958  *   This flow rule is created by request external to PMD.
3959  * @param[out] error
3960  *   Perform verbose error reporting if not NULL.
3961  * @return
3962  *   0 on success, negative value otherwise
3963  */
3964 static int
3965 flow_create_split_meter(struct rte_eth_dev *dev,
3966 			   struct rte_flow *flow,
3967 			   const struct rte_flow_attr *attr,
3968 			   const struct rte_flow_item items[],
3969 			   const struct rte_flow_action actions[],
3970 			   bool external, struct rte_flow_error *error)
3971 {
3972 	struct mlx5_priv *priv = dev->data->dev_private;
3973 	struct rte_flow_action *sfx_actions = NULL;
3974 	struct rte_flow_action *pre_actions = NULL;
3975 	struct rte_flow_item *sfx_items = NULL;
3976 	const  struct rte_flow_item *sfx_port_id_item;
3977 	struct mlx5_flow *dev_flow = NULL;
3978 	struct rte_flow_attr sfx_attr = *attr;
3979 	uint32_t mtr = 0;
3980 	uint32_t mtr_tag_id = 0;
3981 	size_t act_size;
3982 	size_t item_size;
3983 	int actions_n = 0;
3984 	int ret;
3985 
3986 	if (priv->mtr_en)
3987 		actions_n = flow_check_meter_action(actions, &mtr);
3988 	if (mtr) {
3989 		struct mlx5_rte_flow_item_tag *tag_spec;
3990 		struct mlx5_rte_flow_item_tag *tag_mask;
3991 		/* The five prefix actions: meter, decap, encap, tag, end. */
3992 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3993 			   sizeof(struct rte_flow_action_set_tag);
3994 		/* tag, end. */
3995 #define METER_SUFFIX_ITEM 3
3996 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3997 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
3998 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3999 		if (!sfx_actions)
4000 			return rte_flow_error_set(error, ENOMEM,
4001 						  RTE_FLOW_ERROR_TYPE_ACTION,
4002 						  NULL, "no memory to split "
4003 						  "meter flow");
4004 		pre_actions = sfx_actions + actions_n;
4005 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4006 						     pre_actions);
4007 		if (!mtr_tag_id) {
4008 			ret = -rte_errno;
4009 			goto exit;
4010 		}
4011 		/* Add the prefix subflow. */
4012 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4013 						  pre_actions, external, error);
4014 		if (ret) {
4015 			ret = -rte_errno;
4016 			goto exit;
4017 		}
4018 		dev_flow->mtr_flow_id = mtr_tag_id;
4019 		/* Prepare the suffix flow match pattern. */
4020 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4021 			     act_size);
4022 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4023 			    METER_SUFFIX_ITEM);
4024 		tag_spec->data = dev_flow->mtr_flow_id << MLX5_MTR_COLOR_BITS;
4025 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4026 						    error);
4027 		tag_mask = tag_spec + 1;
4028 		tag_mask->data = 0xffffff00;
4029 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4030 		sfx_items->spec = tag_spec;
4031 		sfx_items->last = NULL;
4032 		sfx_items->mask = tag_mask;
4033 		sfx_items++;
4034 		sfx_port_id_item = find_port_id_item(items);
4035 		if (sfx_port_id_item) {
4036 			memcpy(sfx_items, sfx_port_id_item,
4037 			       sizeof(*sfx_items));
4038 			sfx_items++;
4039 		}
4040 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4041 		sfx_items -= sfx_port_id_item ? 2 : 1;
4042 		/* Setting the sfx group atrr. */
4043 		sfx_attr.group = sfx_attr.transfer ?
4044 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4045 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4046 	}
4047 	/* Add the prefix subflow. */
4048 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4049 					 sfx_items ? sfx_items : items,
4050 					 sfx_actions ? sfx_actions : actions,
4051 					 external, error);
4052 exit:
4053 	if (sfx_actions)
4054 		rte_free(sfx_actions);
4055 	return ret;
4056 }
4057 
4058 /**
4059  * Split the flow to subflow set. The splitters might be linked
4060  * in the chain, like this:
4061  * flow_create_split_outer() calls:
4062  *   flow_create_split_meter() calls:
4063  *     flow_create_split_metadata(meter_subflow_0) calls:
4064  *       flow_create_split_inner(metadata_subflow_0)
4065  *       flow_create_split_inner(metadata_subflow_1)
4066  *       flow_create_split_inner(metadata_subflow_2)
4067  *     flow_create_split_metadata(meter_subflow_1) calls:
4068  *       flow_create_split_inner(metadata_subflow_0)
4069  *       flow_create_split_inner(metadata_subflow_1)
4070  *       flow_create_split_inner(metadata_subflow_2)
4071  *
4072  * This provide flexible way to add new levels of flow splitting.
4073  * The all of successfully created subflows are included to the
4074  * parent flow dev_flow list.
4075  *
4076  * @param dev
4077  *   Pointer to Ethernet device.
4078  * @param[in] flow
4079  *   Parent flow structure pointer.
4080  * @param[in] attr
4081  *   Flow rule attributes.
4082  * @param[in] items
4083  *   Pattern specification (list terminated by the END pattern item).
4084  * @param[in] actions
4085  *   Associated actions (list terminated by the END action).
4086  * @param[in] external
4087  *   This flow rule is created by request external to PMD.
4088  * @param[out] error
4089  *   Perform verbose error reporting if not NULL.
4090  * @return
4091  *   0 on success, negative value otherwise
4092  */
4093 static int
4094 flow_create_split_outer(struct rte_eth_dev *dev,
4095 			struct rte_flow *flow,
4096 			const struct rte_flow_attr *attr,
4097 			const struct rte_flow_item items[],
4098 			const struct rte_flow_action actions[],
4099 			bool external, struct rte_flow_error *error)
4100 {
4101 	int ret;
4102 
4103 	ret = flow_create_split_meter(dev, flow, attr, items,
4104 					 actions, external, error);
4105 	MLX5_ASSERT(ret <= 0);
4106 	return ret;
4107 }
4108 
4109 /**
4110  * Create a flow and add it to @p list.
4111  *
4112  * @param dev
4113  *   Pointer to Ethernet device.
4114  * @param list
4115  *   Pointer to a TAILQ flow list. If this parameter NULL,
4116  *   no list insertion occurred, flow is just created,
4117  *   this is caller's responsibility to track the
4118  *   created flow.
4119  * @param[in] attr
4120  *   Flow rule attributes.
4121  * @param[in] items
4122  *   Pattern specification (list terminated by the END pattern item).
4123  * @param[in] actions
4124  *   Associated actions (list terminated by the END action).
4125  * @param[in] external
4126  *   This flow rule is created by request external to PMD.
4127  * @param[out] error
4128  *   Perform verbose error reporting if not NULL.
4129  *
4130  * @return
4131  *   A flow on success, NULL otherwise and rte_errno is set.
4132  */
4133 static struct rte_flow *
4134 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4135 		 const struct rte_flow_attr *attr,
4136 		 const struct rte_flow_item items[],
4137 		 const struct rte_flow_action actions[],
4138 		 bool external, struct rte_flow_error *error)
4139 {
4140 	struct mlx5_priv *priv = dev->data->dev_private;
4141 	struct rte_flow *flow = NULL;
4142 	struct mlx5_flow *dev_flow;
4143 	const struct rte_flow_action_rss *rss;
4144 	union {
4145 		struct rte_flow_expand_rss buf;
4146 		uint8_t buffer[2048];
4147 	} expand_buffer;
4148 	union {
4149 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4150 		uint8_t buffer[2048];
4151 	} actions_rx;
4152 	union {
4153 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4154 		uint8_t buffer[2048];
4155 	} actions_hairpin_tx;
4156 	union {
4157 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4158 		uint8_t buffer[2048];
4159 	} items_tx;
4160 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4161 	const struct rte_flow_action *p_actions_rx = actions;
4162 	uint32_t i;
4163 	uint32_t flow_size;
4164 	int hairpin_flow = 0;
4165 	uint32_t hairpin_id = 0;
4166 	struct rte_flow_attr attr_tx = { .priority = 0 };
4167 	int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4168 				    error);
4169 
4170 	if (ret < 0)
4171 		return NULL;
4172 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4173 	if (hairpin_flow > 0) {
4174 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4175 			rte_errno = EINVAL;
4176 			return NULL;
4177 		}
4178 		flow_hairpin_split(dev, actions, actions_rx.actions,
4179 				   actions_hairpin_tx.actions, items_tx.items,
4180 				   &hairpin_id);
4181 		p_actions_rx = actions_rx.actions;
4182 	}
4183 	flow_size = sizeof(struct rte_flow);
4184 	rss = flow_get_rss_action(p_actions_rx);
4185 	if (rss)
4186 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4187 					    sizeof(void *));
4188 	else
4189 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4190 	flow = rte_calloc(__func__, 1, flow_size, 0);
4191 	if (!flow) {
4192 		rte_errno = ENOMEM;
4193 		goto error_before_flow;
4194 	}
4195 	flow->drv_type = flow_get_drv_type(dev, attr);
4196 	if (hairpin_id != 0)
4197 		flow->hairpin_flow_id = hairpin_id;
4198 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4199 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4200 	flow->rss.queue = (void *)(flow + 1);
4201 	if (rss) {
4202 		/*
4203 		 * The following information is required by
4204 		 * mlx5_flow_hashfields_adjust() in advance.
4205 		 */
4206 		flow->rss.level = rss->level;
4207 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4208 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4209 	}
4210 	LIST_INIT(&flow->dev_flows);
4211 	if (rss && rss->types) {
4212 		unsigned int graph_root;
4213 
4214 		graph_root = find_graph_root(items, rss->level);
4215 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4216 					  items, rss->types,
4217 					  mlx5_support_expansion,
4218 					  graph_root);
4219 		MLX5_ASSERT(ret > 0 &&
4220 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4221 	} else {
4222 		buf->entries = 1;
4223 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4224 	}
4225 	for (i = 0; i < buf->entries; ++i) {
4226 		/*
4227 		 * The splitter may create multiple dev_flows,
4228 		 * depending on configuration. In the simplest
4229 		 * case it just creates unmodified original flow.
4230 		 */
4231 		ret = flow_create_split_outer(dev, flow, attr,
4232 					      buf->entry[i].pattern,
4233 					      p_actions_rx, external,
4234 					      error);
4235 		if (ret < 0)
4236 			goto error;
4237 	}
4238 	/* Create the tx flow. */
4239 	if (hairpin_flow) {
4240 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4241 		attr_tx.ingress = 0;
4242 		attr_tx.egress = 1;
4243 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4244 					    actions_hairpin_tx.actions, error);
4245 		if (!dev_flow)
4246 			goto error;
4247 		dev_flow->flow = flow;
4248 		dev_flow->external = 0;
4249 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4250 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4251 					 items_tx.items,
4252 					 actions_hairpin_tx.actions, error);
4253 		if (ret < 0)
4254 			goto error;
4255 	}
4256 	/*
4257 	 * Update the metadata register copy table. If extensive
4258 	 * metadata feature is enabled and registers are supported
4259 	 * we might create the extra rte_flow for each unique
4260 	 * MARK/FLAG action ID.
4261 	 *
4262 	 * The table is updated for ingress Flows only, because
4263 	 * the egress Flows belong to the different device and
4264 	 * copy table should be updated in peer NIC Rx domain.
4265 	 */
4266 	if (attr->ingress &&
4267 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4268 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4269 		if (ret)
4270 			goto error;
4271 	}
4272 	if (dev->data->dev_started) {
4273 		ret = flow_drv_apply(dev, flow, error);
4274 		if (ret < 0)
4275 			goto error;
4276 	}
4277 	if (list)
4278 		TAILQ_INSERT_TAIL(list, flow, next);
4279 	flow_rxq_flags_set(dev, flow);
4280 	return flow;
4281 error_before_flow:
4282 	if (hairpin_id)
4283 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4284 				     hairpin_id);
4285 	return NULL;
4286 error:
4287 	MLX5_ASSERT(flow);
4288 	flow_mreg_del_copy_action(dev, flow);
4289 	ret = rte_errno; /* Save rte_errno before cleanup. */
4290 	if (flow->hairpin_flow_id)
4291 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4292 				     flow->hairpin_flow_id);
4293 	MLX5_ASSERT(flow);
4294 	flow_drv_destroy(dev, flow);
4295 	rte_free(flow);
4296 	rte_errno = ret; /* Restore rte_errno. */
4297 	return NULL;
4298 }
4299 
4300 /**
4301  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4302  * incoming packets to table 1.
4303  *
4304  * Other flow rules, requested for group n, will be created in
4305  * e-switch table n+1.
4306  * Jump action to e-switch group n will be created to group n+1.
4307  *
4308  * Used when working in switchdev mode, to utilise advantages of table 1
4309  * and above.
4310  *
4311  * @param dev
4312  *   Pointer to Ethernet device.
4313  *
4314  * @return
4315  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4316  */
4317 struct rte_flow *
4318 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4319 {
4320 	const struct rte_flow_attr attr = {
4321 		.group = 0,
4322 		.priority = 0,
4323 		.ingress = 1,
4324 		.egress = 0,
4325 		.transfer = 1,
4326 	};
4327 	const struct rte_flow_item pattern = {
4328 		.type = RTE_FLOW_ITEM_TYPE_END,
4329 	};
4330 	struct rte_flow_action_jump jump = {
4331 		.group = 1,
4332 	};
4333 	const struct rte_flow_action actions[] = {
4334 		{
4335 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4336 			.conf = &jump,
4337 		},
4338 		{
4339 			.type = RTE_FLOW_ACTION_TYPE_END,
4340 		},
4341 	};
4342 	struct mlx5_priv *priv = dev->data->dev_private;
4343 	struct rte_flow_error error;
4344 
4345 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4346 				actions, false, &error);
4347 }
4348 
4349 /**
4350  * Create a flow.
4351  *
4352  * @see rte_flow_create()
4353  * @see rte_flow_ops
4354  */
4355 struct rte_flow *
4356 mlx5_flow_create(struct rte_eth_dev *dev,
4357 		 const struct rte_flow_attr *attr,
4358 		 const struct rte_flow_item items[],
4359 		 const struct rte_flow_action actions[],
4360 		 struct rte_flow_error *error)
4361 {
4362 	struct mlx5_priv *priv = dev->data->dev_private;
4363 
4364 	return flow_list_create(dev, &priv->flows,
4365 				attr, items, actions, true, error);
4366 }
4367 
4368 /**
4369  * Destroy a flow in a list.
4370  *
4371  * @param dev
4372  *   Pointer to Ethernet device.
4373  * @param list
4374  *   Pointer to a TAILQ flow list. If this parameter NULL,
4375  *   there is no flow removal from the list.
4376  * @param[in] flow
4377  *   Flow to destroy.
4378  */
4379 static void
4380 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4381 		  struct rte_flow *flow)
4382 {
4383 	struct mlx5_priv *priv = dev->data->dev_private;
4384 
4385 	/*
4386 	 * Update RX queue flags only if port is started, otherwise it is
4387 	 * already clean.
4388 	 */
4389 	if (dev->data->dev_started)
4390 		flow_rxq_flags_trim(dev, flow);
4391 	if (flow->hairpin_flow_id)
4392 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4393 				     flow->hairpin_flow_id);
4394 	flow_drv_destroy(dev, flow);
4395 	if (list)
4396 		TAILQ_REMOVE(list, flow, next);
4397 	flow_mreg_del_copy_action(dev, flow);
4398 	rte_free(flow->fdir);
4399 	rte_free(flow);
4400 }
4401 
4402 /**
4403  * Destroy all flows.
4404  *
4405  * @param dev
4406  *   Pointer to Ethernet device.
4407  * @param list
4408  *   Pointer to a TAILQ flow list.
4409  */
4410 void
4411 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4412 {
4413 	while (!TAILQ_EMPTY(list)) {
4414 		struct rte_flow *flow;
4415 
4416 		flow = TAILQ_FIRST(list);
4417 		flow_list_destroy(dev, list, flow);
4418 	}
4419 }
4420 
4421 /**
4422  * Remove all flows.
4423  *
4424  * @param dev
4425  *   Pointer to Ethernet device.
4426  * @param list
4427  *   Pointer to a TAILQ flow list.
4428  */
4429 void
4430 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4431 {
4432 	struct rte_flow *flow;
4433 
4434 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4435 		flow_drv_remove(dev, flow);
4436 		flow_mreg_stop_copy_action(dev, flow);
4437 	}
4438 	flow_mreg_del_default_copy_action(dev);
4439 	flow_rxq_flags_clear(dev);
4440 }
4441 
4442 /**
4443  * Add all flows.
4444  *
4445  * @param dev
4446  *   Pointer to Ethernet device.
4447  * @param list
4448  *   Pointer to a TAILQ flow list.
4449  *
4450  * @return
4451  *   0 on success, a negative errno value otherwise and rte_errno is set.
4452  */
4453 int
4454 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4455 {
4456 	struct rte_flow *flow;
4457 	struct rte_flow_error error;
4458 	int ret = 0;
4459 
4460 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4461 	ret = flow_mreg_add_default_copy_action(dev, &error);
4462 	if (ret < 0)
4463 		return -rte_errno;
4464 	/* Apply Flows created by application. */
4465 	TAILQ_FOREACH(flow, list, next) {
4466 		ret = flow_mreg_start_copy_action(dev, flow);
4467 		if (ret < 0)
4468 			goto error;
4469 		ret = flow_drv_apply(dev, flow, &error);
4470 		if (ret < 0)
4471 			goto error;
4472 		flow_rxq_flags_set(dev, flow);
4473 	}
4474 	return 0;
4475 error:
4476 	ret = rte_errno; /* Save rte_errno before cleanup. */
4477 	mlx5_flow_stop(dev, list);
4478 	rte_errno = ret; /* Restore rte_errno. */
4479 	return -rte_errno;
4480 }
4481 
4482 /**
4483  * Verify the flow list is empty
4484  *
4485  * @param dev
4486  *  Pointer to Ethernet device.
4487  *
4488  * @return the number of flows not released.
4489  */
4490 int
4491 mlx5_flow_verify(struct rte_eth_dev *dev)
4492 {
4493 	struct mlx5_priv *priv = dev->data->dev_private;
4494 	struct rte_flow *flow;
4495 	int ret = 0;
4496 
4497 	TAILQ_FOREACH(flow, &priv->flows, next) {
4498 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4499 			dev->data->port_id, (void *)flow);
4500 		++ret;
4501 	}
4502 	return ret;
4503 }
4504 
4505 /**
4506  * Enable default hairpin egress flow.
4507  *
4508  * @param dev
4509  *   Pointer to Ethernet device.
4510  * @param queue
4511  *   The queue index.
4512  *
4513  * @return
4514  *   0 on success, a negative errno value otherwise and rte_errno is set.
4515  */
4516 int
4517 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4518 			    uint32_t queue)
4519 {
4520 	struct mlx5_priv *priv = dev->data->dev_private;
4521 	const struct rte_flow_attr attr = {
4522 		.egress = 1,
4523 		.priority = 0,
4524 	};
4525 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4526 		.queue = queue,
4527 	};
4528 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4529 		.queue = UINT32_MAX,
4530 	};
4531 	struct rte_flow_item items[] = {
4532 		{
4533 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4534 			.spec = &queue_spec,
4535 			.last = NULL,
4536 			.mask = &queue_mask,
4537 		},
4538 		{
4539 			.type = RTE_FLOW_ITEM_TYPE_END,
4540 		},
4541 	};
4542 	struct rte_flow_action_jump jump = {
4543 		.group = MLX5_HAIRPIN_TX_TABLE,
4544 	};
4545 	struct rte_flow_action actions[2];
4546 	struct rte_flow *flow;
4547 	struct rte_flow_error error;
4548 
4549 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4550 	actions[0].conf = &jump;
4551 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4552 	flow = flow_list_create(dev, &priv->ctrl_flows,
4553 				&attr, items, actions, false, &error);
4554 	if (!flow) {
4555 		DRV_LOG(DEBUG,
4556 			"Failed to create ctrl flow: rte_errno(%d),"
4557 			" type(%d), message(%s)",
4558 			rte_errno, error.type,
4559 			error.message ? error.message : " (no stated reason)");
4560 		return -rte_errno;
4561 	}
4562 	return 0;
4563 }
4564 
4565 /**
4566  * Enable a control flow configured from the control plane.
4567  *
4568  * @param dev
4569  *   Pointer to Ethernet device.
4570  * @param eth_spec
4571  *   An Ethernet flow spec to apply.
4572  * @param eth_mask
4573  *   An Ethernet flow mask to apply.
4574  * @param vlan_spec
4575  *   A VLAN flow spec to apply.
4576  * @param vlan_mask
4577  *   A VLAN flow mask to apply.
4578  *
4579  * @return
4580  *   0 on success, a negative errno value otherwise and rte_errno is set.
4581  */
4582 int
4583 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4584 		    struct rte_flow_item_eth *eth_spec,
4585 		    struct rte_flow_item_eth *eth_mask,
4586 		    struct rte_flow_item_vlan *vlan_spec,
4587 		    struct rte_flow_item_vlan *vlan_mask)
4588 {
4589 	struct mlx5_priv *priv = dev->data->dev_private;
4590 	const struct rte_flow_attr attr = {
4591 		.ingress = 1,
4592 		.priority = MLX5_FLOW_PRIO_RSVD,
4593 	};
4594 	struct rte_flow_item items[] = {
4595 		{
4596 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4597 			.spec = eth_spec,
4598 			.last = NULL,
4599 			.mask = eth_mask,
4600 		},
4601 		{
4602 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4603 					      RTE_FLOW_ITEM_TYPE_END,
4604 			.spec = vlan_spec,
4605 			.last = NULL,
4606 			.mask = vlan_mask,
4607 		},
4608 		{
4609 			.type = RTE_FLOW_ITEM_TYPE_END,
4610 		},
4611 	};
4612 	uint16_t queue[priv->reta_idx_n];
4613 	struct rte_flow_action_rss action_rss = {
4614 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4615 		.level = 0,
4616 		.types = priv->rss_conf.rss_hf,
4617 		.key_len = priv->rss_conf.rss_key_len,
4618 		.queue_num = priv->reta_idx_n,
4619 		.key = priv->rss_conf.rss_key,
4620 		.queue = queue,
4621 	};
4622 	struct rte_flow_action actions[] = {
4623 		{
4624 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4625 			.conf = &action_rss,
4626 		},
4627 		{
4628 			.type = RTE_FLOW_ACTION_TYPE_END,
4629 		},
4630 	};
4631 	struct rte_flow *flow;
4632 	struct rte_flow_error error;
4633 	unsigned int i;
4634 
4635 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4636 		return 0;
4637 	}
4638 	for (i = 0; i != priv->reta_idx_n; ++i)
4639 		queue[i] = (*priv->reta_idx)[i];
4640 	flow = flow_list_create(dev, &priv->ctrl_flows,
4641 				&attr, items, actions, false, &error);
4642 	if (!flow)
4643 		return -rte_errno;
4644 	return 0;
4645 }
4646 
4647 /**
4648  * Enable a flow control configured from the control plane.
4649  *
4650  * @param dev
4651  *   Pointer to Ethernet device.
4652  * @param eth_spec
4653  *   An Ethernet flow spec to apply.
4654  * @param eth_mask
4655  *   An Ethernet flow mask to apply.
4656  *
4657  * @return
4658  *   0 on success, a negative errno value otherwise and rte_errno is set.
4659  */
4660 int
4661 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4662 	       struct rte_flow_item_eth *eth_spec,
4663 	       struct rte_flow_item_eth *eth_mask)
4664 {
4665 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4666 }
4667 
4668 /**
4669  * Destroy a flow.
4670  *
4671  * @see rte_flow_destroy()
4672  * @see rte_flow_ops
4673  */
4674 int
4675 mlx5_flow_destroy(struct rte_eth_dev *dev,
4676 		  struct rte_flow *flow,
4677 		  struct rte_flow_error *error __rte_unused)
4678 {
4679 	struct mlx5_priv *priv = dev->data->dev_private;
4680 
4681 	flow_list_destroy(dev, &priv->flows, flow);
4682 	return 0;
4683 }
4684 
4685 /**
4686  * Destroy all flows.
4687  *
4688  * @see rte_flow_flush()
4689  * @see rte_flow_ops
4690  */
4691 int
4692 mlx5_flow_flush(struct rte_eth_dev *dev,
4693 		struct rte_flow_error *error __rte_unused)
4694 {
4695 	struct mlx5_priv *priv = dev->data->dev_private;
4696 
4697 	mlx5_flow_list_flush(dev, &priv->flows);
4698 	return 0;
4699 }
4700 
4701 /**
4702  * Isolated mode.
4703  *
4704  * @see rte_flow_isolate()
4705  * @see rte_flow_ops
4706  */
4707 int
4708 mlx5_flow_isolate(struct rte_eth_dev *dev,
4709 		  int enable,
4710 		  struct rte_flow_error *error)
4711 {
4712 	struct mlx5_priv *priv = dev->data->dev_private;
4713 
4714 	if (dev->data->dev_started) {
4715 		rte_flow_error_set(error, EBUSY,
4716 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4717 				   NULL,
4718 				   "port must be stopped first");
4719 		return -rte_errno;
4720 	}
4721 	priv->isolated = !!enable;
4722 	if (enable)
4723 		dev->dev_ops = &mlx5_dev_ops_isolate;
4724 	else
4725 		dev->dev_ops = &mlx5_dev_ops;
4726 	return 0;
4727 }
4728 
4729 /**
4730  * Query a flow.
4731  *
4732  * @see rte_flow_query()
4733  * @see rte_flow_ops
4734  */
4735 static int
4736 flow_drv_query(struct rte_eth_dev *dev,
4737 	       struct rte_flow *flow,
4738 	       const struct rte_flow_action *actions,
4739 	       void *data,
4740 	       struct rte_flow_error *error)
4741 {
4742 	const struct mlx5_flow_driver_ops *fops;
4743 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4744 
4745 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4746 	fops = flow_get_drv_ops(ftype);
4747 
4748 	return fops->query(dev, flow, actions, data, error);
4749 }
4750 
4751 /**
4752  * Query a flow.
4753  *
4754  * @see rte_flow_query()
4755  * @see rte_flow_ops
4756  */
4757 int
4758 mlx5_flow_query(struct rte_eth_dev *dev,
4759 		struct rte_flow *flow,
4760 		const struct rte_flow_action *actions,
4761 		void *data,
4762 		struct rte_flow_error *error)
4763 {
4764 	int ret;
4765 
4766 	ret = flow_drv_query(dev, flow, actions, data, error);
4767 	if (ret < 0)
4768 		return ret;
4769 	return 0;
4770 }
4771 
4772 /**
4773  * Convert a flow director filter to a generic flow.
4774  *
4775  * @param dev
4776  *   Pointer to Ethernet device.
4777  * @param fdir_filter
4778  *   Flow director filter to add.
4779  * @param attributes
4780  *   Generic flow parameters structure.
4781  *
4782  * @return
4783  *   0 on success, a negative errno value otherwise and rte_errno is set.
4784  */
4785 static int
4786 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4787 			 const struct rte_eth_fdir_filter *fdir_filter,
4788 			 struct mlx5_fdir *attributes)
4789 {
4790 	struct mlx5_priv *priv = dev->data->dev_private;
4791 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4792 	const struct rte_eth_fdir_masks *mask =
4793 		&dev->data->dev_conf.fdir_conf.mask;
4794 
4795 	/* Validate queue number. */
4796 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4797 		DRV_LOG(ERR, "port %u invalid queue number %d",
4798 			dev->data->port_id, fdir_filter->action.rx_queue);
4799 		rte_errno = EINVAL;
4800 		return -rte_errno;
4801 	}
4802 	attributes->attr.ingress = 1;
4803 	attributes->items[0] = (struct rte_flow_item) {
4804 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4805 		.spec = &attributes->l2,
4806 		.mask = &attributes->l2_mask,
4807 	};
4808 	switch (fdir_filter->action.behavior) {
4809 	case RTE_ETH_FDIR_ACCEPT:
4810 		attributes->actions[0] = (struct rte_flow_action){
4811 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4812 			.conf = &attributes->queue,
4813 		};
4814 		break;
4815 	case RTE_ETH_FDIR_REJECT:
4816 		attributes->actions[0] = (struct rte_flow_action){
4817 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4818 		};
4819 		break;
4820 	default:
4821 		DRV_LOG(ERR, "port %u invalid behavior %d",
4822 			dev->data->port_id,
4823 			fdir_filter->action.behavior);
4824 		rte_errno = ENOTSUP;
4825 		return -rte_errno;
4826 	}
4827 	attributes->queue.index = fdir_filter->action.rx_queue;
4828 	/* Handle L3. */
4829 	switch (fdir_filter->input.flow_type) {
4830 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4831 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4832 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4833 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4834 			.src_addr = input->flow.ip4_flow.src_ip,
4835 			.dst_addr = input->flow.ip4_flow.dst_ip,
4836 			.time_to_live = input->flow.ip4_flow.ttl,
4837 			.type_of_service = input->flow.ip4_flow.tos,
4838 		};
4839 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4840 			.src_addr = mask->ipv4_mask.src_ip,
4841 			.dst_addr = mask->ipv4_mask.dst_ip,
4842 			.time_to_live = mask->ipv4_mask.ttl,
4843 			.type_of_service = mask->ipv4_mask.tos,
4844 			.next_proto_id = mask->ipv4_mask.proto,
4845 		};
4846 		attributes->items[1] = (struct rte_flow_item){
4847 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4848 			.spec = &attributes->l3,
4849 			.mask = &attributes->l3_mask,
4850 		};
4851 		break;
4852 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4853 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4854 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4855 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4856 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4857 			.proto = input->flow.ipv6_flow.proto,
4858 		};
4859 
4860 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4861 		       input->flow.ipv6_flow.src_ip,
4862 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4863 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4864 		       input->flow.ipv6_flow.dst_ip,
4865 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4866 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4867 		       mask->ipv6_mask.src_ip,
4868 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4869 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4870 		       mask->ipv6_mask.dst_ip,
4871 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4872 		attributes->items[1] = (struct rte_flow_item){
4873 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4874 			.spec = &attributes->l3,
4875 			.mask = &attributes->l3_mask,
4876 		};
4877 		break;
4878 	default:
4879 		DRV_LOG(ERR, "port %u invalid flow type%d",
4880 			dev->data->port_id, fdir_filter->input.flow_type);
4881 		rte_errno = ENOTSUP;
4882 		return -rte_errno;
4883 	}
4884 	/* Handle L4. */
4885 	switch (fdir_filter->input.flow_type) {
4886 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4887 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4888 			.src_port = input->flow.udp4_flow.src_port,
4889 			.dst_port = input->flow.udp4_flow.dst_port,
4890 		};
4891 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4892 			.src_port = mask->src_port_mask,
4893 			.dst_port = mask->dst_port_mask,
4894 		};
4895 		attributes->items[2] = (struct rte_flow_item){
4896 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4897 			.spec = &attributes->l4,
4898 			.mask = &attributes->l4_mask,
4899 		};
4900 		break;
4901 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4902 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4903 			.src_port = input->flow.tcp4_flow.src_port,
4904 			.dst_port = input->flow.tcp4_flow.dst_port,
4905 		};
4906 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4907 			.src_port = mask->src_port_mask,
4908 			.dst_port = mask->dst_port_mask,
4909 		};
4910 		attributes->items[2] = (struct rte_flow_item){
4911 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4912 			.spec = &attributes->l4,
4913 			.mask = &attributes->l4_mask,
4914 		};
4915 		break;
4916 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4917 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4918 			.src_port = input->flow.udp6_flow.src_port,
4919 			.dst_port = input->flow.udp6_flow.dst_port,
4920 		};
4921 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4922 			.src_port = mask->src_port_mask,
4923 			.dst_port = mask->dst_port_mask,
4924 		};
4925 		attributes->items[2] = (struct rte_flow_item){
4926 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4927 			.spec = &attributes->l4,
4928 			.mask = &attributes->l4_mask,
4929 		};
4930 		break;
4931 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4932 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4933 			.src_port = input->flow.tcp6_flow.src_port,
4934 			.dst_port = input->flow.tcp6_flow.dst_port,
4935 		};
4936 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4937 			.src_port = mask->src_port_mask,
4938 			.dst_port = mask->dst_port_mask,
4939 		};
4940 		attributes->items[2] = (struct rte_flow_item){
4941 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4942 			.spec = &attributes->l4,
4943 			.mask = &attributes->l4_mask,
4944 		};
4945 		break;
4946 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4947 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4948 		break;
4949 	default:
4950 		DRV_LOG(ERR, "port %u invalid flow type%d",
4951 			dev->data->port_id, fdir_filter->input.flow_type);
4952 		rte_errno = ENOTSUP;
4953 		return -rte_errno;
4954 	}
4955 	return 0;
4956 }
4957 
4958 #define FLOW_FDIR_CMP(f1, f2, fld) \
4959 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4960 
4961 /**
4962  * Compare two FDIR flows. If items and actions are identical, the two flows are
4963  * regarded as same.
4964  *
4965  * @param dev
4966  *   Pointer to Ethernet device.
4967  * @param f1
4968  *   FDIR flow to compare.
4969  * @param f2
4970  *   FDIR flow to compare.
4971  *
4972  * @return
4973  *   Zero on match, 1 otherwise.
4974  */
4975 static int
4976 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4977 {
4978 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4979 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4980 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4981 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4982 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4983 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4984 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4985 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4986 		return 1;
4987 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4988 	    FLOW_FDIR_CMP(f1, f2, queue))
4989 		return 1;
4990 	return 0;
4991 }
4992 
4993 /**
4994  * Search device flow list to find out a matched FDIR flow.
4995  *
4996  * @param dev
4997  *   Pointer to Ethernet device.
4998  * @param fdir_flow
4999  *   FDIR flow to lookup.
5000  *
5001  * @return
5002  *   Pointer of flow if found, NULL otherwise.
5003  */
5004 static struct rte_flow *
5005 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5006 {
5007 	struct mlx5_priv *priv = dev->data->dev_private;
5008 	struct rte_flow *flow = NULL;
5009 
5010 	MLX5_ASSERT(fdir_flow);
5011 	TAILQ_FOREACH(flow, &priv->flows, next) {
5012 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5013 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5014 				dev->data->port_id, (void *)flow);
5015 			break;
5016 		}
5017 	}
5018 	return flow;
5019 }
5020 
5021 /**
5022  * Add new flow director filter and store it in list.
5023  *
5024  * @param dev
5025  *   Pointer to Ethernet device.
5026  * @param fdir_filter
5027  *   Flow director filter to add.
5028  *
5029  * @return
5030  *   0 on success, a negative errno value otherwise and rte_errno is set.
5031  */
5032 static int
5033 flow_fdir_filter_add(struct rte_eth_dev *dev,
5034 		     const struct rte_eth_fdir_filter *fdir_filter)
5035 {
5036 	struct mlx5_priv *priv = dev->data->dev_private;
5037 	struct mlx5_fdir *fdir_flow;
5038 	struct rte_flow *flow;
5039 	int ret;
5040 
5041 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5042 	if (!fdir_flow) {
5043 		rte_errno = ENOMEM;
5044 		return -rte_errno;
5045 	}
5046 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5047 	if (ret)
5048 		goto error;
5049 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5050 	if (flow) {
5051 		rte_errno = EEXIST;
5052 		goto error;
5053 	}
5054 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5055 				fdir_flow->items, fdir_flow->actions, true,
5056 				NULL);
5057 	if (!flow)
5058 		goto error;
5059 	MLX5_ASSERT(!flow->fdir);
5060 	flow->fdir = fdir_flow;
5061 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5062 		dev->data->port_id, (void *)flow);
5063 	return 0;
5064 error:
5065 	rte_free(fdir_flow);
5066 	return -rte_errno;
5067 }
5068 
5069 /**
5070  * Delete specific filter.
5071  *
5072  * @param dev
5073  *   Pointer to Ethernet device.
5074  * @param fdir_filter
5075  *   Filter to be deleted.
5076  *
5077  * @return
5078  *   0 on success, a negative errno value otherwise and rte_errno is set.
5079  */
5080 static int
5081 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5082 			const struct rte_eth_fdir_filter *fdir_filter)
5083 {
5084 	struct mlx5_priv *priv = dev->data->dev_private;
5085 	struct rte_flow *flow;
5086 	struct mlx5_fdir fdir_flow = {
5087 		.attr.group = 0,
5088 	};
5089 	int ret;
5090 
5091 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5092 	if (ret)
5093 		return -rte_errno;
5094 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5095 	if (!flow) {
5096 		rte_errno = ENOENT;
5097 		return -rte_errno;
5098 	}
5099 	flow_list_destroy(dev, &priv->flows, flow);
5100 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5101 		dev->data->port_id, (void *)flow);
5102 	return 0;
5103 }
5104 
5105 /**
5106  * Update queue for specific filter.
5107  *
5108  * @param dev
5109  *   Pointer to Ethernet device.
5110  * @param fdir_filter
5111  *   Filter to be updated.
5112  *
5113  * @return
5114  *   0 on success, a negative errno value otherwise and rte_errno is set.
5115  */
5116 static int
5117 flow_fdir_filter_update(struct rte_eth_dev *dev,
5118 			const struct rte_eth_fdir_filter *fdir_filter)
5119 {
5120 	int ret;
5121 
5122 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5123 	if (ret)
5124 		return ret;
5125 	return flow_fdir_filter_add(dev, fdir_filter);
5126 }
5127 
5128 /**
5129  * Flush all filters.
5130  *
5131  * @param dev
5132  *   Pointer to Ethernet device.
5133  */
5134 static void
5135 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5136 {
5137 	struct mlx5_priv *priv = dev->data->dev_private;
5138 
5139 	mlx5_flow_list_flush(dev, &priv->flows);
5140 }
5141 
5142 /**
5143  * Get flow director information.
5144  *
5145  * @param dev
5146  *   Pointer to Ethernet device.
5147  * @param[out] fdir_info
5148  *   Resulting flow director information.
5149  */
5150 static void
5151 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5152 {
5153 	struct rte_eth_fdir_masks *mask =
5154 		&dev->data->dev_conf.fdir_conf.mask;
5155 
5156 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5157 	fdir_info->guarant_spc = 0;
5158 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5159 	fdir_info->max_flexpayload = 0;
5160 	fdir_info->flow_types_mask[0] = 0;
5161 	fdir_info->flex_payload_unit = 0;
5162 	fdir_info->max_flex_payload_segment_num = 0;
5163 	fdir_info->flex_payload_limit = 0;
5164 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5165 }
5166 
5167 /**
5168  * Deal with flow director operations.
5169  *
5170  * @param dev
5171  *   Pointer to Ethernet device.
5172  * @param filter_op
5173  *   Operation to perform.
5174  * @param arg
5175  *   Pointer to operation-specific structure.
5176  *
5177  * @return
5178  *   0 on success, a negative errno value otherwise and rte_errno is set.
5179  */
5180 static int
5181 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5182 		    void *arg)
5183 {
5184 	enum rte_fdir_mode fdir_mode =
5185 		dev->data->dev_conf.fdir_conf.mode;
5186 
5187 	if (filter_op == RTE_ETH_FILTER_NOP)
5188 		return 0;
5189 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5190 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5191 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5192 			dev->data->port_id, fdir_mode);
5193 		rte_errno = EINVAL;
5194 		return -rte_errno;
5195 	}
5196 	switch (filter_op) {
5197 	case RTE_ETH_FILTER_ADD:
5198 		return flow_fdir_filter_add(dev, arg);
5199 	case RTE_ETH_FILTER_UPDATE:
5200 		return flow_fdir_filter_update(dev, arg);
5201 	case RTE_ETH_FILTER_DELETE:
5202 		return flow_fdir_filter_delete(dev, arg);
5203 	case RTE_ETH_FILTER_FLUSH:
5204 		flow_fdir_filter_flush(dev);
5205 		break;
5206 	case RTE_ETH_FILTER_INFO:
5207 		flow_fdir_info_get(dev, arg);
5208 		break;
5209 	default:
5210 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5211 			dev->data->port_id, filter_op);
5212 		rte_errno = EINVAL;
5213 		return -rte_errno;
5214 	}
5215 	return 0;
5216 }
5217 
5218 /**
5219  * Manage filter operations.
5220  *
5221  * @param dev
5222  *   Pointer to Ethernet device structure.
5223  * @param filter_type
5224  *   Filter type.
5225  * @param filter_op
5226  *   Operation to perform.
5227  * @param arg
5228  *   Pointer to operation-specific structure.
5229  *
5230  * @return
5231  *   0 on success, a negative errno value otherwise and rte_errno is set.
5232  */
5233 int
5234 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5235 		     enum rte_filter_type filter_type,
5236 		     enum rte_filter_op filter_op,
5237 		     void *arg)
5238 {
5239 	switch (filter_type) {
5240 	case RTE_ETH_FILTER_GENERIC:
5241 		if (filter_op != RTE_ETH_FILTER_GET) {
5242 			rte_errno = EINVAL;
5243 			return -rte_errno;
5244 		}
5245 		*(const void **)arg = &mlx5_flow_ops;
5246 		return 0;
5247 	case RTE_ETH_FILTER_FDIR:
5248 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5249 	default:
5250 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5251 			dev->data->port_id, filter_type);
5252 		rte_errno = ENOTSUP;
5253 		return -rte_errno;
5254 	}
5255 	return 0;
5256 }
5257 
5258 /**
5259  * Create the needed meter and suffix tables.
5260  *
5261  * @param[in] dev
5262  *   Pointer to Ethernet device.
5263  * @param[in] fm
5264  *   Pointer to the flow meter.
5265  *
5266  * @return
5267  *   Pointer to table set on success, NULL otherwise.
5268  */
5269 struct mlx5_meter_domains_infos *
5270 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5271 			  const struct mlx5_flow_meter *fm)
5272 {
5273 	const struct mlx5_flow_driver_ops *fops;
5274 
5275 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5276 	return fops->create_mtr_tbls(dev, fm);
5277 }
5278 
5279 /**
5280  * Destroy the meter table set.
5281  *
5282  * @param[in] dev
5283  *   Pointer to Ethernet device.
5284  * @param[in] tbl
5285  *   Pointer to the meter table set.
5286  *
5287  * @return
5288  *   0 on success.
5289  */
5290 int
5291 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5292 			   struct mlx5_meter_domains_infos *tbls)
5293 {
5294 	const struct mlx5_flow_driver_ops *fops;
5295 
5296 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5297 	return fops->destroy_mtr_tbls(dev, tbls);
5298 }
5299 
5300 /**
5301  * Create policer rules.
5302  *
5303  * @param[in] dev
5304  *   Pointer to Ethernet device.
5305  * @param[in] fm
5306  *   Pointer to flow meter structure.
5307  * @param[in] attr
5308  *   Pointer to flow attributes.
5309  *
5310  * @return
5311  *   0 on success, -1 otherwise.
5312  */
5313 int
5314 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5315 			       struct mlx5_flow_meter *fm,
5316 			       const struct rte_flow_attr *attr)
5317 {
5318 	const struct mlx5_flow_driver_ops *fops;
5319 
5320 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5321 	return fops->create_policer_rules(dev, fm, attr);
5322 }
5323 
5324 /**
5325  * Destroy policer rules.
5326  *
5327  * @param[in] fm
5328  *   Pointer to flow meter structure.
5329  * @param[in] attr
5330  *   Pointer to flow attributes.
5331  *
5332  * @return
5333  *   0 on success, -1 otherwise.
5334  */
5335 int
5336 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5337 				struct mlx5_flow_meter *fm,
5338 				const struct rte_flow_attr *attr)
5339 {
5340 	const struct mlx5_flow_driver_ops *fops;
5341 
5342 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5343 	return fops->destroy_policer_rules(dev, fm, attr);
5344 }
5345 
5346 /**
5347  * Allocate a counter.
5348  *
5349  * @param[in] dev
5350  *   Pointer to Ethernet device structure.
5351  *
5352  * @return
5353  *   Pointer to allocated counter  on success, NULL otherwise.
5354  */
5355 struct mlx5_flow_counter *
5356 mlx5_counter_alloc(struct rte_eth_dev *dev)
5357 {
5358 	const struct mlx5_flow_driver_ops *fops;
5359 	struct rte_flow_attr attr = { .transfer = 0 };
5360 
5361 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5362 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5363 		return fops->counter_alloc(dev);
5364 	}
5365 	DRV_LOG(ERR,
5366 		"port %u counter allocate is not supported.",
5367 		 dev->data->port_id);
5368 	return NULL;
5369 }
5370 
5371 /**
5372  * Free a counter.
5373  *
5374  * @param[in] dev
5375  *   Pointer to Ethernet device structure.
5376  * @param[in] cnt
5377  *   Pointer to counter to be free.
5378  */
5379 void
5380 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5381 {
5382 	const struct mlx5_flow_driver_ops *fops;
5383 	struct rte_flow_attr attr = { .transfer = 0 };
5384 
5385 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5386 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5387 		fops->counter_free(dev, cnt);
5388 		return;
5389 	}
5390 	DRV_LOG(ERR,
5391 		"port %u counter free is not supported.",
5392 		 dev->data->port_id);
5393 }
5394 
5395 /**
5396  * Query counter statistics.
5397  *
5398  * @param[in] dev
5399  *   Pointer to Ethernet device structure.
5400  * @param[in] cnt
5401  *   Pointer to counter to query.
5402  * @param[in] clear
5403  *   Set to clear counter statistics.
5404  * @param[out] pkts
5405  *   The counter hits packets number to save.
5406  * @param[out] bytes
5407  *   The counter hits bytes number to save.
5408  *
5409  * @return
5410  *   0 on success, a negative errno value otherwise.
5411  */
5412 int
5413 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5414 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5415 {
5416 	const struct mlx5_flow_driver_ops *fops;
5417 	struct rte_flow_attr attr = { .transfer = 0 };
5418 
5419 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5420 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5421 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5422 	}
5423 	DRV_LOG(ERR,
5424 		"port %u counter query is not supported.",
5425 		 dev->data->port_id);
5426 	return -ENOTSUP;
5427 }
5428 
5429 #define MLX5_POOL_QUERY_FREQ_US 1000000
5430 
5431 /**
5432  * Set the periodic procedure for triggering asynchronous batch queries for all
5433  * the counter pools.
5434  *
5435  * @param[in] sh
5436  *   Pointer to mlx5_ibv_shared object.
5437  */
5438 void
5439 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5440 {
5441 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5442 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5443 	uint32_t us;
5444 
5445 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5446 	pools_n += rte_atomic16_read(&cont->n_valid);
5447 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5448 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5449 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5450 		sh->cmng.query_thread_on = 0;
5451 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5452 	} else {
5453 		sh->cmng.query_thread_on = 1;
5454 	}
5455 }
5456 
5457 /**
5458  * The periodic procedure for triggering asynchronous batch queries for all the
5459  * counter pools. This function is probably called by the host thread.
5460  *
5461  * @param[in] arg
5462  *   The parameter for the alarm process.
5463  */
5464 void
5465 mlx5_flow_query_alarm(void *arg)
5466 {
5467 	struct mlx5_ibv_shared *sh = arg;
5468 	struct mlx5_devx_obj *dcs;
5469 	uint16_t offset;
5470 	int ret;
5471 	uint8_t batch = sh->cmng.batch;
5472 	uint16_t pool_index = sh->cmng.pool_index;
5473 	struct mlx5_pools_container *cont;
5474 	struct mlx5_pools_container *mcont;
5475 	struct mlx5_flow_counter_pool *pool;
5476 
5477 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5478 		goto set_alarm;
5479 next_container:
5480 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5481 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5482 	/* Check if resize was done and need to flip a container. */
5483 	if (cont != mcont) {
5484 		if (cont->pools) {
5485 			/* Clean the old container. */
5486 			rte_free(cont->pools);
5487 			memset(cont, 0, sizeof(*cont));
5488 		}
5489 		rte_cio_wmb();
5490 		 /* Flip the host container. */
5491 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5492 		cont = mcont;
5493 	}
5494 	if (!cont->pools) {
5495 		/* 2 empty containers case is unexpected. */
5496 		if (unlikely(batch != sh->cmng.batch))
5497 			goto set_alarm;
5498 		batch ^= 0x1;
5499 		pool_index = 0;
5500 		goto next_container;
5501 	}
5502 	pool = cont->pools[pool_index];
5503 	if (pool->raw_hw)
5504 		/* There is a pool query in progress. */
5505 		goto set_alarm;
5506 	pool->raw_hw =
5507 		LIST_FIRST(&sh->cmng.free_stat_raws);
5508 	if (!pool->raw_hw)
5509 		/* No free counter statistics raw memory. */
5510 		goto set_alarm;
5511 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5512 							      (&pool->a64_dcs);
5513 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5514 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5515 					       offset, NULL, NULL,
5516 					       pool->raw_hw->mem_mng->dm->id,
5517 					       (void *)(uintptr_t)
5518 					       (pool->raw_hw->data + offset),
5519 					       sh->devx_comp,
5520 					       (uint64_t)(uintptr_t)pool);
5521 	if (ret) {
5522 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5523 			" %d", pool->min_dcs->id);
5524 		pool->raw_hw = NULL;
5525 		goto set_alarm;
5526 	}
5527 	pool->raw_hw->min_dcs_id = dcs->id;
5528 	LIST_REMOVE(pool->raw_hw, next);
5529 	sh->cmng.pending_queries++;
5530 	pool_index++;
5531 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5532 		batch ^= 0x1;
5533 		pool_index = 0;
5534 	}
5535 set_alarm:
5536 	sh->cmng.batch = batch;
5537 	sh->cmng.pool_index = pool_index;
5538 	mlx5_set_query_alarm(sh);
5539 }
5540 
5541 /**
5542  * Handler for the HW respond about ready values from an asynchronous batch
5543  * query. This function is probably called by the host thread.
5544  *
5545  * @param[in] sh
5546  *   The pointer to the shared IB device context.
5547  * @param[in] async_id
5548  *   The Devx async ID.
5549  * @param[in] status
5550  *   The status of the completion.
5551  */
5552 void
5553 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5554 				  uint64_t async_id, int status)
5555 {
5556 	struct mlx5_flow_counter_pool *pool =
5557 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5558 	struct mlx5_counter_stats_raw *raw_to_free;
5559 
5560 	if (unlikely(status)) {
5561 		raw_to_free = pool->raw_hw;
5562 	} else {
5563 		raw_to_free = pool->raw;
5564 		rte_spinlock_lock(&pool->sl);
5565 		pool->raw = pool->raw_hw;
5566 		rte_spinlock_unlock(&pool->sl);
5567 		rte_atomic64_add(&pool->query_gen, 1);
5568 		/* Be sure the new raw counters data is updated in memory. */
5569 		rte_cio_wmb();
5570 	}
5571 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5572 	pool->raw_hw = NULL;
5573 	sh->cmng.pending_queries--;
5574 }
5575 
5576 /**
5577  * Translate the rte_flow group index to HW table value.
5578  *
5579  * @param[in] attributes
5580  *   Pointer to flow attributes
5581  * @param[in] external
5582  *   Value is part of flow rule created by request external to PMD.
5583  * @param[in] group
5584  *   rte_flow group index value.
5585  * @param[out] fdb_def_rule
5586  *   Whether fdb jump to table 1 is configured.
5587  * @param[out] table
5588  *   HW table value.
5589  * @param[out] error
5590  *   Pointer to error structure.
5591  *
5592  * @return
5593  *   0 on success, a negative errno value otherwise and rte_errno is set.
5594  */
5595 int
5596 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5597 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5598 			 struct rte_flow_error *error)
5599 {
5600 	if (attributes->transfer && external && fdb_def_rule) {
5601 		if (group == UINT32_MAX)
5602 			return rte_flow_error_set
5603 						(error, EINVAL,
5604 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5605 						 NULL,
5606 						 "group index not supported");
5607 		*table = group + 1;
5608 	} else {
5609 		*table = group;
5610 	}
5611 	return 0;
5612 }
5613 
5614 /**
5615  * Discover availability of metadata reg_c's.
5616  *
5617  * Iteratively use test flows to check availability.
5618  *
5619  * @param[in] dev
5620  *   Pointer to the Ethernet device structure.
5621  *
5622  * @return
5623  *   0 on success, a negative errno value otherwise and rte_errno is set.
5624  */
5625 int
5626 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5627 {
5628 	struct mlx5_priv *priv = dev->data->dev_private;
5629 	struct mlx5_dev_config *config = &priv->config;
5630 	enum modify_reg idx;
5631 	int n = 0;
5632 
5633 	/* reg_c[0] and reg_c[1] are reserved. */
5634 	config->flow_mreg_c[n++] = REG_C_0;
5635 	config->flow_mreg_c[n++] = REG_C_1;
5636 	/* Discover availability of other reg_c's. */
5637 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5638 		struct rte_flow_attr attr = {
5639 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5640 			.priority = MLX5_FLOW_PRIO_RSVD,
5641 			.ingress = 1,
5642 		};
5643 		struct rte_flow_item items[] = {
5644 			[0] = {
5645 				.type = RTE_FLOW_ITEM_TYPE_END,
5646 			},
5647 		};
5648 		struct rte_flow_action actions[] = {
5649 			[0] = {
5650 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5651 				.conf = &(struct mlx5_flow_action_copy_mreg){
5652 					.src = REG_C_1,
5653 					.dst = idx,
5654 				},
5655 			},
5656 			[1] = {
5657 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5658 				.conf = &(struct rte_flow_action_jump){
5659 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5660 				},
5661 			},
5662 			[2] = {
5663 				.type = RTE_FLOW_ACTION_TYPE_END,
5664 			},
5665 		};
5666 		struct rte_flow *flow;
5667 		struct rte_flow_error error;
5668 
5669 		if (!config->dv_flow_en)
5670 			break;
5671 		/* Create internal flow, validation skips copy action. */
5672 		flow = flow_list_create(dev, NULL, &attr, items,
5673 					actions, false, &error);
5674 		if (!flow)
5675 			continue;
5676 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5677 			config->flow_mreg_c[n++] = idx;
5678 		flow_list_destroy(dev, NULL, flow);
5679 	}
5680 	for (; n < MLX5_MREG_C_NUM; ++n)
5681 		config->flow_mreg_c[n] = REG_NONE;
5682 	return 0;
5683 }
5684 
5685 /**
5686  * Dump flow raw hw data to file
5687  *
5688  * @param[in] dev
5689  *    The pointer to Ethernet device.
5690  * @param[in] file
5691  *   A pointer to a file for output.
5692  * @param[out] error
5693  *   Perform verbose error reporting if not NULL. PMDs initialize this
5694  *   structure in case of error only.
5695  * @return
5696  *   0 on success, a nagative value otherwise.
5697  */
5698 int
5699 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5700 		   FILE *file,
5701 		   struct rte_flow_error *error __rte_unused)
5702 {
5703 	struct mlx5_priv *priv = dev->data->dev_private;
5704 	struct mlx5_ibv_shared *sh = priv->sh;
5705 
5706 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5707 				       sh->tx_domain, file);
5708 }
5709