xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision f6d7202402c9ac81aa676fb23de8b379f7428401)
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.h"
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
34 #include "mlx5_prm.h"
35 #include "mlx5_rxtx.h"
36 
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
40 
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
46 
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
48 
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
53 #endif
54 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
56 };
57 
58 enum mlx5_expansion {
59 	MLX5_EXPANSION_ROOT,
60 	MLX5_EXPANSION_ROOT_OUTER,
61 	MLX5_EXPANSION_ROOT_ETH_VLAN,
62 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 	MLX5_EXPANSION_OUTER_ETH,
64 	MLX5_EXPANSION_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_VLAN,
66 	MLX5_EXPANSION_OUTER_IPV4,
67 	MLX5_EXPANSION_OUTER_IPV4_UDP,
68 	MLX5_EXPANSION_OUTER_IPV4_TCP,
69 	MLX5_EXPANSION_OUTER_IPV6,
70 	MLX5_EXPANSION_OUTER_IPV6_UDP,
71 	MLX5_EXPANSION_OUTER_IPV6_TCP,
72 	MLX5_EXPANSION_VXLAN,
73 	MLX5_EXPANSION_VXLAN_GPE,
74 	MLX5_EXPANSION_GRE,
75 	MLX5_EXPANSION_MPLS,
76 	MLX5_EXPANSION_ETH,
77 	MLX5_EXPANSION_ETH_VLAN,
78 	MLX5_EXPANSION_VLAN,
79 	MLX5_EXPANSION_IPV4,
80 	MLX5_EXPANSION_IPV4_UDP,
81 	MLX5_EXPANSION_IPV4_TCP,
82 	MLX5_EXPANSION_IPV6,
83 	MLX5_EXPANSION_IPV6_UDP,
84 	MLX5_EXPANSION_IPV6_TCP,
85 };
86 
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 	[MLX5_EXPANSION_ROOT] = {
90 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91 						 MLX5_EXPANSION_IPV4,
92 						 MLX5_EXPANSION_IPV6),
93 		.type = RTE_FLOW_ITEM_TYPE_END,
94 	},
95 	[MLX5_EXPANSION_ROOT_OUTER] = {
96 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 						 MLX5_EXPANSION_OUTER_IPV4,
98 						 MLX5_EXPANSION_OUTER_IPV6),
99 		.type = RTE_FLOW_ITEM_TYPE_END,
100 	},
101 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 		.type = RTE_FLOW_ITEM_TYPE_END,
104 	},
105 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 		.type = RTE_FLOW_ITEM_TYPE_END,
108 	},
109 	[MLX5_EXPANSION_OUTER_ETH] = {
110 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 						 MLX5_EXPANSION_OUTER_IPV6,
112 						 MLX5_EXPANSION_MPLS),
113 		.type = RTE_FLOW_ITEM_TYPE_ETH,
114 		.rss_types = 0,
115 	},
116 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 		.type = RTE_FLOW_ITEM_TYPE_ETH,
119 		.rss_types = 0,
120 	},
121 	[MLX5_EXPANSION_OUTER_VLAN] = {
122 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 						 MLX5_EXPANSION_OUTER_IPV6),
124 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
125 	},
126 	[MLX5_EXPANSION_OUTER_IPV4] = {
127 		.next = RTE_FLOW_EXPAND_RSS_NEXT
128 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
129 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
130 			 MLX5_EXPANSION_GRE,
131 			 MLX5_EXPANSION_IPV4,
132 			 MLX5_EXPANSION_IPV6),
133 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
134 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135 			ETH_RSS_NONFRAG_IPV4_OTHER,
136 	},
137 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139 						 MLX5_EXPANSION_VXLAN_GPE),
140 		.type = RTE_FLOW_ITEM_TYPE_UDP,
141 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
142 	},
143 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144 		.type = RTE_FLOW_ITEM_TYPE_TCP,
145 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
146 	},
147 	[MLX5_EXPANSION_OUTER_IPV6] = {
148 		.next = RTE_FLOW_EXPAND_RSS_NEXT
149 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
150 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
151 			 MLX5_EXPANSION_IPV4,
152 			 MLX5_EXPANSION_IPV6),
153 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
154 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155 			ETH_RSS_NONFRAG_IPV6_OTHER,
156 	},
157 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159 						 MLX5_EXPANSION_VXLAN_GPE),
160 		.type = RTE_FLOW_ITEM_TYPE_UDP,
161 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
162 	},
163 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164 		.type = RTE_FLOW_ITEM_TYPE_TCP,
165 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
166 	},
167 	[MLX5_EXPANSION_VXLAN] = {
168 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
169 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
170 	},
171 	[MLX5_EXPANSION_VXLAN_GPE] = {
172 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173 						 MLX5_EXPANSION_IPV4,
174 						 MLX5_EXPANSION_IPV6),
175 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
176 	},
177 	[MLX5_EXPANSION_GRE] = {
178 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 		.type = RTE_FLOW_ITEM_TYPE_GRE,
180 	},
181 	[MLX5_EXPANSION_MPLS] = {
182 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 						 MLX5_EXPANSION_IPV6),
184 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
185 	},
186 	[MLX5_EXPANSION_ETH] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_ETH,
190 	},
191 	[MLX5_EXPANSION_ETH_VLAN] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 		.type = RTE_FLOW_ITEM_TYPE_ETH,
194 	},
195 	[MLX5_EXPANSION_VLAN] = {
196 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 						 MLX5_EXPANSION_IPV6),
198 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
199 	},
200 	[MLX5_EXPANSION_IPV4] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 						 MLX5_EXPANSION_IPV4_TCP),
203 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
204 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 			ETH_RSS_NONFRAG_IPV4_OTHER,
206 	},
207 	[MLX5_EXPANSION_IPV4_UDP] = {
208 		.type = RTE_FLOW_ITEM_TYPE_UDP,
209 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
210 	},
211 	[MLX5_EXPANSION_IPV4_TCP] = {
212 		.type = RTE_FLOW_ITEM_TYPE_TCP,
213 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
214 	},
215 	[MLX5_EXPANSION_IPV6] = {
216 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 						 MLX5_EXPANSION_IPV6_TCP),
218 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
219 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 			ETH_RSS_NONFRAG_IPV6_OTHER,
221 	},
222 	[MLX5_EXPANSION_IPV6_UDP] = {
223 		.type = RTE_FLOW_ITEM_TYPE_UDP,
224 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
225 	},
226 	[MLX5_EXPANSION_IPV6_TCP] = {
227 		.type = RTE_FLOW_ITEM_TYPE_TCP,
228 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
229 	},
230 };
231 
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 	.validate = mlx5_flow_validate,
234 	.create = mlx5_flow_create,
235 	.destroy = mlx5_flow_destroy,
236 	.flush = mlx5_flow_flush,
237 	.isolate = mlx5_flow_isolate,
238 	.query = mlx5_flow_query,
239 	.dev_dump = mlx5_flow_dev_dump,
240 };
241 
242 /* Convert FDIR request to Generic flow. */
243 struct mlx5_fdir {
244 	struct rte_flow_attr attr;
245 	struct rte_flow_item items[4];
246 	struct rte_flow_item_eth l2;
247 	struct rte_flow_item_eth l2_mask;
248 	union {
249 		struct rte_flow_item_ipv4 ipv4;
250 		struct rte_flow_item_ipv6 ipv6;
251 	} l3;
252 	union {
253 		struct rte_flow_item_ipv4 ipv4;
254 		struct rte_flow_item_ipv6 ipv6;
255 	} l3_mask;
256 	union {
257 		struct rte_flow_item_udp udp;
258 		struct rte_flow_item_tcp tcp;
259 	} l4;
260 	union {
261 		struct rte_flow_item_udp udp;
262 		struct rte_flow_item_tcp tcp;
263 	} l4_mask;
264 	struct rte_flow_action actions[2];
265 	struct rte_flow_action_queue queue;
266 };
267 
268 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
269 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
270 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
271 };
272 
273 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
274 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
275 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
276 	{ 9, 10, 11 }, { 12, 13, 14 },
277 };
278 
279 /* Tunnel information. */
280 struct mlx5_flow_tunnel_info {
281 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
282 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
283 };
284 
285 static struct mlx5_flow_tunnel_info tunnels_info[] = {
286 	{
287 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
288 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
289 	},
290 	{
291 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
292 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
293 	},
294 	{
295 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
296 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
297 	},
298 	{
299 		.tunnel = MLX5_FLOW_LAYER_GRE,
300 		.ptype = RTE_PTYPE_TUNNEL_GRE,
301 	},
302 	{
303 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
304 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
305 	},
306 	{
307 		.tunnel = MLX5_FLOW_LAYER_MPLS,
308 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
309 	},
310 	{
311 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
312 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
313 	},
314 	{
315 		.tunnel = MLX5_FLOW_LAYER_IPIP,
316 		.ptype = RTE_PTYPE_TUNNEL_IP,
317 	},
318 	{
319 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
320 		.ptype = RTE_PTYPE_TUNNEL_IP,
321 	},
322 };
323 
324 /**
325  * Translate tag ID to register.
326  *
327  * @param[in] dev
328  *   Pointer to the Ethernet device structure.
329  * @param[in] feature
330  *   The feature that request the register.
331  * @param[in] id
332  *   The request register ID.
333  * @param[out] error
334  *   Error description in case of any.
335  *
336  * @return
337  *   The request register on success, a negative errno
338  *   value otherwise and rte_errno is set.
339  */
340 enum modify_reg
341 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
342 		     enum mlx5_feature_name feature,
343 		     uint32_t id,
344 		     struct rte_flow_error *error)
345 {
346 	struct mlx5_priv *priv = dev->data->dev_private;
347 	struct mlx5_dev_config *config = &priv->config;
348 	enum modify_reg start_reg;
349 
350 	switch (feature) {
351 	case MLX5_HAIRPIN_RX:
352 		return REG_B;
353 	case MLX5_HAIRPIN_TX:
354 		return REG_A;
355 	case MLX5_METADATA_RX:
356 		switch (config->dv_xmeta_en) {
357 		case MLX5_XMETA_MODE_LEGACY:
358 			return REG_B;
359 		case MLX5_XMETA_MODE_META16:
360 			return REG_C_0;
361 		case MLX5_XMETA_MODE_META32:
362 			return REG_C_1;
363 		}
364 		break;
365 	case MLX5_METADATA_TX:
366 		return REG_A;
367 	case MLX5_METADATA_FDB:
368 		switch (config->dv_xmeta_en) {
369 		case MLX5_XMETA_MODE_LEGACY:
370 			return REG_NONE;
371 		case MLX5_XMETA_MODE_META16:
372 			return REG_C_0;
373 		case MLX5_XMETA_MODE_META32:
374 			return REG_C_1;
375 		}
376 		break;
377 	case MLX5_FLOW_MARK:
378 		switch (config->dv_xmeta_en) {
379 		case MLX5_XMETA_MODE_LEGACY:
380 			return REG_NONE;
381 		case MLX5_XMETA_MODE_META16:
382 			return REG_C_1;
383 		case MLX5_XMETA_MODE_META32:
384 			return REG_C_0;
385 		}
386 		break;
387 	case MLX5_COPY_MARK:
388 	case MLX5_MTR_SFX:
389 		/*
390 		 * Metadata COPY_MARK register using is in meter suffix sub
391 		 * flow while with meter. It's safe to share the same register.
392 		 */
393 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
394 	case MLX5_MTR_COLOR:
395 		RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
396 		return priv->mtr_color_reg;
397 	case MLX5_APP_TAG:
398 		/*
399 		 * If meter is enable, it will engage two registers for color
400 		 * match and flow match. If meter color match is not using the
401 		 * REG_C_2, need to skip the REG_C_x be used by meter color
402 		 * match.
403 		 * If meter is disable, free to use all available registers.
404 		 */
405 		if (priv->mtr_color_reg != REG_NONE)
406 			start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
407 				    REG_C_4;
408 		else
409 			start_reg = REG_C_2;
410 		if (id > (REG_C_7 - start_reg))
411 			return rte_flow_error_set(error, EINVAL,
412 						  RTE_FLOW_ERROR_TYPE_ITEM,
413 						  NULL, "invalid tag id");
414 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
415 			return rte_flow_error_set(error, ENOTSUP,
416 						  RTE_FLOW_ERROR_TYPE_ITEM,
417 						  NULL, "unsupported tag id");
418 		/*
419 		 * This case means meter is using the REG_C_x great than 2.
420 		 * Take care not to conflict with meter color REG_C_x.
421 		 * If the available index REG_C_y >= REG_C_x, skip the
422 		 * color register.
423 		 */
424 		if (start_reg == REG_C_3 && config->flow_mreg_c
425 		    [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
426 			if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
427 			    REG_NONE)
428 				return config->flow_mreg_c
429 						[id + 1 + REG_C_3 - REG_C_0];
430 			return rte_flow_error_set(error, ENOTSUP,
431 						  RTE_FLOW_ERROR_TYPE_ITEM,
432 						  NULL, "unsupported tag id");
433 		}
434 		return config->flow_mreg_c[id + start_reg - REG_C_0];
435 	}
436 	assert(false);
437 	return rte_flow_error_set(error, EINVAL,
438 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
439 				  NULL, "invalid feature name");
440 }
441 
442 /**
443  * Check extensive flow metadata register support.
444  *
445  * @param dev
446  *   Pointer to rte_eth_dev structure.
447  *
448  * @return
449  *   True if device supports extensive flow metadata register, otherwise false.
450  */
451 bool
452 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
453 {
454 	struct mlx5_priv *priv = dev->data->dev_private;
455 	struct mlx5_dev_config *config = &priv->config;
456 
457 	/*
458 	 * Having available reg_c can be regarded inclusively as supporting
459 	 * extensive flow metadata register, which could mean,
460 	 * - metadata register copy action by modify header.
461 	 * - 16 modify header actions is supported.
462 	 * - reg_c's are preserved across different domain (FDB and NIC) on
463 	 *   packet loopback by flow lookup miss.
464 	 */
465 	return config->flow_mreg_c[2] != REG_NONE;
466 }
467 
468 /**
469  * Discover the maximum number of priority available.
470  *
471  * @param[in] dev
472  *   Pointer to the Ethernet device structure.
473  *
474  * @return
475  *   number of supported flow priority on success, a negative errno
476  *   value otherwise and rte_errno is set.
477  */
478 int
479 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
480 {
481 	struct mlx5_priv *priv = dev->data->dev_private;
482 	struct {
483 		struct ibv_flow_attr attr;
484 		struct ibv_flow_spec_eth eth;
485 		struct ibv_flow_spec_action_drop drop;
486 	} flow_attr = {
487 		.attr = {
488 			.num_of_specs = 2,
489 			.port = (uint8_t)priv->ibv_port,
490 		},
491 		.eth = {
492 			.type = IBV_FLOW_SPEC_ETH,
493 			.size = sizeof(struct ibv_flow_spec_eth),
494 		},
495 		.drop = {
496 			.size = sizeof(struct ibv_flow_spec_action_drop),
497 			.type = IBV_FLOW_SPEC_ACTION_DROP,
498 		},
499 	};
500 	struct ibv_flow *flow;
501 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
502 	uint16_t vprio[] = { 8, 16 };
503 	int i;
504 	int priority = 0;
505 
506 	if (!drop) {
507 		rte_errno = ENOTSUP;
508 		return -rte_errno;
509 	}
510 	for (i = 0; i != RTE_DIM(vprio); i++) {
511 		flow_attr.attr.priority = vprio[i] - 1;
512 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
513 		if (!flow)
514 			break;
515 		claim_zero(mlx5_glue->destroy_flow(flow));
516 		priority = vprio[i];
517 	}
518 	mlx5_hrxq_drop_release(dev);
519 	switch (priority) {
520 	case 8:
521 		priority = RTE_DIM(priority_map_3);
522 		break;
523 	case 16:
524 		priority = RTE_DIM(priority_map_5);
525 		break;
526 	default:
527 		rte_errno = ENOTSUP;
528 		DRV_LOG(ERR,
529 			"port %u verbs maximum priority: %d expected 8/16",
530 			dev->data->port_id, priority);
531 		return -rte_errno;
532 	}
533 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
534 		dev->data->port_id, priority);
535 	return priority;
536 }
537 
538 /**
539  * Adjust flow priority based on the highest layer and the request priority.
540  *
541  * @param[in] dev
542  *   Pointer to the Ethernet device structure.
543  * @param[in] priority
544  *   The rule base priority.
545  * @param[in] subpriority
546  *   The priority based on the items.
547  *
548  * @return
549  *   The new priority.
550  */
551 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
552 				   uint32_t subpriority)
553 {
554 	uint32_t res = 0;
555 	struct mlx5_priv *priv = dev->data->dev_private;
556 
557 	switch (priv->config.flow_prio) {
558 	case RTE_DIM(priority_map_3):
559 		res = priority_map_3[priority][subpriority];
560 		break;
561 	case RTE_DIM(priority_map_5):
562 		res = priority_map_5[priority][subpriority];
563 		break;
564 	}
565 	return  res;
566 }
567 
568 /**
569  * Verify the @p item specifications (spec, last, mask) are compatible with the
570  * NIC capabilities.
571  *
572  * @param[in] item
573  *   Item specification.
574  * @param[in] mask
575  *   @p item->mask or flow default bit-masks.
576  * @param[in] nic_mask
577  *   Bit-masks covering supported fields by the NIC to compare with user mask.
578  * @param[in] size
579  *   Bit-masks size in bytes.
580  * @param[out] error
581  *   Pointer to error structure.
582  *
583  * @return
584  *   0 on success, a negative errno value otherwise and rte_errno is set.
585  */
586 int
587 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
588 			  const uint8_t *mask,
589 			  const uint8_t *nic_mask,
590 			  unsigned int size,
591 			  struct rte_flow_error *error)
592 {
593 	unsigned int i;
594 
595 	assert(nic_mask);
596 	for (i = 0; i < size; ++i)
597 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
598 			return rte_flow_error_set(error, ENOTSUP,
599 						  RTE_FLOW_ERROR_TYPE_ITEM,
600 						  item,
601 						  "mask enables non supported"
602 						  " bits");
603 	if (!item->spec && (item->mask || item->last))
604 		return rte_flow_error_set(error, EINVAL,
605 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
606 					  "mask/last without a spec is not"
607 					  " supported");
608 	if (item->spec && item->last) {
609 		uint8_t spec[size];
610 		uint8_t last[size];
611 		unsigned int i;
612 		int ret;
613 
614 		for (i = 0; i < size; ++i) {
615 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
616 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
617 		}
618 		ret = memcmp(spec, last, size);
619 		if (ret != 0)
620 			return rte_flow_error_set(error, EINVAL,
621 						  RTE_FLOW_ERROR_TYPE_ITEM,
622 						  item,
623 						  "range is not valid");
624 	}
625 	return 0;
626 }
627 
628 /**
629  * Adjust the hash fields according to the @p flow information.
630  *
631  * @param[in] dev_flow.
632  *   Pointer to the mlx5_flow.
633  * @param[in] tunnel
634  *   1 when the hash field is for a tunnel item.
635  * @param[in] layer_types
636  *   ETH_RSS_* types.
637  * @param[in] hash_fields
638  *   Item hash fields.
639  *
640  * @return
641  *   The hash fields that should be used.
642  */
643 uint64_t
644 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
645 			    int tunnel __rte_unused, uint64_t layer_types,
646 			    uint64_t hash_fields)
647 {
648 	struct rte_flow *flow = dev_flow->flow;
649 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
650 	int rss_request_inner = flow->rss.level >= 2;
651 
652 	/* Check RSS hash level for tunnel. */
653 	if (tunnel && rss_request_inner)
654 		hash_fields |= IBV_RX_HASH_INNER;
655 	else if (tunnel || rss_request_inner)
656 		return 0;
657 #endif
658 	/* Check if requested layer matches RSS hash fields. */
659 	if (!(flow->rss.types & layer_types))
660 		return 0;
661 	return hash_fields;
662 }
663 
664 /**
665  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
666  * if several tunnel rules are used on this queue, the tunnel ptype will be
667  * cleared.
668  *
669  * @param rxq_ctrl
670  *   Rx queue to update.
671  */
672 static void
673 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
674 {
675 	unsigned int i;
676 	uint32_t tunnel_ptype = 0;
677 
678 	/* Look up for the ptype to use. */
679 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
680 		if (!rxq_ctrl->flow_tunnels_n[i])
681 			continue;
682 		if (!tunnel_ptype) {
683 			tunnel_ptype = tunnels_info[i].ptype;
684 		} else {
685 			tunnel_ptype = 0;
686 			break;
687 		}
688 	}
689 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
690 }
691 
692 /**
693  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
694  * flow.
695  *
696  * @param[in] dev
697  *   Pointer to the Ethernet device structure.
698  * @param[in] dev_flow
699  *   Pointer to device flow structure.
700  */
701 static void
702 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
703 {
704 	struct mlx5_priv *priv = dev->data->dev_private;
705 	struct rte_flow *flow = dev_flow->flow;
706 	const int mark = !!(dev_flow->actions &
707 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
708 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
709 	unsigned int i;
710 
711 	for (i = 0; i != flow->rss.queue_num; ++i) {
712 		int idx = (*flow->rss.queue)[i];
713 		struct mlx5_rxq_ctrl *rxq_ctrl =
714 			container_of((*priv->rxqs)[idx],
715 				     struct mlx5_rxq_ctrl, rxq);
716 
717 		/*
718 		 * To support metadata register copy on Tx loopback,
719 		 * this must be always enabled (metadata may arive
720 		 * from other port - not from local flows only.
721 		 */
722 		if (priv->config.dv_flow_en &&
723 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
724 		    mlx5_flow_ext_mreg_supported(dev)) {
725 			rxq_ctrl->rxq.mark = 1;
726 			rxq_ctrl->flow_mark_n = 1;
727 		} else if (mark) {
728 			rxq_ctrl->rxq.mark = 1;
729 			rxq_ctrl->flow_mark_n++;
730 		}
731 		if (tunnel) {
732 			unsigned int j;
733 
734 			/* Increase the counter matching the flow. */
735 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
736 				if ((tunnels_info[j].tunnel &
737 				     dev_flow->layers) ==
738 				    tunnels_info[j].tunnel) {
739 					rxq_ctrl->flow_tunnels_n[j]++;
740 					break;
741 				}
742 			}
743 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
744 		}
745 	}
746 }
747 
748 /**
749  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
750  *
751  * @param[in] dev
752  *   Pointer to the Ethernet device structure.
753  * @param[in] flow
754  *   Pointer to flow structure.
755  */
756 static void
757 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
758 {
759 	struct mlx5_flow *dev_flow;
760 
761 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
762 		flow_drv_rxq_flags_set(dev, dev_flow);
763 }
764 
765 /**
766  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
767  * device flow if no other flow uses it with the same kind of request.
768  *
769  * @param dev
770  *   Pointer to Ethernet device.
771  * @param[in] dev_flow
772  *   Pointer to the device flow.
773  */
774 static void
775 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
776 {
777 	struct mlx5_priv *priv = dev->data->dev_private;
778 	struct rte_flow *flow = dev_flow->flow;
779 	const int mark = !!(dev_flow->actions &
780 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
781 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
782 	unsigned int i;
783 
784 	assert(dev->data->dev_started);
785 	for (i = 0; i != flow->rss.queue_num; ++i) {
786 		int idx = (*flow->rss.queue)[i];
787 		struct mlx5_rxq_ctrl *rxq_ctrl =
788 			container_of((*priv->rxqs)[idx],
789 				     struct mlx5_rxq_ctrl, rxq);
790 
791 		if (priv->config.dv_flow_en &&
792 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
793 		    mlx5_flow_ext_mreg_supported(dev)) {
794 			rxq_ctrl->rxq.mark = 1;
795 			rxq_ctrl->flow_mark_n = 1;
796 		} else if (mark) {
797 			rxq_ctrl->flow_mark_n--;
798 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
799 		}
800 		if (tunnel) {
801 			unsigned int j;
802 
803 			/* Decrease the counter matching the flow. */
804 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
805 				if ((tunnels_info[j].tunnel &
806 				     dev_flow->layers) ==
807 				    tunnels_info[j].tunnel) {
808 					rxq_ctrl->flow_tunnels_n[j]--;
809 					break;
810 				}
811 			}
812 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
813 		}
814 	}
815 }
816 
817 /**
818  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
819  * @p flow if no other flow uses it with the same kind of request.
820  *
821  * @param dev
822  *   Pointer to Ethernet device.
823  * @param[in] flow
824  *   Pointer to the flow.
825  */
826 static void
827 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
828 {
829 	struct mlx5_flow *dev_flow;
830 
831 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
832 		flow_drv_rxq_flags_trim(dev, dev_flow);
833 }
834 
835 /**
836  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
837  *
838  * @param dev
839  *   Pointer to Ethernet device.
840  */
841 static void
842 flow_rxq_flags_clear(struct rte_eth_dev *dev)
843 {
844 	struct mlx5_priv *priv = dev->data->dev_private;
845 	unsigned int i;
846 
847 	for (i = 0; i != priv->rxqs_n; ++i) {
848 		struct mlx5_rxq_ctrl *rxq_ctrl;
849 		unsigned int j;
850 
851 		if (!(*priv->rxqs)[i])
852 			continue;
853 		rxq_ctrl = container_of((*priv->rxqs)[i],
854 					struct mlx5_rxq_ctrl, rxq);
855 		rxq_ctrl->flow_mark_n = 0;
856 		rxq_ctrl->rxq.mark = 0;
857 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
858 			rxq_ctrl->flow_tunnels_n[j] = 0;
859 		rxq_ctrl->rxq.tunnel = 0;
860 	}
861 }
862 
863 /*
864  * return a pointer to the desired action in the list of actions.
865  *
866  * @param[in] actions
867  *   The list of actions to search the action in.
868  * @param[in] action
869  *   The action to find.
870  *
871  * @return
872  *   Pointer to the action in the list, if found. NULL otherwise.
873  */
874 const struct rte_flow_action *
875 mlx5_flow_find_action(const struct rte_flow_action *actions,
876 		      enum rte_flow_action_type action)
877 {
878 	if (actions == NULL)
879 		return NULL;
880 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
881 		if (actions->type == action)
882 			return actions;
883 	return NULL;
884 }
885 
886 /*
887  * Validate the flag action.
888  *
889  * @param[in] action_flags
890  *   Bit-fields that holds the actions detected until now.
891  * @param[in] attr
892  *   Attributes of flow that includes this action.
893  * @param[out] error
894  *   Pointer to error structure.
895  *
896  * @return
897  *   0 on success, a negative errno value otherwise and rte_errno is set.
898  */
899 int
900 mlx5_flow_validate_action_flag(uint64_t action_flags,
901 			       const struct rte_flow_attr *attr,
902 			       struct rte_flow_error *error)
903 {
904 
905 	if (action_flags & MLX5_FLOW_ACTION_DROP)
906 		return rte_flow_error_set(error, EINVAL,
907 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
908 					  "can't drop and flag in same flow");
909 	if (action_flags & MLX5_FLOW_ACTION_MARK)
910 		return rte_flow_error_set(error, EINVAL,
911 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
912 					  "can't mark and flag in same flow");
913 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
914 		return rte_flow_error_set(error, EINVAL,
915 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
916 					  "can't have 2 flag"
917 					  " actions in same flow");
918 	if (attr->egress)
919 		return rte_flow_error_set(error, ENOTSUP,
920 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
921 					  "flag action not supported for "
922 					  "egress");
923 	return 0;
924 }
925 
926 /*
927  * Validate the mark action.
928  *
929  * @param[in] action
930  *   Pointer to the queue action.
931  * @param[in] action_flags
932  *   Bit-fields that holds the actions detected until now.
933  * @param[in] attr
934  *   Attributes of flow that includes this action.
935  * @param[out] error
936  *   Pointer to error structure.
937  *
938  * @return
939  *   0 on success, a negative errno value otherwise and rte_errno is set.
940  */
941 int
942 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
943 			       uint64_t action_flags,
944 			       const struct rte_flow_attr *attr,
945 			       struct rte_flow_error *error)
946 {
947 	const struct rte_flow_action_mark *mark = action->conf;
948 
949 	if (!mark)
950 		return rte_flow_error_set(error, EINVAL,
951 					  RTE_FLOW_ERROR_TYPE_ACTION,
952 					  action,
953 					  "configuration cannot be null");
954 	if (mark->id >= MLX5_FLOW_MARK_MAX)
955 		return rte_flow_error_set(error, EINVAL,
956 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
957 					  &mark->id,
958 					  "mark id must in 0 <= id < "
959 					  RTE_STR(MLX5_FLOW_MARK_MAX));
960 	if (action_flags & MLX5_FLOW_ACTION_DROP)
961 		return rte_flow_error_set(error, EINVAL,
962 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
963 					  "can't drop and mark in same flow");
964 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
965 		return rte_flow_error_set(error, EINVAL,
966 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
967 					  "can't flag and mark in same flow");
968 	if (action_flags & MLX5_FLOW_ACTION_MARK)
969 		return rte_flow_error_set(error, EINVAL,
970 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
971 					  "can't have 2 mark actions in same"
972 					  " flow");
973 	if (attr->egress)
974 		return rte_flow_error_set(error, ENOTSUP,
975 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
976 					  "mark action not supported for "
977 					  "egress");
978 	return 0;
979 }
980 
981 /*
982  * Validate the drop action.
983  *
984  * @param[in] action_flags
985  *   Bit-fields that holds the actions detected until now.
986  * @param[in] attr
987  *   Attributes of flow that includes this action.
988  * @param[out] error
989  *   Pointer to error structure.
990  *
991  * @return
992  *   0 on success, a negative errno value otherwise and rte_errno is set.
993  */
994 int
995 mlx5_flow_validate_action_drop(uint64_t action_flags,
996 			       const struct rte_flow_attr *attr,
997 			       struct rte_flow_error *error)
998 {
999 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1000 		return rte_flow_error_set(error, EINVAL,
1001 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1002 					  "can't drop and flag in same flow");
1003 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1004 		return rte_flow_error_set(error, EINVAL,
1005 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1006 					  "can't drop and mark in same flow");
1007 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
1008 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1009 		return rte_flow_error_set(error, EINVAL,
1010 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1011 					  "can't have 2 fate actions in"
1012 					  " same flow");
1013 	if (attr->egress)
1014 		return rte_flow_error_set(error, ENOTSUP,
1015 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1016 					  "drop action not supported for "
1017 					  "egress");
1018 	return 0;
1019 }
1020 
1021 /*
1022  * Validate the queue action.
1023  *
1024  * @param[in] action
1025  *   Pointer to the queue action.
1026  * @param[in] action_flags
1027  *   Bit-fields that holds the actions detected until now.
1028  * @param[in] dev
1029  *   Pointer to the Ethernet device structure.
1030  * @param[in] attr
1031  *   Attributes of flow that includes this action.
1032  * @param[out] error
1033  *   Pointer to error structure.
1034  *
1035  * @return
1036  *   0 on success, a negative errno value otherwise and rte_errno is set.
1037  */
1038 int
1039 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1040 				uint64_t action_flags,
1041 				struct rte_eth_dev *dev,
1042 				const struct rte_flow_attr *attr,
1043 				struct rte_flow_error *error)
1044 {
1045 	struct mlx5_priv *priv = dev->data->dev_private;
1046 	const struct rte_flow_action_queue *queue = action->conf;
1047 
1048 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1049 		return rte_flow_error_set(error, EINVAL,
1050 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1051 					  "can't have 2 fate actions in"
1052 					  " same flow");
1053 	if (!priv->rxqs_n)
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056 					  NULL, "No Rx queues configured");
1057 	if (queue->index >= priv->rxqs_n)
1058 		return rte_flow_error_set(error, EINVAL,
1059 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1060 					  &queue->index,
1061 					  "queue index out of range");
1062 	if (!(*priv->rxqs)[queue->index])
1063 		return rte_flow_error_set(error, EINVAL,
1064 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1065 					  &queue->index,
1066 					  "queue is not configured");
1067 	if (attr->egress)
1068 		return rte_flow_error_set(error, ENOTSUP,
1069 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1070 					  "queue action not supported for "
1071 					  "egress");
1072 	return 0;
1073 }
1074 
1075 /*
1076  * Validate the rss action.
1077  *
1078  * @param[in] action
1079  *   Pointer to the queue action.
1080  * @param[in] action_flags
1081  *   Bit-fields that holds the actions detected until now.
1082  * @param[in] dev
1083  *   Pointer to the Ethernet device structure.
1084  * @param[in] attr
1085  *   Attributes of flow that includes this action.
1086  * @param[in] item_flags
1087  *   Items that were detected.
1088  * @param[out] error
1089  *   Pointer to error structure.
1090  *
1091  * @return
1092  *   0 on success, a negative errno value otherwise and rte_errno is set.
1093  */
1094 int
1095 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1096 			      uint64_t action_flags,
1097 			      struct rte_eth_dev *dev,
1098 			      const struct rte_flow_attr *attr,
1099 			      uint64_t item_flags,
1100 			      struct rte_flow_error *error)
1101 {
1102 	struct mlx5_priv *priv = dev->data->dev_private;
1103 	const struct rte_flow_action_rss *rss = action->conf;
1104 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1105 	unsigned int i;
1106 
1107 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1108 		return rte_flow_error_set(error, EINVAL,
1109 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1110 					  "can't have 2 fate actions"
1111 					  " in same flow");
1112 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1113 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 					  &rss->func,
1117 					  "RSS hash function not supported");
1118 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1119 	if (rss->level > 2)
1120 #else
1121 	if (rss->level > 1)
1122 #endif
1123 		return rte_flow_error_set(error, ENOTSUP,
1124 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1125 					  &rss->level,
1126 					  "tunnel RSS is not supported");
1127 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1128 	if (rss->key_len == 0 && rss->key != NULL)
1129 		return rte_flow_error_set(error, ENOTSUP,
1130 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1131 					  &rss->key_len,
1132 					  "RSS hash key length 0");
1133 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1134 		return rte_flow_error_set(error, ENOTSUP,
1135 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136 					  &rss->key_len,
1137 					  "RSS hash key too small");
1138 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1139 		return rte_flow_error_set(error, ENOTSUP,
1140 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1141 					  &rss->key_len,
1142 					  "RSS hash key too large");
1143 	if (rss->queue_num > priv->config.ind_table_max_size)
1144 		return rte_flow_error_set(error, ENOTSUP,
1145 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1146 					  &rss->queue_num,
1147 					  "number of queues too large");
1148 	if (rss->types & MLX5_RSS_HF_MASK)
1149 		return rte_flow_error_set(error, ENOTSUP,
1150 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1151 					  &rss->types,
1152 					  "some RSS protocols are not"
1153 					  " supported");
1154 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1155 	    !(rss->types & ETH_RSS_IP))
1156 		return rte_flow_error_set(error, EINVAL,
1157 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1158 					  "L3 partial RSS requested but L3 RSS"
1159 					  " type not specified");
1160 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1161 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1162 		return rte_flow_error_set(error, EINVAL,
1163 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1164 					  "L4 partial RSS requested but L4 RSS"
1165 					  " type not specified");
1166 	if (!priv->rxqs_n)
1167 		return rte_flow_error_set(error, EINVAL,
1168 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1169 					  NULL, "No Rx queues configured");
1170 	if (!rss->queue_num)
1171 		return rte_flow_error_set(error, EINVAL,
1172 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 					  NULL, "No queues configured");
1174 	for (i = 0; i != rss->queue_num; ++i) {
1175 		if (rss->queue[i] >= priv->rxqs_n)
1176 			return rte_flow_error_set
1177 				(error, EINVAL,
1178 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1179 				 &rss->queue[i], "queue index out of range");
1180 		if (!(*priv->rxqs)[rss->queue[i]])
1181 			return rte_flow_error_set
1182 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183 				 &rss->queue[i], "queue is not configured");
1184 	}
1185 	if (attr->egress)
1186 		return rte_flow_error_set(error, ENOTSUP,
1187 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1188 					  "rss action not supported for "
1189 					  "egress");
1190 	if (rss->level > 1 &&  !tunnel)
1191 		return rte_flow_error_set(error, EINVAL,
1192 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1193 					  "inner RSS is not supported for "
1194 					  "non-tunnel flows");
1195 	return 0;
1196 }
1197 
1198 /*
1199  * Validate the count action.
1200  *
1201  * @param[in] dev
1202  *   Pointer to the Ethernet device structure.
1203  * @param[in] attr
1204  *   Attributes of flow that includes this action.
1205  * @param[out] error
1206  *   Pointer to error structure.
1207  *
1208  * @return
1209  *   0 on success, a negative errno value otherwise and rte_errno is set.
1210  */
1211 int
1212 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1213 				const struct rte_flow_attr *attr,
1214 				struct rte_flow_error *error)
1215 {
1216 	if (attr->egress)
1217 		return rte_flow_error_set(error, ENOTSUP,
1218 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1219 					  "count action not supported for "
1220 					  "egress");
1221 	return 0;
1222 }
1223 
1224 /**
1225  * Verify the @p attributes will be correctly understood by the NIC and store
1226  * them in the @p flow if everything is correct.
1227  *
1228  * @param[in] dev
1229  *   Pointer to the Ethernet device structure.
1230  * @param[in] attributes
1231  *   Pointer to flow attributes
1232  * @param[out] error
1233  *   Pointer to error structure.
1234  *
1235  * @return
1236  *   0 on success, a negative errno value otherwise and rte_errno is set.
1237  */
1238 int
1239 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1240 			      const struct rte_flow_attr *attributes,
1241 			      struct rte_flow_error *error)
1242 {
1243 	struct mlx5_priv *priv = dev->data->dev_private;
1244 	uint32_t priority_max = priv->config.flow_prio - 1;
1245 
1246 	if (attributes->group)
1247 		return rte_flow_error_set(error, ENOTSUP,
1248 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1249 					  NULL, "groups is not supported");
1250 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1251 	    attributes->priority >= priority_max)
1252 		return rte_flow_error_set(error, ENOTSUP,
1253 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1254 					  NULL, "priority out of range");
1255 	if (attributes->egress)
1256 		return rte_flow_error_set(error, ENOTSUP,
1257 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1258 					  "egress is not supported");
1259 	if (attributes->transfer && !priv->config.dv_esw_en)
1260 		return rte_flow_error_set(error, ENOTSUP,
1261 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1262 					  NULL, "transfer is not supported");
1263 	if (!attributes->ingress)
1264 		return rte_flow_error_set(error, EINVAL,
1265 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1266 					  NULL,
1267 					  "ingress attribute is mandatory");
1268 	return 0;
1269 }
1270 
1271 /**
1272  * Validate ICMP6 item.
1273  *
1274  * @param[in] item
1275  *   Item specification.
1276  * @param[in] item_flags
1277  *   Bit-fields that holds the items detected until now.
1278  * @param[out] error
1279  *   Pointer to error structure.
1280  *
1281  * @return
1282  *   0 on success, a negative errno value otherwise and rte_errno is set.
1283  */
1284 int
1285 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1286 			       uint64_t item_flags,
1287 			       uint8_t target_protocol,
1288 			       struct rte_flow_error *error)
1289 {
1290 	const struct rte_flow_item_icmp6 *mask = item->mask;
1291 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1292 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1293 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1294 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1295 				      MLX5_FLOW_LAYER_OUTER_L4;
1296 	int ret;
1297 
1298 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1299 		return rte_flow_error_set(error, EINVAL,
1300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1301 					  "protocol filtering not compatible"
1302 					  " with ICMP6 layer");
1303 	if (!(item_flags & l3m))
1304 		return rte_flow_error_set(error, EINVAL,
1305 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1306 					  "IPv6 is mandatory to filter on"
1307 					  " ICMP6");
1308 	if (item_flags & l4m)
1309 		return rte_flow_error_set(error, EINVAL,
1310 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1311 					  "multiple L4 layers not supported");
1312 	if (!mask)
1313 		mask = &rte_flow_item_icmp6_mask;
1314 	ret = mlx5_flow_item_acceptable
1315 		(item, (const uint8_t *)mask,
1316 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1317 		 sizeof(struct rte_flow_item_icmp6), error);
1318 	if (ret < 0)
1319 		return ret;
1320 	return 0;
1321 }
1322 
1323 /**
1324  * Validate ICMP item.
1325  *
1326  * @param[in] item
1327  *   Item specification.
1328  * @param[in] item_flags
1329  *   Bit-fields that holds the items detected until now.
1330  * @param[out] error
1331  *   Pointer to error structure.
1332  *
1333  * @return
1334  *   0 on success, a negative errno value otherwise and rte_errno is set.
1335  */
1336 int
1337 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1338 			     uint64_t item_flags,
1339 			     uint8_t target_protocol,
1340 			     struct rte_flow_error *error)
1341 {
1342 	const struct rte_flow_item_icmp *mask = item->mask;
1343 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1344 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1345 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1346 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1347 				      MLX5_FLOW_LAYER_OUTER_L4;
1348 	int ret;
1349 
1350 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1351 		return rte_flow_error_set(error, EINVAL,
1352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 					  "protocol filtering not compatible"
1354 					  " with ICMP layer");
1355 	if (!(item_flags & l3m))
1356 		return rte_flow_error_set(error, EINVAL,
1357 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1358 					  "IPv4 is mandatory to filter"
1359 					  " on ICMP");
1360 	if (item_flags & l4m)
1361 		return rte_flow_error_set(error, EINVAL,
1362 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1363 					  "multiple L4 layers not supported");
1364 	if (!mask)
1365 		mask = &rte_flow_item_icmp_mask;
1366 	ret = mlx5_flow_item_acceptable
1367 		(item, (const uint8_t *)mask,
1368 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1369 		 sizeof(struct rte_flow_item_icmp), error);
1370 	if (ret < 0)
1371 		return ret;
1372 	return 0;
1373 }
1374 
1375 /**
1376  * Validate Ethernet item.
1377  *
1378  * @param[in] item
1379  *   Item specification.
1380  * @param[in] item_flags
1381  *   Bit-fields that holds the items detected until now.
1382  * @param[out] error
1383  *   Pointer to error structure.
1384  *
1385  * @return
1386  *   0 on success, a negative errno value otherwise and rte_errno is set.
1387  */
1388 int
1389 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1390 			    uint64_t item_flags,
1391 			    struct rte_flow_error *error)
1392 {
1393 	const struct rte_flow_item_eth *mask = item->mask;
1394 	const struct rte_flow_item_eth nic_mask = {
1395 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1396 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1397 		.type = RTE_BE16(0xffff),
1398 	};
1399 	int ret;
1400 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1401 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1402 				       MLX5_FLOW_LAYER_OUTER_L2;
1403 
1404 	if (item_flags & ethm)
1405 		return rte_flow_error_set(error, ENOTSUP,
1406 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1407 					  "multiple L2 layers not supported");
1408 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1409 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1410 		return rte_flow_error_set(error, EINVAL,
1411 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1412 					  "L2 layer should not follow "
1413 					  "L3 layers");
1414 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1415 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1416 		return rte_flow_error_set(error, EINVAL,
1417 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1418 					  "L2 layer should not follow VLAN");
1419 	if (!mask)
1420 		mask = &rte_flow_item_eth_mask;
1421 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1422 					(const uint8_t *)&nic_mask,
1423 					sizeof(struct rte_flow_item_eth),
1424 					error);
1425 	return ret;
1426 }
1427 
1428 /**
1429  * Validate VLAN item.
1430  *
1431  * @param[in] item
1432  *   Item specification.
1433  * @param[in] item_flags
1434  *   Bit-fields that holds the items detected until now.
1435  * @param[in] dev
1436  *   Ethernet device flow is being created on.
1437  * @param[out] error
1438  *   Pointer to error structure.
1439  *
1440  * @return
1441  *   0 on success, a negative errno value otherwise and rte_errno is set.
1442  */
1443 int
1444 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1445 			     uint64_t item_flags,
1446 			     struct rte_eth_dev *dev,
1447 			     struct rte_flow_error *error)
1448 {
1449 	const struct rte_flow_item_vlan *spec = item->spec;
1450 	const struct rte_flow_item_vlan *mask = item->mask;
1451 	const struct rte_flow_item_vlan nic_mask = {
1452 		.tci = RTE_BE16(UINT16_MAX),
1453 		.inner_type = RTE_BE16(UINT16_MAX),
1454 	};
1455 	uint16_t vlan_tag = 0;
1456 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1457 	int ret;
1458 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1459 					MLX5_FLOW_LAYER_INNER_L4) :
1460 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1461 					MLX5_FLOW_LAYER_OUTER_L4);
1462 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1463 					MLX5_FLOW_LAYER_OUTER_VLAN;
1464 
1465 	if (item_flags & vlanm)
1466 		return rte_flow_error_set(error, EINVAL,
1467 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1468 					  "multiple VLAN layers not supported");
1469 	else if ((item_flags & l34m) != 0)
1470 		return rte_flow_error_set(error, EINVAL,
1471 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1472 					  "VLAN cannot follow L3/L4 layer");
1473 	if (!mask)
1474 		mask = &rte_flow_item_vlan_mask;
1475 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1476 					(const uint8_t *)&nic_mask,
1477 					sizeof(struct rte_flow_item_vlan),
1478 					error);
1479 	if (ret)
1480 		return ret;
1481 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1482 		struct mlx5_priv *priv = dev->data->dev_private;
1483 
1484 		if (priv->vmwa_context) {
1485 			/*
1486 			 * Non-NULL context means we have a virtual machine
1487 			 * and SR-IOV enabled, we have to create VLAN interface
1488 			 * to make hypervisor to setup E-Switch vport
1489 			 * context correctly. We avoid creating the multiple
1490 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1491 			 */
1492 			return rte_flow_error_set(error, EINVAL,
1493 						  RTE_FLOW_ERROR_TYPE_ITEM,
1494 						  item,
1495 						  "VLAN tag mask is not"
1496 						  " supported in virtual"
1497 						  " environment");
1498 		}
1499 	}
1500 	if (spec) {
1501 		vlan_tag = spec->tci;
1502 		vlan_tag &= mask->tci;
1503 	}
1504 	/*
1505 	 * From verbs perspective an empty VLAN is equivalent
1506 	 * to a packet without VLAN layer.
1507 	 */
1508 	if (!vlan_tag)
1509 		return rte_flow_error_set(error, EINVAL,
1510 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1511 					  item->spec,
1512 					  "VLAN cannot be empty");
1513 	return 0;
1514 }
1515 
1516 /**
1517  * Validate IPV4 item.
1518  *
1519  * @param[in] item
1520  *   Item specification.
1521  * @param[in] item_flags
1522  *   Bit-fields that holds the items detected until now.
1523  * @param[in] acc_mask
1524  *   Acceptable mask, if NULL default internal default mask
1525  *   will be used to check whether item fields are supported.
1526  * @param[out] error
1527  *   Pointer to error structure.
1528  *
1529  * @return
1530  *   0 on success, a negative errno value otherwise and rte_errno is set.
1531  */
1532 int
1533 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1534 			     uint64_t item_flags,
1535 			     uint64_t last_item,
1536 			     uint16_t ether_type,
1537 			     const struct rte_flow_item_ipv4 *acc_mask,
1538 			     struct rte_flow_error *error)
1539 {
1540 	const struct rte_flow_item_ipv4 *mask = item->mask;
1541 	const struct rte_flow_item_ipv4 *spec = item->spec;
1542 	const struct rte_flow_item_ipv4 nic_mask = {
1543 		.hdr = {
1544 			.src_addr = RTE_BE32(0xffffffff),
1545 			.dst_addr = RTE_BE32(0xffffffff),
1546 			.type_of_service = 0xff,
1547 			.next_proto_id = 0xff,
1548 		},
1549 	};
1550 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1551 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1552 				      MLX5_FLOW_LAYER_OUTER_L3;
1553 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1554 				      MLX5_FLOW_LAYER_OUTER_L4;
1555 	int ret;
1556 	uint8_t next_proto = 0xFF;
1557 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1558 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1559 				  MLX5_FLOW_LAYER_INNER_VLAN);
1560 
1561 	if ((last_item & l2_vlan) && ether_type &&
1562 	    ether_type != RTE_ETHER_TYPE_IPV4)
1563 		return rte_flow_error_set(error, EINVAL,
1564 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1565 					  "IPv4 cannot follow L2/VLAN layer "
1566 					  "which ether type is not IPv4");
1567 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1568 		if (mask && spec)
1569 			next_proto = mask->hdr.next_proto_id &
1570 				     spec->hdr.next_proto_id;
1571 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1572 			return rte_flow_error_set(error, EINVAL,
1573 						  RTE_FLOW_ERROR_TYPE_ITEM,
1574 						  item,
1575 						  "multiple tunnel "
1576 						  "not supported");
1577 	}
1578 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 					  "wrong tunnel type - IPv6 specified "
1582 					  "but IPv4 item provided");
1583 	if (item_flags & l3m)
1584 		return rte_flow_error_set(error, ENOTSUP,
1585 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1586 					  "multiple L3 layers not supported");
1587 	else if (item_flags & l4m)
1588 		return rte_flow_error_set(error, EINVAL,
1589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1590 					  "L3 cannot follow an L4 layer.");
1591 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1592 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1595 					  "L3 cannot follow an NVGRE layer.");
1596 	if (!mask)
1597 		mask = &rte_flow_item_ipv4_mask;
1598 	else if (mask->hdr.next_proto_id != 0 &&
1599 		 mask->hdr.next_proto_id != 0xff)
1600 		return rte_flow_error_set(error, EINVAL,
1601 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1602 					  "partial mask is not supported"
1603 					  " for protocol");
1604 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1605 					acc_mask ? (const uint8_t *)acc_mask
1606 						 : (const uint8_t *)&nic_mask,
1607 					sizeof(struct rte_flow_item_ipv4),
1608 					error);
1609 	if (ret < 0)
1610 		return ret;
1611 	return 0;
1612 }
1613 
1614 /**
1615  * Validate IPV6 item.
1616  *
1617  * @param[in] item
1618  *   Item specification.
1619  * @param[in] item_flags
1620  *   Bit-fields that holds the items detected until now.
1621  * @param[in] acc_mask
1622  *   Acceptable mask, if NULL default internal default mask
1623  *   will be used to check whether item fields are supported.
1624  * @param[out] error
1625  *   Pointer to error structure.
1626  *
1627  * @return
1628  *   0 on success, a negative errno value otherwise and rte_errno is set.
1629  */
1630 int
1631 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1632 			     uint64_t item_flags,
1633 			     uint64_t last_item,
1634 			     uint16_t ether_type,
1635 			     const struct rte_flow_item_ipv6 *acc_mask,
1636 			     struct rte_flow_error *error)
1637 {
1638 	const struct rte_flow_item_ipv6 *mask = item->mask;
1639 	const struct rte_flow_item_ipv6 *spec = item->spec;
1640 	const struct rte_flow_item_ipv6 nic_mask = {
1641 		.hdr = {
1642 			.src_addr =
1643 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1644 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1645 			.dst_addr =
1646 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1647 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1648 			.vtc_flow = RTE_BE32(0xffffffff),
1649 			.proto = 0xff,
1650 			.hop_limits = 0xff,
1651 		},
1652 	};
1653 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1654 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1655 				      MLX5_FLOW_LAYER_OUTER_L3;
1656 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1657 				      MLX5_FLOW_LAYER_OUTER_L4;
1658 	int ret;
1659 	uint8_t next_proto = 0xFF;
1660 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1661 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1662 				  MLX5_FLOW_LAYER_INNER_VLAN);
1663 
1664 	if ((last_item & l2_vlan) && ether_type &&
1665 	    ether_type != RTE_ETHER_TYPE_IPV6)
1666 		return rte_flow_error_set(error, EINVAL,
1667 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1668 					  "IPv6 cannot follow L2/VLAN layer "
1669 					  "which ether type is not IPv6");
1670 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1671 		if (mask && spec)
1672 			next_proto = mask->hdr.proto & spec->hdr.proto;
1673 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1674 			return rte_flow_error_set(error, EINVAL,
1675 						  RTE_FLOW_ERROR_TYPE_ITEM,
1676 						  item,
1677 						  "multiple tunnel "
1678 						  "not supported");
1679 	}
1680 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 					  "wrong tunnel type - IPv4 specified "
1684 					  "but IPv6 item provided");
1685 	if (item_flags & l3m)
1686 		return rte_flow_error_set(error, ENOTSUP,
1687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1688 					  "multiple L3 layers not supported");
1689 	else if (item_flags & l4m)
1690 		return rte_flow_error_set(error, EINVAL,
1691 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1692 					  "L3 cannot follow an L4 layer.");
1693 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1694 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1695 		return rte_flow_error_set(error, EINVAL,
1696 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1697 					  "L3 cannot follow an NVGRE layer.");
1698 	if (!mask)
1699 		mask = &rte_flow_item_ipv6_mask;
1700 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1701 					acc_mask ? (const uint8_t *)acc_mask
1702 						 : (const uint8_t *)&nic_mask,
1703 					sizeof(struct rte_flow_item_ipv6),
1704 					error);
1705 	if (ret < 0)
1706 		return ret;
1707 	return 0;
1708 }
1709 
1710 /**
1711  * Validate UDP item.
1712  *
1713  * @param[in] item
1714  *   Item specification.
1715  * @param[in] item_flags
1716  *   Bit-fields that holds the items detected until now.
1717  * @param[in] target_protocol
1718  *   The next protocol in the previous item.
1719  * @param[in] flow_mask
1720  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1721  * @param[out] error
1722  *   Pointer to error structure.
1723  *
1724  * @return
1725  *   0 on success, a negative errno value otherwise and rte_errno is set.
1726  */
1727 int
1728 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1729 			    uint64_t item_flags,
1730 			    uint8_t target_protocol,
1731 			    struct rte_flow_error *error)
1732 {
1733 	const struct rte_flow_item_udp *mask = item->mask;
1734 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1735 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1736 				      MLX5_FLOW_LAYER_OUTER_L3;
1737 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1738 				      MLX5_FLOW_LAYER_OUTER_L4;
1739 	int ret;
1740 
1741 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1742 		return rte_flow_error_set(error, EINVAL,
1743 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1744 					  "protocol filtering not compatible"
1745 					  " with UDP layer");
1746 	if (!(item_flags & l3m))
1747 		return rte_flow_error_set(error, EINVAL,
1748 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1749 					  "L3 is mandatory to filter on L4");
1750 	if (item_flags & l4m)
1751 		return rte_flow_error_set(error, EINVAL,
1752 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1753 					  "multiple L4 layers not supported");
1754 	if (!mask)
1755 		mask = &rte_flow_item_udp_mask;
1756 	ret = mlx5_flow_item_acceptable
1757 		(item, (const uint8_t *)mask,
1758 		 (const uint8_t *)&rte_flow_item_udp_mask,
1759 		 sizeof(struct rte_flow_item_udp), error);
1760 	if (ret < 0)
1761 		return ret;
1762 	return 0;
1763 }
1764 
1765 /**
1766  * Validate TCP item.
1767  *
1768  * @param[in] item
1769  *   Item specification.
1770  * @param[in] item_flags
1771  *   Bit-fields that holds the items detected until now.
1772  * @param[in] target_protocol
1773  *   The next protocol in the previous item.
1774  * @param[out] error
1775  *   Pointer to error structure.
1776  *
1777  * @return
1778  *   0 on success, a negative errno value otherwise and rte_errno is set.
1779  */
1780 int
1781 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1782 			    uint64_t item_flags,
1783 			    uint8_t target_protocol,
1784 			    const struct rte_flow_item_tcp *flow_mask,
1785 			    struct rte_flow_error *error)
1786 {
1787 	const struct rte_flow_item_tcp *mask = item->mask;
1788 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1789 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1790 				      MLX5_FLOW_LAYER_OUTER_L3;
1791 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1792 				      MLX5_FLOW_LAYER_OUTER_L4;
1793 	int ret;
1794 
1795 	assert(flow_mask);
1796 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1797 		return rte_flow_error_set(error, EINVAL,
1798 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1799 					  "protocol filtering not compatible"
1800 					  " with TCP layer");
1801 	if (!(item_flags & l3m))
1802 		return rte_flow_error_set(error, EINVAL,
1803 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1804 					  "L3 is mandatory to filter on L4");
1805 	if (item_flags & l4m)
1806 		return rte_flow_error_set(error, EINVAL,
1807 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1808 					  "multiple L4 layers not supported");
1809 	if (!mask)
1810 		mask = &rte_flow_item_tcp_mask;
1811 	ret = mlx5_flow_item_acceptable
1812 		(item, (const uint8_t *)mask,
1813 		 (const uint8_t *)flow_mask,
1814 		 sizeof(struct rte_flow_item_tcp), error);
1815 	if (ret < 0)
1816 		return ret;
1817 	return 0;
1818 }
1819 
1820 /**
1821  * Validate VXLAN item.
1822  *
1823  * @param[in] item
1824  *   Item specification.
1825  * @param[in] item_flags
1826  *   Bit-fields that holds the items detected until now.
1827  * @param[in] target_protocol
1828  *   The next protocol in the previous item.
1829  * @param[out] error
1830  *   Pointer to error structure.
1831  *
1832  * @return
1833  *   0 on success, a negative errno value otherwise and rte_errno is set.
1834  */
1835 int
1836 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1837 			      uint64_t item_flags,
1838 			      struct rte_flow_error *error)
1839 {
1840 	const struct rte_flow_item_vxlan *spec = item->spec;
1841 	const struct rte_flow_item_vxlan *mask = item->mask;
1842 	int ret;
1843 	union vni {
1844 		uint32_t vlan_id;
1845 		uint8_t vni[4];
1846 	} id = { .vlan_id = 0, };
1847 	uint32_t vlan_id = 0;
1848 
1849 
1850 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1851 		return rte_flow_error_set(error, ENOTSUP,
1852 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 					  "multiple tunnel layers not"
1854 					  " supported");
1855 	/*
1856 	 * Verify only UDPv4 is present as defined in
1857 	 * https://tools.ietf.org/html/rfc7348
1858 	 */
1859 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1860 		return rte_flow_error_set(error, EINVAL,
1861 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1862 					  "no outer UDP layer found");
1863 	if (!mask)
1864 		mask = &rte_flow_item_vxlan_mask;
1865 	ret = mlx5_flow_item_acceptable
1866 		(item, (const uint8_t *)mask,
1867 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1868 		 sizeof(struct rte_flow_item_vxlan),
1869 		 error);
1870 	if (ret < 0)
1871 		return ret;
1872 	if (spec) {
1873 		memcpy(&id.vni[1], spec->vni, 3);
1874 		vlan_id = id.vlan_id;
1875 		memcpy(&id.vni[1], mask->vni, 3);
1876 		vlan_id &= id.vlan_id;
1877 	}
1878 	/*
1879 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1880 	 * only this layer is defined in the Verbs specification it is
1881 	 * interpreted as wildcard and all packets will match this
1882 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1883 	 * udp), all packets matching the layers before will also
1884 	 * match this rule.  To avoid such situation, VNI 0 is
1885 	 * currently refused.
1886 	 */
1887 	if (!vlan_id)
1888 		return rte_flow_error_set(error, ENOTSUP,
1889 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1890 					  "VXLAN vni cannot be 0");
1891 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1892 		return rte_flow_error_set(error, ENOTSUP,
1893 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1894 					  "VXLAN tunnel must be fully defined");
1895 	return 0;
1896 }
1897 
1898 /**
1899  * Validate VXLAN_GPE item.
1900  *
1901  * @param[in] item
1902  *   Item specification.
1903  * @param[in] item_flags
1904  *   Bit-fields that holds the items detected until now.
1905  * @param[in] priv
1906  *   Pointer to the private data structure.
1907  * @param[in] target_protocol
1908  *   The next protocol in the previous item.
1909  * @param[out] error
1910  *   Pointer to error structure.
1911  *
1912  * @return
1913  *   0 on success, a negative errno value otherwise and rte_errno is set.
1914  */
1915 int
1916 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1917 				  uint64_t item_flags,
1918 				  struct rte_eth_dev *dev,
1919 				  struct rte_flow_error *error)
1920 {
1921 	struct mlx5_priv *priv = dev->data->dev_private;
1922 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1923 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1924 	int ret;
1925 	union vni {
1926 		uint32_t vlan_id;
1927 		uint8_t vni[4];
1928 	} id = { .vlan_id = 0, };
1929 	uint32_t vlan_id = 0;
1930 
1931 	if (!priv->config.l3_vxlan_en)
1932 		return rte_flow_error_set(error, ENOTSUP,
1933 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1934 					  "L3 VXLAN is not enabled by device"
1935 					  " parameter and/or not configured in"
1936 					  " firmware");
1937 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1938 		return rte_flow_error_set(error, ENOTSUP,
1939 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1940 					  "multiple tunnel layers not"
1941 					  " supported");
1942 	/*
1943 	 * Verify only UDPv4 is present as defined in
1944 	 * https://tools.ietf.org/html/rfc7348
1945 	 */
1946 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1947 		return rte_flow_error_set(error, EINVAL,
1948 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 					  "no outer UDP layer found");
1950 	if (!mask)
1951 		mask = &rte_flow_item_vxlan_gpe_mask;
1952 	ret = mlx5_flow_item_acceptable
1953 		(item, (const uint8_t *)mask,
1954 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1955 		 sizeof(struct rte_flow_item_vxlan_gpe),
1956 		 error);
1957 	if (ret < 0)
1958 		return ret;
1959 	if (spec) {
1960 		if (spec->protocol)
1961 			return rte_flow_error_set(error, ENOTSUP,
1962 						  RTE_FLOW_ERROR_TYPE_ITEM,
1963 						  item,
1964 						  "VxLAN-GPE protocol"
1965 						  " not supported");
1966 		memcpy(&id.vni[1], spec->vni, 3);
1967 		vlan_id = id.vlan_id;
1968 		memcpy(&id.vni[1], mask->vni, 3);
1969 		vlan_id &= id.vlan_id;
1970 	}
1971 	/*
1972 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1973 	 * layer is defined in the Verbs specification it is interpreted as
1974 	 * wildcard and all packets will match this rule, if it follows a full
1975 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1976 	 * before will also match this rule.  To avoid such situation, VNI 0
1977 	 * is currently refused.
1978 	 */
1979 	if (!vlan_id)
1980 		return rte_flow_error_set(error, ENOTSUP,
1981 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1982 					  "VXLAN-GPE vni cannot be 0");
1983 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1984 		return rte_flow_error_set(error, ENOTSUP,
1985 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1986 					  "VXLAN-GPE tunnel must be fully"
1987 					  " defined");
1988 	return 0;
1989 }
1990 /**
1991  * Validate GRE Key item.
1992  *
1993  * @param[in] item
1994  *   Item specification.
1995  * @param[in] item_flags
1996  *   Bit flags to mark detected items.
1997  * @param[in] gre_item
1998  *   Pointer to gre_item
1999  * @param[out] error
2000  *   Pointer to error structure.
2001  *
2002  * @return
2003  *   0 on success, a negative errno value otherwise and rte_errno is set.
2004  */
2005 int
2006 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2007 				uint64_t item_flags,
2008 				const struct rte_flow_item *gre_item,
2009 				struct rte_flow_error *error)
2010 {
2011 	const rte_be32_t *mask = item->mask;
2012 	int ret = 0;
2013 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2014 	const struct rte_flow_item_gre *gre_spec;
2015 	const struct rte_flow_item_gre *gre_mask;
2016 
2017 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2018 		return rte_flow_error_set(error, ENOTSUP,
2019 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2020 					  "Multiple GRE key not support");
2021 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2022 		return rte_flow_error_set(error, ENOTSUP,
2023 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2024 					  "No preceding GRE header");
2025 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2026 		return rte_flow_error_set(error, ENOTSUP,
2027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 					  "GRE key following a wrong item");
2029 	gre_mask = gre_item->mask;
2030 	if (!gre_mask)
2031 		gre_mask = &rte_flow_item_gre_mask;
2032 	gre_spec = gre_item->spec;
2033 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2034 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2035 		return rte_flow_error_set(error, EINVAL,
2036 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2037 					  "Key bit must be on");
2038 
2039 	if (!mask)
2040 		mask = &gre_key_default_mask;
2041 	ret = mlx5_flow_item_acceptable
2042 		(item, (const uint8_t *)mask,
2043 		 (const uint8_t *)&gre_key_default_mask,
2044 		 sizeof(rte_be32_t), error);
2045 	return ret;
2046 }
2047 
2048 /**
2049  * Validate GRE item.
2050  *
2051  * @param[in] item
2052  *   Item specification.
2053  * @param[in] item_flags
2054  *   Bit flags to mark detected items.
2055  * @param[in] target_protocol
2056  *   The next protocol in the previous item.
2057  * @param[out] error
2058  *   Pointer to error structure.
2059  *
2060  * @return
2061  *   0 on success, a negative errno value otherwise and rte_errno is set.
2062  */
2063 int
2064 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2065 			    uint64_t item_flags,
2066 			    uint8_t target_protocol,
2067 			    struct rte_flow_error *error)
2068 {
2069 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2070 	const struct rte_flow_item_gre *mask = item->mask;
2071 	int ret;
2072 	const struct rte_flow_item_gre nic_mask = {
2073 		.c_rsvd0_ver = RTE_BE16(0xB000),
2074 		.protocol = RTE_BE16(UINT16_MAX),
2075 	};
2076 
2077 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2078 		return rte_flow_error_set(error, EINVAL,
2079 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2080 					  "protocol filtering not compatible"
2081 					  " with this GRE layer");
2082 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2083 		return rte_flow_error_set(error, ENOTSUP,
2084 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2085 					  "multiple tunnel layers not"
2086 					  " supported");
2087 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2088 		return rte_flow_error_set(error, ENOTSUP,
2089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2090 					  "L3 Layer is missing");
2091 	if (!mask)
2092 		mask = &rte_flow_item_gre_mask;
2093 	ret = mlx5_flow_item_acceptable
2094 		(item, (const uint8_t *)mask,
2095 		 (const uint8_t *)&nic_mask,
2096 		 sizeof(struct rte_flow_item_gre), error);
2097 	if (ret < 0)
2098 		return ret;
2099 #ifndef HAVE_MLX5DV_DR
2100 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2101 	if (spec && (spec->protocol & mask->protocol))
2102 		return rte_flow_error_set(error, ENOTSUP,
2103 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2104 					  "without MPLS support the"
2105 					  " specification cannot be used for"
2106 					  " filtering");
2107 #endif
2108 #endif
2109 	return 0;
2110 }
2111 
2112 /**
2113  * Validate Geneve item.
2114  *
2115  * @param[in] item
2116  *   Item specification.
2117  * @param[in] itemFlags
2118  *   Bit-fields that holds the items detected until now.
2119  * @param[in] enPriv
2120  *   Pointer to the private data structure.
2121  * @param[out] error
2122  *   Pointer to error structure.
2123  *
2124  * @return
2125  *   0 on success, a negative errno value otherwise and rte_errno is set.
2126  */
2127 
2128 int
2129 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2130 			       uint64_t item_flags,
2131 			       struct rte_eth_dev *dev,
2132 			       struct rte_flow_error *error)
2133 {
2134 	struct mlx5_priv *priv = dev->data->dev_private;
2135 	const struct rte_flow_item_geneve *spec = item->spec;
2136 	const struct rte_flow_item_geneve *mask = item->mask;
2137 	int ret;
2138 	uint16_t gbhdr;
2139 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2140 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2141 	const struct rte_flow_item_geneve nic_mask = {
2142 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2143 		.vni = "\xff\xff\xff",
2144 		.protocol = RTE_BE16(UINT16_MAX),
2145 	};
2146 
2147 	if (!(priv->config.hca_attr.flex_parser_protocols &
2148 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2149 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2150 		return rte_flow_error_set(error, ENOTSUP,
2151 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2152 					  "L3 Geneve is not enabled by device"
2153 					  " parameter and/or not configured in"
2154 					  " firmware");
2155 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2156 		return rte_flow_error_set(error, ENOTSUP,
2157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 					  "multiple tunnel layers not"
2159 					  " supported");
2160 	/*
2161 	 * Verify only UDPv4 is present as defined in
2162 	 * https://tools.ietf.org/html/rfc7348
2163 	 */
2164 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2165 		return rte_flow_error_set(error, EINVAL,
2166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 					  "no outer UDP layer found");
2168 	if (!mask)
2169 		mask = &rte_flow_item_geneve_mask;
2170 	ret = mlx5_flow_item_acceptable
2171 				  (item, (const uint8_t *)mask,
2172 				   (const uint8_t *)&nic_mask,
2173 				   sizeof(struct rte_flow_item_geneve), error);
2174 	if (ret)
2175 		return ret;
2176 	if (spec) {
2177 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2178 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2179 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2180 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2181 			return rte_flow_error_set(error, ENOTSUP,
2182 						  RTE_FLOW_ERROR_TYPE_ITEM,
2183 						  item,
2184 						  "Geneve protocol unsupported"
2185 						  " fields are being used");
2186 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2187 			return rte_flow_error_set
2188 					(error, ENOTSUP,
2189 					 RTE_FLOW_ERROR_TYPE_ITEM,
2190 					 item,
2191 					 "Unsupported Geneve options length");
2192 	}
2193 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2194 		return rte_flow_error_set
2195 				    (error, ENOTSUP,
2196 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 				     "Geneve tunnel must be fully defined");
2198 	return 0;
2199 }
2200 
2201 /**
2202  * Validate MPLS item.
2203  *
2204  * @param[in] dev
2205  *   Pointer to the rte_eth_dev structure.
2206  * @param[in] item
2207  *   Item specification.
2208  * @param[in] item_flags
2209  *   Bit-fields that holds the items detected until now.
2210  * @param[in] prev_layer
2211  *   The protocol layer indicated in previous item.
2212  * @param[out] error
2213  *   Pointer to error structure.
2214  *
2215  * @return
2216  *   0 on success, a negative errno value otherwise and rte_errno is set.
2217  */
2218 int
2219 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2220 			     const struct rte_flow_item *item __rte_unused,
2221 			     uint64_t item_flags __rte_unused,
2222 			     uint64_t prev_layer __rte_unused,
2223 			     struct rte_flow_error *error)
2224 {
2225 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2226 	const struct rte_flow_item_mpls *mask = item->mask;
2227 	struct mlx5_priv *priv = dev->data->dev_private;
2228 	int ret;
2229 
2230 	if (!priv->config.mpls_en)
2231 		return rte_flow_error_set(error, ENOTSUP,
2232 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 					  "MPLS not supported or"
2234 					  " disabled in firmware"
2235 					  " configuration.");
2236 	/* MPLS over IP, UDP, GRE is allowed */
2237 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2238 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2239 			    MLX5_FLOW_LAYER_GRE)))
2240 		return rte_flow_error_set(error, EINVAL,
2241 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2242 					  "protocol filtering not compatible"
2243 					  " with MPLS layer");
2244 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2245 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2246 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2247 		return rte_flow_error_set(error, ENOTSUP,
2248 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2249 					  "multiple tunnel layers not"
2250 					  " supported");
2251 	if (!mask)
2252 		mask = &rte_flow_item_mpls_mask;
2253 	ret = mlx5_flow_item_acceptable
2254 		(item, (const uint8_t *)mask,
2255 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2256 		 sizeof(struct rte_flow_item_mpls), error);
2257 	if (ret < 0)
2258 		return ret;
2259 	return 0;
2260 #endif
2261 	return rte_flow_error_set(error, ENOTSUP,
2262 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2263 				  "MPLS is not supported by Verbs, please"
2264 				  " update.");
2265 }
2266 
2267 /**
2268  * Validate NVGRE item.
2269  *
2270  * @param[in] item
2271  *   Item specification.
2272  * @param[in] item_flags
2273  *   Bit flags to mark detected items.
2274  * @param[in] target_protocol
2275  *   The next protocol in the previous item.
2276  * @param[out] error
2277  *   Pointer to error structure.
2278  *
2279  * @return
2280  *   0 on success, a negative errno value otherwise and rte_errno is set.
2281  */
2282 int
2283 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2284 			      uint64_t item_flags,
2285 			      uint8_t target_protocol,
2286 			      struct rte_flow_error *error)
2287 {
2288 	const struct rte_flow_item_nvgre *mask = item->mask;
2289 	int ret;
2290 
2291 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2292 		return rte_flow_error_set(error, EINVAL,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "protocol filtering not compatible"
2295 					  " with this GRE layer");
2296 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2297 		return rte_flow_error_set(error, ENOTSUP,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "multiple tunnel layers not"
2300 					  " supported");
2301 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2302 		return rte_flow_error_set(error, ENOTSUP,
2303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2304 					  "L3 Layer is missing");
2305 	if (!mask)
2306 		mask = &rte_flow_item_nvgre_mask;
2307 	ret = mlx5_flow_item_acceptable
2308 		(item, (const uint8_t *)mask,
2309 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2310 		 sizeof(struct rte_flow_item_nvgre), error);
2311 	if (ret < 0)
2312 		return ret;
2313 	return 0;
2314 }
2315 
2316 /* Allocate unique ID for the split Q/RSS subflows. */
2317 static uint32_t
2318 flow_qrss_get_id(struct rte_eth_dev *dev)
2319 {
2320 	struct mlx5_priv *priv = dev->data->dev_private;
2321 	uint32_t qrss_id, ret;
2322 
2323 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2324 	if (ret)
2325 		return 0;
2326 	assert(qrss_id);
2327 	return qrss_id;
2328 }
2329 
2330 /* Free unique ID for the split Q/RSS subflows. */
2331 static void
2332 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2333 {
2334 	struct mlx5_priv *priv = dev->data->dev_private;
2335 
2336 	if (qrss_id)
2337 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2338 }
2339 
2340 /**
2341  * Release resource related QUEUE/RSS action split.
2342  *
2343  * @param dev
2344  *   Pointer to Ethernet device.
2345  * @param flow
2346  *   Flow to release id's from.
2347  */
2348 static void
2349 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2350 			     struct rte_flow *flow)
2351 {
2352 	struct mlx5_flow *dev_flow;
2353 
2354 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2355 		if (dev_flow->qrss_id)
2356 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2357 }
2358 
2359 static int
2360 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2361 		   const struct rte_flow_attr *attr __rte_unused,
2362 		   const struct rte_flow_item items[] __rte_unused,
2363 		   const struct rte_flow_action actions[] __rte_unused,
2364 		   bool external __rte_unused,
2365 		   struct rte_flow_error *error)
2366 {
2367 	return rte_flow_error_set(error, ENOTSUP,
2368 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2369 }
2370 
2371 static struct mlx5_flow *
2372 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2373 		  const struct rte_flow_item items[] __rte_unused,
2374 		  const struct rte_flow_action actions[] __rte_unused,
2375 		  struct rte_flow_error *error)
2376 {
2377 	rte_flow_error_set(error, ENOTSUP,
2378 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2379 	return NULL;
2380 }
2381 
2382 static int
2383 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2384 		    struct mlx5_flow *dev_flow __rte_unused,
2385 		    const struct rte_flow_attr *attr __rte_unused,
2386 		    const struct rte_flow_item items[] __rte_unused,
2387 		    const struct rte_flow_action actions[] __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 int
2395 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2396 		struct rte_flow *flow __rte_unused,
2397 		struct rte_flow_error *error)
2398 {
2399 	return rte_flow_error_set(error, ENOTSUP,
2400 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2401 }
2402 
2403 static void
2404 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2405 		 struct rte_flow *flow __rte_unused)
2406 {
2407 }
2408 
2409 static void
2410 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2411 		  struct rte_flow *flow __rte_unused)
2412 {
2413 }
2414 
2415 static int
2416 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2417 		struct rte_flow *flow __rte_unused,
2418 		const struct rte_flow_action *actions __rte_unused,
2419 		void *data __rte_unused,
2420 		struct rte_flow_error *error)
2421 {
2422 	return rte_flow_error_set(error, ENOTSUP,
2423 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2424 }
2425 
2426 /* Void driver to protect from null pointer reference. */
2427 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2428 	.validate = flow_null_validate,
2429 	.prepare = flow_null_prepare,
2430 	.translate = flow_null_translate,
2431 	.apply = flow_null_apply,
2432 	.remove = flow_null_remove,
2433 	.destroy = flow_null_destroy,
2434 	.query = flow_null_query,
2435 };
2436 
2437 /**
2438  * Select flow driver type according to flow attributes and device
2439  * configuration.
2440  *
2441  * @param[in] dev
2442  *   Pointer to the dev structure.
2443  * @param[in] attr
2444  *   Pointer to the flow attributes.
2445  *
2446  * @return
2447  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2448  */
2449 static enum mlx5_flow_drv_type
2450 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2451 {
2452 	struct mlx5_priv *priv = dev->data->dev_private;
2453 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2454 
2455 	if (attr->transfer && priv->config.dv_esw_en)
2456 		type = MLX5_FLOW_TYPE_DV;
2457 	if (!attr->transfer)
2458 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2459 						 MLX5_FLOW_TYPE_VERBS;
2460 	return type;
2461 }
2462 
2463 #define flow_get_drv_ops(type) flow_drv_ops[type]
2464 
2465 /**
2466  * Flow driver validation API. This abstracts calling driver specific functions.
2467  * The type of flow driver is determined according to flow attributes.
2468  *
2469  * @param[in] dev
2470  *   Pointer to the dev structure.
2471  * @param[in] attr
2472  *   Pointer to the flow attributes.
2473  * @param[in] items
2474  *   Pointer to the list of items.
2475  * @param[in] actions
2476  *   Pointer to the list of actions.
2477  * @param[in] external
2478  *   This flow rule is created by request external to PMD.
2479  * @param[out] error
2480  *   Pointer to the error structure.
2481  *
2482  * @return
2483  *   0 on success, a negative errno value otherwise and rte_errno is set.
2484  */
2485 static inline int
2486 flow_drv_validate(struct rte_eth_dev *dev,
2487 		  const struct rte_flow_attr *attr,
2488 		  const struct rte_flow_item items[],
2489 		  const struct rte_flow_action actions[],
2490 		  bool external, struct rte_flow_error *error)
2491 {
2492 	const struct mlx5_flow_driver_ops *fops;
2493 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2494 
2495 	fops = flow_get_drv_ops(type);
2496 	return fops->validate(dev, attr, items, actions, external, error);
2497 }
2498 
2499 /**
2500  * Flow driver preparation API. This abstracts calling driver specific
2501  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2502  * calculates the size of memory required for device flow, allocates the memory,
2503  * initializes the device flow and returns the pointer.
2504  *
2505  * @note
2506  *   This function initializes device flow structure such as dv or verbs in
2507  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2508  *   rest. For example, adding returning device flow to flow->dev_flow list and
2509  *   setting backward reference to the flow should be done out of this function.
2510  *   layers field is not filled either.
2511  *
2512  * @param[in] attr
2513  *   Pointer to the flow attributes.
2514  * @param[in] items
2515  *   Pointer to the list of items.
2516  * @param[in] actions
2517  *   Pointer to the list of actions.
2518  * @param[out] error
2519  *   Pointer to the error structure.
2520  *
2521  * @return
2522  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2523  */
2524 static inline struct mlx5_flow *
2525 flow_drv_prepare(const struct rte_flow *flow,
2526 		 const struct rte_flow_attr *attr,
2527 		 const struct rte_flow_item items[],
2528 		 const struct rte_flow_action actions[],
2529 		 struct rte_flow_error *error)
2530 {
2531 	const struct mlx5_flow_driver_ops *fops;
2532 	enum mlx5_flow_drv_type type = flow->drv_type;
2533 
2534 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2535 	fops = flow_get_drv_ops(type);
2536 	return fops->prepare(attr, items, actions, error);
2537 }
2538 
2539 /**
2540  * Flow driver translation API. This abstracts calling driver specific
2541  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2542  * translates a generic flow into a driver flow. flow_drv_prepare() must
2543  * precede.
2544  *
2545  * @note
2546  *   dev_flow->layers could be filled as a result of parsing during translation
2547  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2548  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2549  *   flow->actions could be overwritten even though all the expanded dev_flows
2550  *   have the same actions.
2551  *
2552  * @param[in] dev
2553  *   Pointer to the rte dev structure.
2554  * @param[in, out] dev_flow
2555  *   Pointer to the mlx5 flow.
2556  * @param[in] attr
2557  *   Pointer to the flow attributes.
2558  * @param[in] items
2559  *   Pointer to the list of items.
2560  * @param[in] actions
2561  *   Pointer to the list of actions.
2562  * @param[out] error
2563  *   Pointer to the error structure.
2564  *
2565  * @return
2566  *   0 on success, a negative errno value otherwise and rte_errno is set.
2567  */
2568 static inline int
2569 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2570 		   const struct rte_flow_attr *attr,
2571 		   const struct rte_flow_item items[],
2572 		   const struct rte_flow_action actions[],
2573 		   struct rte_flow_error *error)
2574 {
2575 	const struct mlx5_flow_driver_ops *fops;
2576 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2577 
2578 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2579 	fops = flow_get_drv_ops(type);
2580 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2581 }
2582 
2583 /**
2584  * Flow driver apply API. This abstracts calling driver specific functions.
2585  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2586  * translated driver flows on to device. flow_drv_translate() must precede.
2587  *
2588  * @param[in] dev
2589  *   Pointer to Ethernet device structure.
2590  * @param[in, out] flow
2591  *   Pointer to flow structure.
2592  * @param[out] error
2593  *   Pointer to error structure.
2594  *
2595  * @return
2596  *   0 on success, a negative errno value otherwise and rte_errno is set.
2597  */
2598 static inline int
2599 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2600 	       struct rte_flow_error *error)
2601 {
2602 	const struct mlx5_flow_driver_ops *fops;
2603 	enum mlx5_flow_drv_type type = flow->drv_type;
2604 
2605 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2606 	fops = flow_get_drv_ops(type);
2607 	return fops->apply(dev, flow, error);
2608 }
2609 
2610 /**
2611  * Flow driver remove API. This abstracts calling driver specific functions.
2612  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2613  * on device. All the resources of the flow should be freed by calling
2614  * flow_drv_destroy().
2615  *
2616  * @param[in] dev
2617  *   Pointer to Ethernet device.
2618  * @param[in, out] flow
2619  *   Pointer to flow structure.
2620  */
2621 static inline void
2622 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2623 {
2624 	const struct mlx5_flow_driver_ops *fops;
2625 	enum mlx5_flow_drv_type type = flow->drv_type;
2626 
2627 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2628 	fops = flow_get_drv_ops(type);
2629 	fops->remove(dev, flow);
2630 }
2631 
2632 /**
2633  * Flow driver destroy API. This abstracts calling driver specific functions.
2634  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2635  * on device and releases resources of the flow.
2636  *
2637  * @param[in] dev
2638  *   Pointer to Ethernet device.
2639  * @param[in, out] flow
2640  *   Pointer to flow structure.
2641  */
2642 static inline void
2643 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2644 {
2645 	const struct mlx5_flow_driver_ops *fops;
2646 	enum mlx5_flow_drv_type type = flow->drv_type;
2647 
2648 	flow_mreg_split_qrss_release(dev, flow);
2649 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2650 	fops = flow_get_drv_ops(type);
2651 	fops->destroy(dev, flow);
2652 }
2653 
2654 /**
2655  * Validate a flow supported by the NIC.
2656  *
2657  * @see rte_flow_validate()
2658  * @see rte_flow_ops
2659  */
2660 int
2661 mlx5_flow_validate(struct rte_eth_dev *dev,
2662 		   const struct rte_flow_attr *attr,
2663 		   const struct rte_flow_item items[],
2664 		   const struct rte_flow_action actions[],
2665 		   struct rte_flow_error *error)
2666 {
2667 	int ret;
2668 
2669 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2670 	if (ret < 0)
2671 		return ret;
2672 	return 0;
2673 }
2674 
2675 /**
2676  * Get port id item from the item list.
2677  *
2678  * @param[in] item
2679  *   Pointer to the list of items.
2680  *
2681  * @return
2682  *   Pointer to the port id item if exist, else return NULL.
2683  */
2684 static const struct rte_flow_item *
2685 find_port_id_item(const struct rte_flow_item *item)
2686 {
2687 	assert(item);
2688 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2689 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2690 			return item;
2691 	}
2692 	return NULL;
2693 }
2694 
2695 /**
2696  * Get RSS action from the action list.
2697  *
2698  * @param[in] actions
2699  *   Pointer to the list of actions.
2700  *
2701  * @return
2702  *   Pointer to the RSS action if exist, else return NULL.
2703  */
2704 static const struct rte_flow_action_rss*
2705 flow_get_rss_action(const struct rte_flow_action actions[])
2706 {
2707 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2708 		switch (actions->type) {
2709 		case RTE_FLOW_ACTION_TYPE_RSS:
2710 			return (const struct rte_flow_action_rss *)
2711 			       actions->conf;
2712 		default:
2713 			break;
2714 		}
2715 	}
2716 	return NULL;
2717 }
2718 
2719 static unsigned int
2720 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2721 {
2722 	const struct rte_flow_item *item;
2723 	unsigned int has_vlan = 0;
2724 
2725 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2726 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2727 			has_vlan = 1;
2728 			break;
2729 		}
2730 	}
2731 	if (has_vlan)
2732 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2733 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2734 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2735 			       MLX5_EXPANSION_ROOT_OUTER;
2736 }
2737 
2738 /**
2739  * Get QUEUE/RSS action from the action list.
2740  *
2741  * @param[in] actions
2742  *   Pointer to the list of actions.
2743  * @param[out] qrss
2744  *   Pointer to the return pointer.
2745  * @param[out] qrss_type
2746  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2747  *   if no QUEUE/RSS is found.
2748  *
2749  * @return
2750  *   Total number of actions.
2751  */
2752 static int
2753 flow_parse_qrss_action(const struct rte_flow_action actions[],
2754 		       const struct rte_flow_action **qrss)
2755 {
2756 	int actions_n = 0;
2757 
2758 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2759 		switch (actions->type) {
2760 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2761 		case RTE_FLOW_ACTION_TYPE_RSS:
2762 			*qrss = actions;
2763 			break;
2764 		default:
2765 			break;
2766 		}
2767 		actions_n++;
2768 	}
2769 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2770 	return actions_n + 1;
2771 }
2772 
2773 /**
2774  * Check meter action from the action list.
2775  *
2776  * @param[in] actions
2777  *   Pointer to the list of actions.
2778  * @param[out] mtr
2779  *   Pointer to the meter exist flag.
2780  *
2781  * @return
2782  *   Total number of actions.
2783  */
2784 static int
2785 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2786 {
2787 	int actions_n = 0;
2788 
2789 	assert(mtr);
2790 	*mtr = 0;
2791 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2792 		switch (actions->type) {
2793 		case RTE_FLOW_ACTION_TYPE_METER:
2794 			*mtr = 1;
2795 			break;
2796 		default:
2797 			break;
2798 		}
2799 		actions_n++;
2800 	}
2801 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2802 	return actions_n + 1;
2803 }
2804 
2805 /**
2806  * Check if the flow should be splited due to hairpin.
2807  * The reason for the split is that in current HW we can't
2808  * support encap on Rx, so if a flow have encap we move it
2809  * to Tx.
2810  *
2811  * @param dev
2812  *   Pointer to Ethernet device.
2813  * @param[in] attr
2814  *   Flow rule attributes.
2815  * @param[in] actions
2816  *   Associated actions (list terminated by the END action).
2817  *
2818  * @return
2819  *   > 0 the number of actions and the flow should be split,
2820  *   0 when no split required.
2821  */
2822 static int
2823 flow_check_hairpin_split(struct rte_eth_dev *dev,
2824 			 const struct rte_flow_attr *attr,
2825 			 const struct rte_flow_action actions[])
2826 {
2827 	int queue_action = 0;
2828 	int action_n = 0;
2829 	int encap = 0;
2830 	const struct rte_flow_action_queue *queue;
2831 	const struct rte_flow_action_rss *rss;
2832 	const struct rte_flow_action_raw_encap *raw_encap;
2833 
2834 	if (!attr->ingress)
2835 		return 0;
2836 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2837 		switch (actions->type) {
2838 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2839 			queue = actions->conf;
2840 			if (queue == NULL)
2841 				return 0;
2842 			if (mlx5_rxq_get_type(dev, queue->index) !=
2843 			    MLX5_RXQ_TYPE_HAIRPIN)
2844 				return 0;
2845 			queue_action = 1;
2846 			action_n++;
2847 			break;
2848 		case RTE_FLOW_ACTION_TYPE_RSS:
2849 			rss = actions->conf;
2850 			if (rss == NULL || rss->queue_num == 0)
2851 				return 0;
2852 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2853 			    MLX5_RXQ_TYPE_HAIRPIN)
2854 				return 0;
2855 			queue_action = 1;
2856 			action_n++;
2857 			break;
2858 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2859 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2860 			encap = 1;
2861 			action_n++;
2862 			break;
2863 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2864 			raw_encap = actions->conf;
2865 			if (raw_encap->size >
2866 			    (sizeof(struct rte_flow_item_eth) +
2867 			     sizeof(struct rte_flow_item_ipv4)))
2868 				encap = 1;
2869 			action_n++;
2870 			break;
2871 		default:
2872 			action_n++;
2873 			break;
2874 		}
2875 	}
2876 	if (encap == 1 && queue_action)
2877 		return action_n;
2878 	return 0;
2879 }
2880 
2881 /* Declare flow create/destroy prototype in advance. */
2882 static struct rte_flow *
2883 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2884 		 const struct rte_flow_attr *attr,
2885 		 const struct rte_flow_item items[],
2886 		 const struct rte_flow_action actions[],
2887 		 bool external, struct rte_flow_error *error);
2888 
2889 static void
2890 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2891 		  struct rte_flow *flow);
2892 
2893 /**
2894  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2895  *
2896  * As mark_id is unique, if there's already a registered flow for the mark_id,
2897  * return by increasing the reference counter of the resource. Otherwise, create
2898  * the resource (mcp_res) and flow.
2899  *
2900  * Flow looks like,
2901  *   - If ingress port is ANY and reg_c[1] is mark_id,
2902  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2903  *
2904  * For default flow (zero mark_id), flow is like,
2905  *   - If ingress port is ANY,
2906  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2907  *
2908  * @param dev
2909  *   Pointer to Ethernet device.
2910  * @param mark_id
2911  *   ID of MARK action, zero means default flow for META.
2912  * @param[out] error
2913  *   Perform verbose error reporting if not NULL.
2914  *
2915  * @return
2916  *   Associated resource on success, NULL otherwise and rte_errno is set.
2917  */
2918 static struct mlx5_flow_mreg_copy_resource *
2919 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2920 			  struct rte_flow_error *error)
2921 {
2922 	struct mlx5_priv *priv = dev->data->dev_private;
2923 	struct rte_flow_attr attr = {
2924 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2925 		.ingress = 1,
2926 	};
2927 	struct mlx5_rte_flow_item_tag tag_spec = {
2928 		.data = mark_id,
2929 	};
2930 	struct rte_flow_item items[] = {
2931 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2932 	};
2933 	struct rte_flow_action_mark ftag = {
2934 		.id = mark_id,
2935 	};
2936 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2937 		.dst = REG_B,
2938 		.src = 0,
2939 	};
2940 	struct rte_flow_action_jump jump = {
2941 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2942 	};
2943 	struct rte_flow_action actions[] = {
2944 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2945 	};
2946 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2947 	int ret;
2948 
2949 	/* Fill the register fileds in the flow. */
2950 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2951 	if (ret < 0)
2952 		return NULL;
2953 	tag_spec.id = ret;
2954 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2955 	if (ret < 0)
2956 		return NULL;
2957 	cp_mreg.src = ret;
2958 	/* Check if already registered. */
2959 	assert(priv->mreg_cp_tbl);
2960 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2961 	if (mcp_res) {
2962 		/* For non-default rule. */
2963 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2964 			mcp_res->refcnt++;
2965 		assert(mark_id != MLX5_DEFAULT_COPY_ID || mcp_res->refcnt == 1);
2966 		return mcp_res;
2967 	}
2968 	/* Provide the full width of FLAG specific value. */
2969 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2970 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2971 	/* Build a new flow. */
2972 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2973 		items[0] = (struct rte_flow_item){
2974 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2975 			.spec = &tag_spec,
2976 		};
2977 		items[1] = (struct rte_flow_item){
2978 			.type = RTE_FLOW_ITEM_TYPE_END,
2979 		};
2980 		actions[0] = (struct rte_flow_action){
2981 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2982 			.conf = &ftag,
2983 		};
2984 		actions[1] = (struct rte_flow_action){
2985 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2986 			.conf = &cp_mreg,
2987 		};
2988 		actions[2] = (struct rte_flow_action){
2989 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2990 			.conf = &jump,
2991 		};
2992 		actions[3] = (struct rte_flow_action){
2993 			.type = RTE_FLOW_ACTION_TYPE_END,
2994 		};
2995 	} else {
2996 		/* Default rule, wildcard match. */
2997 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2998 		items[0] = (struct rte_flow_item){
2999 			.type = RTE_FLOW_ITEM_TYPE_END,
3000 		};
3001 		actions[0] = (struct rte_flow_action){
3002 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3003 			.conf = &cp_mreg,
3004 		};
3005 		actions[1] = (struct rte_flow_action){
3006 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3007 			.conf = &jump,
3008 		};
3009 		actions[2] = (struct rte_flow_action){
3010 			.type = RTE_FLOW_ACTION_TYPE_END,
3011 		};
3012 	}
3013 	/* Build a new entry. */
3014 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3015 	if (!mcp_res) {
3016 		rte_errno = ENOMEM;
3017 		return NULL;
3018 	}
3019 	/*
3020 	 * The copy Flows are not included in any list. There
3021 	 * ones are referenced from other Flows and can not
3022 	 * be applied, removed, deleted in ardbitrary order
3023 	 * by list traversing.
3024 	 */
3025 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3026 					 actions, false, error);
3027 	if (!mcp_res->flow)
3028 		goto error;
3029 	mcp_res->refcnt++;
3030 	mcp_res->hlist_ent.key = mark_id;
3031 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3032 				&mcp_res->hlist_ent);
3033 	assert(!ret);
3034 	if (ret)
3035 		goto error;
3036 	return mcp_res;
3037 error:
3038 	if (mcp_res->flow)
3039 		flow_list_destroy(dev, NULL, mcp_res->flow);
3040 	rte_free(mcp_res);
3041 	return NULL;
3042 }
3043 
3044 /**
3045  * Release flow in RX_CP_TBL.
3046  *
3047  * @param dev
3048  *   Pointer to Ethernet device.
3049  * @flow
3050  *   Parent flow for wich copying is provided.
3051  */
3052 static void
3053 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3054 			  struct rte_flow *flow)
3055 {
3056 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3057 	struct mlx5_priv *priv = dev->data->dev_private;
3058 
3059 	if (!mcp_res || !priv->mreg_cp_tbl)
3060 		return;
3061 	if (flow->copy_applied) {
3062 		assert(mcp_res->appcnt);
3063 		flow->copy_applied = 0;
3064 		--mcp_res->appcnt;
3065 		if (!mcp_res->appcnt)
3066 			flow_drv_remove(dev, mcp_res->flow);
3067 	}
3068 	/*
3069 	 * We do not check availability of metadata registers here,
3070 	 * because copy resources are not allocated in this case.
3071 	 */
3072 	if (--mcp_res->refcnt)
3073 		return;
3074 	assert(mcp_res->flow);
3075 	flow_list_destroy(dev, NULL, mcp_res->flow);
3076 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3077 	rte_free(mcp_res);
3078 	flow->mreg_copy = NULL;
3079 }
3080 
3081 /**
3082  * Start flow in RX_CP_TBL.
3083  *
3084  * @param dev
3085  *   Pointer to Ethernet device.
3086  * @flow
3087  *   Parent flow for wich copying is provided.
3088  *
3089  * @return
3090  *   0 on success, a negative errno value otherwise and rte_errno is set.
3091  */
3092 static int
3093 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3094 			    struct rte_flow *flow)
3095 {
3096 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3097 	int ret;
3098 
3099 	if (!mcp_res || flow->copy_applied)
3100 		return 0;
3101 	if (!mcp_res->appcnt) {
3102 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3103 		if (ret)
3104 			return ret;
3105 	}
3106 	++mcp_res->appcnt;
3107 	flow->copy_applied = 1;
3108 	return 0;
3109 }
3110 
3111 /**
3112  * Stop flow in RX_CP_TBL.
3113  *
3114  * @param dev
3115  *   Pointer to Ethernet device.
3116  * @flow
3117  *   Parent flow for wich copying is provided.
3118  */
3119 static void
3120 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3121 			   struct rte_flow *flow)
3122 {
3123 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3124 
3125 	if (!mcp_res || !flow->copy_applied)
3126 		return;
3127 	assert(mcp_res->appcnt);
3128 	--mcp_res->appcnt;
3129 	flow->copy_applied = 0;
3130 	if (!mcp_res->appcnt)
3131 		flow_drv_remove(dev, mcp_res->flow);
3132 }
3133 
3134 /**
3135  * Remove the default copy action from RX_CP_TBL.
3136  *
3137  * @param dev
3138  *   Pointer to Ethernet device.
3139  */
3140 static void
3141 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3142 {
3143 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3144 	struct mlx5_priv *priv = dev->data->dev_private;
3145 
3146 	/* Check if default flow is registered. */
3147 	if (!priv->mreg_cp_tbl)
3148 		return;
3149 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3150 					    MLX5_DEFAULT_COPY_ID);
3151 	if (!mcp_res)
3152 		return;
3153 	assert(mcp_res->flow);
3154 	flow_list_destroy(dev, NULL, mcp_res->flow);
3155 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3156 	rte_free(mcp_res);
3157 }
3158 
3159 /**
3160  * Add the default copy action in in RX_CP_TBL.
3161  *
3162  * @param dev
3163  *   Pointer to Ethernet device.
3164  * @param[out] error
3165  *   Perform verbose error reporting if not NULL.
3166  *
3167  * @return
3168  *   0 for success, negative value otherwise and rte_errno is set.
3169  */
3170 static int
3171 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3172 				  struct rte_flow_error *error)
3173 {
3174 	struct mlx5_priv *priv = dev->data->dev_private;
3175 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3176 
3177 	/* Check whether extensive metadata feature is engaged. */
3178 	if (!priv->config.dv_flow_en ||
3179 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3180 	    !mlx5_flow_ext_mreg_supported(dev) ||
3181 	    !priv->sh->dv_regc0_mask)
3182 		return 0;
3183 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3184 	if (!mcp_res)
3185 		return -rte_errno;
3186 	return 0;
3187 }
3188 
3189 /**
3190  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3191  *
3192  * All the flow having Q/RSS action should be split by
3193  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3194  * performs the following,
3195  *   - CQE->flow_tag := reg_c[1] (MARK)
3196  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3197  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3198  * but there should be a flow per each MARK ID set by MARK action.
3199  *
3200  * For the aforementioned reason, if there's a MARK action in flow's action
3201  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3202  * the MARK ID to CQE's flow_tag like,
3203  *   - If reg_c[1] is mark_id,
3204  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3205  *
3206  * For SET_META action which stores value in reg_c[0], as the destination is
3207  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3208  * MARK ID means the default flow. The default flow looks like,
3209  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3210  *
3211  * @param dev
3212  *   Pointer to Ethernet device.
3213  * @param flow
3214  *   Pointer to flow structure.
3215  * @param[in] actions
3216  *   Pointer to the list of actions.
3217  * @param[out] error
3218  *   Perform verbose error reporting if not NULL.
3219  *
3220  * @return
3221  *   0 on success, negative value otherwise and rte_errno is set.
3222  */
3223 static int
3224 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3225 			    struct rte_flow *flow,
3226 			    const struct rte_flow_action *actions,
3227 			    struct rte_flow_error *error)
3228 {
3229 	struct mlx5_priv *priv = dev->data->dev_private;
3230 	struct mlx5_dev_config *config = &priv->config;
3231 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3232 	const struct rte_flow_action_mark *mark;
3233 
3234 	/* Check whether extensive metadata feature is engaged. */
3235 	if (!config->dv_flow_en ||
3236 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3237 	    !mlx5_flow_ext_mreg_supported(dev) ||
3238 	    !priv->sh->dv_regc0_mask)
3239 		return 0;
3240 	/* Find MARK action. */
3241 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3242 		switch (actions->type) {
3243 		case RTE_FLOW_ACTION_TYPE_FLAG:
3244 			mcp_res = flow_mreg_add_copy_action
3245 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3246 			if (!mcp_res)
3247 				return -rte_errno;
3248 			flow->mreg_copy = mcp_res;
3249 			if (dev->data->dev_started) {
3250 				mcp_res->appcnt++;
3251 				flow->copy_applied = 1;
3252 			}
3253 			return 0;
3254 		case RTE_FLOW_ACTION_TYPE_MARK:
3255 			mark = (const struct rte_flow_action_mark *)
3256 				actions->conf;
3257 			mcp_res =
3258 				flow_mreg_add_copy_action(dev, mark->id, error);
3259 			if (!mcp_res)
3260 				return -rte_errno;
3261 			flow->mreg_copy = mcp_res;
3262 			if (dev->data->dev_started) {
3263 				mcp_res->appcnt++;
3264 				flow->copy_applied = 1;
3265 			}
3266 			return 0;
3267 		default:
3268 			break;
3269 		}
3270 	}
3271 	return 0;
3272 }
3273 
3274 #define MLX5_MAX_SPLIT_ACTIONS 24
3275 #define MLX5_MAX_SPLIT_ITEMS 24
3276 
3277 /**
3278  * Split the hairpin flow.
3279  * Since HW can't support encap on Rx we move the encap to Tx.
3280  * If the count action is after the encap then we also
3281  * move the count action. in this case the count will also measure
3282  * the outer bytes.
3283  *
3284  * @param dev
3285  *   Pointer to Ethernet device.
3286  * @param[in] actions
3287  *   Associated actions (list terminated by the END action).
3288  * @param[out] actions_rx
3289  *   Rx flow actions.
3290  * @param[out] actions_tx
3291  *   Tx flow actions..
3292  * @param[out] pattern_tx
3293  *   The pattern items for the Tx flow.
3294  * @param[out] flow_id
3295  *   The flow ID connected to this flow.
3296  *
3297  * @return
3298  *   0 on success.
3299  */
3300 static int
3301 flow_hairpin_split(struct rte_eth_dev *dev,
3302 		   const struct rte_flow_action actions[],
3303 		   struct rte_flow_action actions_rx[],
3304 		   struct rte_flow_action actions_tx[],
3305 		   struct rte_flow_item pattern_tx[],
3306 		   uint32_t *flow_id)
3307 {
3308 	struct mlx5_priv *priv = dev->data->dev_private;
3309 	const struct rte_flow_action_raw_encap *raw_encap;
3310 	const struct rte_flow_action_raw_decap *raw_decap;
3311 	struct mlx5_rte_flow_action_set_tag *set_tag;
3312 	struct rte_flow_action *tag_action;
3313 	struct mlx5_rte_flow_item_tag *tag_item;
3314 	struct rte_flow_item *item;
3315 	char *addr;
3316 	int encap = 0;
3317 
3318 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3319 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3320 		switch (actions->type) {
3321 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3322 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3323 			rte_memcpy(actions_tx, actions,
3324 			       sizeof(struct rte_flow_action));
3325 			actions_tx++;
3326 			break;
3327 		case RTE_FLOW_ACTION_TYPE_COUNT:
3328 			if (encap) {
3329 				rte_memcpy(actions_tx, actions,
3330 					   sizeof(struct rte_flow_action));
3331 				actions_tx++;
3332 			} else {
3333 				rte_memcpy(actions_rx, actions,
3334 					   sizeof(struct rte_flow_action));
3335 				actions_rx++;
3336 			}
3337 			break;
3338 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3339 			raw_encap = actions->conf;
3340 			if (raw_encap->size >
3341 			    (sizeof(struct rte_flow_item_eth) +
3342 			     sizeof(struct rte_flow_item_ipv4))) {
3343 				memcpy(actions_tx, actions,
3344 				       sizeof(struct rte_flow_action));
3345 				actions_tx++;
3346 				encap = 1;
3347 			} else {
3348 				rte_memcpy(actions_rx, actions,
3349 					   sizeof(struct rte_flow_action));
3350 				actions_rx++;
3351 			}
3352 			break;
3353 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3354 			raw_decap = actions->conf;
3355 			if (raw_decap->size <
3356 			    (sizeof(struct rte_flow_item_eth) +
3357 			     sizeof(struct rte_flow_item_ipv4))) {
3358 				memcpy(actions_tx, actions,
3359 				       sizeof(struct rte_flow_action));
3360 				actions_tx++;
3361 			} else {
3362 				rte_memcpy(actions_rx, actions,
3363 					   sizeof(struct rte_flow_action));
3364 				actions_rx++;
3365 			}
3366 			break;
3367 		default:
3368 			rte_memcpy(actions_rx, actions,
3369 				   sizeof(struct rte_flow_action));
3370 			actions_rx++;
3371 			break;
3372 		}
3373 	}
3374 	/* Add set meta action and end action for the Rx flow. */
3375 	tag_action = actions_rx;
3376 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3377 	actions_rx++;
3378 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3379 	actions_rx++;
3380 	set_tag = (void *)actions_rx;
3381 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3382 	assert(set_tag->id > REG_NONE);
3383 	set_tag->data = *flow_id;
3384 	tag_action->conf = set_tag;
3385 	/* Create Tx item list. */
3386 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3387 	addr = (void *)&pattern_tx[2];
3388 	item = pattern_tx;
3389 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3390 	tag_item = (void *)addr;
3391 	tag_item->data = *flow_id;
3392 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3393 	assert(set_tag->id > REG_NONE);
3394 	item->spec = tag_item;
3395 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3396 	tag_item = (void *)addr;
3397 	tag_item->data = UINT32_MAX;
3398 	tag_item->id = UINT16_MAX;
3399 	item->mask = tag_item;
3400 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3401 	item->last = NULL;
3402 	item++;
3403 	item->type = RTE_FLOW_ITEM_TYPE_END;
3404 	return 0;
3405 }
3406 
3407 /**
3408  * The last stage of splitting chain, just creates the subflow
3409  * without any modification.
3410  *
3411  * @param dev
3412  *   Pointer to Ethernet device.
3413  * @param[in] flow
3414  *   Parent flow structure pointer.
3415  * @param[in, out] sub_flow
3416  *   Pointer to return the created subflow, may be NULL.
3417  * @param[in] attr
3418  *   Flow rule attributes.
3419  * @param[in] items
3420  *   Pattern specification (list terminated by the END pattern item).
3421  * @param[in] actions
3422  *   Associated actions (list terminated by the END action).
3423  * @param[in] external
3424  *   This flow rule is created by request external to PMD.
3425  * @param[out] error
3426  *   Perform verbose error reporting if not NULL.
3427  * @return
3428  *   0 on success, negative value otherwise
3429  */
3430 static int
3431 flow_create_split_inner(struct rte_eth_dev *dev,
3432 			struct rte_flow *flow,
3433 			struct mlx5_flow **sub_flow,
3434 			const struct rte_flow_attr *attr,
3435 			const struct rte_flow_item items[],
3436 			const struct rte_flow_action actions[],
3437 			bool external, struct rte_flow_error *error)
3438 {
3439 	struct mlx5_flow *dev_flow;
3440 
3441 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3442 	if (!dev_flow)
3443 		return -rte_errno;
3444 	dev_flow->flow = flow;
3445 	dev_flow->external = external;
3446 	/* Subflow object was created, we must include one in the list. */
3447 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3448 	if (sub_flow)
3449 		*sub_flow = dev_flow;
3450 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3451 }
3452 
3453 /**
3454  * Split the meter flow.
3455  *
3456  * As meter flow will split to three sub flow, other than meter
3457  * action, the other actions make sense to only meter accepts
3458  * the packet. If it need to be dropped, no other additional
3459  * actions should be take.
3460  *
3461  * One kind of special action which decapsulates the L3 tunnel
3462  * header will be in the prefix sub flow, as not to take the
3463  * L3 tunnel header into account.
3464  *
3465  * @param dev
3466  *   Pointer to Ethernet device.
3467  * @param[in] actions
3468  *   Associated actions (list terminated by the END action).
3469  * @param[out] actions_sfx
3470  *   Suffix flow actions.
3471  * @param[out] actions_pre
3472  *   Prefix flow actions.
3473  * @param[out] pattern_sfx
3474  *   The pattern items for the suffix flow.
3475  * @param[out] tag_sfx
3476  *   Pointer to suffix flow tag.
3477  *
3478  * @return
3479  *   0 on success.
3480  */
3481 static int
3482 flow_meter_split_prep(struct rte_eth_dev *dev,
3483 		 const struct rte_flow_action actions[],
3484 		 struct rte_flow_action actions_sfx[],
3485 		 struct rte_flow_action actions_pre[])
3486 {
3487 	struct rte_flow_action *tag_action;
3488 	struct mlx5_rte_flow_action_set_tag *set_tag;
3489 	struct rte_flow_error error;
3490 	const struct rte_flow_action_raw_encap *raw_encap;
3491 	const struct rte_flow_action_raw_decap *raw_decap;
3492 	uint32_t tag_id;
3493 
3494 	/* Add the extra tag action first. */
3495 	tag_action = actions_pre;
3496 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3497 	actions_pre++;
3498 	/* Prepare the actions for prefix and suffix flow. */
3499 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3500 		switch (actions->type) {
3501 		case RTE_FLOW_ACTION_TYPE_METER:
3502 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3503 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3504 			memcpy(actions_pre, actions,
3505 			       sizeof(struct rte_flow_action));
3506 			actions_pre++;
3507 			break;
3508 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3509 			raw_encap = actions->conf;
3510 			if (raw_encap->size >
3511 			    (sizeof(struct rte_flow_item_eth) +
3512 			     sizeof(struct rte_flow_item_ipv4))) {
3513 				memcpy(actions_sfx, actions,
3514 				       sizeof(struct rte_flow_action));
3515 				actions_sfx++;
3516 			} else {
3517 				rte_memcpy(actions_pre, actions,
3518 					   sizeof(struct rte_flow_action));
3519 				actions_pre++;
3520 			}
3521 			break;
3522 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3523 			raw_decap = actions->conf;
3524 			/* Size 0 decap means 50 bytes as vxlan decap. */
3525 			if (raw_decap->size && (raw_decap->size <
3526 			    (sizeof(struct rte_flow_item_eth) +
3527 			     sizeof(struct rte_flow_item_ipv4)))) {
3528 				memcpy(actions_sfx, actions,
3529 				       sizeof(struct rte_flow_action));
3530 				actions_sfx++;
3531 			} else {
3532 				rte_memcpy(actions_pre, actions,
3533 					   sizeof(struct rte_flow_action));
3534 				actions_pre++;
3535 			}
3536 			break;
3537 		default:
3538 			memcpy(actions_sfx, actions,
3539 				sizeof(struct rte_flow_action));
3540 			actions_sfx++;
3541 			break;
3542 		}
3543 	}
3544 	/* Add end action to the actions. */
3545 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3546 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3547 	actions_pre++;
3548 	/* Set the tag. */
3549 	set_tag = (void *)actions_pre;
3550 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3551 	/*
3552 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3553 	 */
3554 	tag_id = flow_qrss_get_id(dev);
3555 	set_tag->data = rte_cpu_to_be_32(tag_id);
3556 	tag_action->conf = set_tag;
3557 	return tag_id;
3558 }
3559 
3560 /**
3561  * Split action list having QUEUE/RSS for metadata register copy.
3562  *
3563  * Once Q/RSS action is detected in user's action list, the flow action
3564  * should be split in order to copy metadata registers, which will happen in
3565  * RX_CP_TBL like,
3566  *   - CQE->flow_tag := reg_c[1] (MARK)
3567  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3568  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3569  * This is because the last action of each flow must be a terminal action
3570  * (QUEUE, RSS or DROP).
3571  *
3572  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3573  * stored and kept in the mlx5_flow structure per each sub_flow.
3574  *
3575  * The Q/RSS action is replaced with,
3576  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3577  * And the following JUMP action is added at the end,
3578  *   - JUMP, to RX_CP_TBL.
3579  *
3580  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3581  * flow_create_split_metadata() routine. The flow will look like,
3582  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3583  *
3584  * @param dev
3585  *   Pointer to Ethernet device.
3586  * @param[out] split_actions
3587  *   Pointer to store split actions to jump to CP_TBL.
3588  * @param[in] actions
3589  *   Pointer to the list of original flow actions.
3590  * @param[in] qrss
3591  *   Pointer to the Q/RSS action.
3592  * @param[in] actions_n
3593  *   Number of original actions.
3594  * @param[out] error
3595  *   Perform verbose error reporting if not NULL.
3596  *
3597  * @return
3598  *   non-zero unique flow_id on success, otherwise 0 and
3599  *   error/rte_error are set.
3600  */
3601 static uint32_t
3602 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3603 			  struct rte_flow_action *split_actions,
3604 			  const struct rte_flow_action *actions,
3605 			  const struct rte_flow_action *qrss,
3606 			  int actions_n, struct rte_flow_error *error)
3607 {
3608 	struct mlx5_rte_flow_action_set_tag *set_tag;
3609 	struct rte_flow_action_jump *jump;
3610 	const int qrss_idx = qrss - actions;
3611 	uint32_t flow_id = 0;
3612 	int ret = 0;
3613 
3614 	/*
3615 	 * Given actions will be split
3616 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3617 	 * - Add jump to mreg CP_TBL.
3618 	 * As a result, there will be one more action.
3619 	 */
3620 	++actions_n;
3621 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3622 	set_tag = (void *)(split_actions + actions_n);
3623 	/*
3624 	 * If tag action is not set to void(it means we are not the meter
3625 	 * suffix flow), add the tag action. Since meter suffix flow already
3626 	 * has the tag added.
3627 	 */
3628 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3629 		/*
3630 		 * Allocate the new subflow ID. This one is unique within
3631 		 * device and not shared with representors. Otherwise,
3632 		 * we would have to resolve multi-thread access synch
3633 		 * issue. Each flow on the shared device is appended
3634 		 * with source vport identifier, so the resulting
3635 		 * flows will be unique in the shared (by master and
3636 		 * representors) domain even if they have coinciding
3637 		 * IDs.
3638 		 */
3639 		flow_id = flow_qrss_get_id(dev);
3640 		if (!flow_id)
3641 			return rte_flow_error_set(error, ENOMEM,
3642 						  RTE_FLOW_ERROR_TYPE_ACTION,
3643 						  NULL, "can't allocate id "
3644 						  "for split Q/RSS subflow");
3645 		/* Internal SET_TAG action to set flow ID. */
3646 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3647 			.data = flow_id,
3648 		};
3649 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3650 		if (ret < 0)
3651 			return ret;
3652 		set_tag->id = ret;
3653 		/* Construct new actions array. */
3654 		/* Replace QUEUE/RSS action. */
3655 		split_actions[qrss_idx] = (struct rte_flow_action){
3656 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3657 			.conf = set_tag,
3658 		};
3659 	}
3660 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3661 	jump = (void *)(set_tag + 1);
3662 	*jump = (struct rte_flow_action_jump){
3663 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3664 	};
3665 	split_actions[actions_n - 2] = (struct rte_flow_action){
3666 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3667 		.conf = jump,
3668 	};
3669 	split_actions[actions_n - 1] = (struct rte_flow_action){
3670 		.type = RTE_FLOW_ACTION_TYPE_END,
3671 	};
3672 	return flow_id;
3673 }
3674 
3675 /**
3676  * Extend the given action list for Tx metadata copy.
3677  *
3678  * Copy the given action list to the ext_actions and add flow metadata register
3679  * copy action in order to copy reg_a set by WQE to reg_c[0].
3680  *
3681  * @param[out] ext_actions
3682  *   Pointer to the extended action list.
3683  * @param[in] actions
3684  *   Pointer to the list of actions.
3685  * @param[in] actions_n
3686  *   Number of actions in the list.
3687  * @param[out] error
3688  *   Perform verbose error reporting if not NULL.
3689  *
3690  * @return
3691  *   0 on success, negative value otherwise
3692  */
3693 static int
3694 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3695 		       struct rte_flow_action *ext_actions,
3696 		       const struct rte_flow_action *actions,
3697 		       int actions_n, struct rte_flow_error *error)
3698 {
3699 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3700 		(struct mlx5_flow_action_copy_mreg *)
3701 			(ext_actions + actions_n + 1);
3702 	int ret;
3703 
3704 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3705 	if (ret < 0)
3706 		return ret;
3707 	cp_mreg->dst = ret;
3708 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3709 	if (ret < 0)
3710 		return ret;
3711 	cp_mreg->src = ret;
3712 	memcpy(ext_actions, actions,
3713 			sizeof(*ext_actions) * actions_n);
3714 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3715 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3716 		.conf = cp_mreg,
3717 	};
3718 	ext_actions[actions_n] = (struct rte_flow_action){
3719 		.type = RTE_FLOW_ACTION_TYPE_END,
3720 	};
3721 	return 0;
3722 }
3723 
3724 /**
3725  * The splitting for metadata feature.
3726  *
3727  * - Q/RSS action on NIC Rx should be split in order to pass by
3728  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3729  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3730  *
3731  * - All the actions on NIC Tx should have a mreg copy action to
3732  *   copy reg_a from WQE to reg_c[0].
3733  *
3734  * @param dev
3735  *   Pointer to Ethernet device.
3736  * @param[in] flow
3737  *   Parent flow structure pointer.
3738  * @param[in] attr
3739  *   Flow rule attributes.
3740  * @param[in] items
3741  *   Pattern specification (list terminated by the END pattern item).
3742  * @param[in] actions
3743  *   Associated actions (list terminated by the END action).
3744  * @param[in] external
3745  *   This flow rule is created by request external to PMD.
3746  * @param[out] error
3747  *   Perform verbose error reporting if not NULL.
3748  * @return
3749  *   0 on success, negative value otherwise
3750  */
3751 static int
3752 flow_create_split_metadata(struct rte_eth_dev *dev,
3753 			   struct rte_flow *flow,
3754 			   const struct rte_flow_attr *attr,
3755 			   const struct rte_flow_item items[],
3756 			   const struct rte_flow_action actions[],
3757 			   bool external, struct rte_flow_error *error)
3758 {
3759 	struct mlx5_priv *priv = dev->data->dev_private;
3760 	struct mlx5_dev_config *config = &priv->config;
3761 	const struct rte_flow_action *qrss = NULL;
3762 	struct rte_flow_action *ext_actions = NULL;
3763 	struct mlx5_flow *dev_flow = NULL;
3764 	uint32_t qrss_id = 0;
3765 	int mtr_sfx = 0;
3766 	size_t act_size;
3767 	int actions_n;
3768 	int ret;
3769 
3770 	/* Check whether extensive metadata feature is engaged. */
3771 	if (!config->dv_flow_en ||
3772 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3773 	    !mlx5_flow_ext_mreg_supported(dev))
3774 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3775 					       actions, external, error);
3776 	actions_n = flow_parse_qrss_action(actions, &qrss);
3777 	if (qrss) {
3778 		/* Exclude hairpin flows from splitting. */
3779 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3780 			const struct rte_flow_action_queue *queue;
3781 
3782 			queue = qrss->conf;
3783 			if (mlx5_rxq_get_type(dev, queue->index) ==
3784 			    MLX5_RXQ_TYPE_HAIRPIN)
3785 				qrss = NULL;
3786 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3787 			const struct rte_flow_action_rss *rss;
3788 
3789 			rss = qrss->conf;
3790 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3791 			    MLX5_RXQ_TYPE_HAIRPIN)
3792 				qrss = NULL;
3793 		}
3794 	}
3795 	if (qrss) {
3796 		/* Check if it is in meter suffix table. */
3797 		mtr_sfx = attr->group == (attr->transfer ?
3798 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3799 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3800 		/*
3801 		 * Q/RSS action on NIC Rx should be split in order to pass by
3802 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3803 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3804 		 */
3805 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3806 			   sizeof(struct rte_flow_action_set_tag) +
3807 			   sizeof(struct rte_flow_action_jump);
3808 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3809 		if (!ext_actions)
3810 			return rte_flow_error_set(error, ENOMEM,
3811 						  RTE_FLOW_ERROR_TYPE_ACTION,
3812 						  NULL, "no memory to split "
3813 						  "metadata flow");
3814 		/*
3815 		 * If we are the suffix flow of meter, tag already exist.
3816 		 * Set the tag action to void.
3817 		 */
3818 		if (mtr_sfx)
3819 			ext_actions[qrss - actions].type =
3820 						RTE_FLOW_ACTION_TYPE_VOID;
3821 		else
3822 			ext_actions[qrss - actions].type =
3823 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3824 		/*
3825 		 * Create the new actions list with removed Q/RSS action
3826 		 * and appended set tag and jump to register copy table
3827 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3828 		 * in advance, because it is needed for set tag action.
3829 		 */
3830 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3831 						    qrss, actions_n, error);
3832 		if (!mtr_sfx && !qrss_id) {
3833 			ret = -rte_errno;
3834 			goto exit;
3835 		}
3836 	} else if (attr->egress && !attr->transfer) {
3837 		/*
3838 		 * All the actions on NIC Tx should have a metadata register
3839 		 * copy action to copy reg_a from WQE to reg_c[meta]
3840 		 */
3841 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3842 			   sizeof(struct mlx5_flow_action_copy_mreg);
3843 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3844 		if (!ext_actions)
3845 			return rte_flow_error_set(error, ENOMEM,
3846 						  RTE_FLOW_ERROR_TYPE_ACTION,
3847 						  NULL, "no memory to split "
3848 						  "metadata flow");
3849 		/* Create the action list appended with copy register. */
3850 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3851 					     actions_n, error);
3852 		if (ret < 0)
3853 			goto exit;
3854 	}
3855 	/* Add the unmodified original or prefix subflow. */
3856 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3857 				      ext_actions ? ext_actions : actions,
3858 				      external, error);
3859 	if (ret < 0)
3860 		goto exit;
3861 	assert(dev_flow);
3862 	if (qrss) {
3863 		const struct rte_flow_attr q_attr = {
3864 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3865 			.ingress = 1,
3866 		};
3867 		/* Internal PMD action to set register. */
3868 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3869 			.data = qrss_id,
3870 			.id = 0,
3871 		};
3872 		struct rte_flow_item q_items[] = {
3873 			{
3874 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3875 				.spec = &q_tag_spec,
3876 				.last = NULL,
3877 				.mask = NULL,
3878 			},
3879 			{
3880 				.type = RTE_FLOW_ITEM_TYPE_END,
3881 			},
3882 		};
3883 		struct rte_flow_action q_actions[] = {
3884 			{
3885 				.type = qrss->type,
3886 				.conf = qrss->conf,
3887 			},
3888 			{
3889 				.type = RTE_FLOW_ACTION_TYPE_END,
3890 			},
3891 		};
3892 		uint64_t hash_fields = dev_flow->hash_fields;
3893 
3894 		/*
3895 		 * Configure the tag item only if there is no meter subflow.
3896 		 * Since tag is already marked in the meter suffix subflow
3897 		 * we can just use the meter suffix items as is.
3898 		 */
3899 		if (qrss_id) {
3900 			/* Not meter subflow. */
3901 			assert(!mtr_sfx);
3902 			/*
3903 			 * Put unique id in prefix flow due to it is destroyed
3904 			 * after suffix flow and id will be freed after there
3905 			 * is no actual flows with this id and identifier
3906 			 * reallocation becomes possible (for example, for
3907 			 * other flows in other threads).
3908 			 */
3909 			dev_flow->qrss_id = qrss_id;
3910 			qrss_id = 0;
3911 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3912 						   error);
3913 			if (ret < 0)
3914 				goto exit;
3915 			q_tag_spec.id = ret;
3916 		}
3917 		dev_flow = NULL;
3918 		/* Add suffix subflow to execute Q/RSS. */
3919 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3920 					      &q_attr, mtr_sfx ? items :
3921 					      q_items, q_actions,
3922 					      external, error);
3923 		if (ret < 0)
3924 			goto exit;
3925 		assert(dev_flow);
3926 		dev_flow->hash_fields = hash_fields;
3927 	}
3928 
3929 exit:
3930 	/*
3931 	 * We do not destroy the partially created sub_flows in case of error.
3932 	 * These ones are included into parent flow list and will be destroyed
3933 	 * by flow_drv_destroy.
3934 	 */
3935 	flow_qrss_free_id(dev, qrss_id);
3936 	rte_free(ext_actions);
3937 	return ret;
3938 }
3939 
3940 /**
3941  * The splitting for meter feature.
3942  *
3943  * - The meter flow will be split to two flows as prefix and
3944  *   suffix flow. The packets make sense only it pass the prefix
3945  *   meter action.
3946  *
3947  * - Reg_C_5 is used for the packet to match betweend prefix and
3948  *   suffix flow.
3949  *
3950  * @param dev
3951  *   Pointer to Ethernet device.
3952  * @param[in] flow
3953  *   Parent flow structure pointer.
3954  * @param[in] attr
3955  *   Flow rule attributes.
3956  * @param[in] items
3957  *   Pattern specification (list terminated by the END pattern item).
3958  * @param[in] actions
3959  *   Associated actions (list terminated by the END action).
3960  * @param[in] external
3961  *   This flow rule is created by request external to PMD.
3962  * @param[out] error
3963  *   Perform verbose error reporting if not NULL.
3964  * @return
3965  *   0 on success, negative value otherwise
3966  */
3967 static int
3968 flow_create_split_meter(struct rte_eth_dev *dev,
3969 			   struct rte_flow *flow,
3970 			   const struct rte_flow_attr *attr,
3971 			   const struct rte_flow_item items[],
3972 			   const struct rte_flow_action actions[],
3973 			   bool external, struct rte_flow_error *error)
3974 {
3975 	struct mlx5_priv *priv = dev->data->dev_private;
3976 	struct rte_flow_action *sfx_actions = NULL;
3977 	struct rte_flow_action *pre_actions = NULL;
3978 	struct rte_flow_item *sfx_items = NULL;
3979 	const  struct rte_flow_item *sfx_port_id_item;
3980 	struct mlx5_flow *dev_flow = NULL;
3981 	struct rte_flow_attr sfx_attr = *attr;
3982 	uint32_t mtr = 0;
3983 	uint32_t mtr_tag_id = 0;
3984 	size_t act_size;
3985 	size_t item_size;
3986 	int actions_n = 0;
3987 	int ret;
3988 
3989 	if (priv->mtr_en)
3990 		actions_n = flow_check_meter_action(actions, &mtr);
3991 	if (mtr) {
3992 		struct mlx5_rte_flow_item_tag *tag_spec;
3993 		/* The five prefix actions: meter, decap, encap, tag, end. */
3994 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3995 			   sizeof(struct rte_flow_action_set_tag);
3996 		/* tag, end. */
3997 #define METER_SUFFIX_ITEM 3
3998 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3999 			    sizeof(struct mlx5_rte_flow_item_tag);
4000 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4001 		if (!sfx_actions)
4002 			return rte_flow_error_set(error, ENOMEM,
4003 						  RTE_FLOW_ERROR_TYPE_ACTION,
4004 						  NULL, "no memory to split "
4005 						  "meter flow");
4006 		pre_actions = sfx_actions + actions_n;
4007 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4008 						     pre_actions);
4009 		if (!mtr_tag_id) {
4010 			ret = -rte_errno;
4011 			goto exit;
4012 		}
4013 		/* Add the prefix subflow. */
4014 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4015 						  pre_actions, external, error);
4016 		if (ret) {
4017 			ret = -rte_errno;
4018 			goto exit;
4019 		}
4020 		dev_flow->mtr_flow_id = mtr_tag_id;
4021 		/* Prepare the suffix flow match pattern. */
4022 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4023 			     act_size);
4024 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4025 			    METER_SUFFIX_ITEM);
4026 		tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
4027 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4028 						    error);
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 = NULL;
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 -= METER_SUFFIX_ITEM;
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 	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 	int ret;
4163 	uint32_t i;
4164 	uint32_t flow_size;
4165 	int hairpin_flow = 0;
4166 	uint32_t hairpin_id = 0;
4167 	struct rte_flow_attr attr_tx = { .priority = 0 };
4168 
4169 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4170 	if (hairpin_flow > 0) {
4171 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4172 			rte_errno = EINVAL;
4173 			return NULL;
4174 		}
4175 		flow_hairpin_split(dev, actions, actions_rx.actions,
4176 				   actions_hairpin_tx.actions, items_tx.items,
4177 				   &hairpin_id);
4178 		p_actions_rx = actions_rx.actions;
4179 	}
4180 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4181 				error);
4182 	if (ret < 0)
4183 		goto error_before_flow;
4184 	flow_size = sizeof(struct rte_flow);
4185 	rss = flow_get_rss_action(p_actions_rx);
4186 	if (rss)
4187 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4188 					    sizeof(void *));
4189 	else
4190 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4191 	flow = rte_calloc(__func__, 1, flow_size, 0);
4192 	if (!flow) {
4193 		rte_errno = ENOMEM;
4194 		goto error_before_flow;
4195 	}
4196 	flow->drv_type = flow_get_drv_type(dev, attr);
4197 	if (hairpin_id != 0)
4198 		flow->hairpin_flow_id = hairpin_id;
4199 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4200 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
4201 	flow->rss.queue = (void *)(flow + 1);
4202 	if (rss) {
4203 		/*
4204 		 * The following information is required by
4205 		 * mlx5_flow_hashfields_adjust() in advance.
4206 		 */
4207 		flow->rss.level = rss->level;
4208 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4209 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4210 	}
4211 	LIST_INIT(&flow->dev_flows);
4212 	if (rss && rss->types) {
4213 		unsigned int graph_root;
4214 
4215 		graph_root = find_graph_root(items, rss->level);
4216 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4217 					  items, rss->types,
4218 					  mlx5_support_expansion,
4219 					  graph_root);
4220 		assert(ret > 0 &&
4221 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4222 	} else {
4223 		buf->entries = 1;
4224 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4225 	}
4226 	for (i = 0; i < buf->entries; ++i) {
4227 		/*
4228 		 * The splitter may create multiple dev_flows,
4229 		 * depending on configuration. In the simplest
4230 		 * case it just creates unmodified original flow.
4231 		 */
4232 		ret = flow_create_split_outer(dev, flow, attr,
4233 					      buf->entry[i].pattern,
4234 					      p_actions_rx, external,
4235 					      error);
4236 		if (ret < 0)
4237 			goto error;
4238 	}
4239 	/* Create the tx flow. */
4240 	if (hairpin_flow) {
4241 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4242 		attr_tx.ingress = 0;
4243 		attr_tx.egress = 1;
4244 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4245 					    actions_hairpin_tx.actions, error);
4246 		if (!dev_flow)
4247 			goto error;
4248 		dev_flow->flow = flow;
4249 		dev_flow->external = 0;
4250 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4251 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4252 					 items_tx.items,
4253 					 actions_hairpin_tx.actions, error);
4254 		if (ret < 0)
4255 			goto error;
4256 	}
4257 	/*
4258 	 * Update the metadata register copy table. If extensive
4259 	 * metadata feature is enabled and registers are supported
4260 	 * we might create the extra rte_flow for each unique
4261 	 * MARK/FLAG action ID.
4262 	 *
4263 	 * The table is updated for ingress Flows only, because
4264 	 * the egress Flows belong to the different device and
4265 	 * copy table should be updated in peer NIC Rx domain.
4266 	 */
4267 	if (attr->ingress &&
4268 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4269 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4270 		if (ret)
4271 			goto error;
4272 	}
4273 	if (dev->data->dev_started) {
4274 		ret = flow_drv_apply(dev, flow, error);
4275 		if (ret < 0)
4276 			goto error;
4277 	}
4278 	if (list)
4279 		TAILQ_INSERT_TAIL(list, flow, next);
4280 	flow_rxq_flags_set(dev, flow);
4281 	return flow;
4282 error_before_flow:
4283 	if (hairpin_id)
4284 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4285 				     hairpin_id);
4286 	return NULL;
4287 error:
4288 	assert(flow);
4289 	flow_mreg_del_copy_action(dev, flow);
4290 	ret = rte_errno; /* Save rte_errno before cleanup. */
4291 	if (flow->hairpin_flow_id)
4292 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4293 				     flow->hairpin_flow_id);
4294 	assert(flow);
4295 	flow_drv_destroy(dev, flow);
4296 	rte_free(flow);
4297 	rte_errno = ret; /* Restore rte_errno. */
4298 	return NULL;
4299 }
4300 
4301 /**
4302  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4303  * incoming packets to table 1.
4304  *
4305  * Other flow rules, requested for group n, will be created in
4306  * e-switch table n+1.
4307  * Jump action to e-switch group n will be created to group n+1.
4308  *
4309  * Used when working in switchdev mode, to utilise advantages of table 1
4310  * and above.
4311  *
4312  * @param dev
4313  *   Pointer to Ethernet device.
4314  *
4315  * @return
4316  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4317  */
4318 struct rte_flow *
4319 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4320 {
4321 	const struct rte_flow_attr attr = {
4322 		.group = 0,
4323 		.priority = 0,
4324 		.ingress = 1,
4325 		.egress = 0,
4326 		.transfer = 1,
4327 	};
4328 	const struct rte_flow_item pattern = {
4329 		.type = RTE_FLOW_ITEM_TYPE_END,
4330 	};
4331 	struct rte_flow_action_jump jump = {
4332 		.group = 1,
4333 	};
4334 	const struct rte_flow_action actions[] = {
4335 		{
4336 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4337 			.conf = &jump,
4338 		},
4339 		{
4340 			.type = RTE_FLOW_ACTION_TYPE_END,
4341 		},
4342 	};
4343 	struct mlx5_priv *priv = dev->data->dev_private;
4344 	struct rte_flow_error error;
4345 
4346 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4347 				actions, false, &error);
4348 }
4349 
4350 /**
4351  * Create a flow.
4352  *
4353  * @see rte_flow_create()
4354  * @see rte_flow_ops
4355  */
4356 struct rte_flow *
4357 mlx5_flow_create(struct rte_eth_dev *dev,
4358 		 const struct rte_flow_attr *attr,
4359 		 const struct rte_flow_item items[],
4360 		 const struct rte_flow_action actions[],
4361 		 struct rte_flow_error *error)
4362 {
4363 	struct mlx5_priv *priv = dev->data->dev_private;
4364 
4365 	return flow_list_create(dev, &priv->flows,
4366 				attr, items, actions, true, error);
4367 }
4368 
4369 /**
4370  * Destroy a flow in a list.
4371  *
4372  * @param dev
4373  *   Pointer to Ethernet device.
4374  * @param list
4375  *   Pointer to a TAILQ flow list. If this parameter NULL,
4376  *   there is no flow removal from the list.
4377  * @param[in] flow
4378  *   Flow to destroy.
4379  */
4380 static void
4381 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4382 		  struct rte_flow *flow)
4383 {
4384 	struct mlx5_priv *priv = dev->data->dev_private;
4385 
4386 	/*
4387 	 * Update RX queue flags only if port is started, otherwise it is
4388 	 * already clean.
4389 	 */
4390 	if (dev->data->dev_started)
4391 		flow_rxq_flags_trim(dev, flow);
4392 	if (flow->hairpin_flow_id)
4393 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4394 				     flow->hairpin_flow_id);
4395 	flow_drv_destroy(dev, flow);
4396 	if (list)
4397 		TAILQ_REMOVE(list, flow, next);
4398 	flow_mreg_del_copy_action(dev, flow);
4399 	rte_free(flow->fdir);
4400 	rte_free(flow);
4401 }
4402 
4403 /**
4404  * Destroy all flows.
4405  *
4406  * @param dev
4407  *   Pointer to Ethernet device.
4408  * @param list
4409  *   Pointer to a TAILQ flow list.
4410  */
4411 void
4412 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4413 {
4414 	while (!TAILQ_EMPTY(list)) {
4415 		struct rte_flow *flow;
4416 
4417 		flow = TAILQ_FIRST(list);
4418 		flow_list_destroy(dev, list, flow);
4419 	}
4420 }
4421 
4422 /**
4423  * Remove all flows.
4424  *
4425  * @param dev
4426  *   Pointer to Ethernet device.
4427  * @param list
4428  *   Pointer to a TAILQ flow list.
4429  */
4430 void
4431 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4432 {
4433 	struct rte_flow *flow;
4434 
4435 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4436 		flow_drv_remove(dev, flow);
4437 		flow_mreg_stop_copy_action(dev, flow);
4438 	}
4439 	flow_mreg_del_default_copy_action(dev);
4440 	flow_rxq_flags_clear(dev);
4441 }
4442 
4443 /**
4444  * Add all flows.
4445  *
4446  * @param dev
4447  *   Pointer to Ethernet device.
4448  * @param list
4449  *   Pointer to a TAILQ flow list.
4450  *
4451  * @return
4452  *   0 on success, a negative errno value otherwise and rte_errno is set.
4453  */
4454 int
4455 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4456 {
4457 	struct rte_flow *flow;
4458 	struct rte_flow_error error;
4459 	int ret = 0;
4460 
4461 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4462 	ret = flow_mreg_add_default_copy_action(dev, &error);
4463 	if (ret < 0)
4464 		return -rte_errno;
4465 	/* Apply Flows created by application. */
4466 	TAILQ_FOREACH(flow, list, next) {
4467 		ret = flow_mreg_start_copy_action(dev, flow);
4468 		if (ret < 0)
4469 			goto error;
4470 		ret = flow_drv_apply(dev, flow, &error);
4471 		if (ret < 0)
4472 			goto error;
4473 		flow_rxq_flags_set(dev, flow);
4474 	}
4475 	return 0;
4476 error:
4477 	ret = rte_errno; /* Save rte_errno before cleanup. */
4478 	mlx5_flow_stop(dev, list);
4479 	rte_errno = ret; /* Restore rte_errno. */
4480 	return -rte_errno;
4481 }
4482 
4483 /**
4484  * Verify the flow list is empty
4485  *
4486  * @param dev
4487  *  Pointer to Ethernet device.
4488  *
4489  * @return the number of flows not released.
4490  */
4491 int
4492 mlx5_flow_verify(struct rte_eth_dev *dev)
4493 {
4494 	struct mlx5_priv *priv = dev->data->dev_private;
4495 	struct rte_flow *flow;
4496 	int ret = 0;
4497 
4498 	TAILQ_FOREACH(flow, &priv->flows, next) {
4499 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4500 			dev->data->port_id, (void *)flow);
4501 		++ret;
4502 	}
4503 	return ret;
4504 }
4505 
4506 /**
4507  * Enable default hairpin egress flow.
4508  *
4509  * @param dev
4510  *   Pointer to Ethernet device.
4511  * @param queue
4512  *   The queue index.
4513  *
4514  * @return
4515  *   0 on success, a negative errno value otherwise and rte_errno is set.
4516  */
4517 int
4518 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4519 			    uint32_t queue)
4520 {
4521 	struct mlx5_priv *priv = dev->data->dev_private;
4522 	const struct rte_flow_attr attr = {
4523 		.egress = 1,
4524 		.priority = 0,
4525 	};
4526 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4527 		.queue = queue,
4528 	};
4529 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4530 		.queue = UINT32_MAX,
4531 	};
4532 	struct rte_flow_item items[] = {
4533 		{
4534 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4535 			.spec = &queue_spec,
4536 			.last = NULL,
4537 			.mask = &queue_mask,
4538 		},
4539 		{
4540 			.type = RTE_FLOW_ITEM_TYPE_END,
4541 		},
4542 	};
4543 	struct rte_flow_action_jump jump = {
4544 		.group = MLX5_HAIRPIN_TX_TABLE,
4545 	};
4546 	struct rte_flow_action actions[2];
4547 	struct rte_flow *flow;
4548 	struct rte_flow_error error;
4549 
4550 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4551 	actions[0].conf = &jump;
4552 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4553 	flow = flow_list_create(dev, &priv->ctrl_flows,
4554 				&attr, items, actions, false, &error);
4555 	if (!flow) {
4556 		DRV_LOG(DEBUG,
4557 			"Failed to create ctrl flow: rte_errno(%d),"
4558 			" type(%d), message(%s)",
4559 			rte_errno, error.type,
4560 			error.message ? error.message : " (no stated reason)");
4561 		return -rte_errno;
4562 	}
4563 	return 0;
4564 }
4565 
4566 /**
4567  * Enable a control flow configured from the control plane.
4568  *
4569  * @param dev
4570  *   Pointer to Ethernet device.
4571  * @param eth_spec
4572  *   An Ethernet flow spec to apply.
4573  * @param eth_mask
4574  *   An Ethernet flow mask to apply.
4575  * @param vlan_spec
4576  *   A VLAN flow spec to apply.
4577  * @param vlan_mask
4578  *   A VLAN flow mask to apply.
4579  *
4580  * @return
4581  *   0 on success, a negative errno value otherwise and rte_errno is set.
4582  */
4583 int
4584 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4585 		    struct rte_flow_item_eth *eth_spec,
4586 		    struct rte_flow_item_eth *eth_mask,
4587 		    struct rte_flow_item_vlan *vlan_spec,
4588 		    struct rte_flow_item_vlan *vlan_mask)
4589 {
4590 	struct mlx5_priv *priv = dev->data->dev_private;
4591 	const struct rte_flow_attr attr = {
4592 		.ingress = 1,
4593 		.priority = MLX5_FLOW_PRIO_RSVD,
4594 	};
4595 	struct rte_flow_item items[] = {
4596 		{
4597 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4598 			.spec = eth_spec,
4599 			.last = NULL,
4600 			.mask = eth_mask,
4601 		},
4602 		{
4603 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4604 					      RTE_FLOW_ITEM_TYPE_END,
4605 			.spec = vlan_spec,
4606 			.last = NULL,
4607 			.mask = vlan_mask,
4608 		},
4609 		{
4610 			.type = RTE_FLOW_ITEM_TYPE_END,
4611 		},
4612 	};
4613 	uint16_t queue[priv->reta_idx_n];
4614 	struct rte_flow_action_rss action_rss = {
4615 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4616 		.level = 0,
4617 		.types = priv->rss_conf.rss_hf,
4618 		.key_len = priv->rss_conf.rss_key_len,
4619 		.queue_num = priv->reta_idx_n,
4620 		.key = priv->rss_conf.rss_key,
4621 		.queue = queue,
4622 	};
4623 	struct rte_flow_action actions[] = {
4624 		{
4625 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4626 			.conf = &action_rss,
4627 		},
4628 		{
4629 			.type = RTE_FLOW_ACTION_TYPE_END,
4630 		},
4631 	};
4632 	struct rte_flow *flow;
4633 	struct rte_flow_error error;
4634 	unsigned int i;
4635 
4636 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4637 		return 0;
4638 	}
4639 	for (i = 0; i != priv->reta_idx_n; ++i)
4640 		queue[i] = (*priv->reta_idx)[i];
4641 	flow = flow_list_create(dev, &priv->ctrl_flows,
4642 				&attr, items, actions, false, &error);
4643 	if (!flow)
4644 		return -rte_errno;
4645 	return 0;
4646 }
4647 
4648 /**
4649  * Enable a flow control configured from the control plane.
4650  *
4651  * @param dev
4652  *   Pointer to Ethernet device.
4653  * @param eth_spec
4654  *   An Ethernet flow spec to apply.
4655  * @param eth_mask
4656  *   An Ethernet flow mask to apply.
4657  *
4658  * @return
4659  *   0 on success, a negative errno value otherwise and rte_errno is set.
4660  */
4661 int
4662 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4663 	       struct rte_flow_item_eth *eth_spec,
4664 	       struct rte_flow_item_eth *eth_mask)
4665 {
4666 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4667 }
4668 
4669 /**
4670  * Destroy a flow.
4671  *
4672  * @see rte_flow_destroy()
4673  * @see rte_flow_ops
4674  */
4675 int
4676 mlx5_flow_destroy(struct rte_eth_dev *dev,
4677 		  struct rte_flow *flow,
4678 		  struct rte_flow_error *error __rte_unused)
4679 {
4680 	struct mlx5_priv *priv = dev->data->dev_private;
4681 
4682 	flow_list_destroy(dev, &priv->flows, flow);
4683 	return 0;
4684 }
4685 
4686 /**
4687  * Destroy all flows.
4688  *
4689  * @see rte_flow_flush()
4690  * @see rte_flow_ops
4691  */
4692 int
4693 mlx5_flow_flush(struct rte_eth_dev *dev,
4694 		struct rte_flow_error *error __rte_unused)
4695 {
4696 	struct mlx5_priv *priv = dev->data->dev_private;
4697 
4698 	mlx5_flow_list_flush(dev, &priv->flows);
4699 	return 0;
4700 }
4701 
4702 /**
4703  * Isolated mode.
4704  *
4705  * @see rte_flow_isolate()
4706  * @see rte_flow_ops
4707  */
4708 int
4709 mlx5_flow_isolate(struct rte_eth_dev *dev,
4710 		  int enable,
4711 		  struct rte_flow_error *error)
4712 {
4713 	struct mlx5_priv *priv = dev->data->dev_private;
4714 
4715 	if (dev->data->dev_started) {
4716 		rte_flow_error_set(error, EBUSY,
4717 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4718 				   NULL,
4719 				   "port must be stopped first");
4720 		return -rte_errno;
4721 	}
4722 	priv->isolated = !!enable;
4723 	if (enable)
4724 		dev->dev_ops = &mlx5_dev_ops_isolate;
4725 	else
4726 		dev->dev_ops = &mlx5_dev_ops;
4727 	return 0;
4728 }
4729 
4730 /**
4731  * Query a flow.
4732  *
4733  * @see rte_flow_query()
4734  * @see rte_flow_ops
4735  */
4736 static int
4737 flow_drv_query(struct rte_eth_dev *dev,
4738 	       struct rte_flow *flow,
4739 	       const struct rte_flow_action *actions,
4740 	       void *data,
4741 	       struct rte_flow_error *error)
4742 {
4743 	const struct mlx5_flow_driver_ops *fops;
4744 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4745 
4746 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4747 	fops = flow_get_drv_ops(ftype);
4748 
4749 	return fops->query(dev, flow, actions, data, error);
4750 }
4751 
4752 /**
4753  * Query a flow.
4754  *
4755  * @see rte_flow_query()
4756  * @see rte_flow_ops
4757  */
4758 int
4759 mlx5_flow_query(struct rte_eth_dev *dev,
4760 		struct rte_flow *flow,
4761 		const struct rte_flow_action *actions,
4762 		void *data,
4763 		struct rte_flow_error *error)
4764 {
4765 	int ret;
4766 
4767 	ret = flow_drv_query(dev, flow, actions, data, error);
4768 	if (ret < 0)
4769 		return ret;
4770 	return 0;
4771 }
4772 
4773 /**
4774  * Convert a flow director filter to a generic flow.
4775  *
4776  * @param dev
4777  *   Pointer to Ethernet device.
4778  * @param fdir_filter
4779  *   Flow director filter to add.
4780  * @param attributes
4781  *   Generic flow parameters structure.
4782  *
4783  * @return
4784  *   0 on success, a negative errno value otherwise and rte_errno is set.
4785  */
4786 static int
4787 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4788 			 const struct rte_eth_fdir_filter *fdir_filter,
4789 			 struct mlx5_fdir *attributes)
4790 {
4791 	struct mlx5_priv *priv = dev->data->dev_private;
4792 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4793 	const struct rte_eth_fdir_masks *mask =
4794 		&dev->data->dev_conf.fdir_conf.mask;
4795 
4796 	/* Validate queue number. */
4797 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4798 		DRV_LOG(ERR, "port %u invalid queue number %d",
4799 			dev->data->port_id, fdir_filter->action.rx_queue);
4800 		rte_errno = EINVAL;
4801 		return -rte_errno;
4802 	}
4803 	attributes->attr.ingress = 1;
4804 	attributes->items[0] = (struct rte_flow_item) {
4805 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4806 		.spec = &attributes->l2,
4807 		.mask = &attributes->l2_mask,
4808 	};
4809 	switch (fdir_filter->action.behavior) {
4810 	case RTE_ETH_FDIR_ACCEPT:
4811 		attributes->actions[0] = (struct rte_flow_action){
4812 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4813 			.conf = &attributes->queue,
4814 		};
4815 		break;
4816 	case RTE_ETH_FDIR_REJECT:
4817 		attributes->actions[0] = (struct rte_flow_action){
4818 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4819 		};
4820 		break;
4821 	default:
4822 		DRV_LOG(ERR, "port %u invalid behavior %d",
4823 			dev->data->port_id,
4824 			fdir_filter->action.behavior);
4825 		rte_errno = ENOTSUP;
4826 		return -rte_errno;
4827 	}
4828 	attributes->queue.index = fdir_filter->action.rx_queue;
4829 	/* Handle L3. */
4830 	switch (fdir_filter->input.flow_type) {
4831 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4832 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4833 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4834 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4835 			.src_addr = input->flow.ip4_flow.src_ip,
4836 			.dst_addr = input->flow.ip4_flow.dst_ip,
4837 			.time_to_live = input->flow.ip4_flow.ttl,
4838 			.type_of_service = input->flow.ip4_flow.tos,
4839 		};
4840 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4841 			.src_addr = mask->ipv4_mask.src_ip,
4842 			.dst_addr = mask->ipv4_mask.dst_ip,
4843 			.time_to_live = mask->ipv4_mask.ttl,
4844 			.type_of_service = mask->ipv4_mask.tos,
4845 			.next_proto_id = mask->ipv4_mask.proto,
4846 		};
4847 		attributes->items[1] = (struct rte_flow_item){
4848 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4849 			.spec = &attributes->l3,
4850 			.mask = &attributes->l3_mask,
4851 		};
4852 		break;
4853 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4854 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4855 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4856 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4857 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4858 			.proto = input->flow.ipv6_flow.proto,
4859 		};
4860 
4861 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4862 		       input->flow.ipv6_flow.src_ip,
4863 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4864 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4865 		       input->flow.ipv6_flow.dst_ip,
4866 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4867 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4868 		       mask->ipv6_mask.src_ip,
4869 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4870 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4871 		       mask->ipv6_mask.dst_ip,
4872 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4873 		attributes->items[1] = (struct rte_flow_item){
4874 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4875 			.spec = &attributes->l3,
4876 			.mask = &attributes->l3_mask,
4877 		};
4878 		break;
4879 	default:
4880 		DRV_LOG(ERR, "port %u invalid flow type%d",
4881 			dev->data->port_id, fdir_filter->input.flow_type);
4882 		rte_errno = ENOTSUP;
4883 		return -rte_errno;
4884 	}
4885 	/* Handle L4. */
4886 	switch (fdir_filter->input.flow_type) {
4887 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4888 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4889 			.src_port = input->flow.udp4_flow.src_port,
4890 			.dst_port = input->flow.udp4_flow.dst_port,
4891 		};
4892 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4893 			.src_port = mask->src_port_mask,
4894 			.dst_port = mask->dst_port_mask,
4895 		};
4896 		attributes->items[2] = (struct rte_flow_item){
4897 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4898 			.spec = &attributes->l4,
4899 			.mask = &attributes->l4_mask,
4900 		};
4901 		break;
4902 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4903 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4904 			.src_port = input->flow.tcp4_flow.src_port,
4905 			.dst_port = input->flow.tcp4_flow.dst_port,
4906 		};
4907 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4908 			.src_port = mask->src_port_mask,
4909 			.dst_port = mask->dst_port_mask,
4910 		};
4911 		attributes->items[2] = (struct rte_flow_item){
4912 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4913 			.spec = &attributes->l4,
4914 			.mask = &attributes->l4_mask,
4915 		};
4916 		break;
4917 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4918 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4919 			.src_port = input->flow.udp6_flow.src_port,
4920 			.dst_port = input->flow.udp6_flow.dst_port,
4921 		};
4922 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4923 			.src_port = mask->src_port_mask,
4924 			.dst_port = mask->dst_port_mask,
4925 		};
4926 		attributes->items[2] = (struct rte_flow_item){
4927 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4928 			.spec = &attributes->l4,
4929 			.mask = &attributes->l4_mask,
4930 		};
4931 		break;
4932 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4933 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4934 			.src_port = input->flow.tcp6_flow.src_port,
4935 			.dst_port = input->flow.tcp6_flow.dst_port,
4936 		};
4937 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4938 			.src_port = mask->src_port_mask,
4939 			.dst_port = mask->dst_port_mask,
4940 		};
4941 		attributes->items[2] = (struct rte_flow_item){
4942 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4943 			.spec = &attributes->l4,
4944 			.mask = &attributes->l4_mask,
4945 		};
4946 		break;
4947 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4948 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4949 		break;
4950 	default:
4951 		DRV_LOG(ERR, "port %u invalid flow type%d",
4952 			dev->data->port_id, fdir_filter->input.flow_type);
4953 		rte_errno = ENOTSUP;
4954 		return -rte_errno;
4955 	}
4956 	return 0;
4957 }
4958 
4959 #define FLOW_FDIR_CMP(f1, f2, fld) \
4960 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4961 
4962 /**
4963  * Compare two FDIR flows. If items and actions are identical, the two flows are
4964  * regarded as same.
4965  *
4966  * @param dev
4967  *   Pointer to Ethernet device.
4968  * @param f1
4969  *   FDIR flow to compare.
4970  * @param f2
4971  *   FDIR flow to compare.
4972  *
4973  * @return
4974  *   Zero on match, 1 otherwise.
4975  */
4976 static int
4977 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4978 {
4979 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4980 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4981 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4982 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4983 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4984 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4985 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4986 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4987 		return 1;
4988 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4989 	    FLOW_FDIR_CMP(f1, f2, queue))
4990 		return 1;
4991 	return 0;
4992 }
4993 
4994 /**
4995  * Search device flow list to find out a matched FDIR flow.
4996  *
4997  * @param dev
4998  *   Pointer to Ethernet device.
4999  * @param fdir_flow
5000  *   FDIR flow to lookup.
5001  *
5002  * @return
5003  *   Pointer of flow if found, NULL otherwise.
5004  */
5005 static struct rte_flow *
5006 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5007 {
5008 	struct mlx5_priv *priv = dev->data->dev_private;
5009 	struct rte_flow *flow = NULL;
5010 
5011 	assert(fdir_flow);
5012 	TAILQ_FOREACH(flow, &priv->flows, next) {
5013 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5014 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5015 				dev->data->port_id, (void *)flow);
5016 			break;
5017 		}
5018 	}
5019 	return flow;
5020 }
5021 
5022 /**
5023  * Add new flow director filter and store it in list.
5024  *
5025  * @param dev
5026  *   Pointer to Ethernet device.
5027  * @param fdir_filter
5028  *   Flow director filter to add.
5029  *
5030  * @return
5031  *   0 on success, a negative errno value otherwise and rte_errno is set.
5032  */
5033 static int
5034 flow_fdir_filter_add(struct rte_eth_dev *dev,
5035 		     const struct rte_eth_fdir_filter *fdir_filter)
5036 {
5037 	struct mlx5_priv *priv = dev->data->dev_private;
5038 	struct mlx5_fdir *fdir_flow;
5039 	struct rte_flow *flow;
5040 	int ret;
5041 
5042 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5043 	if (!fdir_flow) {
5044 		rte_errno = ENOMEM;
5045 		return -rte_errno;
5046 	}
5047 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5048 	if (ret)
5049 		goto error;
5050 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5051 	if (flow) {
5052 		rte_errno = EEXIST;
5053 		goto error;
5054 	}
5055 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5056 				fdir_flow->items, fdir_flow->actions, true,
5057 				NULL);
5058 	if (!flow)
5059 		goto error;
5060 	assert(!flow->fdir);
5061 	flow->fdir = fdir_flow;
5062 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5063 		dev->data->port_id, (void *)flow);
5064 	return 0;
5065 error:
5066 	rte_free(fdir_flow);
5067 	return -rte_errno;
5068 }
5069 
5070 /**
5071  * Delete specific filter.
5072  *
5073  * @param dev
5074  *   Pointer to Ethernet device.
5075  * @param fdir_filter
5076  *   Filter to be deleted.
5077  *
5078  * @return
5079  *   0 on success, a negative errno value otherwise and rte_errno is set.
5080  */
5081 static int
5082 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5083 			const struct rte_eth_fdir_filter *fdir_filter)
5084 {
5085 	struct mlx5_priv *priv = dev->data->dev_private;
5086 	struct rte_flow *flow;
5087 	struct mlx5_fdir fdir_flow = {
5088 		.attr.group = 0,
5089 	};
5090 	int ret;
5091 
5092 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5093 	if (ret)
5094 		return -rte_errno;
5095 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5096 	if (!flow) {
5097 		rte_errno = ENOENT;
5098 		return -rte_errno;
5099 	}
5100 	flow_list_destroy(dev, &priv->flows, flow);
5101 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5102 		dev->data->port_id, (void *)flow);
5103 	return 0;
5104 }
5105 
5106 /**
5107  * Update queue for specific filter.
5108  *
5109  * @param dev
5110  *   Pointer to Ethernet device.
5111  * @param fdir_filter
5112  *   Filter to be updated.
5113  *
5114  * @return
5115  *   0 on success, a negative errno value otherwise and rte_errno is set.
5116  */
5117 static int
5118 flow_fdir_filter_update(struct rte_eth_dev *dev,
5119 			const struct rte_eth_fdir_filter *fdir_filter)
5120 {
5121 	int ret;
5122 
5123 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5124 	if (ret)
5125 		return ret;
5126 	return flow_fdir_filter_add(dev, fdir_filter);
5127 }
5128 
5129 /**
5130  * Flush all filters.
5131  *
5132  * @param dev
5133  *   Pointer to Ethernet device.
5134  */
5135 static void
5136 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5137 {
5138 	struct mlx5_priv *priv = dev->data->dev_private;
5139 
5140 	mlx5_flow_list_flush(dev, &priv->flows);
5141 }
5142 
5143 /**
5144  * Get flow director information.
5145  *
5146  * @param dev
5147  *   Pointer to Ethernet device.
5148  * @param[out] fdir_info
5149  *   Resulting flow director information.
5150  */
5151 static void
5152 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5153 {
5154 	struct rte_eth_fdir_masks *mask =
5155 		&dev->data->dev_conf.fdir_conf.mask;
5156 
5157 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5158 	fdir_info->guarant_spc = 0;
5159 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5160 	fdir_info->max_flexpayload = 0;
5161 	fdir_info->flow_types_mask[0] = 0;
5162 	fdir_info->flex_payload_unit = 0;
5163 	fdir_info->max_flex_payload_segment_num = 0;
5164 	fdir_info->flex_payload_limit = 0;
5165 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5166 }
5167 
5168 /**
5169  * Deal with flow director operations.
5170  *
5171  * @param dev
5172  *   Pointer to Ethernet device.
5173  * @param filter_op
5174  *   Operation to perform.
5175  * @param arg
5176  *   Pointer to operation-specific structure.
5177  *
5178  * @return
5179  *   0 on success, a negative errno value otherwise and rte_errno is set.
5180  */
5181 static int
5182 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5183 		    void *arg)
5184 {
5185 	enum rte_fdir_mode fdir_mode =
5186 		dev->data->dev_conf.fdir_conf.mode;
5187 
5188 	if (filter_op == RTE_ETH_FILTER_NOP)
5189 		return 0;
5190 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5191 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5192 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5193 			dev->data->port_id, fdir_mode);
5194 		rte_errno = EINVAL;
5195 		return -rte_errno;
5196 	}
5197 	switch (filter_op) {
5198 	case RTE_ETH_FILTER_ADD:
5199 		return flow_fdir_filter_add(dev, arg);
5200 	case RTE_ETH_FILTER_UPDATE:
5201 		return flow_fdir_filter_update(dev, arg);
5202 	case RTE_ETH_FILTER_DELETE:
5203 		return flow_fdir_filter_delete(dev, arg);
5204 	case RTE_ETH_FILTER_FLUSH:
5205 		flow_fdir_filter_flush(dev);
5206 		break;
5207 	case RTE_ETH_FILTER_INFO:
5208 		flow_fdir_info_get(dev, arg);
5209 		break;
5210 	default:
5211 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5212 			dev->data->port_id, filter_op);
5213 		rte_errno = EINVAL;
5214 		return -rte_errno;
5215 	}
5216 	return 0;
5217 }
5218 
5219 /**
5220  * Manage filter operations.
5221  *
5222  * @param dev
5223  *   Pointer to Ethernet device structure.
5224  * @param filter_type
5225  *   Filter type.
5226  * @param filter_op
5227  *   Operation to perform.
5228  * @param arg
5229  *   Pointer to operation-specific structure.
5230  *
5231  * @return
5232  *   0 on success, a negative errno value otherwise and rte_errno is set.
5233  */
5234 int
5235 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5236 		     enum rte_filter_type filter_type,
5237 		     enum rte_filter_op filter_op,
5238 		     void *arg)
5239 {
5240 	switch (filter_type) {
5241 	case RTE_ETH_FILTER_GENERIC:
5242 		if (filter_op != RTE_ETH_FILTER_GET) {
5243 			rte_errno = EINVAL;
5244 			return -rte_errno;
5245 		}
5246 		*(const void **)arg = &mlx5_flow_ops;
5247 		return 0;
5248 	case RTE_ETH_FILTER_FDIR:
5249 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5250 	default:
5251 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5252 			dev->data->port_id, filter_type);
5253 		rte_errno = ENOTSUP;
5254 		return -rte_errno;
5255 	}
5256 	return 0;
5257 }
5258 
5259 /**
5260  * Create the needed meter and suffix tables.
5261  *
5262  * @param[in] dev
5263  *   Pointer to Ethernet device.
5264  * @param[in] fm
5265  *   Pointer to the flow meter.
5266  *
5267  * @return
5268  *   Pointer to table set on success, NULL otherwise.
5269  */
5270 struct mlx5_meter_domains_infos *
5271 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5272 			  const struct mlx5_flow_meter *fm)
5273 {
5274 	const struct mlx5_flow_driver_ops *fops;
5275 
5276 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5277 	return fops->create_mtr_tbls(dev, fm);
5278 }
5279 
5280 /**
5281  * Destroy the meter table set.
5282  *
5283  * @param[in] dev
5284  *   Pointer to Ethernet device.
5285  * @param[in] tbl
5286  *   Pointer to the meter table set.
5287  *
5288  * @return
5289  *   0 on success.
5290  */
5291 int
5292 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5293 			   struct mlx5_meter_domains_infos *tbls)
5294 {
5295 	const struct mlx5_flow_driver_ops *fops;
5296 
5297 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5298 	return fops->destroy_mtr_tbls(dev, tbls);
5299 }
5300 
5301 /**
5302  * Create policer rules.
5303  *
5304  * @param[in] dev
5305  *   Pointer to Ethernet device.
5306  * @param[in] fm
5307  *   Pointer to flow meter structure.
5308  * @param[in] attr
5309  *   Pointer to flow attributes.
5310  *
5311  * @return
5312  *   0 on success, -1 otherwise.
5313  */
5314 int
5315 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5316 			       struct mlx5_flow_meter *fm,
5317 			       const struct rte_flow_attr *attr)
5318 {
5319 	const struct mlx5_flow_driver_ops *fops;
5320 
5321 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5322 	return fops->create_policer_rules(dev, fm, attr);
5323 }
5324 
5325 /**
5326  * Destroy policer rules.
5327  *
5328  * @param[in] fm
5329  *   Pointer to flow meter structure.
5330  * @param[in] attr
5331  *   Pointer to flow attributes.
5332  *
5333  * @return
5334  *   0 on success, -1 otherwise.
5335  */
5336 int
5337 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5338 				struct mlx5_flow_meter *fm,
5339 				const struct rte_flow_attr *attr)
5340 {
5341 	const struct mlx5_flow_driver_ops *fops;
5342 
5343 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5344 	return fops->destroy_policer_rules(dev, fm, attr);
5345 }
5346 
5347 /**
5348  * Allocate a counter.
5349  *
5350  * @param[in] dev
5351  *   Pointer to Ethernet device structure.
5352  *
5353  * @return
5354  *   Pointer to allocated counter  on success, NULL otherwise.
5355  */
5356 struct mlx5_flow_counter *
5357 mlx5_counter_alloc(struct rte_eth_dev *dev)
5358 {
5359 	const struct mlx5_flow_driver_ops *fops;
5360 	struct rte_flow_attr attr = { .transfer = 0 };
5361 
5362 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5363 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5364 		return fops->counter_alloc(dev);
5365 	}
5366 	DRV_LOG(ERR,
5367 		"port %u counter allocate is not supported.",
5368 		 dev->data->port_id);
5369 	return NULL;
5370 }
5371 
5372 /**
5373  * Free a counter.
5374  *
5375  * @param[in] dev
5376  *   Pointer to Ethernet device structure.
5377  * @param[in] cnt
5378  *   Pointer to counter to be free.
5379  */
5380 void
5381 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5382 {
5383 	const struct mlx5_flow_driver_ops *fops;
5384 	struct rte_flow_attr attr = { .transfer = 0 };
5385 
5386 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5387 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5388 		fops->counter_free(dev, cnt);
5389 		return;
5390 	}
5391 	DRV_LOG(ERR,
5392 		"port %u counter free is not supported.",
5393 		 dev->data->port_id);
5394 }
5395 
5396 /**
5397  * Query counter statistics.
5398  *
5399  * @param[in] dev
5400  *   Pointer to Ethernet device structure.
5401  * @param[in] cnt
5402  *   Pointer to counter to query.
5403  * @param[in] clear
5404  *   Set to clear counter statistics.
5405  * @param[out] pkts
5406  *   The counter hits packets number to save.
5407  * @param[out] bytes
5408  *   The counter hits bytes number to save.
5409  *
5410  * @return
5411  *   0 on success, a negative errno value otherwise.
5412  */
5413 int
5414 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5415 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5416 {
5417 	const struct mlx5_flow_driver_ops *fops;
5418 	struct rte_flow_attr attr = { .transfer = 0 };
5419 
5420 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5421 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5422 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5423 	}
5424 	DRV_LOG(ERR,
5425 		"port %u counter query is not supported.",
5426 		 dev->data->port_id);
5427 	return -ENOTSUP;
5428 }
5429 
5430 #define MLX5_POOL_QUERY_FREQ_US 1000000
5431 
5432 /**
5433  * Set the periodic procedure for triggering asynchronous batch queries for all
5434  * the counter pools.
5435  *
5436  * @param[in] sh
5437  *   Pointer to mlx5_ibv_shared object.
5438  */
5439 void
5440 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5441 {
5442 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5443 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5444 	uint32_t us;
5445 
5446 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5447 	pools_n += rte_atomic16_read(&cont->n_valid);
5448 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5449 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5450 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5451 		sh->cmng.query_thread_on = 0;
5452 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5453 	} else {
5454 		sh->cmng.query_thread_on = 1;
5455 	}
5456 }
5457 
5458 /**
5459  * The periodic procedure for triggering asynchronous batch queries for all the
5460  * counter pools. This function is probably called by the host thread.
5461  *
5462  * @param[in] arg
5463  *   The parameter for the alarm process.
5464  */
5465 void
5466 mlx5_flow_query_alarm(void *arg)
5467 {
5468 	struct mlx5_ibv_shared *sh = arg;
5469 	struct mlx5_devx_obj *dcs;
5470 	uint16_t offset;
5471 	int ret;
5472 	uint8_t batch = sh->cmng.batch;
5473 	uint16_t pool_index = sh->cmng.pool_index;
5474 	struct mlx5_pools_container *cont;
5475 	struct mlx5_pools_container *mcont;
5476 	struct mlx5_flow_counter_pool *pool;
5477 
5478 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5479 		goto set_alarm;
5480 next_container:
5481 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5482 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5483 	/* Check if resize was done and need to flip a container. */
5484 	if (cont != mcont) {
5485 		if (cont->pools) {
5486 			/* Clean the old container. */
5487 			rte_free(cont->pools);
5488 			memset(cont, 0, sizeof(*cont));
5489 		}
5490 		rte_cio_wmb();
5491 		 /* Flip the host container. */
5492 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5493 		cont = mcont;
5494 	}
5495 	if (!cont->pools) {
5496 		/* 2 empty containers case is unexpected. */
5497 		if (unlikely(batch != sh->cmng.batch))
5498 			goto set_alarm;
5499 		batch ^= 0x1;
5500 		pool_index = 0;
5501 		goto next_container;
5502 	}
5503 	pool = cont->pools[pool_index];
5504 	if (pool->raw_hw)
5505 		/* There is a pool query in progress. */
5506 		goto set_alarm;
5507 	pool->raw_hw =
5508 		LIST_FIRST(&sh->cmng.free_stat_raws);
5509 	if (!pool->raw_hw)
5510 		/* No free counter statistics raw memory. */
5511 		goto set_alarm;
5512 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5513 							      (&pool->a64_dcs);
5514 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5515 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5516 					       offset, NULL, NULL,
5517 					       pool->raw_hw->mem_mng->dm->id,
5518 					       (void *)(uintptr_t)
5519 					       (pool->raw_hw->data + offset),
5520 					       sh->devx_comp,
5521 					       (uint64_t)(uintptr_t)pool);
5522 	if (ret) {
5523 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5524 			" %d", pool->min_dcs->id);
5525 		pool->raw_hw = NULL;
5526 		goto set_alarm;
5527 	}
5528 	pool->raw_hw->min_dcs_id = dcs->id;
5529 	LIST_REMOVE(pool->raw_hw, next);
5530 	sh->cmng.pending_queries++;
5531 	pool_index++;
5532 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5533 		batch ^= 0x1;
5534 		pool_index = 0;
5535 	}
5536 set_alarm:
5537 	sh->cmng.batch = batch;
5538 	sh->cmng.pool_index = pool_index;
5539 	mlx5_set_query_alarm(sh);
5540 }
5541 
5542 /**
5543  * Handler for the HW respond about ready values from an asynchronous batch
5544  * query. This function is probably called by the host thread.
5545  *
5546  * @param[in] sh
5547  *   The pointer to the shared IB device context.
5548  * @param[in] async_id
5549  *   The Devx async ID.
5550  * @param[in] status
5551  *   The status of the completion.
5552  */
5553 void
5554 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5555 				  uint64_t async_id, int status)
5556 {
5557 	struct mlx5_flow_counter_pool *pool =
5558 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5559 	struct mlx5_counter_stats_raw *raw_to_free;
5560 
5561 	if (unlikely(status)) {
5562 		raw_to_free = pool->raw_hw;
5563 	} else {
5564 		raw_to_free = pool->raw;
5565 		rte_spinlock_lock(&pool->sl);
5566 		pool->raw = pool->raw_hw;
5567 		rte_spinlock_unlock(&pool->sl);
5568 		rte_atomic64_add(&pool->query_gen, 1);
5569 		/* Be sure the new raw counters data is updated in memory. */
5570 		rte_cio_wmb();
5571 	}
5572 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5573 	pool->raw_hw = NULL;
5574 	sh->cmng.pending_queries--;
5575 }
5576 
5577 /**
5578  * Translate the rte_flow group index to HW table value.
5579  *
5580  * @param[in] attributes
5581  *   Pointer to flow attributes
5582  * @param[in] external
5583  *   Value is part of flow rule created by request external to PMD.
5584  * @param[in] group
5585  *   rte_flow group index value.
5586  * @param[out] table
5587  *   HW table value.
5588  * @param[out] error
5589  *   Pointer to error structure.
5590  *
5591  * @return
5592  *   0 on success, a negative errno value otherwise and rte_errno is set.
5593  */
5594 int
5595 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5596 			 uint32_t group, uint32_t *table,
5597 			 struct rte_flow_error *error)
5598 {
5599 	if (attributes->transfer && external) {
5600 		if (group == UINT32_MAX)
5601 			return rte_flow_error_set
5602 						(error, EINVAL,
5603 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5604 						 NULL,
5605 						 "group index not supported");
5606 		*table = group + 1;
5607 	} else {
5608 		*table = group;
5609 	}
5610 	return 0;
5611 }
5612 
5613 /**
5614  * Discover availability of metadata reg_c's.
5615  *
5616  * Iteratively use test flows to check availability.
5617  *
5618  * @param[in] dev
5619  *   Pointer to the Ethernet device structure.
5620  *
5621  * @return
5622  *   0 on success, a negative errno value otherwise and rte_errno is set.
5623  */
5624 int
5625 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5626 {
5627 	struct mlx5_priv *priv = dev->data->dev_private;
5628 	struct mlx5_dev_config *config = &priv->config;
5629 	enum modify_reg idx;
5630 	int n = 0;
5631 
5632 	/* reg_c[0] and reg_c[1] are reserved. */
5633 	config->flow_mreg_c[n++] = REG_C_0;
5634 	config->flow_mreg_c[n++] = REG_C_1;
5635 	/* Discover availability of other reg_c's. */
5636 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5637 		struct rte_flow_attr attr = {
5638 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5639 			.priority = MLX5_FLOW_PRIO_RSVD,
5640 			.ingress = 1,
5641 		};
5642 		struct rte_flow_item items[] = {
5643 			[0] = {
5644 				.type = RTE_FLOW_ITEM_TYPE_END,
5645 			},
5646 		};
5647 		struct rte_flow_action actions[] = {
5648 			[0] = {
5649 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5650 				.conf = &(struct mlx5_flow_action_copy_mreg){
5651 					.src = REG_C_1,
5652 					.dst = idx,
5653 				},
5654 			},
5655 			[1] = {
5656 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5657 				.conf = &(struct rte_flow_action_jump){
5658 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5659 				},
5660 			},
5661 			[2] = {
5662 				.type = RTE_FLOW_ACTION_TYPE_END,
5663 			},
5664 		};
5665 		struct rte_flow *flow;
5666 		struct rte_flow_error error;
5667 
5668 		if (!config->dv_flow_en)
5669 			break;
5670 		/* Create internal flow, validation skips copy action. */
5671 		flow = flow_list_create(dev, NULL, &attr, items,
5672 					actions, false, &error);
5673 		if (!flow)
5674 			continue;
5675 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5676 			config->flow_mreg_c[n++] = idx;
5677 		flow_list_destroy(dev, NULL, flow);
5678 	}
5679 	for (; n < MLX5_MREG_C_NUM; ++n)
5680 		config->flow_mreg_c[n] = REG_NONE;
5681 	return 0;
5682 }
5683 
5684 /**
5685  * Dump flow raw hw data to file
5686  *
5687  * @param[in] dev
5688  *    The pointer to Ethernet device.
5689  * @param[in] file
5690  *   A pointer to a file for output.
5691  * @param[out] error
5692  *   Perform verbose error reporting if not NULL. PMDs initialize this
5693  *   structure in case of error only.
5694  * @return
5695  *   0 on success, a nagative value otherwise.
5696  */
5697 int
5698 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5699 		   FILE *file,
5700 		   struct rte_flow_error *error __rte_unused)
5701 {
5702 	struct mlx5_priv *priv = dev->data->dev_private;
5703 
5704 	return mlx5_devx_cmd_flow_dump(priv->sh, file);
5705 }
5706