xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision fdf7471cccb8be023037c218d1402c0549eb2c8e)
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 };
240 
241 /* Convert FDIR request to Generic flow. */
242 struct mlx5_fdir {
243 	struct rte_flow_attr attr;
244 	struct rte_flow_item items[4];
245 	struct rte_flow_item_eth l2;
246 	struct rte_flow_item_eth l2_mask;
247 	union {
248 		struct rte_flow_item_ipv4 ipv4;
249 		struct rte_flow_item_ipv6 ipv6;
250 	} l3;
251 	union {
252 		struct rte_flow_item_ipv4 ipv4;
253 		struct rte_flow_item_ipv6 ipv6;
254 	} l3_mask;
255 	union {
256 		struct rte_flow_item_udp udp;
257 		struct rte_flow_item_tcp tcp;
258 	} l4;
259 	union {
260 		struct rte_flow_item_udp udp;
261 		struct rte_flow_item_tcp tcp;
262 	} l4_mask;
263 	struct rte_flow_action actions[2];
264 	struct rte_flow_action_queue queue;
265 };
266 
267 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
268 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
269 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
270 };
271 
272 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
273 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
274 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
275 	{ 9, 10, 11 }, { 12, 13, 14 },
276 };
277 
278 /* Tunnel information. */
279 struct mlx5_flow_tunnel_info {
280 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
281 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
282 };
283 
284 static struct mlx5_flow_tunnel_info tunnels_info[] = {
285 	{
286 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
287 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
288 	},
289 	{
290 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
291 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
292 	},
293 	{
294 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
295 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
296 	},
297 	{
298 		.tunnel = MLX5_FLOW_LAYER_GRE,
299 		.ptype = RTE_PTYPE_TUNNEL_GRE,
300 	},
301 	{
302 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
303 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
304 	},
305 	{
306 		.tunnel = MLX5_FLOW_LAYER_MPLS,
307 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
308 	},
309 	{
310 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
311 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
312 	},
313 	{
314 		.tunnel = MLX5_FLOW_LAYER_IPIP,
315 		.ptype = RTE_PTYPE_TUNNEL_IP,
316 	},
317 	{
318 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
319 		.ptype = RTE_PTYPE_TUNNEL_IP,
320 	},
321 };
322 
323 /**
324  * Translate tag ID to register.
325  *
326  * @param[in] dev
327  *   Pointer to the Ethernet device structure.
328  * @param[in] feature
329  *   The feature that request the register.
330  * @param[in] id
331  *   The request register ID.
332  * @param[out] error
333  *   Error description in case of any.
334  *
335  * @return
336  *   The request register on success, a negative errno
337  *   value otherwise and rte_errno is set.
338  */
339 enum modify_reg
340 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
341 		     enum mlx5_feature_name feature,
342 		     uint32_t id,
343 		     struct rte_flow_error *error)
344 {
345 	struct mlx5_priv *priv = dev->data->dev_private;
346 	struct mlx5_dev_config *config = &priv->config;
347 	enum modify_reg start_reg;
348 
349 	switch (feature) {
350 	case MLX5_HAIRPIN_RX:
351 		return REG_B;
352 	case MLX5_HAIRPIN_TX:
353 		return REG_A;
354 	case MLX5_METADATA_RX:
355 		switch (config->dv_xmeta_en) {
356 		case MLX5_XMETA_MODE_LEGACY:
357 			return REG_B;
358 		case MLX5_XMETA_MODE_META16:
359 			return REG_C_0;
360 		case MLX5_XMETA_MODE_META32:
361 			return REG_C_1;
362 		}
363 		break;
364 	case MLX5_METADATA_TX:
365 		return REG_A;
366 	case MLX5_METADATA_FDB:
367 		return REG_C_0;
368 	case MLX5_FLOW_MARK:
369 		switch (config->dv_xmeta_en) {
370 		case MLX5_XMETA_MODE_LEGACY:
371 			return REG_NONE;
372 		case MLX5_XMETA_MODE_META16:
373 			return REG_C_1;
374 		case MLX5_XMETA_MODE_META32:
375 			return REG_C_0;
376 		}
377 		break;
378 	case MLX5_COPY_MARK:
379 	case MLX5_MTR_SFX:
380 		/*
381 		 * Metadata COPY_MARK register using is in meter suffix sub
382 		 * flow while with meter. It's safe to share the same register.
383 		 */
384 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
385 	case MLX5_MTR_COLOR:
386 		RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
387 		return priv->mtr_color_reg;
388 	case MLX5_APP_TAG:
389 		/*
390 		 * If meter is enable, it will engage two registers for color
391 		 * match and flow match. If meter color match is not using the
392 		 * REG_C_2, need to skip the REG_C_x be used by meter color
393 		 * match.
394 		 * If meter is disable, free to use all available registers.
395 		 */
396 		if (priv->mtr_color_reg != REG_NONE)
397 			start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
398 				    REG_C_4;
399 		else
400 			start_reg = REG_C_2;
401 		if (id > (REG_C_7 - start_reg))
402 			return rte_flow_error_set(error, EINVAL,
403 						  RTE_FLOW_ERROR_TYPE_ITEM,
404 						  NULL, "invalid tag id");
405 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
406 			return rte_flow_error_set(error, ENOTSUP,
407 						  RTE_FLOW_ERROR_TYPE_ITEM,
408 						  NULL, "unsupported tag id");
409 		/*
410 		 * This case means meter is using the REG_C_x great than 2.
411 		 * Take care not to conflict with meter color REG_C_x.
412 		 * If the available index REG_C_y >= REG_C_x, skip the
413 		 * color register.
414 		 */
415 		if (start_reg == REG_C_3 && config->flow_mreg_c
416 		    [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
417 			if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
418 			    REG_NONE)
419 				return config->flow_mreg_c
420 						[id + 1 + REG_C_3 - REG_C_0];
421 			return rte_flow_error_set(error, ENOTSUP,
422 						  RTE_FLOW_ERROR_TYPE_ITEM,
423 						  NULL, "unsupported tag id");
424 		}
425 		return config->flow_mreg_c[id + start_reg - REG_C_0];
426 	}
427 	assert(false);
428 	return rte_flow_error_set(error, EINVAL,
429 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
430 				  NULL, "invalid feature name");
431 }
432 
433 /**
434  * Check extensive flow metadata register support.
435  *
436  * @param dev
437  *   Pointer to rte_eth_dev structure.
438  *
439  * @return
440  *   True if device supports extensive flow metadata register, otherwise false.
441  */
442 bool
443 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
444 {
445 	struct mlx5_priv *priv = dev->data->dev_private;
446 	struct mlx5_dev_config *config = &priv->config;
447 
448 	/*
449 	 * Having available reg_c can be regarded inclusively as supporting
450 	 * extensive flow metadata register, which could mean,
451 	 * - metadata register copy action by modify header.
452 	 * - 16 modify header actions is supported.
453 	 * - reg_c's are preserved across different domain (FDB and NIC) on
454 	 *   packet loopback by flow lookup miss.
455 	 */
456 	return config->flow_mreg_c[2] != REG_NONE;
457 }
458 
459 /**
460  * Discover the maximum number of priority available.
461  *
462  * @param[in] dev
463  *   Pointer to the Ethernet device structure.
464  *
465  * @return
466  *   number of supported flow priority on success, a negative errno
467  *   value otherwise and rte_errno is set.
468  */
469 int
470 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
471 {
472 	struct mlx5_priv *priv = dev->data->dev_private;
473 	struct {
474 		struct ibv_flow_attr attr;
475 		struct ibv_flow_spec_eth eth;
476 		struct ibv_flow_spec_action_drop drop;
477 	} flow_attr = {
478 		.attr = {
479 			.num_of_specs = 2,
480 			.port = (uint8_t)priv->ibv_port,
481 		},
482 		.eth = {
483 			.type = IBV_FLOW_SPEC_ETH,
484 			.size = sizeof(struct ibv_flow_spec_eth),
485 		},
486 		.drop = {
487 			.size = sizeof(struct ibv_flow_spec_action_drop),
488 			.type = IBV_FLOW_SPEC_ACTION_DROP,
489 		},
490 	};
491 	struct ibv_flow *flow;
492 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
493 	uint16_t vprio[] = { 8, 16 };
494 	int i;
495 	int priority = 0;
496 
497 	if (!drop) {
498 		rte_errno = ENOTSUP;
499 		return -rte_errno;
500 	}
501 	for (i = 0; i != RTE_DIM(vprio); i++) {
502 		flow_attr.attr.priority = vprio[i] - 1;
503 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
504 		if (!flow)
505 			break;
506 		claim_zero(mlx5_glue->destroy_flow(flow));
507 		priority = vprio[i];
508 	}
509 	mlx5_hrxq_drop_release(dev);
510 	switch (priority) {
511 	case 8:
512 		priority = RTE_DIM(priority_map_3);
513 		break;
514 	case 16:
515 		priority = RTE_DIM(priority_map_5);
516 		break;
517 	default:
518 		rte_errno = ENOTSUP;
519 		DRV_LOG(ERR,
520 			"port %u verbs maximum priority: %d expected 8/16",
521 			dev->data->port_id, priority);
522 		return -rte_errno;
523 	}
524 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
525 		dev->data->port_id, priority);
526 	return priority;
527 }
528 
529 /**
530  * Adjust flow priority based on the highest layer and the request priority.
531  *
532  * @param[in] dev
533  *   Pointer to the Ethernet device structure.
534  * @param[in] priority
535  *   The rule base priority.
536  * @param[in] subpriority
537  *   The priority based on the items.
538  *
539  * @return
540  *   The new priority.
541  */
542 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
543 				   uint32_t subpriority)
544 {
545 	uint32_t res = 0;
546 	struct mlx5_priv *priv = dev->data->dev_private;
547 
548 	switch (priv->config.flow_prio) {
549 	case RTE_DIM(priority_map_3):
550 		res = priority_map_3[priority][subpriority];
551 		break;
552 	case RTE_DIM(priority_map_5):
553 		res = priority_map_5[priority][subpriority];
554 		break;
555 	}
556 	return  res;
557 }
558 
559 /**
560  * Verify the @p item specifications (spec, last, mask) are compatible with the
561  * NIC capabilities.
562  *
563  * @param[in] item
564  *   Item specification.
565  * @param[in] mask
566  *   @p item->mask or flow default bit-masks.
567  * @param[in] nic_mask
568  *   Bit-masks covering supported fields by the NIC to compare with user mask.
569  * @param[in] size
570  *   Bit-masks size in bytes.
571  * @param[out] error
572  *   Pointer to error structure.
573  *
574  * @return
575  *   0 on success, a negative errno value otherwise and rte_errno is set.
576  */
577 int
578 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
579 			  const uint8_t *mask,
580 			  const uint8_t *nic_mask,
581 			  unsigned int size,
582 			  struct rte_flow_error *error)
583 {
584 	unsigned int i;
585 
586 	assert(nic_mask);
587 	for (i = 0; i < size; ++i)
588 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
589 			return rte_flow_error_set(error, ENOTSUP,
590 						  RTE_FLOW_ERROR_TYPE_ITEM,
591 						  item,
592 						  "mask enables non supported"
593 						  " bits");
594 	if (!item->spec && (item->mask || item->last))
595 		return rte_flow_error_set(error, EINVAL,
596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
597 					  "mask/last without a spec is not"
598 					  " supported");
599 	if (item->spec && item->last) {
600 		uint8_t spec[size];
601 		uint8_t last[size];
602 		unsigned int i;
603 		int ret;
604 
605 		for (i = 0; i < size; ++i) {
606 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
607 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
608 		}
609 		ret = memcmp(spec, last, size);
610 		if (ret != 0)
611 			return rte_flow_error_set(error, EINVAL,
612 						  RTE_FLOW_ERROR_TYPE_ITEM,
613 						  item,
614 						  "range is not valid");
615 	}
616 	return 0;
617 }
618 
619 /**
620  * Adjust the hash fields according to the @p flow information.
621  *
622  * @param[in] dev_flow.
623  *   Pointer to the mlx5_flow.
624  * @param[in] tunnel
625  *   1 when the hash field is for a tunnel item.
626  * @param[in] layer_types
627  *   ETH_RSS_* types.
628  * @param[in] hash_fields
629  *   Item hash fields.
630  *
631  * @return
632  *   The hash fields that should be used.
633  */
634 uint64_t
635 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
636 			    int tunnel __rte_unused, uint64_t layer_types,
637 			    uint64_t hash_fields)
638 {
639 	struct rte_flow *flow = dev_flow->flow;
640 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
641 	int rss_request_inner = flow->rss.level >= 2;
642 
643 	/* Check RSS hash level for tunnel. */
644 	if (tunnel && rss_request_inner)
645 		hash_fields |= IBV_RX_HASH_INNER;
646 	else if (tunnel || rss_request_inner)
647 		return 0;
648 #endif
649 	/* Check if requested layer matches RSS hash fields. */
650 	if (!(flow->rss.types & layer_types))
651 		return 0;
652 	return hash_fields;
653 }
654 
655 /**
656  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
657  * if several tunnel rules are used on this queue, the tunnel ptype will be
658  * cleared.
659  *
660  * @param rxq_ctrl
661  *   Rx queue to update.
662  */
663 static void
664 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
665 {
666 	unsigned int i;
667 	uint32_t tunnel_ptype = 0;
668 
669 	/* Look up for the ptype to use. */
670 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
671 		if (!rxq_ctrl->flow_tunnels_n[i])
672 			continue;
673 		if (!tunnel_ptype) {
674 			tunnel_ptype = tunnels_info[i].ptype;
675 		} else {
676 			tunnel_ptype = 0;
677 			break;
678 		}
679 	}
680 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
681 }
682 
683 /**
684  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
685  * flow.
686  *
687  * @param[in] dev
688  *   Pointer to the Ethernet device structure.
689  * @param[in] dev_flow
690  *   Pointer to device flow structure.
691  */
692 static void
693 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
694 {
695 	struct mlx5_priv *priv = dev->data->dev_private;
696 	struct rte_flow *flow = dev_flow->flow;
697 	const int mark = !!(dev_flow->actions &
698 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
699 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
700 	unsigned int i;
701 
702 	for (i = 0; i != flow->rss.queue_num; ++i) {
703 		int idx = (*flow->rss.queue)[i];
704 		struct mlx5_rxq_ctrl *rxq_ctrl =
705 			container_of((*priv->rxqs)[idx],
706 				     struct mlx5_rxq_ctrl, rxq);
707 
708 		/*
709 		 * To support metadata register copy on Tx loopback,
710 		 * this must be always enabled (metadata may arive
711 		 * from other port - not from local flows only.
712 		 */
713 		if (priv->config.dv_flow_en &&
714 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
715 		    mlx5_flow_ext_mreg_supported(dev)) {
716 			rxq_ctrl->rxq.mark = 1;
717 			rxq_ctrl->flow_mark_n = 1;
718 		} else if (mark) {
719 			rxq_ctrl->rxq.mark = 1;
720 			rxq_ctrl->flow_mark_n++;
721 		}
722 		if (tunnel) {
723 			unsigned int j;
724 
725 			/* Increase the counter matching the flow. */
726 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
727 				if ((tunnels_info[j].tunnel &
728 				     dev_flow->layers) ==
729 				    tunnels_info[j].tunnel) {
730 					rxq_ctrl->flow_tunnels_n[j]++;
731 					break;
732 				}
733 			}
734 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
735 		}
736 	}
737 }
738 
739 /**
740  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
741  *
742  * @param[in] dev
743  *   Pointer to the Ethernet device structure.
744  * @param[in] flow
745  *   Pointer to flow structure.
746  */
747 static void
748 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
749 {
750 	struct mlx5_flow *dev_flow;
751 
752 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
753 		flow_drv_rxq_flags_set(dev, dev_flow);
754 }
755 
756 /**
757  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
758  * device flow if no other flow uses it with the same kind of request.
759  *
760  * @param dev
761  *   Pointer to Ethernet device.
762  * @param[in] dev_flow
763  *   Pointer to the device flow.
764  */
765 static void
766 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
767 {
768 	struct mlx5_priv *priv = dev->data->dev_private;
769 	struct rte_flow *flow = dev_flow->flow;
770 	const int mark = !!(dev_flow->actions &
771 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
772 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
773 	unsigned int i;
774 
775 	assert(dev->data->dev_started);
776 	for (i = 0; i != flow->rss.queue_num; ++i) {
777 		int idx = (*flow->rss.queue)[i];
778 		struct mlx5_rxq_ctrl *rxq_ctrl =
779 			container_of((*priv->rxqs)[idx],
780 				     struct mlx5_rxq_ctrl, rxq);
781 
782 		if (priv->config.dv_flow_en &&
783 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
784 		    mlx5_flow_ext_mreg_supported(dev)) {
785 			rxq_ctrl->rxq.mark = 1;
786 			rxq_ctrl->flow_mark_n = 1;
787 		} else if (mark) {
788 			rxq_ctrl->flow_mark_n--;
789 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
790 		}
791 		if (tunnel) {
792 			unsigned int j;
793 
794 			/* Decrease the counter matching the flow. */
795 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
796 				if ((tunnels_info[j].tunnel &
797 				     dev_flow->layers) ==
798 				    tunnels_info[j].tunnel) {
799 					rxq_ctrl->flow_tunnels_n[j]--;
800 					break;
801 				}
802 			}
803 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
804 		}
805 	}
806 }
807 
808 /**
809  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
810  * @p flow if no other flow uses it with the same kind of request.
811  *
812  * @param dev
813  *   Pointer to Ethernet device.
814  * @param[in] flow
815  *   Pointer to the flow.
816  */
817 static void
818 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
819 {
820 	struct mlx5_flow *dev_flow;
821 
822 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
823 		flow_drv_rxq_flags_trim(dev, dev_flow);
824 }
825 
826 /**
827  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
828  *
829  * @param dev
830  *   Pointer to Ethernet device.
831  */
832 static void
833 flow_rxq_flags_clear(struct rte_eth_dev *dev)
834 {
835 	struct mlx5_priv *priv = dev->data->dev_private;
836 	unsigned int i;
837 
838 	for (i = 0; i != priv->rxqs_n; ++i) {
839 		struct mlx5_rxq_ctrl *rxq_ctrl;
840 		unsigned int j;
841 
842 		if (!(*priv->rxqs)[i])
843 			continue;
844 		rxq_ctrl = container_of((*priv->rxqs)[i],
845 					struct mlx5_rxq_ctrl, rxq);
846 		rxq_ctrl->flow_mark_n = 0;
847 		rxq_ctrl->rxq.mark = 0;
848 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
849 			rxq_ctrl->flow_tunnels_n[j] = 0;
850 		rxq_ctrl->rxq.tunnel = 0;
851 	}
852 }
853 
854 /*
855  * return a pointer to the desired action in the list of actions.
856  *
857  * @param[in] actions
858  *   The list of actions to search the action in.
859  * @param[in] action
860  *   The action to find.
861  *
862  * @return
863  *   Pointer to the action in the list, if found. NULL otherwise.
864  */
865 const struct rte_flow_action *
866 mlx5_flow_find_action(const struct rte_flow_action *actions,
867 		      enum rte_flow_action_type action)
868 {
869 	if (actions == NULL)
870 		return NULL;
871 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
872 		if (actions->type == action)
873 			return actions;
874 	return NULL;
875 }
876 
877 /*
878  * Validate the flag action.
879  *
880  * @param[in] action_flags
881  *   Bit-fields that holds the actions detected until now.
882  * @param[in] attr
883  *   Attributes of flow that includes this action.
884  * @param[out] error
885  *   Pointer to error structure.
886  *
887  * @return
888  *   0 on success, a negative errno value otherwise and rte_errno is set.
889  */
890 int
891 mlx5_flow_validate_action_flag(uint64_t action_flags,
892 			       const struct rte_flow_attr *attr,
893 			       struct rte_flow_error *error)
894 {
895 
896 	if (action_flags & MLX5_FLOW_ACTION_DROP)
897 		return rte_flow_error_set(error, EINVAL,
898 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
899 					  "can't drop and flag in same flow");
900 	if (action_flags & MLX5_FLOW_ACTION_MARK)
901 		return rte_flow_error_set(error, EINVAL,
902 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
903 					  "can't mark and flag in same flow");
904 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
905 		return rte_flow_error_set(error, EINVAL,
906 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
907 					  "can't have 2 flag"
908 					  " actions in same flow");
909 	if (attr->egress)
910 		return rte_flow_error_set(error, ENOTSUP,
911 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
912 					  "flag action not supported for "
913 					  "egress");
914 	return 0;
915 }
916 
917 /*
918  * Validate the mark action.
919  *
920  * @param[in] action
921  *   Pointer to the queue action.
922  * @param[in] action_flags
923  *   Bit-fields that holds the actions detected until now.
924  * @param[in] attr
925  *   Attributes of flow that includes this action.
926  * @param[out] error
927  *   Pointer to error structure.
928  *
929  * @return
930  *   0 on success, a negative errno value otherwise and rte_errno is set.
931  */
932 int
933 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
934 			       uint64_t action_flags,
935 			       const struct rte_flow_attr *attr,
936 			       struct rte_flow_error *error)
937 {
938 	const struct rte_flow_action_mark *mark = action->conf;
939 
940 	if (!mark)
941 		return rte_flow_error_set(error, EINVAL,
942 					  RTE_FLOW_ERROR_TYPE_ACTION,
943 					  action,
944 					  "configuration cannot be null");
945 	if (mark->id >= MLX5_FLOW_MARK_MAX)
946 		return rte_flow_error_set(error, EINVAL,
947 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
948 					  &mark->id,
949 					  "mark id must in 0 <= id < "
950 					  RTE_STR(MLX5_FLOW_MARK_MAX));
951 	if (action_flags & MLX5_FLOW_ACTION_DROP)
952 		return rte_flow_error_set(error, EINVAL,
953 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
954 					  "can't drop and mark in same flow");
955 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
956 		return rte_flow_error_set(error, EINVAL,
957 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
958 					  "can't flag and mark in same flow");
959 	if (action_flags & MLX5_FLOW_ACTION_MARK)
960 		return rte_flow_error_set(error, EINVAL,
961 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
962 					  "can't have 2 mark actions in same"
963 					  " flow");
964 	if (attr->egress)
965 		return rte_flow_error_set(error, ENOTSUP,
966 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
967 					  "mark action not supported for "
968 					  "egress");
969 	return 0;
970 }
971 
972 /*
973  * Validate the drop action.
974  *
975  * @param[in] action_flags
976  *   Bit-fields that holds the actions detected until now.
977  * @param[in] attr
978  *   Attributes of flow that includes this action.
979  * @param[out] error
980  *   Pointer to error structure.
981  *
982  * @return
983  *   0 on success, a negative errno value otherwise and rte_errno is set.
984  */
985 int
986 mlx5_flow_validate_action_drop(uint64_t action_flags,
987 			       const struct rte_flow_attr *attr,
988 			       struct rte_flow_error *error)
989 {
990 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
991 		return rte_flow_error_set(error, EINVAL,
992 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
993 					  "can't drop and flag in same flow");
994 	if (action_flags & MLX5_FLOW_ACTION_MARK)
995 		return rte_flow_error_set(error, EINVAL,
996 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
997 					  "can't drop and mark in same flow");
998 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
999 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1000 		return rte_flow_error_set(error, EINVAL,
1001 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1002 					  "can't have 2 fate actions in"
1003 					  " same flow");
1004 	if (attr->egress)
1005 		return rte_flow_error_set(error, ENOTSUP,
1006 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007 					  "drop action not supported for "
1008 					  "egress");
1009 	return 0;
1010 }
1011 
1012 /*
1013  * Validate the queue action.
1014  *
1015  * @param[in] action
1016  *   Pointer to the queue action.
1017  * @param[in] action_flags
1018  *   Bit-fields that holds the actions detected until now.
1019  * @param[in] dev
1020  *   Pointer to the Ethernet device structure.
1021  * @param[in] attr
1022  *   Attributes of flow that includes this action.
1023  * @param[out] error
1024  *   Pointer to error structure.
1025  *
1026  * @return
1027  *   0 on success, a negative errno value otherwise and rte_errno is set.
1028  */
1029 int
1030 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1031 				uint64_t action_flags,
1032 				struct rte_eth_dev *dev,
1033 				const struct rte_flow_attr *attr,
1034 				struct rte_flow_error *error)
1035 {
1036 	struct mlx5_priv *priv = dev->data->dev_private;
1037 	const struct rte_flow_action_queue *queue = action->conf;
1038 
1039 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1040 		return rte_flow_error_set(error, EINVAL,
1041 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1042 					  "can't have 2 fate actions in"
1043 					  " same flow");
1044 	if (!priv->rxqs_n)
1045 		return rte_flow_error_set(error, EINVAL,
1046 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1047 					  NULL, "No Rx queues configured");
1048 	if (queue->index >= priv->rxqs_n)
1049 		return rte_flow_error_set(error, EINVAL,
1050 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1051 					  &queue->index,
1052 					  "queue index out of range");
1053 	if (!(*priv->rxqs)[queue->index])
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056 					  &queue->index,
1057 					  "queue is not configured");
1058 	if (attr->egress)
1059 		return rte_flow_error_set(error, ENOTSUP,
1060 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1061 					  "queue action not supported for "
1062 					  "egress");
1063 	return 0;
1064 }
1065 
1066 /*
1067  * Validate the rss action.
1068  *
1069  * @param[in] action
1070  *   Pointer to the queue action.
1071  * @param[in] action_flags
1072  *   Bit-fields that holds the actions detected until now.
1073  * @param[in] dev
1074  *   Pointer to the Ethernet device structure.
1075  * @param[in] attr
1076  *   Attributes of flow that includes this action.
1077  * @param[in] item_flags
1078  *   Items that were detected.
1079  * @param[out] error
1080  *   Pointer to error structure.
1081  *
1082  * @return
1083  *   0 on success, a negative errno value otherwise and rte_errno is set.
1084  */
1085 int
1086 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1087 			      uint64_t action_flags,
1088 			      struct rte_eth_dev *dev,
1089 			      const struct rte_flow_attr *attr,
1090 			      uint64_t item_flags,
1091 			      struct rte_flow_error *error)
1092 {
1093 	struct mlx5_priv *priv = dev->data->dev_private;
1094 	const struct rte_flow_action_rss *rss = action->conf;
1095 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1096 	unsigned int i;
1097 
1098 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1099 		return rte_flow_error_set(error, EINVAL,
1100 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1101 					  "can't have 2 fate actions"
1102 					  " in same flow");
1103 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1104 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1105 		return rte_flow_error_set(error, ENOTSUP,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &rss->func,
1108 					  "RSS hash function not supported");
1109 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1110 	if (rss->level > 2)
1111 #else
1112 	if (rss->level > 1)
1113 #endif
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 					  &rss->level,
1117 					  "tunnel RSS is not supported");
1118 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1119 	if (rss->key_len == 0 && rss->key != NULL)
1120 		return rte_flow_error_set(error, ENOTSUP,
1121 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1122 					  &rss->key_len,
1123 					  "RSS hash key length 0");
1124 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127 					  &rss->key_len,
1128 					  "RSS hash key too small");
1129 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1130 		return rte_flow_error_set(error, ENOTSUP,
1131 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1132 					  &rss->key_len,
1133 					  "RSS hash key too large");
1134 	if (rss->queue_num > priv->config.ind_table_max_size)
1135 		return rte_flow_error_set(error, ENOTSUP,
1136 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1137 					  &rss->queue_num,
1138 					  "number of queues too large");
1139 	if (rss->types & MLX5_RSS_HF_MASK)
1140 		return rte_flow_error_set(error, ENOTSUP,
1141 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142 					  &rss->types,
1143 					  "some RSS protocols are not"
1144 					  " supported");
1145 	if (!priv->rxqs_n)
1146 		return rte_flow_error_set(error, EINVAL,
1147 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1148 					  NULL, "No Rx queues configured");
1149 	if (!rss->queue_num)
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1152 					  NULL, "No queues configured");
1153 	for (i = 0; i != rss->queue_num; ++i) {
1154 		if (!(*priv->rxqs)[rss->queue[i]])
1155 			return rte_flow_error_set
1156 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1157 				 &rss->queue[i], "queue is not configured");
1158 	}
1159 	if (attr->egress)
1160 		return rte_flow_error_set(error, ENOTSUP,
1161 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1162 					  "rss action not supported for "
1163 					  "egress");
1164 	if (rss->level > 1 &&  !tunnel)
1165 		return rte_flow_error_set(error, EINVAL,
1166 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1167 					  "inner RSS is not supported for "
1168 					  "non-tunnel flows");
1169 	return 0;
1170 }
1171 
1172 /*
1173  * Validate the count action.
1174  *
1175  * @param[in] dev
1176  *   Pointer to the Ethernet device structure.
1177  * @param[in] attr
1178  *   Attributes of flow that includes this action.
1179  * @param[out] error
1180  *   Pointer to error structure.
1181  *
1182  * @return
1183  *   0 on success, a negative errno value otherwise and rte_errno is set.
1184  */
1185 int
1186 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1187 				const struct rte_flow_attr *attr,
1188 				struct rte_flow_error *error)
1189 {
1190 	if (attr->egress)
1191 		return rte_flow_error_set(error, ENOTSUP,
1192 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1193 					  "count action not supported for "
1194 					  "egress");
1195 	return 0;
1196 }
1197 
1198 /**
1199  * Verify the @p attributes will be correctly understood by the NIC and store
1200  * them in the @p flow if everything is correct.
1201  *
1202  * @param[in] dev
1203  *   Pointer to the Ethernet device structure.
1204  * @param[in] attributes
1205  *   Pointer to flow attributes
1206  * @param[out] error
1207  *   Pointer to error structure.
1208  *
1209  * @return
1210  *   0 on success, a negative errno value otherwise and rte_errno is set.
1211  */
1212 int
1213 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1214 			      const struct rte_flow_attr *attributes,
1215 			      struct rte_flow_error *error)
1216 {
1217 	struct mlx5_priv *priv = dev->data->dev_private;
1218 	uint32_t priority_max = priv->config.flow_prio - 1;
1219 
1220 	if (attributes->group)
1221 		return rte_flow_error_set(error, ENOTSUP,
1222 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1223 					  NULL, "groups is not supported");
1224 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1225 	    attributes->priority >= priority_max)
1226 		return rte_flow_error_set(error, ENOTSUP,
1227 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1228 					  NULL, "priority out of range");
1229 	if (attributes->egress)
1230 		return rte_flow_error_set(error, ENOTSUP,
1231 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1232 					  "egress is not supported");
1233 	if (attributes->transfer && !priv->config.dv_esw_en)
1234 		return rte_flow_error_set(error, ENOTSUP,
1235 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1236 					  NULL, "transfer is not supported");
1237 	if (!attributes->ingress)
1238 		return rte_flow_error_set(error, EINVAL,
1239 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1240 					  NULL,
1241 					  "ingress attribute is mandatory");
1242 	return 0;
1243 }
1244 
1245 /**
1246  * Validate ICMP6 item.
1247  *
1248  * @param[in] item
1249  *   Item specification.
1250  * @param[in] item_flags
1251  *   Bit-fields that holds the items detected until now.
1252  * @param[out] error
1253  *   Pointer to error structure.
1254  *
1255  * @return
1256  *   0 on success, a negative errno value otherwise and rte_errno is set.
1257  */
1258 int
1259 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1260 			       uint64_t item_flags,
1261 			       uint8_t target_protocol,
1262 			       struct rte_flow_error *error)
1263 {
1264 	const struct rte_flow_item_icmp6 *mask = item->mask;
1265 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1266 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1267 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1268 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1269 				      MLX5_FLOW_LAYER_OUTER_L4;
1270 	int ret;
1271 
1272 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1273 		return rte_flow_error_set(error, EINVAL,
1274 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1275 					  "protocol filtering not compatible"
1276 					  " with ICMP6 layer");
1277 	if (!(item_flags & l3m))
1278 		return rte_flow_error_set(error, EINVAL,
1279 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1280 					  "IPv6 is mandatory to filter on"
1281 					  " ICMP6");
1282 	if (item_flags & l4m)
1283 		return rte_flow_error_set(error, EINVAL,
1284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1285 					  "multiple L4 layers not supported");
1286 	if (!mask)
1287 		mask = &rte_flow_item_icmp6_mask;
1288 	ret = mlx5_flow_item_acceptable
1289 		(item, (const uint8_t *)mask,
1290 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1291 		 sizeof(struct rte_flow_item_icmp6), error);
1292 	if (ret < 0)
1293 		return ret;
1294 	return 0;
1295 }
1296 
1297 /**
1298  * Validate ICMP item.
1299  *
1300  * @param[in] item
1301  *   Item specification.
1302  * @param[in] item_flags
1303  *   Bit-fields that holds the items detected until now.
1304  * @param[out] error
1305  *   Pointer to error structure.
1306  *
1307  * @return
1308  *   0 on success, a negative errno value otherwise and rte_errno is set.
1309  */
1310 int
1311 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1312 			     uint64_t item_flags,
1313 			     uint8_t target_protocol,
1314 			     struct rte_flow_error *error)
1315 {
1316 	const struct rte_flow_item_icmp *mask = item->mask;
1317 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1318 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1319 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1320 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1321 				      MLX5_FLOW_LAYER_OUTER_L4;
1322 	int ret;
1323 
1324 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1325 		return rte_flow_error_set(error, EINVAL,
1326 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1327 					  "protocol filtering not compatible"
1328 					  " with ICMP layer");
1329 	if (!(item_flags & l3m))
1330 		return rte_flow_error_set(error, EINVAL,
1331 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1332 					  "IPv4 is mandatory to filter"
1333 					  " on ICMP");
1334 	if (item_flags & l4m)
1335 		return rte_flow_error_set(error, EINVAL,
1336 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1337 					  "multiple L4 layers not supported");
1338 	if (!mask)
1339 		mask = &rte_flow_item_icmp_mask;
1340 	ret = mlx5_flow_item_acceptable
1341 		(item, (const uint8_t *)mask,
1342 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1343 		 sizeof(struct rte_flow_item_icmp), error);
1344 	if (ret < 0)
1345 		return ret;
1346 	return 0;
1347 }
1348 
1349 /**
1350  * Validate Ethernet item.
1351  *
1352  * @param[in] item
1353  *   Item specification.
1354  * @param[in] item_flags
1355  *   Bit-fields that holds the items detected until now.
1356  * @param[out] error
1357  *   Pointer to error structure.
1358  *
1359  * @return
1360  *   0 on success, a negative errno value otherwise and rte_errno is set.
1361  */
1362 int
1363 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1364 			    uint64_t item_flags,
1365 			    struct rte_flow_error *error)
1366 {
1367 	const struct rte_flow_item_eth *mask = item->mask;
1368 	const struct rte_flow_item_eth nic_mask = {
1369 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1370 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1371 		.type = RTE_BE16(0xffff),
1372 	};
1373 	int ret;
1374 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1375 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1376 				       MLX5_FLOW_LAYER_OUTER_L2;
1377 
1378 	if (item_flags & ethm)
1379 		return rte_flow_error_set(error, ENOTSUP,
1380 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1381 					  "multiple L2 layers not supported");
1382 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1383 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1384 		return rte_flow_error_set(error, EINVAL,
1385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1386 					  "L2 layer should not follow "
1387 					  "L3 layers");
1388 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1389 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1390 		return rte_flow_error_set(error, EINVAL,
1391 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1392 					  "L2 layer should not follow VLAN");
1393 	if (!mask)
1394 		mask = &rte_flow_item_eth_mask;
1395 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1396 					(const uint8_t *)&nic_mask,
1397 					sizeof(struct rte_flow_item_eth),
1398 					error);
1399 	return ret;
1400 }
1401 
1402 /**
1403  * Validate VLAN item.
1404  *
1405  * @param[in] item
1406  *   Item specification.
1407  * @param[in] item_flags
1408  *   Bit-fields that holds the items detected until now.
1409  * @param[in] dev
1410  *   Ethernet device flow is being created on.
1411  * @param[out] error
1412  *   Pointer to error structure.
1413  *
1414  * @return
1415  *   0 on success, a negative errno value otherwise and rte_errno is set.
1416  */
1417 int
1418 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1419 			     uint64_t item_flags,
1420 			     struct rte_eth_dev *dev,
1421 			     struct rte_flow_error *error)
1422 {
1423 	const struct rte_flow_item_vlan *spec = item->spec;
1424 	const struct rte_flow_item_vlan *mask = item->mask;
1425 	const struct rte_flow_item_vlan nic_mask = {
1426 		.tci = RTE_BE16(UINT16_MAX),
1427 		.inner_type = RTE_BE16(UINT16_MAX),
1428 	};
1429 	uint16_t vlan_tag = 0;
1430 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1431 	int ret;
1432 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1433 					MLX5_FLOW_LAYER_INNER_L4) :
1434 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1435 					MLX5_FLOW_LAYER_OUTER_L4);
1436 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1437 					MLX5_FLOW_LAYER_OUTER_VLAN;
1438 
1439 	if (item_flags & vlanm)
1440 		return rte_flow_error_set(error, EINVAL,
1441 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1442 					  "multiple VLAN layers not supported");
1443 	else if ((item_flags & l34m) != 0)
1444 		return rte_flow_error_set(error, EINVAL,
1445 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1446 					  "VLAN cannot follow L3/L4 layer");
1447 	if (!mask)
1448 		mask = &rte_flow_item_vlan_mask;
1449 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1450 					(const uint8_t *)&nic_mask,
1451 					sizeof(struct rte_flow_item_vlan),
1452 					error);
1453 	if (ret)
1454 		return ret;
1455 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1456 		struct mlx5_priv *priv = dev->data->dev_private;
1457 
1458 		if (priv->vmwa_context) {
1459 			/*
1460 			 * Non-NULL context means we have a virtual machine
1461 			 * and SR-IOV enabled, we have to create VLAN interface
1462 			 * to make hypervisor to setup E-Switch vport
1463 			 * context correctly. We avoid creating the multiple
1464 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1465 			 */
1466 			return rte_flow_error_set(error, EINVAL,
1467 						  RTE_FLOW_ERROR_TYPE_ITEM,
1468 						  item,
1469 						  "VLAN tag mask is not"
1470 						  " supported in virtual"
1471 						  " environment");
1472 		}
1473 	}
1474 	if (spec) {
1475 		vlan_tag = spec->tci;
1476 		vlan_tag &= mask->tci;
1477 	}
1478 	/*
1479 	 * From verbs perspective an empty VLAN is equivalent
1480 	 * to a packet without VLAN layer.
1481 	 */
1482 	if (!vlan_tag)
1483 		return rte_flow_error_set(error, EINVAL,
1484 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1485 					  item->spec,
1486 					  "VLAN cannot be empty");
1487 	return 0;
1488 }
1489 
1490 /**
1491  * Validate IPV4 item.
1492  *
1493  * @param[in] item
1494  *   Item specification.
1495  * @param[in] item_flags
1496  *   Bit-fields that holds the items detected until now.
1497  * @param[in] acc_mask
1498  *   Acceptable mask, if NULL default internal default mask
1499  *   will be used to check whether item fields are supported.
1500  * @param[out] error
1501  *   Pointer to error structure.
1502  *
1503  * @return
1504  *   0 on success, a negative errno value otherwise and rte_errno is set.
1505  */
1506 int
1507 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1508 			     uint64_t item_flags,
1509 			     uint64_t last_item,
1510 			     uint16_t ether_type,
1511 			     const struct rte_flow_item_ipv4 *acc_mask,
1512 			     struct rte_flow_error *error)
1513 {
1514 	const struct rte_flow_item_ipv4 *mask = item->mask;
1515 	const struct rte_flow_item_ipv4 *spec = item->spec;
1516 	const struct rte_flow_item_ipv4 nic_mask = {
1517 		.hdr = {
1518 			.src_addr = RTE_BE32(0xffffffff),
1519 			.dst_addr = RTE_BE32(0xffffffff),
1520 			.type_of_service = 0xff,
1521 			.next_proto_id = 0xff,
1522 		},
1523 	};
1524 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1525 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1526 				      MLX5_FLOW_LAYER_OUTER_L3;
1527 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1528 				      MLX5_FLOW_LAYER_OUTER_L4;
1529 	int ret;
1530 	uint8_t next_proto = 0xFF;
1531 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1532 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1533 				  MLX5_FLOW_LAYER_INNER_VLAN);
1534 
1535 	if ((last_item & l2_vlan) && ether_type &&
1536 	    ether_type != RTE_ETHER_TYPE_IPV4)
1537 		return rte_flow_error_set(error, EINVAL,
1538 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1539 					  "IPv4 cannot follow L2/VLAN layer "
1540 					  "which ether type is not IPv4");
1541 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1542 		if (mask && spec)
1543 			next_proto = mask->hdr.next_proto_id &
1544 				     spec->hdr.next_proto_id;
1545 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1546 			return rte_flow_error_set(error, EINVAL,
1547 						  RTE_FLOW_ERROR_TYPE_ITEM,
1548 						  item,
1549 						  "multiple tunnel "
1550 						  "not supported");
1551 	}
1552 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1553 		return rte_flow_error_set(error, EINVAL,
1554 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1555 					  "wrong tunnel type - IPv6 specified "
1556 					  "but IPv4 item provided");
1557 	if (item_flags & l3m)
1558 		return rte_flow_error_set(error, ENOTSUP,
1559 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1560 					  "multiple L3 layers not supported");
1561 	else if (item_flags & l4m)
1562 		return rte_flow_error_set(error, EINVAL,
1563 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1564 					  "L3 cannot follow an L4 layer.");
1565 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1566 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1567 		return rte_flow_error_set(error, EINVAL,
1568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1569 					  "L3 cannot follow an NVGRE layer.");
1570 	if (!mask)
1571 		mask = &rte_flow_item_ipv4_mask;
1572 	else if (mask->hdr.next_proto_id != 0 &&
1573 		 mask->hdr.next_proto_id != 0xff)
1574 		return rte_flow_error_set(error, EINVAL,
1575 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1576 					  "partial mask is not supported"
1577 					  " for protocol");
1578 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1579 					acc_mask ? (const uint8_t *)acc_mask
1580 						 : (const uint8_t *)&nic_mask,
1581 					sizeof(struct rte_flow_item_ipv4),
1582 					error);
1583 	if (ret < 0)
1584 		return ret;
1585 	return 0;
1586 }
1587 
1588 /**
1589  * Validate IPV6 item.
1590  *
1591  * @param[in] item
1592  *   Item specification.
1593  * @param[in] item_flags
1594  *   Bit-fields that holds the items detected until now.
1595  * @param[in] acc_mask
1596  *   Acceptable mask, if NULL default internal default mask
1597  *   will be used to check whether item fields are supported.
1598  * @param[out] error
1599  *   Pointer to error structure.
1600  *
1601  * @return
1602  *   0 on success, a negative errno value otherwise and rte_errno is set.
1603  */
1604 int
1605 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1606 			     uint64_t item_flags,
1607 			     uint64_t last_item,
1608 			     uint16_t ether_type,
1609 			     const struct rte_flow_item_ipv6 *acc_mask,
1610 			     struct rte_flow_error *error)
1611 {
1612 	const struct rte_flow_item_ipv6 *mask = item->mask;
1613 	const struct rte_flow_item_ipv6 *spec = item->spec;
1614 	const struct rte_flow_item_ipv6 nic_mask = {
1615 		.hdr = {
1616 			.src_addr =
1617 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1618 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1619 			.dst_addr =
1620 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1621 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1622 			.vtc_flow = RTE_BE32(0xffffffff),
1623 			.proto = 0xff,
1624 			.hop_limits = 0xff,
1625 		},
1626 	};
1627 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1628 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1629 				      MLX5_FLOW_LAYER_OUTER_L3;
1630 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1631 				      MLX5_FLOW_LAYER_OUTER_L4;
1632 	int ret;
1633 	uint8_t next_proto = 0xFF;
1634 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1635 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1636 				  MLX5_FLOW_LAYER_INNER_VLAN);
1637 
1638 	if ((last_item & l2_vlan) && ether_type &&
1639 	    ether_type != RTE_ETHER_TYPE_IPV6)
1640 		return rte_flow_error_set(error, EINVAL,
1641 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1642 					  "IPv6 cannot follow L2/VLAN layer "
1643 					  "which ether type is not IPv6");
1644 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1645 		if (mask && spec)
1646 			next_proto = mask->hdr.proto & spec->hdr.proto;
1647 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1648 			return rte_flow_error_set(error, EINVAL,
1649 						  RTE_FLOW_ERROR_TYPE_ITEM,
1650 						  item,
1651 						  "multiple tunnel "
1652 						  "not supported");
1653 	}
1654 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1655 		return rte_flow_error_set(error, EINVAL,
1656 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1657 					  "wrong tunnel type - IPv4 specified "
1658 					  "but IPv6 item provided");
1659 	if (item_flags & l3m)
1660 		return rte_flow_error_set(error, ENOTSUP,
1661 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1662 					  "multiple L3 layers not supported");
1663 	else if (item_flags & l4m)
1664 		return rte_flow_error_set(error, EINVAL,
1665 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1666 					  "L3 cannot follow an L4 layer.");
1667 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1668 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1669 		return rte_flow_error_set(error, EINVAL,
1670 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1671 					  "L3 cannot follow an NVGRE layer.");
1672 	if (!mask)
1673 		mask = &rte_flow_item_ipv6_mask;
1674 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1675 					acc_mask ? (const uint8_t *)acc_mask
1676 						 : (const uint8_t *)&nic_mask,
1677 					sizeof(struct rte_flow_item_ipv6),
1678 					error);
1679 	if (ret < 0)
1680 		return ret;
1681 	return 0;
1682 }
1683 
1684 /**
1685  * Validate UDP item.
1686  *
1687  * @param[in] item
1688  *   Item specification.
1689  * @param[in] item_flags
1690  *   Bit-fields that holds the items detected until now.
1691  * @param[in] target_protocol
1692  *   The next protocol in the previous item.
1693  * @param[in] flow_mask
1694  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1695  * @param[out] error
1696  *   Pointer to error structure.
1697  *
1698  * @return
1699  *   0 on success, a negative errno value otherwise and rte_errno is set.
1700  */
1701 int
1702 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1703 			    uint64_t item_flags,
1704 			    uint8_t target_protocol,
1705 			    struct rte_flow_error *error)
1706 {
1707 	const struct rte_flow_item_udp *mask = item->mask;
1708 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1709 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1710 				      MLX5_FLOW_LAYER_OUTER_L3;
1711 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1712 				      MLX5_FLOW_LAYER_OUTER_L4;
1713 	int ret;
1714 
1715 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1716 		return rte_flow_error_set(error, EINVAL,
1717 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1718 					  "protocol filtering not compatible"
1719 					  " with UDP layer");
1720 	if (!(item_flags & l3m))
1721 		return rte_flow_error_set(error, EINVAL,
1722 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1723 					  "L3 is mandatory to filter on L4");
1724 	if (item_flags & l4m)
1725 		return rte_flow_error_set(error, EINVAL,
1726 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1727 					  "multiple L4 layers not supported");
1728 	if (!mask)
1729 		mask = &rte_flow_item_udp_mask;
1730 	ret = mlx5_flow_item_acceptable
1731 		(item, (const uint8_t *)mask,
1732 		 (const uint8_t *)&rte_flow_item_udp_mask,
1733 		 sizeof(struct rte_flow_item_udp), error);
1734 	if (ret < 0)
1735 		return ret;
1736 	return 0;
1737 }
1738 
1739 /**
1740  * Validate TCP item.
1741  *
1742  * @param[in] item
1743  *   Item specification.
1744  * @param[in] item_flags
1745  *   Bit-fields that holds the items detected until now.
1746  * @param[in] target_protocol
1747  *   The next protocol in the previous item.
1748  * @param[out] error
1749  *   Pointer to error structure.
1750  *
1751  * @return
1752  *   0 on success, a negative errno value otherwise and rte_errno is set.
1753  */
1754 int
1755 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1756 			    uint64_t item_flags,
1757 			    uint8_t target_protocol,
1758 			    const struct rte_flow_item_tcp *flow_mask,
1759 			    struct rte_flow_error *error)
1760 {
1761 	const struct rte_flow_item_tcp *mask = item->mask;
1762 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1763 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1764 				      MLX5_FLOW_LAYER_OUTER_L3;
1765 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1766 				      MLX5_FLOW_LAYER_OUTER_L4;
1767 	int ret;
1768 
1769 	assert(flow_mask);
1770 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1771 		return rte_flow_error_set(error, EINVAL,
1772 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1773 					  "protocol filtering not compatible"
1774 					  " with TCP layer");
1775 	if (!(item_flags & l3m))
1776 		return rte_flow_error_set(error, EINVAL,
1777 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1778 					  "L3 is mandatory to filter on L4");
1779 	if (item_flags & l4m)
1780 		return rte_flow_error_set(error, EINVAL,
1781 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1782 					  "multiple L4 layers not supported");
1783 	if (!mask)
1784 		mask = &rte_flow_item_tcp_mask;
1785 	ret = mlx5_flow_item_acceptable
1786 		(item, (const uint8_t *)mask,
1787 		 (const uint8_t *)flow_mask,
1788 		 sizeof(struct rte_flow_item_tcp), error);
1789 	if (ret < 0)
1790 		return ret;
1791 	return 0;
1792 }
1793 
1794 /**
1795  * Validate VXLAN item.
1796  *
1797  * @param[in] item
1798  *   Item specification.
1799  * @param[in] item_flags
1800  *   Bit-fields that holds the items detected until now.
1801  * @param[in] target_protocol
1802  *   The next protocol in the previous item.
1803  * @param[out] error
1804  *   Pointer to error structure.
1805  *
1806  * @return
1807  *   0 on success, a negative errno value otherwise and rte_errno is set.
1808  */
1809 int
1810 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1811 			      uint64_t item_flags,
1812 			      struct rte_flow_error *error)
1813 {
1814 	const struct rte_flow_item_vxlan *spec = item->spec;
1815 	const struct rte_flow_item_vxlan *mask = item->mask;
1816 	int ret;
1817 	union vni {
1818 		uint32_t vlan_id;
1819 		uint8_t vni[4];
1820 	} id = { .vlan_id = 0, };
1821 	uint32_t vlan_id = 0;
1822 
1823 
1824 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1825 		return rte_flow_error_set(error, ENOTSUP,
1826 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1827 					  "multiple tunnel layers not"
1828 					  " supported");
1829 	/*
1830 	 * Verify only UDPv4 is present as defined in
1831 	 * https://tools.ietf.org/html/rfc7348
1832 	 */
1833 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1834 		return rte_flow_error_set(error, EINVAL,
1835 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1836 					  "no outer UDP layer found");
1837 	if (!mask)
1838 		mask = &rte_flow_item_vxlan_mask;
1839 	ret = mlx5_flow_item_acceptable
1840 		(item, (const uint8_t *)mask,
1841 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1842 		 sizeof(struct rte_flow_item_vxlan),
1843 		 error);
1844 	if (ret < 0)
1845 		return ret;
1846 	if (spec) {
1847 		memcpy(&id.vni[1], spec->vni, 3);
1848 		vlan_id = id.vlan_id;
1849 		memcpy(&id.vni[1], mask->vni, 3);
1850 		vlan_id &= id.vlan_id;
1851 	}
1852 	/*
1853 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1854 	 * only this layer is defined in the Verbs specification it is
1855 	 * interpreted as wildcard and all packets will match this
1856 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1857 	 * udp), all packets matching the layers before will also
1858 	 * match this rule.  To avoid such situation, VNI 0 is
1859 	 * currently refused.
1860 	 */
1861 	if (!vlan_id)
1862 		return rte_flow_error_set(error, ENOTSUP,
1863 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1864 					  "VXLAN vni cannot be 0");
1865 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1866 		return rte_flow_error_set(error, ENOTSUP,
1867 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1868 					  "VXLAN tunnel must be fully defined");
1869 	return 0;
1870 }
1871 
1872 /**
1873  * Validate VXLAN_GPE item.
1874  *
1875  * @param[in] item
1876  *   Item specification.
1877  * @param[in] item_flags
1878  *   Bit-fields that holds the items detected until now.
1879  * @param[in] priv
1880  *   Pointer to the private data structure.
1881  * @param[in] target_protocol
1882  *   The next protocol in the previous item.
1883  * @param[out] error
1884  *   Pointer to error structure.
1885  *
1886  * @return
1887  *   0 on success, a negative errno value otherwise and rte_errno is set.
1888  */
1889 int
1890 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1891 				  uint64_t item_flags,
1892 				  struct rte_eth_dev *dev,
1893 				  struct rte_flow_error *error)
1894 {
1895 	struct mlx5_priv *priv = dev->data->dev_private;
1896 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1897 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1898 	int ret;
1899 	union vni {
1900 		uint32_t vlan_id;
1901 		uint8_t vni[4];
1902 	} id = { .vlan_id = 0, };
1903 	uint32_t vlan_id = 0;
1904 
1905 	if (!priv->config.l3_vxlan_en)
1906 		return rte_flow_error_set(error, ENOTSUP,
1907 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1908 					  "L3 VXLAN is not enabled by device"
1909 					  " parameter and/or not configured in"
1910 					  " firmware");
1911 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1912 		return rte_flow_error_set(error, ENOTSUP,
1913 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1914 					  "multiple tunnel layers not"
1915 					  " supported");
1916 	/*
1917 	 * Verify only UDPv4 is present as defined in
1918 	 * https://tools.ietf.org/html/rfc7348
1919 	 */
1920 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1921 		return rte_flow_error_set(error, EINVAL,
1922 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1923 					  "no outer UDP layer found");
1924 	if (!mask)
1925 		mask = &rte_flow_item_vxlan_gpe_mask;
1926 	ret = mlx5_flow_item_acceptable
1927 		(item, (const uint8_t *)mask,
1928 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1929 		 sizeof(struct rte_flow_item_vxlan_gpe),
1930 		 error);
1931 	if (ret < 0)
1932 		return ret;
1933 	if (spec) {
1934 		if (spec->protocol)
1935 			return rte_flow_error_set(error, ENOTSUP,
1936 						  RTE_FLOW_ERROR_TYPE_ITEM,
1937 						  item,
1938 						  "VxLAN-GPE protocol"
1939 						  " not supported");
1940 		memcpy(&id.vni[1], spec->vni, 3);
1941 		vlan_id = id.vlan_id;
1942 		memcpy(&id.vni[1], mask->vni, 3);
1943 		vlan_id &= id.vlan_id;
1944 	}
1945 	/*
1946 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1947 	 * layer is defined in the Verbs specification it is interpreted as
1948 	 * wildcard and all packets will match this rule, if it follows a full
1949 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1950 	 * before will also match this rule.  To avoid such situation, VNI 0
1951 	 * is currently refused.
1952 	 */
1953 	if (!vlan_id)
1954 		return rte_flow_error_set(error, ENOTSUP,
1955 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1956 					  "VXLAN-GPE vni cannot be 0");
1957 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1958 		return rte_flow_error_set(error, ENOTSUP,
1959 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1960 					  "VXLAN-GPE tunnel must be fully"
1961 					  " defined");
1962 	return 0;
1963 }
1964 /**
1965  * Validate GRE Key item.
1966  *
1967  * @param[in] item
1968  *   Item specification.
1969  * @param[in] item_flags
1970  *   Bit flags to mark detected items.
1971  * @param[in] gre_item
1972  *   Pointer to gre_item
1973  * @param[out] error
1974  *   Pointer to error structure.
1975  *
1976  * @return
1977  *   0 on success, a negative errno value otherwise and rte_errno is set.
1978  */
1979 int
1980 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1981 				uint64_t item_flags,
1982 				const struct rte_flow_item *gre_item,
1983 				struct rte_flow_error *error)
1984 {
1985 	const rte_be32_t *mask = item->mask;
1986 	int ret = 0;
1987 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1988 	const struct rte_flow_item_gre *gre_spec = gre_item->spec;
1989 	const struct rte_flow_item_gre *gre_mask = gre_item->mask;
1990 
1991 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1992 		return rte_flow_error_set(error, ENOTSUP,
1993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 					  "Multiple GRE key not support");
1995 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1996 		return rte_flow_error_set(error, ENOTSUP,
1997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 					  "No preceding GRE header");
1999 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2000 		return rte_flow_error_set(error, ENOTSUP,
2001 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2002 					  "GRE key following a wrong item");
2003 	if (!gre_mask)
2004 		gre_mask = &rte_flow_item_gre_mask;
2005 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2006 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2007 		return rte_flow_error_set(error, EINVAL,
2008 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 					  "Key bit must be on");
2010 
2011 	if (!mask)
2012 		mask = &gre_key_default_mask;
2013 	ret = mlx5_flow_item_acceptable
2014 		(item, (const uint8_t *)mask,
2015 		 (const uint8_t *)&gre_key_default_mask,
2016 		 sizeof(rte_be32_t), error);
2017 	return ret;
2018 }
2019 
2020 /**
2021  * Validate GRE item.
2022  *
2023  * @param[in] item
2024  *   Item specification.
2025  * @param[in] item_flags
2026  *   Bit flags to mark detected items.
2027  * @param[in] target_protocol
2028  *   The next protocol in the previous item.
2029  * @param[out] error
2030  *   Pointer to error structure.
2031  *
2032  * @return
2033  *   0 on success, a negative errno value otherwise and rte_errno is set.
2034  */
2035 int
2036 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2037 			    uint64_t item_flags,
2038 			    uint8_t target_protocol,
2039 			    struct rte_flow_error *error)
2040 {
2041 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2042 	const struct rte_flow_item_gre *mask = item->mask;
2043 	int ret;
2044 	const struct rte_flow_item_gre nic_mask = {
2045 		.c_rsvd0_ver = RTE_BE16(0xB000),
2046 		.protocol = RTE_BE16(UINT16_MAX),
2047 	};
2048 
2049 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2050 		return rte_flow_error_set(error, EINVAL,
2051 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 					  "protocol filtering not compatible"
2053 					  " with this GRE layer");
2054 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2055 		return rte_flow_error_set(error, ENOTSUP,
2056 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2057 					  "multiple tunnel layers not"
2058 					  " supported");
2059 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2060 		return rte_flow_error_set(error, ENOTSUP,
2061 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2062 					  "L3 Layer is missing");
2063 	if (!mask)
2064 		mask = &rte_flow_item_gre_mask;
2065 	ret = mlx5_flow_item_acceptable
2066 		(item, (const uint8_t *)mask,
2067 		 (const uint8_t *)&nic_mask,
2068 		 sizeof(struct rte_flow_item_gre), error);
2069 	if (ret < 0)
2070 		return ret;
2071 #ifndef HAVE_MLX5DV_DR
2072 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2073 	if (spec && (spec->protocol & mask->protocol))
2074 		return rte_flow_error_set(error, ENOTSUP,
2075 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 					  "without MPLS support the"
2077 					  " specification cannot be used for"
2078 					  " filtering");
2079 #endif
2080 #endif
2081 	return 0;
2082 }
2083 
2084 /**
2085  * Validate Geneve item.
2086  *
2087  * @param[in] item
2088  *   Item specification.
2089  * @param[in] itemFlags
2090  *   Bit-fields that holds the items detected until now.
2091  * @param[in] enPriv
2092  *   Pointer to the private data structure.
2093  * @param[out] error
2094  *   Pointer to error structure.
2095  *
2096  * @return
2097  *   0 on success, a negative errno value otherwise and rte_errno is set.
2098  */
2099 
2100 int
2101 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2102 			       uint64_t item_flags,
2103 			       struct rte_eth_dev *dev,
2104 			       struct rte_flow_error *error)
2105 {
2106 	struct mlx5_priv *priv = dev->data->dev_private;
2107 	const struct rte_flow_item_geneve *spec = item->spec;
2108 	const struct rte_flow_item_geneve *mask = item->mask;
2109 	int ret;
2110 	uint16_t gbhdr;
2111 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2112 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2113 	const struct rte_flow_item_geneve nic_mask = {
2114 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2115 		.vni = "\xff\xff\xff",
2116 		.protocol = RTE_BE16(UINT16_MAX),
2117 	};
2118 
2119 	if (!(priv->config.hca_attr.flex_parser_protocols &
2120 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2121 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2122 		return rte_flow_error_set(error, ENOTSUP,
2123 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2124 					  "L3 Geneve is not enabled by device"
2125 					  " parameter and/or not configured in"
2126 					  " firmware");
2127 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2128 		return rte_flow_error_set(error, ENOTSUP,
2129 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2130 					  "multiple tunnel layers not"
2131 					  " supported");
2132 	/*
2133 	 * Verify only UDPv4 is present as defined in
2134 	 * https://tools.ietf.org/html/rfc7348
2135 	 */
2136 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2137 		return rte_flow_error_set(error, EINVAL,
2138 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2139 					  "no outer UDP layer found");
2140 	if (!mask)
2141 		mask = &rte_flow_item_geneve_mask;
2142 	ret = mlx5_flow_item_acceptable
2143 				  (item, (const uint8_t *)mask,
2144 				   (const uint8_t *)&nic_mask,
2145 				   sizeof(struct rte_flow_item_geneve), error);
2146 	if (ret)
2147 		return ret;
2148 	if (spec) {
2149 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2150 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2151 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2152 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2153 			return rte_flow_error_set(error, ENOTSUP,
2154 						  RTE_FLOW_ERROR_TYPE_ITEM,
2155 						  item,
2156 						  "Geneve protocol unsupported"
2157 						  " fields are being used");
2158 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2159 			return rte_flow_error_set
2160 					(error, ENOTSUP,
2161 					 RTE_FLOW_ERROR_TYPE_ITEM,
2162 					 item,
2163 					 "Unsupported Geneve options length");
2164 	}
2165 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2166 		return rte_flow_error_set
2167 				    (error, ENOTSUP,
2168 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2169 				     "Geneve tunnel must be fully defined");
2170 	return 0;
2171 }
2172 
2173 /**
2174  * Validate MPLS item.
2175  *
2176  * @param[in] dev
2177  *   Pointer to the rte_eth_dev structure.
2178  * @param[in] item
2179  *   Item specification.
2180  * @param[in] item_flags
2181  *   Bit-fields that holds the items detected until now.
2182  * @param[in] prev_layer
2183  *   The protocol layer indicated in previous item.
2184  * @param[out] error
2185  *   Pointer to error structure.
2186  *
2187  * @return
2188  *   0 on success, a negative errno value otherwise and rte_errno is set.
2189  */
2190 int
2191 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2192 			     const struct rte_flow_item *item __rte_unused,
2193 			     uint64_t item_flags __rte_unused,
2194 			     uint64_t prev_layer __rte_unused,
2195 			     struct rte_flow_error *error)
2196 {
2197 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2198 	const struct rte_flow_item_mpls *mask = item->mask;
2199 	struct mlx5_priv *priv = dev->data->dev_private;
2200 	int ret;
2201 
2202 	if (!priv->config.mpls_en)
2203 		return rte_flow_error_set(error, ENOTSUP,
2204 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2205 					  "MPLS not supported or"
2206 					  " disabled in firmware"
2207 					  " configuration.");
2208 	/* MPLS over IP, UDP, GRE is allowed */
2209 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2210 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2211 			    MLX5_FLOW_LAYER_GRE)))
2212 		return rte_flow_error_set(error, EINVAL,
2213 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2214 					  "protocol filtering not compatible"
2215 					  " with MPLS layer");
2216 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2217 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2218 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2219 		return rte_flow_error_set(error, ENOTSUP,
2220 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2221 					  "multiple tunnel layers not"
2222 					  " supported");
2223 	if (!mask)
2224 		mask = &rte_flow_item_mpls_mask;
2225 	ret = mlx5_flow_item_acceptable
2226 		(item, (const uint8_t *)mask,
2227 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2228 		 sizeof(struct rte_flow_item_mpls), error);
2229 	if (ret < 0)
2230 		return ret;
2231 	return 0;
2232 #endif
2233 	return rte_flow_error_set(error, ENOTSUP,
2234 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2235 				  "MPLS is not supported by Verbs, please"
2236 				  " update.");
2237 }
2238 
2239 /**
2240  * Validate NVGRE item.
2241  *
2242  * @param[in] item
2243  *   Item specification.
2244  * @param[in] item_flags
2245  *   Bit flags to mark detected items.
2246  * @param[in] target_protocol
2247  *   The next protocol in the previous item.
2248  * @param[out] error
2249  *   Pointer to error structure.
2250  *
2251  * @return
2252  *   0 on success, a negative errno value otherwise and rte_errno is set.
2253  */
2254 int
2255 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2256 			      uint64_t item_flags,
2257 			      uint8_t target_protocol,
2258 			      struct rte_flow_error *error)
2259 {
2260 	const struct rte_flow_item_nvgre *mask = item->mask;
2261 	int ret;
2262 
2263 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2264 		return rte_flow_error_set(error, EINVAL,
2265 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2266 					  "protocol filtering not compatible"
2267 					  " with this GRE layer");
2268 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2269 		return rte_flow_error_set(error, ENOTSUP,
2270 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 					  "multiple tunnel layers not"
2272 					  " supported");
2273 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2274 		return rte_flow_error_set(error, ENOTSUP,
2275 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 					  "L3 Layer is missing");
2277 	if (!mask)
2278 		mask = &rte_flow_item_nvgre_mask;
2279 	ret = mlx5_flow_item_acceptable
2280 		(item, (const uint8_t *)mask,
2281 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2282 		 sizeof(struct rte_flow_item_nvgre), error);
2283 	if (ret < 0)
2284 		return ret;
2285 	return 0;
2286 }
2287 
2288 /* Allocate unique ID for the split Q/RSS subflows. */
2289 static uint32_t
2290 flow_qrss_get_id(struct rte_eth_dev *dev)
2291 {
2292 	struct mlx5_priv *priv = dev->data->dev_private;
2293 	uint32_t qrss_id, ret;
2294 
2295 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2296 	if (ret)
2297 		return 0;
2298 	assert(qrss_id);
2299 	return qrss_id;
2300 }
2301 
2302 /* Free unique ID for the split Q/RSS subflows. */
2303 static void
2304 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2305 {
2306 	struct mlx5_priv *priv = dev->data->dev_private;
2307 
2308 	if (qrss_id)
2309 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2310 }
2311 
2312 /**
2313  * Release resource related QUEUE/RSS action split.
2314  *
2315  * @param dev
2316  *   Pointer to Ethernet device.
2317  * @param flow
2318  *   Flow to release id's from.
2319  */
2320 static void
2321 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2322 			     struct rte_flow *flow)
2323 {
2324 	struct mlx5_flow *dev_flow;
2325 
2326 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2327 		if (dev_flow->qrss_id)
2328 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2329 }
2330 
2331 static int
2332 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2333 		   const struct rte_flow_attr *attr __rte_unused,
2334 		   const struct rte_flow_item items[] __rte_unused,
2335 		   const struct rte_flow_action actions[] __rte_unused,
2336 		   bool external __rte_unused,
2337 		   struct rte_flow_error *error)
2338 {
2339 	return rte_flow_error_set(error, ENOTSUP,
2340 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2341 }
2342 
2343 static struct mlx5_flow *
2344 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2345 		  const struct rte_flow_item items[] __rte_unused,
2346 		  const struct rte_flow_action actions[] __rte_unused,
2347 		  struct rte_flow_error *error)
2348 {
2349 	rte_flow_error_set(error, ENOTSUP,
2350 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2351 	return NULL;
2352 }
2353 
2354 static int
2355 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2356 		    struct mlx5_flow *dev_flow __rte_unused,
2357 		    const struct rte_flow_attr *attr __rte_unused,
2358 		    const struct rte_flow_item items[] __rte_unused,
2359 		    const struct rte_flow_action actions[] __rte_unused,
2360 		    struct rte_flow_error *error)
2361 {
2362 	return rte_flow_error_set(error, ENOTSUP,
2363 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2364 }
2365 
2366 static int
2367 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2368 		struct rte_flow *flow __rte_unused,
2369 		struct rte_flow_error *error)
2370 {
2371 	return rte_flow_error_set(error, ENOTSUP,
2372 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2373 }
2374 
2375 static void
2376 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2377 		 struct rte_flow *flow __rte_unused)
2378 {
2379 }
2380 
2381 static void
2382 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2383 		  struct rte_flow *flow __rte_unused)
2384 {
2385 }
2386 
2387 static int
2388 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2389 		struct rte_flow *flow __rte_unused,
2390 		const struct rte_flow_action *actions __rte_unused,
2391 		void *data __rte_unused,
2392 		struct rte_flow_error *error)
2393 {
2394 	return rte_flow_error_set(error, ENOTSUP,
2395 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2396 }
2397 
2398 /* Void driver to protect from null pointer reference. */
2399 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2400 	.validate = flow_null_validate,
2401 	.prepare = flow_null_prepare,
2402 	.translate = flow_null_translate,
2403 	.apply = flow_null_apply,
2404 	.remove = flow_null_remove,
2405 	.destroy = flow_null_destroy,
2406 	.query = flow_null_query,
2407 };
2408 
2409 /**
2410  * Select flow driver type according to flow attributes and device
2411  * configuration.
2412  *
2413  * @param[in] dev
2414  *   Pointer to the dev structure.
2415  * @param[in] attr
2416  *   Pointer to the flow attributes.
2417  *
2418  * @return
2419  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2420  */
2421 static enum mlx5_flow_drv_type
2422 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2423 {
2424 	struct mlx5_priv *priv = dev->data->dev_private;
2425 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2426 
2427 	if (attr->transfer && priv->config.dv_esw_en)
2428 		type = MLX5_FLOW_TYPE_DV;
2429 	if (!attr->transfer)
2430 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2431 						 MLX5_FLOW_TYPE_VERBS;
2432 	return type;
2433 }
2434 
2435 #define flow_get_drv_ops(type) flow_drv_ops[type]
2436 
2437 /**
2438  * Flow driver validation API. This abstracts calling driver specific functions.
2439  * The type of flow driver is determined according to flow attributes.
2440  *
2441  * @param[in] dev
2442  *   Pointer to the dev structure.
2443  * @param[in] attr
2444  *   Pointer to the flow attributes.
2445  * @param[in] items
2446  *   Pointer to the list of items.
2447  * @param[in] actions
2448  *   Pointer to the list of actions.
2449  * @param[in] external
2450  *   This flow rule is created by request external to PMD.
2451  * @param[out] error
2452  *   Pointer to the error structure.
2453  *
2454  * @return
2455  *   0 on success, a negative errno value otherwise and rte_errno is set.
2456  */
2457 static inline int
2458 flow_drv_validate(struct rte_eth_dev *dev,
2459 		  const struct rte_flow_attr *attr,
2460 		  const struct rte_flow_item items[],
2461 		  const struct rte_flow_action actions[],
2462 		  bool external, struct rte_flow_error *error)
2463 {
2464 	const struct mlx5_flow_driver_ops *fops;
2465 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2466 
2467 	fops = flow_get_drv_ops(type);
2468 	return fops->validate(dev, attr, items, actions, external, error);
2469 }
2470 
2471 /**
2472  * Flow driver preparation API. This abstracts calling driver specific
2473  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2474  * calculates the size of memory required for device flow, allocates the memory,
2475  * initializes the device flow and returns the pointer.
2476  *
2477  * @note
2478  *   This function initializes device flow structure such as dv or verbs in
2479  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2480  *   rest. For example, adding returning device flow to flow->dev_flow list and
2481  *   setting backward reference to the flow should be done out of this function.
2482  *   layers field is not filled either.
2483  *
2484  * @param[in] attr
2485  *   Pointer to the flow attributes.
2486  * @param[in] items
2487  *   Pointer to the list of items.
2488  * @param[in] actions
2489  *   Pointer to the list of actions.
2490  * @param[out] error
2491  *   Pointer to the error structure.
2492  *
2493  * @return
2494  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2495  */
2496 static inline struct mlx5_flow *
2497 flow_drv_prepare(const struct rte_flow *flow,
2498 		 const struct rte_flow_attr *attr,
2499 		 const struct rte_flow_item items[],
2500 		 const struct rte_flow_action actions[],
2501 		 struct rte_flow_error *error)
2502 {
2503 	const struct mlx5_flow_driver_ops *fops;
2504 	enum mlx5_flow_drv_type type = flow->drv_type;
2505 
2506 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2507 	fops = flow_get_drv_ops(type);
2508 	return fops->prepare(attr, items, actions, error);
2509 }
2510 
2511 /**
2512  * Flow driver translation API. This abstracts calling driver specific
2513  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2514  * translates a generic flow into a driver flow. flow_drv_prepare() must
2515  * precede.
2516  *
2517  * @note
2518  *   dev_flow->layers could be filled as a result of parsing during translation
2519  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2520  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2521  *   flow->actions could be overwritten even though all the expanded dev_flows
2522  *   have the same actions.
2523  *
2524  * @param[in] dev
2525  *   Pointer to the rte dev structure.
2526  * @param[in, out] dev_flow
2527  *   Pointer to the mlx5 flow.
2528  * @param[in] attr
2529  *   Pointer to the flow attributes.
2530  * @param[in] items
2531  *   Pointer to the list of items.
2532  * @param[in] actions
2533  *   Pointer to the list of actions.
2534  * @param[out] error
2535  *   Pointer to the error structure.
2536  *
2537  * @return
2538  *   0 on success, a negative errno value otherwise and rte_errno is set.
2539  */
2540 static inline int
2541 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2542 		   const struct rte_flow_attr *attr,
2543 		   const struct rte_flow_item items[],
2544 		   const struct rte_flow_action actions[],
2545 		   struct rte_flow_error *error)
2546 {
2547 	const struct mlx5_flow_driver_ops *fops;
2548 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2549 
2550 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2551 	fops = flow_get_drv_ops(type);
2552 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2553 }
2554 
2555 /**
2556  * Flow driver apply API. This abstracts calling driver specific functions.
2557  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2558  * translated driver flows on to device. flow_drv_translate() must precede.
2559  *
2560  * @param[in] dev
2561  *   Pointer to Ethernet device structure.
2562  * @param[in, out] flow
2563  *   Pointer to flow structure.
2564  * @param[out] error
2565  *   Pointer to error structure.
2566  *
2567  * @return
2568  *   0 on success, a negative errno value otherwise and rte_errno is set.
2569  */
2570 static inline int
2571 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2572 	       struct rte_flow_error *error)
2573 {
2574 	const struct mlx5_flow_driver_ops *fops;
2575 	enum mlx5_flow_drv_type type = flow->drv_type;
2576 
2577 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2578 	fops = flow_get_drv_ops(type);
2579 	return fops->apply(dev, flow, error);
2580 }
2581 
2582 /**
2583  * Flow driver remove API. This abstracts calling driver specific functions.
2584  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2585  * on device. All the resources of the flow should be freed by calling
2586  * flow_drv_destroy().
2587  *
2588  * @param[in] dev
2589  *   Pointer to Ethernet device.
2590  * @param[in, out] flow
2591  *   Pointer to flow structure.
2592  */
2593 static inline void
2594 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2595 {
2596 	const struct mlx5_flow_driver_ops *fops;
2597 	enum mlx5_flow_drv_type type = flow->drv_type;
2598 
2599 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2600 	fops = flow_get_drv_ops(type);
2601 	fops->remove(dev, flow);
2602 }
2603 
2604 /**
2605  * Flow driver destroy API. This abstracts calling driver specific functions.
2606  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2607  * on device and releases resources of the flow.
2608  *
2609  * @param[in] dev
2610  *   Pointer to Ethernet device.
2611  * @param[in, out] flow
2612  *   Pointer to flow structure.
2613  */
2614 static inline void
2615 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2616 {
2617 	const struct mlx5_flow_driver_ops *fops;
2618 	enum mlx5_flow_drv_type type = flow->drv_type;
2619 
2620 	flow_mreg_split_qrss_release(dev, flow);
2621 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2622 	fops = flow_get_drv_ops(type);
2623 	fops->destroy(dev, flow);
2624 }
2625 
2626 /**
2627  * Validate a flow supported by the NIC.
2628  *
2629  * @see rte_flow_validate()
2630  * @see rte_flow_ops
2631  */
2632 int
2633 mlx5_flow_validate(struct rte_eth_dev *dev,
2634 		   const struct rte_flow_attr *attr,
2635 		   const struct rte_flow_item items[],
2636 		   const struct rte_flow_action actions[],
2637 		   struct rte_flow_error *error)
2638 {
2639 	int ret;
2640 
2641 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2642 	if (ret < 0)
2643 		return ret;
2644 	return 0;
2645 }
2646 
2647 /**
2648  * Get port id item from the item list.
2649  *
2650  * @param[in] item
2651  *   Pointer to the list of items.
2652  *
2653  * @return
2654  *   Pointer to the port id item if exist, else return NULL.
2655  */
2656 static const struct rte_flow_item *
2657 find_port_id_item(const struct rte_flow_item *item)
2658 {
2659 	assert(item);
2660 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2661 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2662 			return item;
2663 	}
2664 	return NULL;
2665 }
2666 
2667 /**
2668  * Get RSS action from the action list.
2669  *
2670  * @param[in] actions
2671  *   Pointer to the list of actions.
2672  *
2673  * @return
2674  *   Pointer to the RSS action if exist, else return NULL.
2675  */
2676 static const struct rte_flow_action_rss*
2677 flow_get_rss_action(const struct rte_flow_action actions[])
2678 {
2679 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2680 		switch (actions->type) {
2681 		case RTE_FLOW_ACTION_TYPE_RSS:
2682 			return (const struct rte_flow_action_rss *)
2683 			       actions->conf;
2684 		default:
2685 			break;
2686 		}
2687 	}
2688 	return NULL;
2689 }
2690 
2691 static unsigned int
2692 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2693 {
2694 	const struct rte_flow_item *item;
2695 	unsigned int has_vlan = 0;
2696 
2697 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2698 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2699 			has_vlan = 1;
2700 			break;
2701 		}
2702 	}
2703 	if (has_vlan)
2704 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2705 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2706 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2707 			       MLX5_EXPANSION_ROOT_OUTER;
2708 }
2709 
2710 /**
2711  * Get QUEUE/RSS action from the action list.
2712  *
2713  * @param[in] actions
2714  *   Pointer to the list of actions.
2715  * @param[out] qrss
2716  *   Pointer to the return pointer.
2717  * @param[out] qrss_type
2718  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2719  *   if no QUEUE/RSS is found.
2720  *
2721  * @return
2722  *   Total number of actions.
2723  */
2724 static int
2725 flow_parse_qrss_action(const struct rte_flow_action actions[],
2726 		       const struct rte_flow_action **qrss)
2727 {
2728 	int actions_n = 0;
2729 
2730 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2731 		switch (actions->type) {
2732 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2733 		case RTE_FLOW_ACTION_TYPE_RSS:
2734 			*qrss = actions;
2735 			break;
2736 		default:
2737 			break;
2738 		}
2739 		actions_n++;
2740 	}
2741 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2742 	return actions_n + 1;
2743 }
2744 
2745 /**
2746  * Check meter action from the action list.
2747  *
2748  * @param[in] actions
2749  *   Pointer to the list of actions.
2750  * @param[out] mtr
2751  *   Pointer to the meter exist flag.
2752  *
2753  * @return
2754  *   Total number of actions.
2755  */
2756 static int
2757 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2758 {
2759 	int actions_n = 0;
2760 
2761 	assert(mtr);
2762 	*mtr = 0;
2763 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2764 		switch (actions->type) {
2765 		case RTE_FLOW_ACTION_TYPE_METER:
2766 			*mtr = 1;
2767 			break;
2768 		default:
2769 			break;
2770 		}
2771 		actions_n++;
2772 	}
2773 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2774 	return actions_n + 1;
2775 }
2776 
2777 /**
2778  * Check if the flow should be splited due to hairpin.
2779  * The reason for the split is that in current HW we can't
2780  * support encap on Rx, so if a flow have encap we move it
2781  * to Tx.
2782  *
2783  * @param dev
2784  *   Pointer to Ethernet device.
2785  * @param[in] attr
2786  *   Flow rule attributes.
2787  * @param[in] actions
2788  *   Associated actions (list terminated by the END action).
2789  *
2790  * @return
2791  *   > 0 the number of actions and the flow should be split,
2792  *   0 when no split required.
2793  */
2794 static int
2795 flow_check_hairpin_split(struct rte_eth_dev *dev,
2796 			 const struct rte_flow_attr *attr,
2797 			 const struct rte_flow_action actions[])
2798 {
2799 	int queue_action = 0;
2800 	int action_n = 0;
2801 	int encap = 0;
2802 	const struct rte_flow_action_queue *queue;
2803 	const struct rte_flow_action_rss *rss;
2804 	const struct rte_flow_action_raw_encap *raw_encap;
2805 
2806 	if (!attr->ingress)
2807 		return 0;
2808 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2809 		switch (actions->type) {
2810 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2811 			queue = actions->conf;
2812 			if (mlx5_rxq_get_type(dev, queue->index) !=
2813 			    MLX5_RXQ_TYPE_HAIRPIN)
2814 				return 0;
2815 			queue_action = 1;
2816 			action_n++;
2817 			break;
2818 		case RTE_FLOW_ACTION_TYPE_RSS:
2819 			rss = actions->conf;
2820 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2821 			    MLX5_RXQ_TYPE_HAIRPIN)
2822 				return 0;
2823 			queue_action = 1;
2824 			action_n++;
2825 			break;
2826 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2827 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2828 			encap = 1;
2829 			action_n++;
2830 			break;
2831 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2832 			raw_encap = actions->conf;
2833 			if (raw_encap->size >
2834 			    (sizeof(struct rte_flow_item_eth) +
2835 			     sizeof(struct rte_flow_item_ipv4)))
2836 				encap = 1;
2837 			action_n++;
2838 			break;
2839 		default:
2840 			action_n++;
2841 			break;
2842 		}
2843 	}
2844 	if (encap == 1 && queue_action)
2845 		return action_n;
2846 	return 0;
2847 }
2848 
2849 /* Declare flow create/destroy prototype in advance. */
2850 static struct rte_flow *
2851 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2852 		 const struct rte_flow_attr *attr,
2853 		 const struct rte_flow_item items[],
2854 		 const struct rte_flow_action actions[],
2855 		 bool external, struct rte_flow_error *error);
2856 
2857 static void
2858 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2859 		  struct rte_flow *flow);
2860 
2861 /**
2862  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2863  *
2864  * As mark_id is unique, if there's already a registered flow for the mark_id,
2865  * return by increasing the reference counter of the resource. Otherwise, create
2866  * the resource (mcp_res) and flow.
2867  *
2868  * Flow looks like,
2869  *   - If ingress port is ANY and reg_c[1] is mark_id,
2870  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2871  *
2872  * For default flow (zero mark_id), flow is like,
2873  *   - If ingress port is ANY,
2874  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2875  *
2876  * @param dev
2877  *   Pointer to Ethernet device.
2878  * @param mark_id
2879  *   ID of MARK action, zero means default flow for META.
2880  * @param[out] error
2881  *   Perform verbose error reporting if not NULL.
2882  *
2883  * @return
2884  *   Associated resource on success, NULL otherwise and rte_errno is set.
2885  */
2886 static struct mlx5_flow_mreg_copy_resource *
2887 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2888 			  struct rte_flow_error *error)
2889 {
2890 	struct mlx5_priv *priv = dev->data->dev_private;
2891 	struct rte_flow_attr attr = {
2892 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2893 		.ingress = 1,
2894 	};
2895 	struct mlx5_rte_flow_item_tag tag_spec = {
2896 		.data = mark_id,
2897 	};
2898 	struct rte_flow_item items[] = {
2899 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2900 	};
2901 	struct rte_flow_action_mark ftag = {
2902 		.id = mark_id,
2903 	};
2904 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2905 		.dst = REG_B,
2906 		.src = 0,
2907 	};
2908 	struct rte_flow_action_jump jump = {
2909 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2910 	};
2911 	struct rte_flow_action actions[] = {
2912 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2913 	};
2914 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2915 	int ret;
2916 
2917 	/* Fill the register fileds in the flow. */
2918 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2919 	if (ret < 0)
2920 		return NULL;
2921 	tag_spec.id = ret;
2922 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2923 	if (ret < 0)
2924 		return NULL;
2925 	cp_mreg.src = ret;
2926 	/* Check if already registered. */
2927 	assert(priv->mreg_cp_tbl);
2928 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2929 	if (mcp_res) {
2930 		/* For non-default rule. */
2931 		if (mark_id)
2932 			mcp_res->refcnt++;
2933 		assert(mark_id || mcp_res->refcnt == 1);
2934 		return mcp_res;
2935 	}
2936 	/* Provide the full width of FLAG specific value. */
2937 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2938 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2939 	/* Build a new flow. */
2940 	if (mark_id) {
2941 		items[0] = (struct rte_flow_item){
2942 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2943 			.spec = &tag_spec,
2944 		};
2945 		items[1] = (struct rte_flow_item){
2946 			.type = RTE_FLOW_ITEM_TYPE_END,
2947 		};
2948 		actions[0] = (struct rte_flow_action){
2949 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2950 			.conf = &ftag,
2951 		};
2952 		actions[1] = (struct rte_flow_action){
2953 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2954 			.conf = &cp_mreg,
2955 		};
2956 		actions[2] = (struct rte_flow_action){
2957 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2958 			.conf = &jump,
2959 		};
2960 		actions[3] = (struct rte_flow_action){
2961 			.type = RTE_FLOW_ACTION_TYPE_END,
2962 		};
2963 	} else {
2964 		/* Default rule, wildcard match. */
2965 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2966 		items[0] = (struct rte_flow_item){
2967 			.type = RTE_FLOW_ITEM_TYPE_END,
2968 		};
2969 		actions[0] = (struct rte_flow_action){
2970 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2971 			.conf = &cp_mreg,
2972 		};
2973 		actions[1] = (struct rte_flow_action){
2974 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2975 			.conf = &jump,
2976 		};
2977 		actions[2] = (struct rte_flow_action){
2978 			.type = RTE_FLOW_ACTION_TYPE_END,
2979 		};
2980 	}
2981 	/* Build a new entry. */
2982 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
2983 	if (!mcp_res) {
2984 		rte_errno = ENOMEM;
2985 		return NULL;
2986 	}
2987 	/*
2988 	 * The copy Flows are not included in any list. There
2989 	 * ones are referenced from other Flows and can not
2990 	 * be applied, removed, deleted in ardbitrary order
2991 	 * by list traversing.
2992 	 */
2993 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
2994 					 actions, false, error);
2995 	if (!mcp_res->flow)
2996 		goto error;
2997 	mcp_res->refcnt++;
2998 	mcp_res->hlist_ent.key = mark_id;
2999 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3000 				&mcp_res->hlist_ent);
3001 	assert(!ret);
3002 	if (ret)
3003 		goto error;
3004 	return mcp_res;
3005 error:
3006 	if (mcp_res->flow)
3007 		flow_list_destroy(dev, NULL, mcp_res->flow);
3008 	rte_free(mcp_res);
3009 	return NULL;
3010 }
3011 
3012 /**
3013  * Release flow in RX_CP_TBL.
3014  *
3015  * @param dev
3016  *   Pointer to Ethernet device.
3017  * @flow
3018  *   Parent flow for wich copying is provided.
3019  */
3020 static void
3021 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3022 			  struct rte_flow *flow)
3023 {
3024 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3025 	struct mlx5_priv *priv = dev->data->dev_private;
3026 
3027 	if (!mcp_res || !priv->mreg_cp_tbl)
3028 		return;
3029 	if (flow->copy_applied) {
3030 		assert(mcp_res->appcnt);
3031 		flow->copy_applied = 0;
3032 		--mcp_res->appcnt;
3033 		if (!mcp_res->appcnt)
3034 			flow_drv_remove(dev, mcp_res->flow);
3035 	}
3036 	/*
3037 	 * We do not check availability of metadata registers here,
3038 	 * because copy resources are allocated in this case.
3039 	 */
3040 	if (--mcp_res->refcnt)
3041 		return;
3042 	assert(mcp_res->flow);
3043 	flow_list_destroy(dev, NULL, mcp_res->flow);
3044 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3045 	rte_free(mcp_res);
3046 	flow->mreg_copy = NULL;
3047 }
3048 
3049 /**
3050  * Start flow in RX_CP_TBL.
3051  *
3052  * @param dev
3053  *   Pointer to Ethernet device.
3054  * @flow
3055  *   Parent flow for wich copying is provided.
3056  *
3057  * @return
3058  *   0 on success, a negative errno value otherwise and rte_errno is set.
3059  */
3060 static int
3061 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3062 			    struct rte_flow *flow)
3063 {
3064 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3065 	int ret;
3066 
3067 	if (!mcp_res || flow->copy_applied)
3068 		return 0;
3069 	if (!mcp_res->appcnt) {
3070 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3071 		if (ret)
3072 			return ret;
3073 	}
3074 	++mcp_res->appcnt;
3075 	flow->copy_applied = 1;
3076 	return 0;
3077 }
3078 
3079 /**
3080  * Stop flow in RX_CP_TBL.
3081  *
3082  * @param dev
3083  *   Pointer to Ethernet device.
3084  * @flow
3085  *   Parent flow for wich copying is provided.
3086  */
3087 static void
3088 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3089 			   struct rte_flow *flow)
3090 {
3091 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3092 
3093 	if (!mcp_res || !flow->copy_applied)
3094 		return;
3095 	assert(mcp_res->appcnt);
3096 	--mcp_res->appcnt;
3097 	flow->copy_applied = 0;
3098 	if (!mcp_res->appcnt)
3099 		flow_drv_remove(dev, mcp_res->flow);
3100 }
3101 
3102 /**
3103  * Remove the default copy action from RX_CP_TBL.
3104  *
3105  * @param dev
3106  *   Pointer to Ethernet device.
3107  */
3108 static void
3109 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3110 {
3111 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3112 	struct mlx5_priv *priv = dev->data->dev_private;
3113 
3114 	/* Check if default flow is registered. */
3115 	if (!priv->mreg_cp_tbl)
3116 		return;
3117 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, 0ULL);
3118 	if (!mcp_res)
3119 		return;
3120 	assert(mcp_res->flow);
3121 	flow_list_destroy(dev, NULL, mcp_res->flow);
3122 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3123 	rte_free(mcp_res);
3124 }
3125 
3126 /**
3127  * Add the default copy action in in RX_CP_TBL.
3128  *
3129  * @param dev
3130  *   Pointer to Ethernet device.
3131  * @param[out] error
3132  *   Perform verbose error reporting if not NULL.
3133  *
3134  * @return
3135  *   0 for success, negative value otherwise and rte_errno is set.
3136  */
3137 static int
3138 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3139 				  struct rte_flow_error *error)
3140 {
3141 	struct mlx5_priv *priv = dev->data->dev_private;
3142 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3143 
3144 	/* Check whether extensive metadata feature is engaged. */
3145 	if (!priv->config.dv_flow_en ||
3146 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3147 	    !mlx5_flow_ext_mreg_supported(dev) ||
3148 	    !priv->sh->dv_regc0_mask)
3149 		return 0;
3150 	mcp_res = flow_mreg_add_copy_action(dev, 0, error);
3151 	if (!mcp_res)
3152 		return -rte_errno;
3153 	return 0;
3154 }
3155 
3156 /**
3157  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3158  *
3159  * All the flow having Q/RSS action should be split by
3160  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3161  * performs the following,
3162  *   - CQE->flow_tag := reg_c[1] (MARK)
3163  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3164  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3165  * but there should be a flow per each MARK ID set by MARK action.
3166  *
3167  * For the aforementioned reason, if there's a MARK action in flow's action
3168  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3169  * the MARK ID to CQE's flow_tag like,
3170  *   - If reg_c[1] is mark_id,
3171  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3172  *
3173  * For SET_META action which stores value in reg_c[0], as the destination is
3174  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3175  * MARK ID means the default flow. The default flow looks like,
3176  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3177  *
3178  * @param dev
3179  *   Pointer to Ethernet device.
3180  * @param flow
3181  *   Pointer to flow structure.
3182  * @param[in] actions
3183  *   Pointer to the list of actions.
3184  * @param[out] error
3185  *   Perform verbose error reporting if not NULL.
3186  *
3187  * @return
3188  *   0 on success, negative value otherwise and rte_errno is set.
3189  */
3190 static int
3191 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3192 			    struct rte_flow *flow,
3193 			    const struct rte_flow_action *actions,
3194 			    struct rte_flow_error *error)
3195 {
3196 	struct mlx5_priv *priv = dev->data->dev_private;
3197 	struct mlx5_dev_config *config = &priv->config;
3198 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3199 	const struct rte_flow_action_mark *mark;
3200 
3201 	/* Check whether extensive metadata feature is engaged. */
3202 	if (!config->dv_flow_en ||
3203 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3204 	    !mlx5_flow_ext_mreg_supported(dev) ||
3205 	    !priv->sh->dv_regc0_mask)
3206 		return 0;
3207 	/* Find MARK action. */
3208 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3209 		switch (actions->type) {
3210 		case RTE_FLOW_ACTION_TYPE_FLAG:
3211 			mcp_res = flow_mreg_add_copy_action
3212 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3213 			if (!mcp_res)
3214 				return -rte_errno;
3215 			flow->mreg_copy = mcp_res;
3216 			if (dev->data->dev_started) {
3217 				mcp_res->appcnt++;
3218 				flow->copy_applied = 1;
3219 			}
3220 			return 0;
3221 		case RTE_FLOW_ACTION_TYPE_MARK:
3222 			mark = (const struct rte_flow_action_mark *)
3223 				actions->conf;
3224 			mcp_res =
3225 				flow_mreg_add_copy_action(dev, mark->id, error);
3226 			if (!mcp_res)
3227 				return -rte_errno;
3228 			flow->mreg_copy = mcp_res;
3229 			if (dev->data->dev_started) {
3230 				mcp_res->appcnt++;
3231 				flow->copy_applied = 1;
3232 			}
3233 			return 0;
3234 		default:
3235 			break;
3236 		}
3237 	}
3238 	return 0;
3239 }
3240 
3241 #define MLX5_MAX_SPLIT_ACTIONS 24
3242 #define MLX5_MAX_SPLIT_ITEMS 24
3243 
3244 /**
3245  * Split the hairpin flow.
3246  * Since HW can't support encap on Rx we move the encap to Tx.
3247  * If the count action is after the encap then we also
3248  * move the count action. in this case the count will also measure
3249  * the outer bytes.
3250  *
3251  * @param dev
3252  *   Pointer to Ethernet device.
3253  * @param[in] actions
3254  *   Associated actions (list terminated by the END action).
3255  * @param[out] actions_rx
3256  *   Rx flow actions.
3257  * @param[out] actions_tx
3258  *   Tx flow actions..
3259  * @param[out] pattern_tx
3260  *   The pattern items for the Tx flow.
3261  * @param[out] flow_id
3262  *   The flow ID connected to this flow.
3263  *
3264  * @return
3265  *   0 on success.
3266  */
3267 static int
3268 flow_hairpin_split(struct rte_eth_dev *dev,
3269 		   const struct rte_flow_action actions[],
3270 		   struct rte_flow_action actions_rx[],
3271 		   struct rte_flow_action actions_tx[],
3272 		   struct rte_flow_item pattern_tx[],
3273 		   uint32_t *flow_id)
3274 {
3275 	struct mlx5_priv *priv = dev->data->dev_private;
3276 	const struct rte_flow_action_raw_encap *raw_encap;
3277 	const struct rte_flow_action_raw_decap *raw_decap;
3278 	struct mlx5_rte_flow_action_set_tag *set_tag;
3279 	struct rte_flow_action *tag_action;
3280 	struct mlx5_rte_flow_item_tag *tag_item;
3281 	struct rte_flow_item *item;
3282 	char *addr;
3283 	int encap = 0;
3284 
3285 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3286 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3287 		switch (actions->type) {
3288 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3289 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3290 			rte_memcpy(actions_tx, actions,
3291 			       sizeof(struct rte_flow_action));
3292 			actions_tx++;
3293 			break;
3294 		case RTE_FLOW_ACTION_TYPE_COUNT:
3295 			if (encap) {
3296 				rte_memcpy(actions_tx, actions,
3297 					   sizeof(struct rte_flow_action));
3298 				actions_tx++;
3299 			} else {
3300 				rte_memcpy(actions_rx, actions,
3301 					   sizeof(struct rte_flow_action));
3302 				actions_rx++;
3303 			}
3304 			break;
3305 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3306 			raw_encap = actions->conf;
3307 			if (raw_encap->size >
3308 			    (sizeof(struct rte_flow_item_eth) +
3309 			     sizeof(struct rte_flow_item_ipv4))) {
3310 				memcpy(actions_tx, actions,
3311 				       sizeof(struct rte_flow_action));
3312 				actions_tx++;
3313 				encap = 1;
3314 			} else {
3315 				rte_memcpy(actions_rx, actions,
3316 					   sizeof(struct rte_flow_action));
3317 				actions_rx++;
3318 			}
3319 			break;
3320 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3321 			raw_decap = actions->conf;
3322 			if (raw_decap->size <
3323 			    (sizeof(struct rte_flow_item_eth) +
3324 			     sizeof(struct rte_flow_item_ipv4))) {
3325 				memcpy(actions_tx, actions,
3326 				       sizeof(struct rte_flow_action));
3327 				actions_tx++;
3328 			} else {
3329 				rte_memcpy(actions_rx, actions,
3330 					   sizeof(struct rte_flow_action));
3331 				actions_rx++;
3332 			}
3333 			break;
3334 		default:
3335 			rte_memcpy(actions_rx, actions,
3336 				   sizeof(struct rte_flow_action));
3337 			actions_rx++;
3338 			break;
3339 		}
3340 	}
3341 	/* Add set meta action and end action for the Rx flow. */
3342 	tag_action = actions_rx;
3343 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3344 	actions_rx++;
3345 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3346 	actions_rx++;
3347 	set_tag = (void *)actions_rx;
3348 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3349 	assert(set_tag->id > REG_NONE);
3350 	set_tag->data = *flow_id;
3351 	tag_action->conf = set_tag;
3352 	/* Create Tx item list. */
3353 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3354 	addr = (void *)&pattern_tx[2];
3355 	item = pattern_tx;
3356 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3357 	tag_item = (void *)addr;
3358 	tag_item->data = *flow_id;
3359 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3360 	assert(set_tag->id > REG_NONE);
3361 	item->spec = tag_item;
3362 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3363 	tag_item = (void *)addr;
3364 	tag_item->data = UINT32_MAX;
3365 	tag_item->id = UINT16_MAX;
3366 	item->mask = tag_item;
3367 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3368 	item->last = NULL;
3369 	item++;
3370 	item->type = RTE_FLOW_ITEM_TYPE_END;
3371 	return 0;
3372 }
3373 
3374 /**
3375  * The last stage of splitting chain, just creates the subflow
3376  * without any modification.
3377  *
3378  * @param dev
3379  *   Pointer to Ethernet device.
3380  * @param[in] flow
3381  *   Parent flow structure pointer.
3382  * @param[in, out] sub_flow
3383  *   Pointer to return the created subflow, may be NULL.
3384  * @param[in] attr
3385  *   Flow rule attributes.
3386  * @param[in] items
3387  *   Pattern specification (list terminated by the END pattern item).
3388  * @param[in] actions
3389  *   Associated actions (list terminated by the END action).
3390  * @param[in] external
3391  *   This flow rule is created by request external to PMD.
3392  * @param[out] error
3393  *   Perform verbose error reporting if not NULL.
3394  * @return
3395  *   0 on success, negative value otherwise
3396  */
3397 static int
3398 flow_create_split_inner(struct rte_eth_dev *dev,
3399 			struct rte_flow *flow,
3400 			struct mlx5_flow **sub_flow,
3401 			const struct rte_flow_attr *attr,
3402 			const struct rte_flow_item items[],
3403 			const struct rte_flow_action actions[],
3404 			bool external, struct rte_flow_error *error)
3405 {
3406 	struct mlx5_flow *dev_flow;
3407 
3408 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3409 	if (!dev_flow)
3410 		return -rte_errno;
3411 	dev_flow->flow = flow;
3412 	dev_flow->external = external;
3413 	/* Subflow object was created, we must include one in the list. */
3414 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3415 	if (sub_flow)
3416 		*sub_flow = dev_flow;
3417 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3418 }
3419 
3420 /**
3421  * Split the meter flow.
3422  *
3423  * As meter flow will split to three sub flow, other than meter
3424  * action, the other actions make sense to only meter accepts
3425  * the packet. If it need to be dropped, no other additional
3426  * actions should be take.
3427  *
3428  * One kind of special action which decapsulates the L3 tunnel
3429  * header will be in the prefix sub flow, as not to take the
3430  * L3 tunnel header into account.
3431  *
3432  * @param dev
3433  *   Pointer to Ethernet device.
3434  * @param[in] actions
3435  *   Associated actions (list terminated by the END action).
3436  * @param[out] actions_sfx
3437  *   Suffix flow actions.
3438  * @param[out] actions_pre
3439  *   Prefix flow actions.
3440  * @param[out] pattern_sfx
3441  *   The pattern items for the suffix flow.
3442  * @param[out] tag_sfx
3443  *   Pointer to suffix flow tag.
3444  *
3445  * @return
3446  *   0 on success.
3447  */
3448 static int
3449 flow_meter_split_prep(struct rte_eth_dev *dev,
3450 		 const struct rte_flow_action actions[],
3451 		 struct rte_flow_action actions_sfx[],
3452 		 struct rte_flow_action actions_pre[])
3453 {
3454 	struct rte_flow_action *tag_action;
3455 	struct mlx5_rte_flow_action_set_tag *set_tag;
3456 	struct rte_flow_error error;
3457 	const struct rte_flow_action_raw_encap *raw_encap;
3458 	const struct rte_flow_action_raw_decap *raw_decap;
3459 	uint32_t tag_id;
3460 
3461 	/* Add the extra tag action first. */
3462 	tag_action = actions_pre;
3463 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3464 	actions_pre++;
3465 	/* Prepare the actions for prefix and suffix flow. */
3466 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3467 		switch (actions->type) {
3468 		case RTE_FLOW_ACTION_TYPE_METER:
3469 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3470 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3471 			memcpy(actions_pre, actions,
3472 			       sizeof(struct rte_flow_action));
3473 			actions_pre++;
3474 			break;
3475 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3476 			raw_encap = actions->conf;
3477 			if (raw_encap->size >
3478 			    (sizeof(struct rte_flow_item_eth) +
3479 			     sizeof(struct rte_flow_item_ipv4))) {
3480 				memcpy(actions_sfx, actions,
3481 				       sizeof(struct rte_flow_action));
3482 				actions_sfx++;
3483 			} else {
3484 				rte_memcpy(actions_pre, actions,
3485 					   sizeof(struct rte_flow_action));
3486 				actions_pre++;
3487 			}
3488 			break;
3489 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3490 			raw_decap = actions->conf;
3491 			/* Size 0 decap means 50 bytes as vxlan decap. */
3492 			if (raw_decap->size && (raw_decap->size <
3493 			    (sizeof(struct rte_flow_item_eth) +
3494 			     sizeof(struct rte_flow_item_ipv4)))) {
3495 				memcpy(actions_sfx, actions,
3496 				       sizeof(struct rte_flow_action));
3497 				actions_sfx++;
3498 			} else {
3499 				rte_memcpy(actions_pre, actions,
3500 					   sizeof(struct rte_flow_action));
3501 				actions_pre++;
3502 			}
3503 			break;
3504 		default:
3505 			memcpy(actions_sfx, actions,
3506 				sizeof(struct rte_flow_action));
3507 			actions_sfx++;
3508 			break;
3509 		}
3510 	}
3511 	/* Add end action to the actions. */
3512 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3513 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3514 	actions_pre++;
3515 	/* Set the tag. */
3516 	set_tag = (void *)actions_pre;
3517 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3518 	/*
3519 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3520 	 */
3521 	tag_id = flow_qrss_get_id(dev);
3522 	set_tag->data = rte_cpu_to_be_32(tag_id);
3523 	tag_action->conf = set_tag;
3524 	return tag_id;
3525 }
3526 
3527 /**
3528  * Split action list having QUEUE/RSS for metadata register copy.
3529  *
3530  * Once Q/RSS action is detected in user's action list, the flow action
3531  * should be split in order to copy metadata registers, which will happen in
3532  * RX_CP_TBL like,
3533  *   - CQE->flow_tag := reg_c[1] (MARK)
3534  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3535  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3536  * This is because the last action of each flow must be a terminal action
3537  * (QUEUE, RSS or DROP).
3538  *
3539  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3540  * stored and kept in the mlx5_flow structure per each sub_flow.
3541  *
3542  * The Q/RSS action is replaced with,
3543  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3544  * And the following JUMP action is added at the end,
3545  *   - JUMP, to RX_CP_TBL.
3546  *
3547  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3548  * flow_create_split_metadata() routine. The flow will look like,
3549  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3550  *
3551  * @param dev
3552  *   Pointer to Ethernet device.
3553  * @param[out] split_actions
3554  *   Pointer to store split actions to jump to CP_TBL.
3555  * @param[in] actions
3556  *   Pointer to the list of original flow actions.
3557  * @param[in] qrss
3558  *   Pointer to the Q/RSS action.
3559  * @param[in] actions_n
3560  *   Number of original actions.
3561  * @param[out] error
3562  *   Perform verbose error reporting if not NULL.
3563  *
3564  * @return
3565  *   non-zero unique flow_id on success, otherwise 0 and
3566  *   error/rte_error are set.
3567  */
3568 static uint32_t
3569 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3570 			  struct rte_flow_action *split_actions,
3571 			  const struct rte_flow_action *actions,
3572 			  const struct rte_flow_action *qrss,
3573 			  int actions_n, struct rte_flow_error *error)
3574 {
3575 	struct mlx5_rte_flow_action_set_tag *set_tag;
3576 	struct rte_flow_action_jump *jump;
3577 	const int qrss_idx = qrss - actions;
3578 	uint32_t flow_id = 0;
3579 	int ret = 0;
3580 
3581 	/*
3582 	 * Given actions will be split
3583 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3584 	 * - Add jump to mreg CP_TBL.
3585 	 * As a result, there will be one more action.
3586 	 */
3587 	++actions_n;
3588 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3589 	set_tag = (void *)(split_actions + actions_n);
3590 	/*
3591 	 * If tag action is not set to void(it means we are not the meter
3592 	 * suffix flow), add the tag action. Since meter suffix flow already
3593 	 * has the tag added.
3594 	 */
3595 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3596 		/*
3597 		 * Allocate the new subflow ID. This one is unique within
3598 		 * device and not shared with representors. Otherwise,
3599 		 * we would have to resolve multi-thread access synch
3600 		 * issue. Each flow on the shared device is appended
3601 		 * with source vport identifier, so the resulting
3602 		 * flows will be unique in the shared (by master and
3603 		 * representors) domain even if they have coinciding
3604 		 * IDs.
3605 		 */
3606 		flow_id = flow_qrss_get_id(dev);
3607 		if (!flow_id)
3608 			return rte_flow_error_set(error, ENOMEM,
3609 						  RTE_FLOW_ERROR_TYPE_ACTION,
3610 						  NULL, "can't allocate id "
3611 						  "for split Q/RSS subflow");
3612 		/* Internal SET_TAG action to set flow ID. */
3613 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3614 			.data = flow_id,
3615 		};
3616 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3617 		if (ret < 0)
3618 			return ret;
3619 		set_tag->id = ret;
3620 		/* Construct new actions array. */
3621 		/* Replace QUEUE/RSS action. */
3622 		split_actions[qrss_idx] = (struct rte_flow_action){
3623 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3624 			.conf = set_tag,
3625 		};
3626 	}
3627 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3628 	jump = (void *)(set_tag + 1);
3629 	*jump = (struct rte_flow_action_jump){
3630 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3631 	};
3632 	split_actions[actions_n - 2] = (struct rte_flow_action){
3633 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3634 		.conf = jump,
3635 	};
3636 	split_actions[actions_n - 1] = (struct rte_flow_action){
3637 		.type = RTE_FLOW_ACTION_TYPE_END,
3638 	};
3639 	return flow_id;
3640 }
3641 
3642 /**
3643  * Extend the given action list for Tx metadata copy.
3644  *
3645  * Copy the given action list to the ext_actions and add flow metadata register
3646  * copy action in order to copy reg_a set by WQE to reg_c[0].
3647  *
3648  * @param[out] ext_actions
3649  *   Pointer to the extended action list.
3650  * @param[in] actions
3651  *   Pointer to the list of actions.
3652  * @param[in] actions_n
3653  *   Number of actions in the list.
3654  * @param[out] error
3655  *   Perform verbose error reporting if not NULL.
3656  *
3657  * @return
3658  *   0 on success, negative value otherwise
3659  */
3660 static int
3661 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3662 		       struct rte_flow_action *ext_actions,
3663 		       const struct rte_flow_action *actions,
3664 		       int actions_n, struct rte_flow_error *error)
3665 {
3666 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3667 		(struct mlx5_flow_action_copy_mreg *)
3668 			(ext_actions + actions_n + 1);
3669 	int ret;
3670 
3671 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3672 	if (ret < 0)
3673 		return ret;
3674 	cp_mreg->dst = ret;
3675 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3676 	if (ret < 0)
3677 		return ret;
3678 	cp_mreg->src = ret;
3679 	memcpy(ext_actions, actions,
3680 			sizeof(*ext_actions) * actions_n);
3681 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3682 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3683 		.conf = cp_mreg,
3684 	};
3685 	ext_actions[actions_n] = (struct rte_flow_action){
3686 		.type = RTE_FLOW_ACTION_TYPE_END,
3687 	};
3688 	return 0;
3689 }
3690 
3691 /**
3692  * The splitting for metadata feature.
3693  *
3694  * - Q/RSS action on NIC Rx should be split in order to pass by
3695  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3696  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3697  *
3698  * - All the actions on NIC Tx should have a mreg copy action to
3699  *   copy reg_a from WQE to reg_c[0].
3700  *
3701  * @param dev
3702  *   Pointer to Ethernet device.
3703  * @param[in] flow
3704  *   Parent flow structure pointer.
3705  * @param[in] attr
3706  *   Flow rule attributes.
3707  * @param[in] items
3708  *   Pattern specification (list terminated by the END pattern item).
3709  * @param[in] actions
3710  *   Associated actions (list terminated by the END action).
3711  * @param[in] external
3712  *   This flow rule is created by request external to PMD.
3713  * @param[out] error
3714  *   Perform verbose error reporting if not NULL.
3715  * @return
3716  *   0 on success, negative value otherwise
3717  */
3718 static int
3719 flow_create_split_metadata(struct rte_eth_dev *dev,
3720 			   struct rte_flow *flow,
3721 			   const struct rte_flow_attr *attr,
3722 			   const struct rte_flow_item items[],
3723 			   const struct rte_flow_action actions[],
3724 			   bool external, struct rte_flow_error *error)
3725 {
3726 	struct mlx5_priv *priv = dev->data->dev_private;
3727 	struct mlx5_dev_config *config = &priv->config;
3728 	const struct rte_flow_action *qrss = NULL;
3729 	struct rte_flow_action *ext_actions = NULL;
3730 	struct mlx5_flow *dev_flow = NULL;
3731 	uint32_t qrss_id = 0;
3732 	int mtr_sfx = 0;
3733 	size_t act_size;
3734 	int actions_n;
3735 	int ret;
3736 
3737 	/* Check whether extensive metadata feature is engaged. */
3738 	if (!config->dv_flow_en ||
3739 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3740 	    !mlx5_flow_ext_mreg_supported(dev))
3741 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3742 					       actions, external, error);
3743 	actions_n = flow_parse_qrss_action(actions, &qrss);
3744 	if (qrss) {
3745 		/* Exclude hairpin flows from splitting. */
3746 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3747 			const struct rte_flow_action_queue *queue;
3748 
3749 			queue = qrss->conf;
3750 			if (mlx5_rxq_get_type(dev, queue->index) ==
3751 			    MLX5_RXQ_TYPE_HAIRPIN)
3752 				qrss = NULL;
3753 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3754 			const struct rte_flow_action_rss *rss;
3755 
3756 			rss = qrss->conf;
3757 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3758 			    MLX5_RXQ_TYPE_HAIRPIN)
3759 				qrss = NULL;
3760 		}
3761 	}
3762 	if (qrss) {
3763 		/* Check if it is in meter suffix table. */
3764 		mtr_sfx = attr->group == (attr->transfer ?
3765 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3766 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3767 		/*
3768 		 * Q/RSS action on NIC Rx should be split in order to pass by
3769 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3770 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3771 		 */
3772 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3773 			   sizeof(struct rte_flow_action_set_tag) +
3774 			   sizeof(struct rte_flow_action_jump);
3775 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3776 		if (!ext_actions)
3777 			return rte_flow_error_set(error, ENOMEM,
3778 						  RTE_FLOW_ERROR_TYPE_ACTION,
3779 						  NULL, "no memory to split "
3780 						  "metadata flow");
3781 		/*
3782 		 * If we are the suffix flow of meter, tag already exist.
3783 		 * Set the tag action to void.
3784 		 */
3785 		if (mtr_sfx)
3786 			ext_actions[qrss - actions].type =
3787 						RTE_FLOW_ACTION_TYPE_VOID;
3788 		else
3789 			ext_actions[qrss - actions].type =
3790 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3791 		/*
3792 		 * Create the new actions list with removed Q/RSS action
3793 		 * and appended set tag and jump to register copy table
3794 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3795 		 * in advance, because it is needed for set tag action.
3796 		 */
3797 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3798 						    qrss, actions_n, error);
3799 		if (!mtr_sfx && !qrss_id) {
3800 			ret = -rte_errno;
3801 			goto exit;
3802 		}
3803 	} else if (attr->egress && !attr->transfer) {
3804 		/*
3805 		 * All the actions on NIC Tx should have a metadata register
3806 		 * copy action to copy reg_a from WQE to reg_c[meta]
3807 		 */
3808 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3809 			   sizeof(struct mlx5_flow_action_copy_mreg);
3810 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3811 		if (!ext_actions)
3812 			return rte_flow_error_set(error, ENOMEM,
3813 						  RTE_FLOW_ERROR_TYPE_ACTION,
3814 						  NULL, "no memory to split "
3815 						  "metadata flow");
3816 		/* Create the action list appended with copy register. */
3817 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3818 					     actions_n, error);
3819 		if (ret < 0)
3820 			goto exit;
3821 	}
3822 	/* Add the unmodified original or prefix subflow. */
3823 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3824 				      ext_actions ? ext_actions : actions,
3825 				      external, error);
3826 	if (ret < 0)
3827 		goto exit;
3828 	assert(dev_flow);
3829 	if (qrss) {
3830 		const struct rte_flow_attr q_attr = {
3831 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3832 			.ingress = 1,
3833 		};
3834 		/* Internal PMD action to set register. */
3835 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3836 			.data = qrss_id,
3837 			.id = 0,
3838 		};
3839 		struct rte_flow_item q_items[] = {
3840 			{
3841 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3842 				.spec = &q_tag_spec,
3843 				.last = NULL,
3844 				.mask = NULL,
3845 			},
3846 			{
3847 				.type = RTE_FLOW_ITEM_TYPE_END,
3848 			},
3849 		};
3850 		struct rte_flow_action q_actions[] = {
3851 			{
3852 				.type = qrss->type,
3853 				.conf = qrss->conf,
3854 			},
3855 			{
3856 				.type = RTE_FLOW_ACTION_TYPE_END,
3857 			},
3858 		};
3859 		uint64_t hash_fields = dev_flow->hash_fields;
3860 		dev_flow = NULL;
3861 		/*
3862 		 * Configure the tag action only if we are not the meter sub
3863 		 * flow. Since tag is already marked in the meter suffix sub
3864 		 * flow.
3865 		 */
3866 		if (qrss_id) {
3867 			/*
3868 			 * Put unique id in prefix flow due to it is destroyed
3869 			 * after prefix flow and id will be freed after there
3870 			 * is no actual flows with this id and identifier
3871 			 * reallocation becomes possible (for example, for
3872 			 * other flows in other threads).
3873 			 */
3874 			dev_flow->qrss_id = qrss_id;
3875 			qrss_id = 0;
3876 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3877 						   error);
3878 			if (ret < 0)
3879 				goto exit;
3880 			q_tag_spec.id = ret;
3881 		}
3882 		/* Add suffix subflow to execute Q/RSS. */
3883 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3884 					      &q_attr, mtr_sfx ? items :
3885 					      q_items, q_actions,
3886 					      external, error);
3887 		if (ret < 0)
3888 			goto exit;
3889 		assert(dev_flow);
3890 		dev_flow->hash_fields = hash_fields;
3891 	}
3892 
3893 exit:
3894 	/*
3895 	 * We do not destroy the partially created sub_flows in case of error.
3896 	 * These ones are included into parent flow list and will be destroyed
3897 	 * by flow_drv_destroy.
3898 	 */
3899 	flow_qrss_free_id(dev, qrss_id);
3900 	rte_free(ext_actions);
3901 	return ret;
3902 }
3903 
3904 /**
3905  * The splitting for meter feature.
3906  *
3907  * - The meter flow will be split to two flows as prefix and
3908  *   suffix flow. The packets make sense only it pass the prefix
3909  *   meter action.
3910  *
3911  * - Reg_C_5 is used for the packet to match betweend prefix and
3912  *   suffix flow.
3913  *
3914  * @param dev
3915  *   Pointer to Ethernet device.
3916  * @param[in] flow
3917  *   Parent flow structure pointer.
3918  * @param[in] attr
3919  *   Flow rule attributes.
3920  * @param[in] items
3921  *   Pattern specification (list terminated by the END pattern item).
3922  * @param[in] actions
3923  *   Associated actions (list terminated by the END action).
3924  * @param[in] external
3925  *   This flow rule is created by request external to PMD.
3926  * @param[out] error
3927  *   Perform verbose error reporting if not NULL.
3928  * @return
3929  *   0 on success, negative value otherwise
3930  */
3931 static int
3932 flow_create_split_meter(struct rte_eth_dev *dev,
3933 			   struct rte_flow *flow,
3934 			   const struct rte_flow_attr *attr,
3935 			   const struct rte_flow_item items[],
3936 			   const struct rte_flow_action actions[],
3937 			   bool external, struct rte_flow_error *error)
3938 {
3939 	struct mlx5_priv *priv = dev->data->dev_private;
3940 	struct rte_flow_action *sfx_actions = NULL;
3941 	struct rte_flow_action *pre_actions = NULL;
3942 	struct rte_flow_item *sfx_items = NULL;
3943 	const  struct rte_flow_item *sfx_port_id_item;
3944 	struct mlx5_flow *dev_flow = NULL;
3945 	struct rte_flow_attr sfx_attr = *attr;
3946 	uint32_t mtr = 0;
3947 	uint32_t mtr_tag_id = 0;
3948 	size_t act_size;
3949 	size_t item_size;
3950 	int actions_n = 0;
3951 	int ret;
3952 
3953 	if (priv->mtr_en)
3954 		actions_n = flow_check_meter_action(actions, &mtr);
3955 	if (mtr) {
3956 		struct mlx5_rte_flow_item_tag *tag_spec;
3957 		/* The five prefix actions: meter, decap, encap, tag, end. */
3958 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3959 			   sizeof(struct rte_flow_action_set_tag);
3960 		/* tag, end. */
3961 #define METER_SUFFIX_ITEM 3
3962 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3963 			    sizeof(struct mlx5_rte_flow_item_tag);
3964 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3965 		if (!sfx_actions)
3966 			return rte_flow_error_set(error, ENOMEM,
3967 						  RTE_FLOW_ERROR_TYPE_ACTION,
3968 						  NULL, "no memory to split "
3969 						  "meter flow");
3970 		pre_actions = sfx_actions + actions_n;
3971 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
3972 						     pre_actions);
3973 		if (!mtr_tag_id) {
3974 			ret = -rte_errno;
3975 			goto exit;
3976 		}
3977 		/* Add the prefix subflow. */
3978 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3979 						  pre_actions, external, error);
3980 		if (ret) {
3981 			ret = -rte_errno;
3982 			goto exit;
3983 		}
3984 		dev_flow->mtr_flow_id = mtr_tag_id;
3985 		/* Prepare the suffix flow match pattern. */
3986 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
3987 			     act_size);
3988 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
3989 			    METER_SUFFIX_ITEM);
3990 		tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
3991 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
3992 						    error);
3993 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3994 		sfx_items->spec = tag_spec;
3995 		sfx_items->last = NULL;
3996 		sfx_items->mask = NULL;
3997 		sfx_items++;
3998 		sfx_port_id_item = find_port_id_item(items);
3999 		if (sfx_port_id_item) {
4000 			memcpy(sfx_items, sfx_port_id_item,
4001 			       sizeof(*sfx_items));
4002 			sfx_items++;
4003 		}
4004 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4005 		sfx_items -= METER_SUFFIX_ITEM;
4006 		/* Setting the sfx group atrr. */
4007 		sfx_attr.group = sfx_attr.transfer ?
4008 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4009 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4010 	}
4011 	/* Add the prefix subflow. */
4012 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4013 					 sfx_items ? sfx_items : items,
4014 					 sfx_actions ? sfx_actions : actions,
4015 					 external, error);
4016 exit:
4017 	if (sfx_actions)
4018 		rte_free(sfx_actions);
4019 	return ret;
4020 }
4021 
4022 /**
4023  * Split the flow to subflow set. The splitters might be linked
4024  * in the chain, like this:
4025  * flow_create_split_outer() calls:
4026  *   flow_create_split_meter() calls:
4027  *     flow_create_split_metadata(meter_subflow_0) calls:
4028  *       flow_create_split_inner(metadata_subflow_0)
4029  *       flow_create_split_inner(metadata_subflow_1)
4030  *       flow_create_split_inner(metadata_subflow_2)
4031  *     flow_create_split_metadata(meter_subflow_1) calls:
4032  *       flow_create_split_inner(metadata_subflow_0)
4033  *       flow_create_split_inner(metadata_subflow_1)
4034  *       flow_create_split_inner(metadata_subflow_2)
4035  *
4036  * This provide flexible way to add new levels of flow splitting.
4037  * The all of successfully created subflows are included to the
4038  * parent flow dev_flow list.
4039  *
4040  * @param dev
4041  *   Pointer to Ethernet device.
4042  * @param[in] flow
4043  *   Parent flow structure pointer.
4044  * @param[in] attr
4045  *   Flow rule attributes.
4046  * @param[in] items
4047  *   Pattern specification (list terminated by the END pattern item).
4048  * @param[in] actions
4049  *   Associated actions (list terminated by the END action).
4050  * @param[in] external
4051  *   This flow rule is created by request external to PMD.
4052  * @param[out] error
4053  *   Perform verbose error reporting if not NULL.
4054  * @return
4055  *   0 on success, negative value otherwise
4056  */
4057 static int
4058 flow_create_split_outer(struct rte_eth_dev *dev,
4059 			struct rte_flow *flow,
4060 			const struct rte_flow_attr *attr,
4061 			const struct rte_flow_item items[],
4062 			const struct rte_flow_action actions[],
4063 			bool external, struct rte_flow_error *error)
4064 {
4065 	int ret;
4066 
4067 	ret = flow_create_split_meter(dev, flow, attr, items,
4068 					 actions, external, error);
4069 	assert(ret <= 0);
4070 	return ret;
4071 }
4072 
4073 /**
4074  * Create a flow and add it to @p list.
4075  *
4076  * @param dev
4077  *   Pointer to Ethernet device.
4078  * @param list
4079  *   Pointer to a TAILQ flow list. If this parameter NULL,
4080  *   no list insertion occurred, flow is just created,
4081  *   this is caller's responsibility to track the
4082  *   created flow.
4083  * @param[in] attr
4084  *   Flow rule attributes.
4085  * @param[in] items
4086  *   Pattern specification (list terminated by the END pattern item).
4087  * @param[in] actions
4088  *   Associated actions (list terminated by the END action).
4089  * @param[in] external
4090  *   This flow rule is created by request external to PMD.
4091  * @param[out] error
4092  *   Perform verbose error reporting if not NULL.
4093  *
4094  * @return
4095  *   A flow on success, NULL otherwise and rte_errno is set.
4096  */
4097 static struct rte_flow *
4098 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4099 		 const struct rte_flow_attr *attr,
4100 		 const struct rte_flow_item items[],
4101 		 const struct rte_flow_action actions[],
4102 		 bool external, struct rte_flow_error *error)
4103 {
4104 	struct mlx5_priv *priv = dev->data->dev_private;
4105 	struct rte_flow *flow = NULL;
4106 	struct mlx5_flow *dev_flow;
4107 	const struct rte_flow_action_rss *rss;
4108 	union {
4109 		struct rte_flow_expand_rss buf;
4110 		uint8_t buffer[2048];
4111 	} expand_buffer;
4112 	union {
4113 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4114 		uint8_t buffer[2048];
4115 	} actions_rx;
4116 	union {
4117 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4118 		uint8_t buffer[2048];
4119 	} actions_hairpin_tx;
4120 	union {
4121 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4122 		uint8_t buffer[2048];
4123 	} items_tx;
4124 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4125 	const struct rte_flow_action *p_actions_rx = actions;
4126 	int ret;
4127 	uint32_t i;
4128 	uint32_t flow_size;
4129 	int hairpin_flow = 0;
4130 	uint32_t hairpin_id = 0;
4131 	struct rte_flow_attr attr_tx = { .priority = 0 };
4132 
4133 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4134 	if (hairpin_flow > 0) {
4135 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4136 			rte_errno = EINVAL;
4137 			return NULL;
4138 		}
4139 		flow_hairpin_split(dev, actions, actions_rx.actions,
4140 				   actions_hairpin_tx.actions, items_tx.items,
4141 				   &hairpin_id);
4142 		p_actions_rx = actions_rx.actions;
4143 	}
4144 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4145 				error);
4146 	if (ret < 0)
4147 		goto error_before_flow;
4148 	flow_size = sizeof(struct rte_flow);
4149 	rss = flow_get_rss_action(p_actions_rx);
4150 	if (rss)
4151 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4152 					    sizeof(void *));
4153 	else
4154 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4155 	flow = rte_calloc(__func__, 1, flow_size, 0);
4156 	if (!flow) {
4157 		rte_errno = ENOMEM;
4158 		goto error_before_flow;
4159 	}
4160 	flow->drv_type = flow_get_drv_type(dev, attr);
4161 	if (hairpin_id != 0)
4162 		flow->hairpin_flow_id = hairpin_id;
4163 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4164 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
4165 	flow->rss.queue = (void *)(flow + 1);
4166 	if (rss) {
4167 		/*
4168 		 * The following information is required by
4169 		 * mlx5_flow_hashfields_adjust() in advance.
4170 		 */
4171 		flow->rss.level = rss->level;
4172 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4173 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4174 	}
4175 	LIST_INIT(&flow->dev_flows);
4176 	if (rss && rss->types) {
4177 		unsigned int graph_root;
4178 
4179 		graph_root = find_graph_root(items, rss->level);
4180 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4181 					  items, rss->types,
4182 					  mlx5_support_expansion,
4183 					  graph_root);
4184 		assert(ret > 0 &&
4185 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4186 	} else {
4187 		buf->entries = 1;
4188 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4189 	}
4190 	for (i = 0; i < buf->entries; ++i) {
4191 		/*
4192 		 * The splitter may create multiple dev_flows,
4193 		 * depending on configuration. In the simplest
4194 		 * case it just creates unmodified original flow.
4195 		 */
4196 		ret = flow_create_split_outer(dev, flow, attr,
4197 					      buf->entry[i].pattern,
4198 					      p_actions_rx, external,
4199 					      error);
4200 		if (ret < 0)
4201 			goto error;
4202 	}
4203 	/* Create the tx flow. */
4204 	if (hairpin_flow) {
4205 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4206 		attr_tx.ingress = 0;
4207 		attr_tx.egress = 1;
4208 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4209 					    actions_hairpin_tx.actions, error);
4210 		if (!dev_flow)
4211 			goto error;
4212 		dev_flow->flow = flow;
4213 		dev_flow->external = 0;
4214 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4215 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4216 					 items_tx.items,
4217 					 actions_hairpin_tx.actions, error);
4218 		if (ret < 0)
4219 			goto error;
4220 	}
4221 	/*
4222 	 * Update the metadata register copy table. If extensive
4223 	 * metadata feature is enabled and registers are supported
4224 	 * we might create the extra rte_flow for each unique
4225 	 * MARK/FLAG action ID.
4226 	 *
4227 	 * The table is updated for ingress Flows only, because
4228 	 * the egress Flows belong to the different device and
4229 	 * copy table should be updated in peer NIC Rx domain.
4230 	 */
4231 	if (attr->ingress &&
4232 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4233 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4234 		if (ret)
4235 			goto error;
4236 	}
4237 	if (dev->data->dev_started) {
4238 		ret = flow_drv_apply(dev, flow, error);
4239 		if (ret < 0)
4240 			goto error;
4241 	}
4242 	if (list)
4243 		TAILQ_INSERT_TAIL(list, flow, next);
4244 	flow_rxq_flags_set(dev, flow);
4245 	return flow;
4246 error_before_flow:
4247 	if (hairpin_id)
4248 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4249 				     hairpin_id);
4250 	return NULL;
4251 error:
4252 	assert(flow);
4253 	flow_mreg_del_copy_action(dev, flow);
4254 	ret = rte_errno; /* Save rte_errno before cleanup. */
4255 	if (flow->hairpin_flow_id)
4256 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4257 				     flow->hairpin_flow_id);
4258 	assert(flow);
4259 	flow_drv_destroy(dev, flow);
4260 	rte_free(flow);
4261 	rte_errno = ret; /* Restore rte_errno. */
4262 	return NULL;
4263 }
4264 
4265 /**
4266  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4267  * incoming packets to table 1.
4268  *
4269  * Other flow rules, requested for group n, will be created in
4270  * e-switch table n+1.
4271  * Jump action to e-switch group n will be created to group n+1.
4272  *
4273  * Used when working in switchdev mode, to utilise advantages of table 1
4274  * and above.
4275  *
4276  * @param dev
4277  *   Pointer to Ethernet device.
4278  *
4279  * @return
4280  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4281  */
4282 struct rte_flow *
4283 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4284 {
4285 	const struct rte_flow_attr attr = {
4286 		.group = 0,
4287 		.priority = 0,
4288 		.ingress = 1,
4289 		.egress = 0,
4290 		.transfer = 1,
4291 	};
4292 	const struct rte_flow_item pattern = {
4293 		.type = RTE_FLOW_ITEM_TYPE_END,
4294 	};
4295 	struct rte_flow_action_jump jump = {
4296 		.group = 1,
4297 	};
4298 	const struct rte_flow_action actions[] = {
4299 		{
4300 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4301 			.conf = &jump,
4302 		},
4303 		{
4304 			.type = RTE_FLOW_ACTION_TYPE_END,
4305 		},
4306 	};
4307 	struct mlx5_priv *priv = dev->data->dev_private;
4308 	struct rte_flow_error error;
4309 
4310 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4311 				actions, false, &error);
4312 }
4313 
4314 /**
4315  * Create a flow.
4316  *
4317  * @see rte_flow_create()
4318  * @see rte_flow_ops
4319  */
4320 struct rte_flow *
4321 mlx5_flow_create(struct rte_eth_dev *dev,
4322 		 const struct rte_flow_attr *attr,
4323 		 const struct rte_flow_item items[],
4324 		 const struct rte_flow_action actions[],
4325 		 struct rte_flow_error *error)
4326 {
4327 	struct mlx5_priv *priv = dev->data->dev_private;
4328 
4329 	return flow_list_create(dev, &priv->flows,
4330 				attr, items, actions, true, error);
4331 }
4332 
4333 /**
4334  * Destroy a flow in a list.
4335  *
4336  * @param dev
4337  *   Pointer to Ethernet device.
4338  * @param list
4339  *   Pointer to a TAILQ flow list. If this parameter NULL,
4340  *   there is no flow removal from the list.
4341  * @param[in] flow
4342  *   Flow to destroy.
4343  */
4344 static void
4345 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4346 		  struct rte_flow *flow)
4347 {
4348 	struct mlx5_priv *priv = dev->data->dev_private;
4349 
4350 	/*
4351 	 * Update RX queue flags only if port is started, otherwise it is
4352 	 * already clean.
4353 	 */
4354 	if (dev->data->dev_started)
4355 		flow_rxq_flags_trim(dev, flow);
4356 	if (flow->hairpin_flow_id)
4357 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4358 				     flow->hairpin_flow_id);
4359 	flow_drv_destroy(dev, flow);
4360 	if (list)
4361 		TAILQ_REMOVE(list, flow, next);
4362 	flow_mreg_del_copy_action(dev, flow);
4363 	rte_free(flow->fdir);
4364 	rte_free(flow);
4365 }
4366 
4367 /**
4368  * Destroy all flows.
4369  *
4370  * @param dev
4371  *   Pointer to Ethernet device.
4372  * @param list
4373  *   Pointer to a TAILQ flow list.
4374  */
4375 void
4376 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4377 {
4378 	while (!TAILQ_EMPTY(list)) {
4379 		struct rte_flow *flow;
4380 
4381 		flow = TAILQ_FIRST(list);
4382 		flow_list_destroy(dev, list, flow);
4383 	}
4384 }
4385 
4386 /**
4387  * Remove all flows.
4388  *
4389  * @param dev
4390  *   Pointer to Ethernet device.
4391  * @param list
4392  *   Pointer to a TAILQ flow list.
4393  */
4394 void
4395 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4396 {
4397 	struct rte_flow *flow;
4398 
4399 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4400 		flow_drv_remove(dev, flow);
4401 		flow_mreg_stop_copy_action(dev, flow);
4402 	}
4403 	flow_mreg_del_default_copy_action(dev);
4404 	flow_rxq_flags_clear(dev);
4405 }
4406 
4407 /**
4408  * Add all flows.
4409  *
4410  * @param dev
4411  *   Pointer to Ethernet device.
4412  * @param list
4413  *   Pointer to a TAILQ flow list.
4414  *
4415  * @return
4416  *   0 on success, a negative errno value otherwise and rte_errno is set.
4417  */
4418 int
4419 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4420 {
4421 	struct rte_flow *flow;
4422 	struct rte_flow_error error;
4423 	int ret = 0;
4424 
4425 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4426 	ret = flow_mreg_add_default_copy_action(dev, &error);
4427 	if (ret < 0)
4428 		return -rte_errno;
4429 	/* Apply Flows created by application. */
4430 	TAILQ_FOREACH(flow, list, next) {
4431 		ret = flow_mreg_start_copy_action(dev, flow);
4432 		if (ret < 0)
4433 			goto error;
4434 		ret = flow_drv_apply(dev, flow, &error);
4435 		if (ret < 0)
4436 			goto error;
4437 		flow_rxq_flags_set(dev, flow);
4438 	}
4439 	return 0;
4440 error:
4441 	ret = rte_errno; /* Save rte_errno before cleanup. */
4442 	mlx5_flow_stop(dev, list);
4443 	rte_errno = ret; /* Restore rte_errno. */
4444 	return -rte_errno;
4445 }
4446 
4447 /**
4448  * Verify the flow list is empty
4449  *
4450  * @param dev
4451  *  Pointer to Ethernet device.
4452  *
4453  * @return the number of flows not released.
4454  */
4455 int
4456 mlx5_flow_verify(struct rte_eth_dev *dev)
4457 {
4458 	struct mlx5_priv *priv = dev->data->dev_private;
4459 	struct rte_flow *flow;
4460 	int ret = 0;
4461 
4462 	TAILQ_FOREACH(flow, &priv->flows, next) {
4463 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4464 			dev->data->port_id, (void *)flow);
4465 		++ret;
4466 	}
4467 	return ret;
4468 }
4469 
4470 /**
4471  * Enable default hairpin egress flow.
4472  *
4473  * @param dev
4474  *   Pointer to Ethernet device.
4475  * @param queue
4476  *   The queue index.
4477  *
4478  * @return
4479  *   0 on success, a negative errno value otherwise and rte_errno is set.
4480  */
4481 int
4482 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4483 			    uint32_t queue)
4484 {
4485 	struct mlx5_priv *priv = dev->data->dev_private;
4486 	const struct rte_flow_attr attr = {
4487 		.egress = 1,
4488 		.priority = 0,
4489 	};
4490 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4491 		.queue = queue,
4492 	};
4493 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4494 		.queue = UINT32_MAX,
4495 	};
4496 	struct rte_flow_item items[] = {
4497 		{
4498 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4499 			.spec = &queue_spec,
4500 			.last = NULL,
4501 			.mask = &queue_mask,
4502 		},
4503 		{
4504 			.type = RTE_FLOW_ITEM_TYPE_END,
4505 		},
4506 	};
4507 	struct rte_flow_action_jump jump = {
4508 		.group = MLX5_HAIRPIN_TX_TABLE,
4509 	};
4510 	struct rte_flow_action actions[2];
4511 	struct rte_flow *flow;
4512 	struct rte_flow_error error;
4513 
4514 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4515 	actions[0].conf = &jump;
4516 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4517 	flow = flow_list_create(dev, &priv->ctrl_flows,
4518 				&attr, items, actions, false, &error);
4519 	if (!flow) {
4520 		DRV_LOG(DEBUG,
4521 			"Failed to create ctrl flow: rte_errno(%d),"
4522 			" type(%d), message(%s)",
4523 			rte_errno, error.type,
4524 			error.message ? error.message : " (no stated reason)");
4525 		return -rte_errno;
4526 	}
4527 	return 0;
4528 }
4529 
4530 /**
4531  * Enable a control flow configured from the control plane.
4532  *
4533  * @param dev
4534  *   Pointer to Ethernet device.
4535  * @param eth_spec
4536  *   An Ethernet flow spec to apply.
4537  * @param eth_mask
4538  *   An Ethernet flow mask to apply.
4539  * @param vlan_spec
4540  *   A VLAN flow spec to apply.
4541  * @param vlan_mask
4542  *   A VLAN flow mask to apply.
4543  *
4544  * @return
4545  *   0 on success, a negative errno value otherwise and rte_errno is set.
4546  */
4547 int
4548 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4549 		    struct rte_flow_item_eth *eth_spec,
4550 		    struct rte_flow_item_eth *eth_mask,
4551 		    struct rte_flow_item_vlan *vlan_spec,
4552 		    struct rte_flow_item_vlan *vlan_mask)
4553 {
4554 	struct mlx5_priv *priv = dev->data->dev_private;
4555 	const struct rte_flow_attr attr = {
4556 		.ingress = 1,
4557 		.priority = MLX5_FLOW_PRIO_RSVD,
4558 	};
4559 	struct rte_flow_item items[] = {
4560 		{
4561 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4562 			.spec = eth_spec,
4563 			.last = NULL,
4564 			.mask = eth_mask,
4565 		},
4566 		{
4567 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4568 					      RTE_FLOW_ITEM_TYPE_END,
4569 			.spec = vlan_spec,
4570 			.last = NULL,
4571 			.mask = vlan_mask,
4572 		},
4573 		{
4574 			.type = RTE_FLOW_ITEM_TYPE_END,
4575 		},
4576 	};
4577 	uint16_t queue[priv->reta_idx_n];
4578 	struct rte_flow_action_rss action_rss = {
4579 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4580 		.level = 0,
4581 		.types = priv->rss_conf.rss_hf,
4582 		.key_len = priv->rss_conf.rss_key_len,
4583 		.queue_num = priv->reta_idx_n,
4584 		.key = priv->rss_conf.rss_key,
4585 		.queue = queue,
4586 	};
4587 	struct rte_flow_action actions[] = {
4588 		{
4589 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4590 			.conf = &action_rss,
4591 		},
4592 		{
4593 			.type = RTE_FLOW_ACTION_TYPE_END,
4594 		},
4595 	};
4596 	struct rte_flow *flow;
4597 	struct rte_flow_error error;
4598 	unsigned int i;
4599 
4600 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4601 		return 0;
4602 	}
4603 	for (i = 0; i != priv->reta_idx_n; ++i)
4604 		queue[i] = (*priv->reta_idx)[i];
4605 	flow = flow_list_create(dev, &priv->ctrl_flows,
4606 				&attr, items, actions, false, &error);
4607 	if (!flow)
4608 		return -rte_errno;
4609 	return 0;
4610 }
4611 
4612 /**
4613  * Enable a flow control configured from the control plane.
4614  *
4615  * @param dev
4616  *   Pointer to Ethernet device.
4617  * @param eth_spec
4618  *   An Ethernet flow spec to apply.
4619  * @param eth_mask
4620  *   An Ethernet flow mask to apply.
4621  *
4622  * @return
4623  *   0 on success, a negative errno value otherwise and rte_errno is set.
4624  */
4625 int
4626 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4627 	       struct rte_flow_item_eth *eth_spec,
4628 	       struct rte_flow_item_eth *eth_mask)
4629 {
4630 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4631 }
4632 
4633 /**
4634  * Destroy a flow.
4635  *
4636  * @see rte_flow_destroy()
4637  * @see rte_flow_ops
4638  */
4639 int
4640 mlx5_flow_destroy(struct rte_eth_dev *dev,
4641 		  struct rte_flow *flow,
4642 		  struct rte_flow_error *error __rte_unused)
4643 {
4644 	struct mlx5_priv *priv = dev->data->dev_private;
4645 
4646 	flow_list_destroy(dev, &priv->flows, flow);
4647 	return 0;
4648 }
4649 
4650 /**
4651  * Destroy all flows.
4652  *
4653  * @see rte_flow_flush()
4654  * @see rte_flow_ops
4655  */
4656 int
4657 mlx5_flow_flush(struct rte_eth_dev *dev,
4658 		struct rte_flow_error *error __rte_unused)
4659 {
4660 	struct mlx5_priv *priv = dev->data->dev_private;
4661 
4662 	mlx5_flow_list_flush(dev, &priv->flows);
4663 	return 0;
4664 }
4665 
4666 /**
4667  * Isolated mode.
4668  *
4669  * @see rte_flow_isolate()
4670  * @see rte_flow_ops
4671  */
4672 int
4673 mlx5_flow_isolate(struct rte_eth_dev *dev,
4674 		  int enable,
4675 		  struct rte_flow_error *error)
4676 {
4677 	struct mlx5_priv *priv = dev->data->dev_private;
4678 
4679 	if (dev->data->dev_started) {
4680 		rte_flow_error_set(error, EBUSY,
4681 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4682 				   NULL,
4683 				   "port must be stopped first");
4684 		return -rte_errno;
4685 	}
4686 	priv->isolated = !!enable;
4687 	if (enable)
4688 		dev->dev_ops = &mlx5_dev_ops_isolate;
4689 	else
4690 		dev->dev_ops = &mlx5_dev_ops;
4691 	return 0;
4692 }
4693 
4694 /**
4695  * Query a flow.
4696  *
4697  * @see rte_flow_query()
4698  * @see rte_flow_ops
4699  */
4700 static int
4701 flow_drv_query(struct rte_eth_dev *dev,
4702 	       struct rte_flow *flow,
4703 	       const struct rte_flow_action *actions,
4704 	       void *data,
4705 	       struct rte_flow_error *error)
4706 {
4707 	const struct mlx5_flow_driver_ops *fops;
4708 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4709 
4710 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4711 	fops = flow_get_drv_ops(ftype);
4712 
4713 	return fops->query(dev, flow, actions, data, error);
4714 }
4715 
4716 /**
4717  * Query a flow.
4718  *
4719  * @see rte_flow_query()
4720  * @see rte_flow_ops
4721  */
4722 int
4723 mlx5_flow_query(struct rte_eth_dev *dev,
4724 		struct rte_flow *flow,
4725 		const struct rte_flow_action *actions,
4726 		void *data,
4727 		struct rte_flow_error *error)
4728 {
4729 	int ret;
4730 
4731 	ret = flow_drv_query(dev, flow, actions, data, error);
4732 	if (ret < 0)
4733 		return ret;
4734 	return 0;
4735 }
4736 
4737 /**
4738  * Convert a flow director filter to a generic flow.
4739  *
4740  * @param dev
4741  *   Pointer to Ethernet device.
4742  * @param fdir_filter
4743  *   Flow director filter to add.
4744  * @param attributes
4745  *   Generic flow parameters structure.
4746  *
4747  * @return
4748  *   0 on success, a negative errno value otherwise and rte_errno is set.
4749  */
4750 static int
4751 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4752 			 const struct rte_eth_fdir_filter *fdir_filter,
4753 			 struct mlx5_fdir *attributes)
4754 {
4755 	struct mlx5_priv *priv = dev->data->dev_private;
4756 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4757 	const struct rte_eth_fdir_masks *mask =
4758 		&dev->data->dev_conf.fdir_conf.mask;
4759 
4760 	/* Validate queue number. */
4761 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4762 		DRV_LOG(ERR, "port %u invalid queue number %d",
4763 			dev->data->port_id, fdir_filter->action.rx_queue);
4764 		rte_errno = EINVAL;
4765 		return -rte_errno;
4766 	}
4767 	attributes->attr.ingress = 1;
4768 	attributes->items[0] = (struct rte_flow_item) {
4769 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4770 		.spec = &attributes->l2,
4771 		.mask = &attributes->l2_mask,
4772 	};
4773 	switch (fdir_filter->action.behavior) {
4774 	case RTE_ETH_FDIR_ACCEPT:
4775 		attributes->actions[0] = (struct rte_flow_action){
4776 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4777 			.conf = &attributes->queue,
4778 		};
4779 		break;
4780 	case RTE_ETH_FDIR_REJECT:
4781 		attributes->actions[0] = (struct rte_flow_action){
4782 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4783 		};
4784 		break;
4785 	default:
4786 		DRV_LOG(ERR, "port %u invalid behavior %d",
4787 			dev->data->port_id,
4788 			fdir_filter->action.behavior);
4789 		rte_errno = ENOTSUP;
4790 		return -rte_errno;
4791 	}
4792 	attributes->queue.index = fdir_filter->action.rx_queue;
4793 	/* Handle L3. */
4794 	switch (fdir_filter->input.flow_type) {
4795 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4796 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4797 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4798 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4799 			.src_addr = input->flow.ip4_flow.src_ip,
4800 			.dst_addr = input->flow.ip4_flow.dst_ip,
4801 			.time_to_live = input->flow.ip4_flow.ttl,
4802 			.type_of_service = input->flow.ip4_flow.tos,
4803 		};
4804 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4805 			.src_addr = mask->ipv4_mask.src_ip,
4806 			.dst_addr = mask->ipv4_mask.dst_ip,
4807 			.time_to_live = mask->ipv4_mask.ttl,
4808 			.type_of_service = mask->ipv4_mask.tos,
4809 			.next_proto_id = mask->ipv4_mask.proto,
4810 		};
4811 		attributes->items[1] = (struct rte_flow_item){
4812 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4813 			.spec = &attributes->l3,
4814 			.mask = &attributes->l3_mask,
4815 		};
4816 		break;
4817 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4818 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4819 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4820 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4821 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4822 			.proto = input->flow.ipv6_flow.proto,
4823 		};
4824 
4825 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4826 		       input->flow.ipv6_flow.src_ip,
4827 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4828 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4829 		       input->flow.ipv6_flow.dst_ip,
4830 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4831 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4832 		       mask->ipv6_mask.src_ip,
4833 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4834 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4835 		       mask->ipv6_mask.dst_ip,
4836 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4837 		attributes->items[1] = (struct rte_flow_item){
4838 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4839 			.spec = &attributes->l3,
4840 			.mask = &attributes->l3_mask,
4841 		};
4842 		break;
4843 	default:
4844 		DRV_LOG(ERR, "port %u invalid flow type%d",
4845 			dev->data->port_id, fdir_filter->input.flow_type);
4846 		rte_errno = ENOTSUP;
4847 		return -rte_errno;
4848 	}
4849 	/* Handle L4. */
4850 	switch (fdir_filter->input.flow_type) {
4851 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4852 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4853 			.src_port = input->flow.udp4_flow.src_port,
4854 			.dst_port = input->flow.udp4_flow.dst_port,
4855 		};
4856 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4857 			.src_port = mask->src_port_mask,
4858 			.dst_port = mask->dst_port_mask,
4859 		};
4860 		attributes->items[2] = (struct rte_flow_item){
4861 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4862 			.spec = &attributes->l4,
4863 			.mask = &attributes->l4_mask,
4864 		};
4865 		break;
4866 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4867 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4868 			.src_port = input->flow.tcp4_flow.src_port,
4869 			.dst_port = input->flow.tcp4_flow.dst_port,
4870 		};
4871 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4872 			.src_port = mask->src_port_mask,
4873 			.dst_port = mask->dst_port_mask,
4874 		};
4875 		attributes->items[2] = (struct rte_flow_item){
4876 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4877 			.spec = &attributes->l4,
4878 			.mask = &attributes->l4_mask,
4879 		};
4880 		break;
4881 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4882 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4883 			.src_port = input->flow.udp6_flow.src_port,
4884 			.dst_port = input->flow.udp6_flow.dst_port,
4885 		};
4886 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4887 			.src_port = mask->src_port_mask,
4888 			.dst_port = mask->dst_port_mask,
4889 		};
4890 		attributes->items[2] = (struct rte_flow_item){
4891 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4892 			.spec = &attributes->l4,
4893 			.mask = &attributes->l4_mask,
4894 		};
4895 		break;
4896 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4897 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4898 			.src_port = input->flow.tcp6_flow.src_port,
4899 			.dst_port = input->flow.tcp6_flow.dst_port,
4900 		};
4901 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4902 			.src_port = mask->src_port_mask,
4903 			.dst_port = mask->dst_port_mask,
4904 		};
4905 		attributes->items[2] = (struct rte_flow_item){
4906 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4907 			.spec = &attributes->l4,
4908 			.mask = &attributes->l4_mask,
4909 		};
4910 		break;
4911 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4912 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4913 		break;
4914 	default:
4915 		DRV_LOG(ERR, "port %u invalid flow type%d",
4916 			dev->data->port_id, fdir_filter->input.flow_type);
4917 		rte_errno = ENOTSUP;
4918 		return -rte_errno;
4919 	}
4920 	return 0;
4921 }
4922 
4923 #define FLOW_FDIR_CMP(f1, f2, fld) \
4924 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4925 
4926 /**
4927  * Compare two FDIR flows. If items and actions are identical, the two flows are
4928  * regarded as same.
4929  *
4930  * @param dev
4931  *   Pointer to Ethernet device.
4932  * @param f1
4933  *   FDIR flow to compare.
4934  * @param f2
4935  *   FDIR flow to compare.
4936  *
4937  * @return
4938  *   Zero on match, 1 otherwise.
4939  */
4940 static int
4941 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4942 {
4943 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4944 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4945 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4946 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4947 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4948 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4949 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4950 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4951 		return 1;
4952 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4953 	    FLOW_FDIR_CMP(f1, f2, queue))
4954 		return 1;
4955 	return 0;
4956 }
4957 
4958 /**
4959  * Search device flow list to find out a matched FDIR flow.
4960  *
4961  * @param dev
4962  *   Pointer to Ethernet device.
4963  * @param fdir_flow
4964  *   FDIR flow to lookup.
4965  *
4966  * @return
4967  *   Pointer of flow if found, NULL otherwise.
4968  */
4969 static struct rte_flow *
4970 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
4971 {
4972 	struct mlx5_priv *priv = dev->data->dev_private;
4973 	struct rte_flow *flow = NULL;
4974 
4975 	assert(fdir_flow);
4976 	TAILQ_FOREACH(flow, &priv->flows, next) {
4977 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
4978 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
4979 				dev->data->port_id, (void *)flow);
4980 			break;
4981 		}
4982 	}
4983 	return flow;
4984 }
4985 
4986 /**
4987  * Add new flow director filter and store it in list.
4988  *
4989  * @param dev
4990  *   Pointer to Ethernet device.
4991  * @param fdir_filter
4992  *   Flow director filter to add.
4993  *
4994  * @return
4995  *   0 on success, a negative errno value otherwise and rte_errno is set.
4996  */
4997 static int
4998 flow_fdir_filter_add(struct rte_eth_dev *dev,
4999 		     const struct rte_eth_fdir_filter *fdir_filter)
5000 {
5001 	struct mlx5_priv *priv = dev->data->dev_private;
5002 	struct mlx5_fdir *fdir_flow;
5003 	struct rte_flow *flow;
5004 	int ret;
5005 
5006 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5007 	if (!fdir_flow) {
5008 		rte_errno = ENOMEM;
5009 		return -rte_errno;
5010 	}
5011 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5012 	if (ret)
5013 		goto error;
5014 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5015 	if (flow) {
5016 		rte_errno = EEXIST;
5017 		goto error;
5018 	}
5019 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5020 				fdir_flow->items, fdir_flow->actions, true,
5021 				NULL);
5022 	if (!flow)
5023 		goto error;
5024 	assert(!flow->fdir);
5025 	flow->fdir = fdir_flow;
5026 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5027 		dev->data->port_id, (void *)flow);
5028 	return 0;
5029 error:
5030 	rte_free(fdir_flow);
5031 	return -rte_errno;
5032 }
5033 
5034 /**
5035  * Delete specific filter.
5036  *
5037  * @param dev
5038  *   Pointer to Ethernet device.
5039  * @param fdir_filter
5040  *   Filter to be deleted.
5041  *
5042  * @return
5043  *   0 on success, a negative errno value otherwise and rte_errno is set.
5044  */
5045 static int
5046 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5047 			const struct rte_eth_fdir_filter *fdir_filter)
5048 {
5049 	struct mlx5_priv *priv = dev->data->dev_private;
5050 	struct rte_flow *flow;
5051 	struct mlx5_fdir fdir_flow = {
5052 		.attr.group = 0,
5053 	};
5054 	int ret;
5055 
5056 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5057 	if (ret)
5058 		return -rte_errno;
5059 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5060 	if (!flow) {
5061 		rte_errno = ENOENT;
5062 		return -rte_errno;
5063 	}
5064 	flow_list_destroy(dev, &priv->flows, flow);
5065 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5066 		dev->data->port_id, (void *)flow);
5067 	return 0;
5068 }
5069 
5070 /**
5071  * Update queue for specific filter.
5072  *
5073  * @param dev
5074  *   Pointer to Ethernet device.
5075  * @param fdir_filter
5076  *   Filter to be updated.
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_update(struct rte_eth_dev *dev,
5083 			const struct rte_eth_fdir_filter *fdir_filter)
5084 {
5085 	int ret;
5086 
5087 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5088 	if (ret)
5089 		return ret;
5090 	return flow_fdir_filter_add(dev, fdir_filter);
5091 }
5092 
5093 /**
5094  * Flush all filters.
5095  *
5096  * @param dev
5097  *   Pointer to Ethernet device.
5098  */
5099 static void
5100 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5101 {
5102 	struct mlx5_priv *priv = dev->data->dev_private;
5103 
5104 	mlx5_flow_list_flush(dev, &priv->flows);
5105 }
5106 
5107 /**
5108  * Get flow director information.
5109  *
5110  * @param dev
5111  *   Pointer to Ethernet device.
5112  * @param[out] fdir_info
5113  *   Resulting flow director information.
5114  */
5115 static void
5116 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5117 {
5118 	struct rte_eth_fdir_masks *mask =
5119 		&dev->data->dev_conf.fdir_conf.mask;
5120 
5121 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5122 	fdir_info->guarant_spc = 0;
5123 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5124 	fdir_info->max_flexpayload = 0;
5125 	fdir_info->flow_types_mask[0] = 0;
5126 	fdir_info->flex_payload_unit = 0;
5127 	fdir_info->max_flex_payload_segment_num = 0;
5128 	fdir_info->flex_payload_limit = 0;
5129 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5130 }
5131 
5132 /**
5133  * Deal with flow director operations.
5134  *
5135  * @param dev
5136  *   Pointer to Ethernet device.
5137  * @param filter_op
5138  *   Operation to perform.
5139  * @param arg
5140  *   Pointer to operation-specific structure.
5141  *
5142  * @return
5143  *   0 on success, a negative errno value otherwise and rte_errno is set.
5144  */
5145 static int
5146 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5147 		    void *arg)
5148 {
5149 	enum rte_fdir_mode fdir_mode =
5150 		dev->data->dev_conf.fdir_conf.mode;
5151 
5152 	if (filter_op == RTE_ETH_FILTER_NOP)
5153 		return 0;
5154 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5155 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5156 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5157 			dev->data->port_id, fdir_mode);
5158 		rte_errno = EINVAL;
5159 		return -rte_errno;
5160 	}
5161 	switch (filter_op) {
5162 	case RTE_ETH_FILTER_ADD:
5163 		return flow_fdir_filter_add(dev, arg);
5164 	case RTE_ETH_FILTER_UPDATE:
5165 		return flow_fdir_filter_update(dev, arg);
5166 	case RTE_ETH_FILTER_DELETE:
5167 		return flow_fdir_filter_delete(dev, arg);
5168 	case RTE_ETH_FILTER_FLUSH:
5169 		flow_fdir_filter_flush(dev);
5170 		break;
5171 	case RTE_ETH_FILTER_INFO:
5172 		flow_fdir_info_get(dev, arg);
5173 		break;
5174 	default:
5175 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5176 			dev->data->port_id, filter_op);
5177 		rte_errno = EINVAL;
5178 		return -rte_errno;
5179 	}
5180 	return 0;
5181 }
5182 
5183 /**
5184  * Manage filter operations.
5185  *
5186  * @param dev
5187  *   Pointer to Ethernet device structure.
5188  * @param filter_type
5189  *   Filter type.
5190  * @param filter_op
5191  *   Operation to perform.
5192  * @param arg
5193  *   Pointer to operation-specific structure.
5194  *
5195  * @return
5196  *   0 on success, a negative errno value otherwise and rte_errno is set.
5197  */
5198 int
5199 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5200 		     enum rte_filter_type filter_type,
5201 		     enum rte_filter_op filter_op,
5202 		     void *arg)
5203 {
5204 	switch (filter_type) {
5205 	case RTE_ETH_FILTER_GENERIC:
5206 		if (filter_op != RTE_ETH_FILTER_GET) {
5207 			rte_errno = EINVAL;
5208 			return -rte_errno;
5209 		}
5210 		*(const void **)arg = &mlx5_flow_ops;
5211 		return 0;
5212 	case RTE_ETH_FILTER_FDIR:
5213 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5214 	default:
5215 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5216 			dev->data->port_id, filter_type);
5217 		rte_errno = ENOTSUP;
5218 		return -rte_errno;
5219 	}
5220 	return 0;
5221 }
5222 
5223 /**
5224  * Create the needed meter and suffix tables.
5225  *
5226  * @param[in] dev
5227  *   Pointer to Ethernet device.
5228  * @param[in] fm
5229  *   Pointer to the flow meter.
5230  *
5231  * @return
5232  *   Pointer to table set on success, NULL otherwise.
5233  */
5234 struct mlx5_meter_domains_infos *
5235 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5236 			  const struct mlx5_flow_meter *fm)
5237 {
5238 	const struct mlx5_flow_driver_ops *fops;
5239 
5240 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5241 	return fops->create_mtr_tbls(dev, fm);
5242 }
5243 
5244 /**
5245  * Destroy the meter table set.
5246  *
5247  * @param[in] dev
5248  *   Pointer to Ethernet device.
5249  * @param[in] tbl
5250  *   Pointer to the meter table set.
5251  *
5252  * @return
5253  *   0 on success.
5254  */
5255 int
5256 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5257 			   struct mlx5_meter_domains_infos *tbls)
5258 {
5259 	const struct mlx5_flow_driver_ops *fops;
5260 
5261 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5262 	return fops->destroy_mtr_tbls(dev, tbls);
5263 }
5264 
5265 /**
5266  * Create policer rules.
5267  *
5268  * @param[in] dev
5269  *   Pointer to Ethernet device.
5270  * @param[in] fm
5271  *   Pointer to flow meter structure.
5272  * @param[in] attr
5273  *   Pointer to flow attributes.
5274  *
5275  * @return
5276  *   0 on success, -1 otherwise.
5277  */
5278 int
5279 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5280 			       struct mlx5_flow_meter *fm,
5281 			       const struct rte_flow_attr *attr)
5282 {
5283 	const struct mlx5_flow_driver_ops *fops;
5284 
5285 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5286 	return fops->create_policer_rules(dev, fm, attr);
5287 }
5288 
5289 /**
5290  * Destroy policer rules.
5291  *
5292  * @param[in] fm
5293  *   Pointer to flow meter structure.
5294  * @param[in] attr
5295  *   Pointer to flow attributes.
5296  *
5297  * @return
5298  *   0 on success, -1 otherwise.
5299  */
5300 int
5301 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5302 				struct mlx5_flow_meter *fm,
5303 				const struct rte_flow_attr *attr)
5304 {
5305 	const struct mlx5_flow_driver_ops *fops;
5306 
5307 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5308 	return fops->destroy_policer_rules(dev, fm, attr);
5309 }
5310 
5311 /**
5312  * Allocate a counter.
5313  *
5314  * @param[in] dev
5315  *   Pointer to Ethernet device structure.
5316  *
5317  * @return
5318  *   Pointer to allocated counter  on success, NULL otherwise.
5319  */
5320 struct mlx5_flow_counter *
5321 mlx5_counter_alloc(struct rte_eth_dev *dev)
5322 {
5323 	const struct mlx5_flow_driver_ops *fops;
5324 	struct rte_flow_attr attr = { .transfer = 0 };
5325 
5326 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5327 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5328 		return fops->counter_alloc(dev);
5329 	}
5330 	DRV_LOG(ERR,
5331 		"port %u counter allocate is not supported.",
5332 		 dev->data->port_id);
5333 	return NULL;
5334 }
5335 
5336 /**
5337  * Free a counter.
5338  *
5339  * @param[in] dev
5340  *   Pointer to Ethernet device structure.
5341  * @param[in] cnt
5342  *   Pointer to counter to be free.
5343  */
5344 void
5345 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5346 {
5347 	const struct mlx5_flow_driver_ops *fops;
5348 	struct rte_flow_attr attr = { .transfer = 0 };
5349 
5350 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5351 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5352 		fops->counter_free(dev, cnt);
5353 		return;
5354 	}
5355 	DRV_LOG(ERR,
5356 		"port %u counter free is not supported.",
5357 		 dev->data->port_id);
5358 }
5359 
5360 /**
5361  * Query counter statistics.
5362  *
5363  * @param[in] dev
5364  *   Pointer to Ethernet device structure.
5365  * @param[in] cnt
5366  *   Pointer to counter to query.
5367  * @param[in] clear
5368  *   Set to clear counter statistics.
5369  * @param[out] pkts
5370  *   The counter hits packets number to save.
5371  * @param[out] bytes
5372  *   The counter hits bytes number to save.
5373  *
5374  * @return
5375  *   0 on success, a negative errno value otherwise.
5376  */
5377 int
5378 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5379 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5380 {
5381 	const struct mlx5_flow_driver_ops *fops;
5382 	struct rte_flow_attr attr = { .transfer = 0 };
5383 
5384 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5385 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5386 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5387 	}
5388 	DRV_LOG(ERR,
5389 		"port %u counter query is not supported.",
5390 		 dev->data->port_id);
5391 	return -ENOTSUP;
5392 }
5393 
5394 #define MLX5_POOL_QUERY_FREQ_US 1000000
5395 
5396 /**
5397  * Set the periodic procedure for triggering asynchronous batch queries for all
5398  * the counter pools.
5399  *
5400  * @param[in] sh
5401  *   Pointer to mlx5_ibv_shared object.
5402  */
5403 void
5404 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5405 {
5406 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5407 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5408 	uint32_t us;
5409 
5410 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5411 	pools_n += rte_atomic16_read(&cont->n_valid);
5412 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5413 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5414 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5415 		sh->cmng.query_thread_on = 0;
5416 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5417 	} else {
5418 		sh->cmng.query_thread_on = 1;
5419 	}
5420 }
5421 
5422 /**
5423  * The periodic procedure for triggering asynchronous batch queries for all the
5424  * counter pools. This function is probably called by the host thread.
5425  *
5426  * @param[in] arg
5427  *   The parameter for the alarm process.
5428  */
5429 void
5430 mlx5_flow_query_alarm(void *arg)
5431 {
5432 	struct mlx5_ibv_shared *sh = arg;
5433 	struct mlx5_devx_obj *dcs;
5434 	uint16_t offset;
5435 	int ret;
5436 	uint8_t batch = sh->cmng.batch;
5437 	uint16_t pool_index = sh->cmng.pool_index;
5438 	struct mlx5_pools_container *cont;
5439 	struct mlx5_pools_container *mcont;
5440 	struct mlx5_flow_counter_pool *pool;
5441 
5442 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5443 		goto set_alarm;
5444 next_container:
5445 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5446 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5447 	/* Check if resize was done and need to flip a container. */
5448 	if (cont != mcont) {
5449 		if (cont->pools) {
5450 			/* Clean the old container. */
5451 			rte_free(cont->pools);
5452 			memset(cont, 0, sizeof(*cont));
5453 		}
5454 		rte_cio_wmb();
5455 		 /* Flip the host container. */
5456 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5457 		cont = mcont;
5458 	}
5459 	if (!cont->pools) {
5460 		/* 2 empty containers case is unexpected. */
5461 		if (unlikely(batch != sh->cmng.batch))
5462 			goto set_alarm;
5463 		batch ^= 0x1;
5464 		pool_index = 0;
5465 		goto next_container;
5466 	}
5467 	pool = cont->pools[pool_index];
5468 	if (pool->raw_hw)
5469 		/* There is a pool query in progress. */
5470 		goto set_alarm;
5471 	pool->raw_hw =
5472 		LIST_FIRST(&sh->cmng.free_stat_raws);
5473 	if (!pool->raw_hw)
5474 		/* No free counter statistics raw memory. */
5475 		goto set_alarm;
5476 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5477 							      (&pool->a64_dcs);
5478 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5479 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5480 					       offset, NULL, NULL,
5481 					       pool->raw_hw->mem_mng->dm->id,
5482 					       (void *)(uintptr_t)
5483 					       (pool->raw_hw->data + offset),
5484 					       sh->devx_comp,
5485 					       (uint64_t)(uintptr_t)pool);
5486 	if (ret) {
5487 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5488 			" %d", pool->min_dcs->id);
5489 		pool->raw_hw = NULL;
5490 		goto set_alarm;
5491 	}
5492 	pool->raw_hw->min_dcs_id = dcs->id;
5493 	LIST_REMOVE(pool->raw_hw, next);
5494 	sh->cmng.pending_queries++;
5495 	pool_index++;
5496 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5497 		batch ^= 0x1;
5498 		pool_index = 0;
5499 	}
5500 set_alarm:
5501 	sh->cmng.batch = batch;
5502 	sh->cmng.pool_index = pool_index;
5503 	mlx5_set_query_alarm(sh);
5504 }
5505 
5506 /**
5507  * Handler for the HW respond about ready values from an asynchronous batch
5508  * query. This function is probably called by the host thread.
5509  *
5510  * @param[in] sh
5511  *   The pointer to the shared IB device context.
5512  * @param[in] async_id
5513  *   The Devx async ID.
5514  * @param[in] status
5515  *   The status of the completion.
5516  */
5517 void
5518 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5519 				  uint64_t async_id, int status)
5520 {
5521 	struct mlx5_flow_counter_pool *pool =
5522 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5523 	struct mlx5_counter_stats_raw *raw_to_free;
5524 
5525 	if (unlikely(status)) {
5526 		raw_to_free = pool->raw_hw;
5527 	} else {
5528 		raw_to_free = pool->raw;
5529 		rte_spinlock_lock(&pool->sl);
5530 		pool->raw = pool->raw_hw;
5531 		rte_spinlock_unlock(&pool->sl);
5532 		rte_atomic64_add(&pool->query_gen, 1);
5533 		/* Be sure the new raw counters data is updated in memory. */
5534 		rte_cio_wmb();
5535 	}
5536 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5537 	pool->raw_hw = NULL;
5538 	sh->cmng.pending_queries--;
5539 }
5540 
5541 /**
5542  * Translate the rte_flow group index to HW table value.
5543  *
5544  * @param[in] attributes
5545  *   Pointer to flow attributes
5546  * @param[in] external
5547  *   Value is part of flow rule created by request external to PMD.
5548  * @param[in] group
5549  *   rte_flow group index value.
5550  * @param[out] table
5551  *   HW table value.
5552  * @param[out] error
5553  *   Pointer to error structure.
5554  *
5555  * @return
5556  *   0 on success, a negative errno value otherwise and rte_errno is set.
5557  */
5558 int
5559 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5560 			 uint32_t group, uint32_t *table,
5561 			 struct rte_flow_error *error)
5562 {
5563 	if (attributes->transfer && external) {
5564 		if (group == UINT32_MAX)
5565 			return rte_flow_error_set
5566 						(error, EINVAL,
5567 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5568 						 NULL,
5569 						 "group index not supported");
5570 		*table = group + 1;
5571 	} else {
5572 		*table = group;
5573 	}
5574 	return 0;
5575 }
5576 
5577 /**
5578  * Discover availability of metadata reg_c's.
5579  *
5580  * Iteratively use test flows to check availability.
5581  *
5582  * @param[in] dev
5583  *   Pointer to the Ethernet device structure.
5584  *
5585  * @return
5586  *   0 on success, a negative errno value otherwise and rte_errno is set.
5587  */
5588 int
5589 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5590 {
5591 	struct mlx5_priv *priv = dev->data->dev_private;
5592 	struct mlx5_dev_config *config = &priv->config;
5593 	enum modify_reg idx;
5594 	int n = 0;
5595 
5596 	/* reg_c[0] and reg_c[1] are reserved. */
5597 	config->flow_mreg_c[n++] = REG_C_0;
5598 	config->flow_mreg_c[n++] = REG_C_1;
5599 	/* Discover availability of other reg_c's. */
5600 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5601 		struct rte_flow_attr attr = {
5602 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5603 			.priority = MLX5_FLOW_PRIO_RSVD,
5604 			.ingress = 1,
5605 		};
5606 		struct rte_flow_item items[] = {
5607 			[0] = {
5608 				.type = RTE_FLOW_ITEM_TYPE_END,
5609 			},
5610 		};
5611 		struct rte_flow_action actions[] = {
5612 			[0] = {
5613 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5614 				.conf = &(struct mlx5_flow_action_copy_mreg){
5615 					.src = REG_C_1,
5616 					.dst = idx,
5617 				},
5618 			},
5619 			[1] = {
5620 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5621 				.conf = &(struct rte_flow_action_jump){
5622 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5623 				},
5624 			},
5625 			[2] = {
5626 				.type = RTE_FLOW_ACTION_TYPE_END,
5627 			},
5628 		};
5629 		struct rte_flow *flow;
5630 		struct rte_flow_error error;
5631 
5632 		if (!config->dv_flow_en)
5633 			break;
5634 		/* Create internal flow, validation skips copy action. */
5635 		flow = flow_list_create(dev, NULL, &attr, items,
5636 					actions, false, &error);
5637 		if (!flow)
5638 			continue;
5639 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5640 			config->flow_mreg_c[n++] = idx;
5641 		flow_list_destroy(dev, NULL, flow);
5642 	}
5643 	for (; n < MLX5_MREG_C_NUM; ++n)
5644 		config->flow_mreg_c[n] = REG_NONE;
5645 	return 0;
5646 }
5647