xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 4a73c86ff616498768aa77ebad8115f99feff2aa)
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 		switch (config->dv_xmeta_en) {
368 		case MLX5_XMETA_MODE_LEGACY:
369 			return REG_NONE;
370 		case MLX5_XMETA_MODE_META16:
371 			return REG_C_0;
372 		case MLX5_XMETA_MODE_META32:
373 			return REG_C_1;
374 		}
375 		break;
376 	case MLX5_FLOW_MARK:
377 		switch (config->dv_xmeta_en) {
378 		case MLX5_XMETA_MODE_LEGACY:
379 			return REG_NONE;
380 		case MLX5_XMETA_MODE_META16:
381 			return REG_C_1;
382 		case MLX5_XMETA_MODE_META32:
383 			return REG_C_0;
384 		}
385 		break;
386 	case MLX5_COPY_MARK:
387 	case MLX5_MTR_SFX:
388 		/*
389 		 * Metadata COPY_MARK register using is in meter suffix sub
390 		 * flow while with meter. It's safe to share the same register.
391 		 */
392 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
393 	case MLX5_MTR_COLOR:
394 		RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
395 		return priv->mtr_color_reg;
396 	case MLX5_APP_TAG:
397 		/*
398 		 * If meter is enable, it will engage two registers for color
399 		 * match and flow match. If meter color match is not using the
400 		 * REG_C_2, need to skip the REG_C_x be used by meter color
401 		 * match.
402 		 * If meter is disable, free to use all available registers.
403 		 */
404 		if (priv->mtr_color_reg != REG_NONE)
405 			start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
406 				    REG_C_4;
407 		else
408 			start_reg = REG_C_2;
409 		if (id > (REG_C_7 - start_reg))
410 			return rte_flow_error_set(error, EINVAL,
411 						  RTE_FLOW_ERROR_TYPE_ITEM,
412 						  NULL, "invalid tag id");
413 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
414 			return rte_flow_error_set(error, ENOTSUP,
415 						  RTE_FLOW_ERROR_TYPE_ITEM,
416 						  NULL, "unsupported tag id");
417 		/*
418 		 * This case means meter is using the REG_C_x great than 2.
419 		 * Take care not to conflict with meter color REG_C_x.
420 		 * If the available index REG_C_y >= REG_C_x, skip the
421 		 * color register.
422 		 */
423 		if (start_reg == REG_C_3 && config->flow_mreg_c
424 		    [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
425 			if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
426 			    REG_NONE)
427 				return config->flow_mreg_c
428 						[id + 1 + REG_C_3 - REG_C_0];
429 			return rte_flow_error_set(error, ENOTSUP,
430 						  RTE_FLOW_ERROR_TYPE_ITEM,
431 						  NULL, "unsupported tag id");
432 		}
433 		return config->flow_mreg_c[id + start_reg - REG_C_0];
434 	}
435 	assert(false);
436 	return rte_flow_error_set(error, EINVAL,
437 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
438 				  NULL, "invalid feature name");
439 }
440 
441 /**
442  * Check extensive flow metadata register support.
443  *
444  * @param dev
445  *   Pointer to rte_eth_dev structure.
446  *
447  * @return
448  *   True if device supports extensive flow metadata register, otherwise false.
449  */
450 bool
451 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
452 {
453 	struct mlx5_priv *priv = dev->data->dev_private;
454 	struct mlx5_dev_config *config = &priv->config;
455 
456 	/*
457 	 * Having available reg_c can be regarded inclusively as supporting
458 	 * extensive flow metadata register, which could mean,
459 	 * - metadata register copy action by modify header.
460 	 * - 16 modify header actions is supported.
461 	 * - reg_c's are preserved across different domain (FDB and NIC) on
462 	 *   packet loopback by flow lookup miss.
463 	 */
464 	return config->flow_mreg_c[2] != REG_NONE;
465 }
466 
467 /**
468  * Discover the maximum number of priority available.
469  *
470  * @param[in] dev
471  *   Pointer to the Ethernet device structure.
472  *
473  * @return
474  *   number of supported flow priority on success, a negative errno
475  *   value otherwise and rte_errno is set.
476  */
477 int
478 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
479 {
480 	struct mlx5_priv *priv = dev->data->dev_private;
481 	struct {
482 		struct ibv_flow_attr attr;
483 		struct ibv_flow_spec_eth eth;
484 		struct ibv_flow_spec_action_drop drop;
485 	} flow_attr = {
486 		.attr = {
487 			.num_of_specs = 2,
488 			.port = (uint8_t)priv->ibv_port,
489 		},
490 		.eth = {
491 			.type = IBV_FLOW_SPEC_ETH,
492 			.size = sizeof(struct ibv_flow_spec_eth),
493 		},
494 		.drop = {
495 			.size = sizeof(struct ibv_flow_spec_action_drop),
496 			.type = IBV_FLOW_SPEC_ACTION_DROP,
497 		},
498 	};
499 	struct ibv_flow *flow;
500 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
501 	uint16_t vprio[] = { 8, 16 };
502 	int i;
503 	int priority = 0;
504 
505 	if (!drop) {
506 		rte_errno = ENOTSUP;
507 		return -rte_errno;
508 	}
509 	for (i = 0; i != RTE_DIM(vprio); i++) {
510 		flow_attr.attr.priority = vprio[i] - 1;
511 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
512 		if (!flow)
513 			break;
514 		claim_zero(mlx5_glue->destroy_flow(flow));
515 		priority = vprio[i];
516 	}
517 	mlx5_hrxq_drop_release(dev);
518 	switch (priority) {
519 	case 8:
520 		priority = RTE_DIM(priority_map_3);
521 		break;
522 	case 16:
523 		priority = RTE_DIM(priority_map_5);
524 		break;
525 	default:
526 		rte_errno = ENOTSUP;
527 		DRV_LOG(ERR,
528 			"port %u verbs maximum priority: %d expected 8/16",
529 			dev->data->port_id, priority);
530 		return -rte_errno;
531 	}
532 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
533 		dev->data->port_id, priority);
534 	return priority;
535 }
536 
537 /**
538  * Adjust flow priority based on the highest layer and the request priority.
539  *
540  * @param[in] dev
541  *   Pointer to the Ethernet device structure.
542  * @param[in] priority
543  *   The rule base priority.
544  * @param[in] subpriority
545  *   The priority based on the items.
546  *
547  * @return
548  *   The new priority.
549  */
550 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
551 				   uint32_t subpriority)
552 {
553 	uint32_t res = 0;
554 	struct mlx5_priv *priv = dev->data->dev_private;
555 
556 	switch (priv->config.flow_prio) {
557 	case RTE_DIM(priority_map_3):
558 		res = priority_map_3[priority][subpriority];
559 		break;
560 	case RTE_DIM(priority_map_5):
561 		res = priority_map_5[priority][subpriority];
562 		break;
563 	}
564 	return  res;
565 }
566 
567 /**
568  * Verify the @p item specifications (spec, last, mask) are compatible with the
569  * NIC capabilities.
570  *
571  * @param[in] item
572  *   Item specification.
573  * @param[in] mask
574  *   @p item->mask or flow default bit-masks.
575  * @param[in] nic_mask
576  *   Bit-masks covering supported fields by the NIC to compare with user mask.
577  * @param[in] size
578  *   Bit-masks size in bytes.
579  * @param[out] error
580  *   Pointer to error structure.
581  *
582  * @return
583  *   0 on success, a negative errno value otherwise and rte_errno is set.
584  */
585 int
586 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
587 			  const uint8_t *mask,
588 			  const uint8_t *nic_mask,
589 			  unsigned int size,
590 			  struct rte_flow_error *error)
591 {
592 	unsigned int i;
593 
594 	assert(nic_mask);
595 	for (i = 0; i < size; ++i)
596 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
597 			return rte_flow_error_set(error, ENOTSUP,
598 						  RTE_FLOW_ERROR_TYPE_ITEM,
599 						  item,
600 						  "mask enables non supported"
601 						  " bits");
602 	if (!item->spec && (item->mask || item->last))
603 		return rte_flow_error_set(error, EINVAL,
604 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
605 					  "mask/last without a spec is not"
606 					  " supported");
607 	if (item->spec && item->last) {
608 		uint8_t spec[size];
609 		uint8_t last[size];
610 		unsigned int i;
611 		int ret;
612 
613 		for (i = 0; i < size; ++i) {
614 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
615 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
616 		}
617 		ret = memcmp(spec, last, size);
618 		if (ret != 0)
619 			return rte_flow_error_set(error, EINVAL,
620 						  RTE_FLOW_ERROR_TYPE_ITEM,
621 						  item,
622 						  "range is not valid");
623 	}
624 	return 0;
625 }
626 
627 /**
628  * Adjust the hash fields according to the @p flow information.
629  *
630  * @param[in] dev_flow.
631  *   Pointer to the mlx5_flow.
632  * @param[in] tunnel
633  *   1 when the hash field is for a tunnel item.
634  * @param[in] layer_types
635  *   ETH_RSS_* types.
636  * @param[in] hash_fields
637  *   Item hash fields.
638  *
639  * @return
640  *   The hash fields that should be used.
641  */
642 uint64_t
643 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
644 			    int tunnel __rte_unused, uint64_t layer_types,
645 			    uint64_t hash_fields)
646 {
647 	struct rte_flow *flow = dev_flow->flow;
648 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
649 	int rss_request_inner = flow->rss.level >= 2;
650 
651 	/* Check RSS hash level for tunnel. */
652 	if (tunnel && rss_request_inner)
653 		hash_fields |= IBV_RX_HASH_INNER;
654 	else if (tunnel || rss_request_inner)
655 		return 0;
656 #endif
657 	/* Check if requested layer matches RSS hash fields. */
658 	if (!(flow->rss.types & layer_types))
659 		return 0;
660 	return hash_fields;
661 }
662 
663 /**
664  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
665  * if several tunnel rules are used on this queue, the tunnel ptype will be
666  * cleared.
667  *
668  * @param rxq_ctrl
669  *   Rx queue to update.
670  */
671 static void
672 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
673 {
674 	unsigned int i;
675 	uint32_t tunnel_ptype = 0;
676 
677 	/* Look up for the ptype to use. */
678 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
679 		if (!rxq_ctrl->flow_tunnels_n[i])
680 			continue;
681 		if (!tunnel_ptype) {
682 			tunnel_ptype = tunnels_info[i].ptype;
683 		} else {
684 			tunnel_ptype = 0;
685 			break;
686 		}
687 	}
688 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
689 }
690 
691 /**
692  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
693  * flow.
694  *
695  * @param[in] dev
696  *   Pointer to the Ethernet device structure.
697  * @param[in] dev_flow
698  *   Pointer to device flow structure.
699  */
700 static void
701 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
702 {
703 	struct mlx5_priv *priv = dev->data->dev_private;
704 	struct rte_flow *flow = dev_flow->flow;
705 	const int mark = !!(dev_flow->actions &
706 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
707 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
708 	unsigned int i;
709 
710 	for (i = 0; i != flow->rss.queue_num; ++i) {
711 		int idx = (*flow->rss.queue)[i];
712 		struct mlx5_rxq_ctrl *rxq_ctrl =
713 			container_of((*priv->rxqs)[idx],
714 				     struct mlx5_rxq_ctrl, rxq);
715 
716 		/*
717 		 * To support metadata register copy on Tx loopback,
718 		 * this must be always enabled (metadata may arive
719 		 * from other port - not from local flows only.
720 		 */
721 		if (priv->config.dv_flow_en &&
722 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
723 		    mlx5_flow_ext_mreg_supported(dev)) {
724 			rxq_ctrl->rxq.mark = 1;
725 			rxq_ctrl->flow_mark_n = 1;
726 		} else if (mark) {
727 			rxq_ctrl->rxq.mark = 1;
728 			rxq_ctrl->flow_mark_n++;
729 		}
730 		if (tunnel) {
731 			unsigned int j;
732 
733 			/* Increase the counter matching the flow. */
734 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
735 				if ((tunnels_info[j].tunnel &
736 				     dev_flow->layers) ==
737 				    tunnels_info[j].tunnel) {
738 					rxq_ctrl->flow_tunnels_n[j]++;
739 					break;
740 				}
741 			}
742 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
743 		}
744 	}
745 }
746 
747 /**
748  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
749  *
750  * @param[in] dev
751  *   Pointer to the Ethernet device structure.
752  * @param[in] flow
753  *   Pointer to flow structure.
754  */
755 static void
756 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
757 {
758 	struct mlx5_flow *dev_flow;
759 
760 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
761 		flow_drv_rxq_flags_set(dev, dev_flow);
762 }
763 
764 /**
765  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
766  * device flow if no other flow uses it with the same kind of request.
767  *
768  * @param dev
769  *   Pointer to Ethernet device.
770  * @param[in] dev_flow
771  *   Pointer to the device flow.
772  */
773 static void
774 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
775 {
776 	struct mlx5_priv *priv = dev->data->dev_private;
777 	struct rte_flow *flow = dev_flow->flow;
778 	const int mark = !!(dev_flow->actions &
779 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
780 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
781 	unsigned int i;
782 
783 	assert(dev->data->dev_started);
784 	for (i = 0; i != flow->rss.queue_num; ++i) {
785 		int idx = (*flow->rss.queue)[i];
786 		struct mlx5_rxq_ctrl *rxq_ctrl =
787 			container_of((*priv->rxqs)[idx],
788 				     struct mlx5_rxq_ctrl, rxq);
789 
790 		if (priv->config.dv_flow_en &&
791 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
792 		    mlx5_flow_ext_mreg_supported(dev)) {
793 			rxq_ctrl->rxq.mark = 1;
794 			rxq_ctrl->flow_mark_n = 1;
795 		} else if (mark) {
796 			rxq_ctrl->flow_mark_n--;
797 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
798 		}
799 		if (tunnel) {
800 			unsigned int j;
801 
802 			/* Decrease the counter matching the flow. */
803 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
804 				if ((tunnels_info[j].tunnel &
805 				     dev_flow->layers) ==
806 				    tunnels_info[j].tunnel) {
807 					rxq_ctrl->flow_tunnels_n[j]--;
808 					break;
809 				}
810 			}
811 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
812 		}
813 	}
814 }
815 
816 /**
817  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
818  * @p flow if no other flow uses it with the same kind of request.
819  *
820  * @param dev
821  *   Pointer to Ethernet device.
822  * @param[in] flow
823  *   Pointer to the flow.
824  */
825 static void
826 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
827 {
828 	struct mlx5_flow *dev_flow;
829 
830 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
831 		flow_drv_rxq_flags_trim(dev, dev_flow);
832 }
833 
834 /**
835  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
836  *
837  * @param dev
838  *   Pointer to Ethernet device.
839  */
840 static void
841 flow_rxq_flags_clear(struct rte_eth_dev *dev)
842 {
843 	struct mlx5_priv *priv = dev->data->dev_private;
844 	unsigned int i;
845 
846 	for (i = 0; i != priv->rxqs_n; ++i) {
847 		struct mlx5_rxq_ctrl *rxq_ctrl;
848 		unsigned int j;
849 
850 		if (!(*priv->rxqs)[i])
851 			continue;
852 		rxq_ctrl = container_of((*priv->rxqs)[i],
853 					struct mlx5_rxq_ctrl, rxq);
854 		rxq_ctrl->flow_mark_n = 0;
855 		rxq_ctrl->rxq.mark = 0;
856 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
857 			rxq_ctrl->flow_tunnels_n[j] = 0;
858 		rxq_ctrl->rxq.tunnel = 0;
859 	}
860 }
861 
862 /*
863  * return a pointer to the desired action in the list of actions.
864  *
865  * @param[in] actions
866  *   The list of actions to search the action in.
867  * @param[in] action
868  *   The action to find.
869  *
870  * @return
871  *   Pointer to the action in the list, if found. NULL otherwise.
872  */
873 const struct rte_flow_action *
874 mlx5_flow_find_action(const struct rte_flow_action *actions,
875 		      enum rte_flow_action_type action)
876 {
877 	if (actions == NULL)
878 		return NULL;
879 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
880 		if (actions->type == action)
881 			return actions;
882 	return NULL;
883 }
884 
885 /*
886  * Validate the flag action.
887  *
888  * @param[in] action_flags
889  *   Bit-fields that holds the actions detected until now.
890  * @param[in] attr
891  *   Attributes of flow that includes this action.
892  * @param[out] error
893  *   Pointer to error structure.
894  *
895  * @return
896  *   0 on success, a negative errno value otherwise and rte_errno is set.
897  */
898 int
899 mlx5_flow_validate_action_flag(uint64_t action_flags,
900 			       const struct rte_flow_attr *attr,
901 			       struct rte_flow_error *error)
902 {
903 
904 	if (action_flags & MLX5_FLOW_ACTION_DROP)
905 		return rte_flow_error_set(error, EINVAL,
906 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
907 					  "can't drop and flag in same flow");
908 	if (action_flags & MLX5_FLOW_ACTION_MARK)
909 		return rte_flow_error_set(error, EINVAL,
910 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
911 					  "can't mark and flag in same flow");
912 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
913 		return rte_flow_error_set(error, EINVAL,
914 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
915 					  "can't have 2 flag"
916 					  " actions in same flow");
917 	if (attr->egress)
918 		return rte_flow_error_set(error, ENOTSUP,
919 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
920 					  "flag action not supported for "
921 					  "egress");
922 	return 0;
923 }
924 
925 /*
926  * Validate the mark action.
927  *
928  * @param[in] action
929  *   Pointer to the queue action.
930  * @param[in] action_flags
931  *   Bit-fields that holds the actions detected until now.
932  * @param[in] attr
933  *   Attributes of flow that includes this action.
934  * @param[out] error
935  *   Pointer to error structure.
936  *
937  * @return
938  *   0 on success, a negative errno value otherwise and rte_errno is set.
939  */
940 int
941 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
942 			       uint64_t action_flags,
943 			       const struct rte_flow_attr *attr,
944 			       struct rte_flow_error *error)
945 {
946 	const struct rte_flow_action_mark *mark = action->conf;
947 
948 	if (!mark)
949 		return rte_flow_error_set(error, EINVAL,
950 					  RTE_FLOW_ERROR_TYPE_ACTION,
951 					  action,
952 					  "configuration cannot be null");
953 	if (mark->id >= MLX5_FLOW_MARK_MAX)
954 		return rte_flow_error_set(error, EINVAL,
955 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
956 					  &mark->id,
957 					  "mark id must in 0 <= id < "
958 					  RTE_STR(MLX5_FLOW_MARK_MAX));
959 	if (action_flags & MLX5_FLOW_ACTION_DROP)
960 		return rte_flow_error_set(error, EINVAL,
961 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
962 					  "can't drop and mark in same flow");
963 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
964 		return rte_flow_error_set(error, EINVAL,
965 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
966 					  "can't flag and mark in same flow");
967 	if (action_flags & MLX5_FLOW_ACTION_MARK)
968 		return rte_flow_error_set(error, EINVAL,
969 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
970 					  "can't have 2 mark actions in same"
971 					  " flow");
972 	if (attr->egress)
973 		return rte_flow_error_set(error, ENOTSUP,
974 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
975 					  "mark action not supported for "
976 					  "egress");
977 	return 0;
978 }
979 
980 /*
981  * Validate the drop action.
982  *
983  * @param[in] action_flags
984  *   Bit-fields that holds the actions detected until now.
985  * @param[in] attr
986  *   Attributes of flow that includes this action.
987  * @param[out] error
988  *   Pointer to error structure.
989  *
990  * @return
991  *   0 on success, a negative errno value otherwise and rte_errno is set.
992  */
993 int
994 mlx5_flow_validate_action_drop(uint64_t action_flags,
995 			       const struct rte_flow_attr *attr,
996 			       struct rte_flow_error *error)
997 {
998 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
999 		return rte_flow_error_set(error, EINVAL,
1000 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1001 					  "can't drop and flag in same flow");
1002 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1003 		return rte_flow_error_set(error, EINVAL,
1004 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1005 					  "can't drop and mark in same flow");
1006 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
1007 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1008 		return rte_flow_error_set(error, EINVAL,
1009 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1010 					  "can't have 2 fate actions in"
1011 					  " same flow");
1012 	if (attr->egress)
1013 		return rte_flow_error_set(error, ENOTSUP,
1014 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1015 					  "drop action not supported for "
1016 					  "egress");
1017 	return 0;
1018 }
1019 
1020 /*
1021  * Validate the queue action.
1022  *
1023  * @param[in] action
1024  *   Pointer to the queue action.
1025  * @param[in] action_flags
1026  *   Bit-fields that holds the actions detected until now.
1027  * @param[in] dev
1028  *   Pointer to the Ethernet device structure.
1029  * @param[in] attr
1030  *   Attributes of flow that includes this action.
1031  * @param[out] error
1032  *   Pointer to error structure.
1033  *
1034  * @return
1035  *   0 on success, a negative errno value otherwise and rte_errno is set.
1036  */
1037 int
1038 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1039 				uint64_t action_flags,
1040 				struct rte_eth_dev *dev,
1041 				const struct rte_flow_attr *attr,
1042 				struct rte_flow_error *error)
1043 {
1044 	struct mlx5_priv *priv = dev->data->dev_private;
1045 	const struct rte_flow_action_queue *queue = action->conf;
1046 
1047 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1048 		return rte_flow_error_set(error, EINVAL,
1049 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1050 					  "can't have 2 fate actions in"
1051 					  " same flow");
1052 	if (!priv->rxqs_n)
1053 		return rte_flow_error_set(error, EINVAL,
1054 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1055 					  NULL, "No Rx queues configured");
1056 	if (queue->index >= priv->rxqs_n)
1057 		return rte_flow_error_set(error, EINVAL,
1058 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1059 					  &queue->index,
1060 					  "queue index out of range");
1061 	if (!(*priv->rxqs)[queue->index])
1062 		return rte_flow_error_set(error, EINVAL,
1063 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1064 					  &queue->index,
1065 					  "queue is not configured");
1066 	if (attr->egress)
1067 		return rte_flow_error_set(error, ENOTSUP,
1068 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1069 					  "queue action not supported for "
1070 					  "egress");
1071 	return 0;
1072 }
1073 
1074 /*
1075  * Validate the rss action.
1076  *
1077  * @param[in] action
1078  *   Pointer to the queue action.
1079  * @param[in] action_flags
1080  *   Bit-fields that holds the actions detected until now.
1081  * @param[in] dev
1082  *   Pointer to the Ethernet device structure.
1083  * @param[in] attr
1084  *   Attributes of flow that includes this action.
1085  * @param[in] item_flags
1086  *   Items that were detected.
1087  * @param[out] error
1088  *   Pointer to error structure.
1089  *
1090  * @return
1091  *   0 on success, a negative errno value otherwise and rte_errno is set.
1092  */
1093 int
1094 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1095 			      uint64_t action_flags,
1096 			      struct rte_eth_dev *dev,
1097 			      const struct rte_flow_attr *attr,
1098 			      uint64_t item_flags,
1099 			      struct rte_flow_error *error)
1100 {
1101 	struct mlx5_priv *priv = dev->data->dev_private;
1102 	const struct rte_flow_action_rss *rss = action->conf;
1103 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1104 	unsigned int i;
1105 
1106 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1107 		return rte_flow_error_set(error, EINVAL,
1108 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1109 					  "can't have 2 fate actions"
1110 					  " in same flow");
1111 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1112 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1113 		return rte_flow_error_set(error, ENOTSUP,
1114 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1115 					  &rss->func,
1116 					  "RSS hash function not supported");
1117 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1118 	if (rss->level > 2)
1119 #else
1120 	if (rss->level > 1)
1121 #endif
1122 		return rte_flow_error_set(error, ENOTSUP,
1123 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1124 					  &rss->level,
1125 					  "tunnel RSS is not supported");
1126 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1127 	if (rss->key_len == 0 && rss->key != NULL)
1128 		return rte_flow_error_set(error, ENOTSUP,
1129 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1130 					  &rss->key_len,
1131 					  "RSS hash key length 0");
1132 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1133 		return rte_flow_error_set(error, ENOTSUP,
1134 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1135 					  &rss->key_len,
1136 					  "RSS hash key too small");
1137 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1138 		return rte_flow_error_set(error, ENOTSUP,
1139 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1140 					  &rss->key_len,
1141 					  "RSS hash key too large");
1142 	if (rss->queue_num > priv->config.ind_table_max_size)
1143 		return rte_flow_error_set(error, ENOTSUP,
1144 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1145 					  &rss->queue_num,
1146 					  "number of queues too large");
1147 	if (rss->types & MLX5_RSS_HF_MASK)
1148 		return rte_flow_error_set(error, ENOTSUP,
1149 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1150 					  &rss->types,
1151 					  "some RSS protocols are not"
1152 					  " supported");
1153 	if (!priv->rxqs_n)
1154 		return rte_flow_error_set(error, EINVAL,
1155 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1156 					  NULL, "No Rx queues configured");
1157 	if (!rss->queue_num)
1158 		return rte_flow_error_set(error, EINVAL,
1159 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1160 					  NULL, "No queues configured");
1161 	for (i = 0; i != rss->queue_num; ++i) {
1162 		if (rss->queue[i] >= priv->rxqs_n)
1163 			return rte_flow_error_set
1164 				(error, EINVAL,
1165 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1166 				 &rss->queue[i], "queue index out of range");
1167 		if (!(*priv->rxqs)[rss->queue[i]])
1168 			return rte_flow_error_set
1169 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 				 &rss->queue[i], "queue is not configured");
1171 	}
1172 	if (attr->egress)
1173 		return rte_flow_error_set(error, ENOTSUP,
1174 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1175 					  "rss action not supported for "
1176 					  "egress");
1177 	if (rss->level > 1 &&  !tunnel)
1178 		return rte_flow_error_set(error, EINVAL,
1179 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1180 					  "inner RSS is not supported for "
1181 					  "non-tunnel flows");
1182 	return 0;
1183 }
1184 
1185 /*
1186  * Validate the count action.
1187  *
1188  * @param[in] dev
1189  *   Pointer to the Ethernet device structure.
1190  * @param[in] attr
1191  *   Attributes of flow that includes this action.
1192  * @param[out] error
1193  *   Pointer to error structure.
1194  *
1195  * @return
1196  *   0 on success, a negative errno value otherwise and rte_errno is set.
1197  */
1198 int
1199 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1200 				const struct rte_flow_attr *attr,
1201 				struct rte_flow_error *error)
1202 {
1203 	if (attr->egress)
1204 		return rte_flow_error_set(error, ENOTSUP,
1205 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1206 					  "count action not supported for "
1207 					  "egress");
1208 	return 0;
1209 }
1210 
1211 /**
1212  * Verify the @p attributes will be correctly understood by the NIC and store
1213  * them in the @p flow if everything is correct.
1214  *
1215  * @param[in] dev
1216  *   Pointer to the Ethernet device structure.
1217  * @param[in] attributes
1218  *   Pointer to flow attributes
1219  * @param[out] error
1220  *   Pointer to error structure.
1221  *
1222  * @return
1223  *   0 on success, a negative errno value otherwise and rte_errno is set.
1224  */
1225 int
1226 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1227 			      const struct rte_flow_attr *attributes,
1228 			      struct rte_flow_error *error)
1229 {
1230 	struct mlx5_priv *priv = dev->data->dev_private;
1231 	uint32_t priority_max = priv->config.flow_prio - 1;
1232 
1233 	if (attributes->group)
1234 		return rte_flow_error_set(error, ENOTSUP,
1235 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1236 					  NULL, "groups is not supported");
1237 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1238 	    attributes->priority >= priority_max)
1239 		return rte_flow_error_set(error, ENOTSUP,
1240 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1241 					  NULL, "priority out of range");
1242 	if (attributes->egress)
1243 		return rte_flow_error_set(error, ENOTSUP,
1244 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1245 					  "egress is not supported");
1246 	if (attributes->transfer && !priv->config.dv_esw_en)
1247 		return rte_flow_error_set(error, ENOTSUP,
1248 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1249 					  NULL, "transfer is not supported");
1250 	if (!attributes->ingress)
1251 		return rte_flow_error_set(error, EINVAL,
1252 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1253 					  NULL,
1254 					  "ingress attribute is mandatory");
1255 	return 0;
1256 }
1257 
1258 /**
1259  * Validate ICMP6 item.
1260  *
1261  * @param[in] item
1262  *   Item specification.
1263  * @param[in] item_flags
1264  *   Bit-fields that holds the items detected until now.
1265  * @param[out] error
1266  *   Pointer to error structure.
1267  *
1268  * @return
1269  *   0 on success, a negative errno value otherwise and rte_errno is set.
1270  */
1271 int
1272 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1273 			       uint64_t item_flags,
1274 			       uint8_t target_protocol,
1275 			       struct rte_flow_error *error)
1276 {
1277 	const struct rte_flow_item_icmp6 *mask = item->mask;
1278 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1279 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1280 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1281 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1282 				      MLX5_FLOW_LAYER_OUTER_L4;
1283 	int ret;
1284 
1285 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1286 		return rte_flow_error_set(error, EINVAL,
1287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1288 					  "protocol filtering not compatible"
1289 					  " with ICMP6 layer");
1290 	if (!(item_flags & l3m))
1291 		return rte_flow_error_set(error, EINVAL,
1292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1293 					  "IPv6 is mandatory to filter on"
1294 					  " ICMP6");
1295 	if (item_flags & l4m)
1296 		return rte_flow_error_set(error, EINVAL,
1297 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1298 					  "multiple L4 layers not supported");
1299 	if (!mask)
1300 		mask = &rte_flow_item_icmp6_mask;
1301 	ret = mlx5_flow_item_acceptable
1302 		(item, (const uint8_t *)mask,
1303 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1304 		 sizeof(struct rte_flow_item_icmp6), error);
1305 	if (ret < 0)
1306 		return ret;
1307 	return 0;
1308 }
1309 
1310 /**
1311  * Validate ICMP item.
1312  *
1313  * @param[in] item
1314  *   Item specification.
1315  * @param[in] item_flags
1316  *   Bit-fields that holds the items detected until now.
1317  * @param[out] error
1318  *   Pointer to error structure.
1319  *
1320  * @return
1321  *   0 on success, a negative errno value otherwise and rte_errno is set.
1322  */
1323 int
1324 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1325 			     uint64_t item_flags,
1326 			     uint8_t target_protocol,
1327 			     struct rte_flow_error *error)
1328 {
1329 	const struct rte_flow_item_icmp *mask = item->mask;
1330 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1331 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1332 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1333 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1334 				      MLX5_FLOW_LAYER_OUTER_L4;
1335 	int ret;
1336 
1337 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1338 		return rte_flow_error_set(error, EINVAL,
1339 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1340 					  "protocol filtering not compatible"
1341 					  " with ICMP layer");
1342 	if (!(item_flags & l3m))
1343 		return rte_flow_error_set(error, EINVAL,
1344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1345 					  "IPv4 is mandatory to filter"
1346 					  " on ICMP");
1347 	if (item_flags & l4m)
1348 		return rte_flow_error_set(error, EINVAL,
1349 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1350 					  "multiple L4 layers not supported");
1351 	if (!mask)
1352 		mask = &rte_flow_item_icmp_mask;
1353 	ret = mlx5_flow_item_acceptable
1354 		(item, (const uint8_t *)mask,
1355 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1356 		 sizeof(struct rte_flow_item_icmp), error);
1357 	if (ret < 0)
1358 		return ret;
1359 	return 0;
1360 }
1361 
1362 /**
1363  * Validate Ethernet item.
1364  *
1365  * @param[in] item
1366  *   Item specification.
1367  * @param[in] item_flags
1368  *   Bit-fields that holds the items detected until now.
1369  * @param[out] error
1370  *   Pointer to error structure.
1371  *
1372  * @return
1373  *   0 on success, a negative errno value otherwise and rte_errno is set.
1374  */
1375 int
1376 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1377 			    uint64_t item_flags,
1378 			    struct rte_flow_error *error)
1379 {
1380 	const struct rte_flow_item_eth *mask = item->mask;
1381 	const struct rte_flow_item_eth nic_mask = {
1382 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1383 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1384 		.type = RTE_BE16(0xffff),
1385 	};
1386 	int ret;
1387 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1388 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1389 				       MLX5_FLOW_LAYER_OUTER_L2;
1390 
1391 	if (item_flags & ethm)
1392 		return rte_flow_error_set(error, ENOTSUP,
1393 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1394 					  "multiple L2 layers not supported");
1395 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1396 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1397 		return rte_flow_error_set(error, EINVAL,
1398 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1399 					  "L2 layer should not follow "
1400 					  "L3 layers");
1401 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1402 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1403 		return rte_flow_error_set(error, EINVAL,
1404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 					  "L2 layer should not follow VLAN");
1406 	if (!mask)
1407 		mask = &rte_flow_item_eth_mask;
1408 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1409 					(const uint8_t *)&nic_mask,
1410 					sizeof(struct rte_flow_item_eth),
1411 					error);
1412 	return ret;
1413 }
1414 
1415 /**
1416  * Validate VLAN item.
1417  *
1418  * @param[in] item
1419  *   Item specification.
1420  * @param[in] item_flags
1421  *   Bit-fields that holds the items detected until now.
1422  * @param[in] dev
1423  *   Ethernet device flow is being created on.
1424  * @param[out] error
1425  *   Pointer to error structure.
1426  *
1427  * @return
1428  *   0 on success, a negative errno value otherwise and rte_errno is set.
1429  */
1430 int
1431 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1432 			     uint64_t item_flags,
1433 			     struct rte_eth_dev *dev,
1434 			     struct rte_flow_error *error)
1435 {
1436 	const struct rte_flow_item_vlan *spec = item->spec;
1437 	const struct rte_flow_item_vlan *mask = item->mask;
1438 	const struct rte_flow_item_vlan nic_mask = {
1439 		.tci = RTE_BE16(UINT16_MAX),
1440 		.inner_type = RTE_BE16(UINT16_MAX),
1441 	};
1442 	uint16_t vlan_tag = 0;
1443 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1444 	int ret;
1445 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1446 					MLX5_FLOW_LAYER_INNER_L4) :
1447 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1448 					MLX5_FLOW_LAYER_OUTER_L4);
1449 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1450 					MLX5_FLOW_LAYER_OUTER_VLAN;
1451 
1452 	if (item_flags & vlanm)
1453 		return rte_flow_error_set(error, EINVAL,
1454 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1455 					  "multiple VLAN layers not supported");
1456 	else if ((item_flags & l34m) != 0)
1457 		return rte_flow_error_set(error, EINVAL,
1458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1459 					  "VLAN cannot follow L3/L4 layer");
1460 	if (!mask)
1461 		mask = &rte_flow_item_vlan_mask;
1462 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1463 					(const uint8_t *)&nic_mask,
1464 					sizeof(struct rte_flow_item_vlan),
1465 					error);
1466 	if (ret)
1467 		return ret;
1468 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1469 		struct mlx5_priv *priv = dev->data->dev_private;
1470 
1471 		if (priv->vmwa_context) {
1472 			/*
1473 			 * Non-NULL context means we have a virtual machine
1474 			 * and SR-IOV enabled, we have to create VLAN interface
1475 			 * to make hypervisor to setup E-Switch vport
1476 			 * context correctly. We avoid creating the multiple
1477 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1478 			 */
1479 			return rte_flow_error_set(error, EINVAL,
1480 						  RTE_FLOW_ERROR_TYPE_ITEM,
1481 						  item,
1482 						  "VLAN tag mask is not"
1483 						  " supported in virtual"
1484 						  " environment");
1485 		}
1486 	}
1487 	if (spec) {
1488 		vlan_tag = spec->tci;
1489 		vlan_tag &= mask->tci;
1490 	}
1491 	/*
1492 	 * From verbs perspective an empty VLAN is equivalent
1493 	 * to a packet without VLAN layer.
1494 	 */
1495 	if (!vlan_tag)
1496 		return rte_flow_error_set(error, EINVAL,
1497 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1498 					  item->spec,
1499 					  "VLAN cannot be empty");
1500 	return 0;
1501 }
1502 
1503 /**
1504  * Validate IPV4 item.
1505  *
1506  * @param[in] item
1507  *   Item specification.
1508  * @param[in] item_flags
1509  *   Bit-fields that holds the items detected until now.
1510  * @param[in] acc_mask
1511  *   Acceptable mask, if NULL default internal default mask
1512  *   will be used to check whether item fields are supported.
1513  * @param[out] error
1514  *   Pointer to error structure.
1515  *
1516  * @return
1517  *   0 on success, a negative errno value otherwise and rte_errno is set.
1518  */
1519 int
1520 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1521 			     uint64_t item_flags,
1522 			     uint64_t last_item,
1523 			     uint16_t ether_type,
1524 			     const struct rte_flow_item_ipv4 *acc_mask,
1525 			     struct rte_flow_error *error)
1526 {
1527 	const struct rte_flow_item_ipv4 *mask = item->mask;
1528 	const struct rte_flow_item_ipv4 *spec = item->spec;
1529 	const struct rte_flow_item_ipv4 nic_mask = {
1530 		.hdr = {
1531 			.src_addr = RTE_BE32(0xffffffff),
1532 			.dst_addr = RTE_BE32(0xffffffff),
1533 			.type_of_service = 0xff,
1534 			.next_proto_id = 0xff,
1535 		},
1536 	};
1537 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1538 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1539 				      MLX5_FLOW_LAYER_OUTER_L3;
1540 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1541 				      MLX5_FLOW_LAYER_OUTER_L4;
1542 	int ret;
1543 	uint8_t next_proto = 0xFF;
1544 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1545 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1546 				  MLX5_FLOW_LAYER_INNER_VLAN);
1547 
1548 	if ((last_item & l2_vlan) && ether_type &&
1549 	    ether_type != RTE_ETHER_TYPE_IPV4)
1550 		return rte_flow_error_set(error, EINVAL,
1551 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1552 					  "IPv4 cannot follow L2/VLAN layer "
1553 					  "which ether type is not IPv4");
1554 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1555 		if (mask && spec)
1556 			next_proto = mask->hdr.next_proto_id &
1557 				     spec->hdr.next_proto_id;
1558 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1559 			return rte_flow_error_set(error, EINVAL,
1560 						  RTE_FLOW_ERROR_TYPE_ITEM,
1561 						  item,
1562 						  "multiple tunnel "
1563 						  "not supported");
1564 	}
1565 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1566 		return rte_flow_error_set(error, EINVAL,
1567 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1568 					  "wrong tunnel type - IPv6 specified "
1569 					  "but IPv4 item provided");
1570 	if (item_flags & l3m)
1571 		return rte_flow_error_set(error, ENOTSUP,
1572 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1573 					  "multiple L3 layers not supported");
1574 	else if (item_flags & l4m)
1575 		return rte_flow_error_set(error, EINVAL,
1576 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1577 					  "L3 cannot follow an L4 layer.");
1578 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1579 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1580 		return rte_flow_error_set(error, EINVAL,
1581 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1582 					  "L3 cannot follow an NVGRE layer.");
1583 	if (!mask)
1584 		mask = &rte_flow_item_ipv4_mask;
1585 	else if (mask->hdr.next_proto_id != 0 &&
1586 		 mask->hdr.next_proto_id != 0xff)
1587 		return rte_flow_error_set(error, EINVAL,
1588 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1589 					  "partial mask is not supported"
1590 					  " for protocol");
1591 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1592 					acc_mask ? (const uint8_t *)acc_mask
1593 						 : (const uint8_t *)&nic_mask,
1594 					sizeof(struct rte_flow_item_ipv4),
1595 					error);
1596 	if (ret < 0)
1597 		return ret;
1598 	return 0;
1599 }
1600 
1601 /**
1602  * Validate IPV6 item.
1603  *
1604  * @param[in] item
1605  *   Item specification.
1606  * @param[in] item_flags
1607  *   Bit-fields that holds the items detected until now.
1608  * @param[in] acc_mask
1609  *   Acceptable mask, if NULL default internal default mask
1610  *   will be used to check whether item fields are supported.
1611  * @param[out] error
1612  *   Pointer to error structure.
1613  *
1614  * @return
1615  *   0 on success, a negative errno value otherwise and rte_errno is set.
1616  */
1617 int
1618 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1619 			     uint64_t item_flags,
1620 			     uint64_t last_item,
1621 			     uint16_t ether_type,
1622 			     const struct rte_flow_item_ipv6 *acc_mask,
1623 			     struct rte_flow_error *error)
1624 {
1625 	const struct rte_flow_item_ipv6 *mask = item->mask;
1626 	const struct rte_flow_item_ipv6 *spec = item->spec;
1627 	const struct rte_flow_item_ipv6 nic_mask = {
1628 		.hdr = {
1629 			.src_addr =
1630 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1631 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1632 			.dst_addr =
1633 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1634 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1635 			.vtc_flow = RTE_BE32(0xffffffff),
1636 			.proto = 0xff,
1637 			.hop_limits = 0xff,
1638 		},
1639 	};
1640 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1641 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1642 				      MLX5_FLOW_LAYER_OUTER_L3;
1643 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1644 				      MLX5_FLOW_LAYER_OUTER_L4;
1645 	int ret;
1646 	uint8_t next_proto = 0xFF;
1647 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1648 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1649 				  MLX5_FLOW_LAYER_INNER_VLAN);
1650 
1651 	if ((last_item & l2_vlan) && ether_type &&
1652 	    ether_type != RTE_ETHER_TYPE_IPV6)
1653 		return rte_flow_error_set(error, EINVAL,
1654 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1655 					  "IPv6 cannot follow L2/VLAN layer "
1656 					  "which ether type is not IPv6");
1657 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1658 		if (mask && spec)
1659 			next_proto = mask->hdr.proto & spec->hdr.proto;
1660 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1661 			return rte_flow_error_set(error, EINVAL,
1662 						  RTE_FLOW_ERROR_TYPE_ITEM,
1663 						  item,
1664 						  "multiple tunnel "
1665 						  "not supported");
1666 	}
1667 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1668 		return rte_flow_error_set(error, EINVAL,
1669 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1670 					  "wrong tunnel type - IPv4 specified "
1671 					  "but IPv6 item provided");
1672 	if (item_flags & l3m)
1673 		return rte_flow_error_set(error, ENOTSUP,
1674 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1675 					  "multiple L3 layers not supported");
1676 	else if (item_flags & l4m)
1677 		return rte_flow_error_set(error, EINVAL,
1678 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 					  "L3 cannot follow an L4 layer.");
1680 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1681 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1682 		return rte_flow_error_set(error, EINVAL,
1683 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1684 					  "L3 cannot follow an NVGRE layer.");
1685 	if (!mask)
1686 		mask = &rte_flow_item_ipv6_mask;
1687 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1688 					acc_mask ? (const uint8_t *)acc_mask
1689 						 : (const uint8_t *)&nic_mask,
1690 					sizeof(struct rte_flow_item_ipv6),
1691 					error);
1692 	if (ret < 0)
1693 		return ret;
1694 	return 0;
1695 }
1696 
1697 /**
1698  * Validate UDP item.
1699  *
1700  * @param[in] item
1701  *   Item specification.
1702  * @param[in] item_flags
1703  *   Bit-fields that holds the items detected until now.
1704  * @param[in] target_protocol
1705  *   The next protocol in the previous item.
1706  * @param[in] flow_mask
1707  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1708  * @param[out] error
1709  *   Pointer to error structure.
1710  *
1711  * @return
1712  *   0 on success, a negative errno value otherwise and rte_errno is set.
1713  */
1714 int
1715 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1716 			    uint64_t item_flags,
1717 			    uint8_t target_protocol,
1718 			    struct rte_flow_error *error)
1719 {
1720 	const struct rte_flow_item_udp *mask = item->mask;
1721 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1722 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1723 				      MLX5_FLOW_LAYER_OUTER_L3;
1724 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1725 				      MLX5_FLOW_LAYER_OUTER_L4;
1726 	int ret;
1727 
1728 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1729 		return rte_flow_error_set(error, EINVAL,
1730 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1731 					  "protocol filtering not compatible"
1732 					  " with UDP layer");
1733 	if (!(item_flags & l3m))
1734 		return rte_flow_error_set(error, EINVAL,
1735 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1736 					  "L3 is mandatory to filter on L4");
1737 	if (item_flags & l4m)
1738 		return rte_flow_error_set(error, EINVAL,
1739 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1740 					  "multiple L4 layers not supported");
1741 	if (!mask)
1742 		mask = &rte_flow_item_udp_mask;
1743 	ret = mlx5_flow_item_acceptable
1744 		(item, (const uint8_t *)mask,
1745 		 (const uint8_t *)&rte_flow_item_udp_mask,
1746 		 sizeof(struct rte_flow_item_udp), error);
1747 	if (ret < 0)
1748 		return ret;
1749 	return 0;
1750 }
1751 
1752 /**
1753  * Validate TCP item.
1754  *
1755  * @param[in] item
1756  *   Item specification.
1757  * @param[in] item_flags
1758  *   Bit-fields that holds the items detected until now.
1759  * @param[in] target_protocol
1760  *   The next protocol in the previous item.
1761  * @param[out] error
1762  *   Pointer to error structure.
1763  *
1764  * @return
1765  *   0 on success, a negative errno value otherwise and rte_errno is set.
1766  */
1767 int
1768 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1769 			    uint64_t item_flags,
1770 			    uint8_t target_protocol,
1771 			    const struct rte_flow_item_tcp *flow_mask,
1772 			    struct rte_flow_error *error)
1773 {
1774 	const struct rte_flow_item_tcp *mask = item->mask;
1775 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1776 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1777 				      MLX5_FLOW_LAYER_OUTER_L3;
1778 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1779 				      MLX5_FLOW_LAYER_OUTER_L4;
1780 	int ret;
1781 
1782 	assert(flow_mask);
1783 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1784 		return rte_flow_error_set(error, EINVAL,
1785 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1786 					  "protocol filtering not compatible"
1787 					  " with TCP layer");
1788 	if (!(item_flags & l3m))
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1791 					  "L3 is mandatory to filter on L4");
1792 	if (item_flags & l4m)
1793 		return rte_flow_error_set(error, EINVAL,
1794 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1795 					  "multiple L4 layers not supported");
1796 	if (!mask)
1797 		mask = &rte_flow_item_tcp_mask;
1798 	ret = mlx5_flow_item_acceptable
1799 		(item, (const uint8_t *)mask,
1800 		 (const uint8_t *)flow_mask,
1801 		 sizeof(struct rte_flow_item_tcp), error);
1802 	if (ret < 0)
1803 		return ret;
1804 	return 0;
1805 }
1806 
1807 /**
1808  * Validate VXLAN item.
1809  *
1810  * @param[in] item
1811  *   Item specification.
1812  * @param[in] item_flags
1813  *   Bit-fields that holds the items detected until now.
1814  * @param[in] target_protocol
1815  *   The next protocol in the previous item.
1816  * @param[out] error
1817  *   Pointer to error structure.
1818  *
1819  * @return
1820  *   0 on success, a negative errno value otherwise and rte_errno is set.
1821  */
1822 int
1823 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1824 			      uint64_t item_flags,
1825 			      struct rte_flow_error *error)
1826 {
1827 	const struct rte_flow_item_vxlan *spec = item->spec;
1828 	const struct rte_flow_item_vxlan *mask = item->mask;
1829 	int ret;
1830 	union vni {
1831 		uint32_t vlan_id;
1832 		uint8_t vni[4];
1833 	} id = { .vlan_id = 0, };
1834 	uint32_t vlan_id = 0;
1835 
1836 
1837 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1838 		return rte_flow_error_set(error, ENOTSUP,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "multiple tunnel layers not"
1841 					  " supported");
1842 	/*
1843 	 * Verify only UDPv4 is present as defined in
1844 	 * https://tools.ietf.org/html/rfc7348
1845 	 */
1846 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1847 		return rte_flow_error_set(error, EINVAL,
1848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1849 					  "no outer UDP layer found");
1850 	if (!mask)
1851 		mask = &rte_flow_item_vxlan_mask;
1852 	ret = mlx5_flow_item_acceptable
1853 		(item, (const uint8_t *)mask,
1854 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1855 		 sizeof(struct rte_flow_item_vxlan),
1856 		 error);
1857 	if (ret < 0)
1858 		return ret;
1859 	if (spec) {
1860 		memcpy(&id.vni[1], spec->vni, 3);
1861 		vlan_id = id.vlan_id;
1862 		memcpy(&id.vni[1], mask->vni, 3);
1863 		vlan_id &= id.vlan_id;
1864 	}
1865 	/*
1866 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1867 	 * only this layer is defined in the Verbs specification it is
1868 	 * interpreted as wildcard and all packets will match this
1869 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1870 	 * udp), all packets matching the layers before will also
1871 	 * match this rule.  To avoid such situation, VNI 0 is
1872 	 * currently refused.
1873 	 */
1874 	if (!vlan_id)
1875 		return rte_flow_error_set(error, ENOTSUP,
1876 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1877 					  "VXLAN vni cannot be 0");
1878 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1879 		return rte_flow_error_set(error, ENOTSUP,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "VXLAN tunnel must be fully defined");
1882 	return 0;
1883 }
1884 
1885 /**
1886  * Validate VXLAN_GPE item.
1887  *
1888  * @param[in] item
1889  *   Item specification.
1890  * @param[in] item_flags
1891  *   Bit-fields that holds the items detected until now.
1892  * @param[in] priv
1893  *   Pointer to the private data structure.
1894  * @param[in] target_protocol
1895  *   The next protocol in the previous item.
1896  * @param[out] error
1897  *   Pointer to error structure.
1898  *
1899  * @return
1900  *   0 on success, a negative errno value otherwise and rte_errno is set.
1901  */
1902 int
1903 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1904 				  uint64_t item_flags,
1905 				  struct rte_eth_dev *dev,
1906 				  struct rte_flow_error *error)
1907 {
1908 	struct mlx5_priv *priv = dev->data->dev_private;
1909 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1910 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1911 	int ret;
1912 	union vni {
1913 		uint32_t vlan_id;
1914 		uint8_t vni[4];
1915 	} id = { .vlan_id = 0, };
1916 	uint32_t vlan_id = 0;
1917 
1918 	if (!priv->config.l3_vxlan_en)
1919 		return rte_flow_error_set(error, ENOTSUP,
1920 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1921 					  "L3 VXLAN is not enabled by device"
1922 					  " parameter and/or not configured in"
1923 					  " firmware");
1924 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1925 		return rte_flow_error_set(error, ENOTSUP,
1926 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 					  "multiple tunnel layers not"
1928 					  " supported");
1929 	/*
1930 	 * Verify only UDPv4 is present as defined in
1931 	 * https://tools.ietf.org/html/rfc7348
1932 	 */
1933 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1934 		return rte_flow_error_set(error, EINVAL,
1935 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1936 					  "no outer UDP layer found");
1937 	if (!mask)
1938 		mask = &rte_flow_item_vxlan_gpe_mask;
1939 	ret = mlx5_flow_item_acceptable
1940 		(item, (const uint8_t *)mask,
1941 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1942 		 sizeof(struct rte_flow_item_vxlan_gpe),
1943 		 error);
1944 	if (ret < 0)
1945 		return ret;
1946 	if (spec) {
1947 		if (spec->protocol)
1948 			return rte_flow_error_set(error, ENOTSUP,
1949 						  RTE_FLOW_ERROR_TYPE_ITEM,
1950 						  item,
1951 						  "VxLAN-GPE protocol"
1952 						  " not supported");
1953 		memcpy(&id.vni[1], spec->vni, 3);
1954 		vlan_id = id.vlan_id;
1955 		memcpy(&id.vni[1], mask->vni, 3);
1956 		vlan_id &= id.vlan_id;
1957 	}
1958 	/*
1959 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1960 	 * layer is defined in the Verbs specification it is interpreted as
1961 	 * wildcard and all packets will match this rule, if it follows a full
1962 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1963 	 * before will also match this rule.  To avoid such situation, VNI 0
1964 	 * is currently refused.
1965 	 */
1966 	if (!vlan_id)
1967 		return rte_flow_error_set(error, ENOTSUP,
1968 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1969 					  "VXLAN-GPE vni cannot be 0");
1970 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1971 		return rte_flow_error_set(error, ENOTSUP,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "VXLAN-GPE tunnel must be fully"
1974 					  " defined");
1975 	return 0;
1976 }
1977 /**
1978  * Validate GRE Key item.
1979  *
1980  * @param[in] item
1981  *   Item specification.
1982  * @param[in] item_flags
1983  *   Bit flags to mark detected items.
1984  * @param[in] gre_item
1985  *   Pointer to gre_item
1986  * @param[out] error
1987  *   Pointer to error structure.
1988  *
1989  * @return
1990  *   0 on success, a negative errno value otherwise and rte_errno is set.
1991  */
1992 int
1993 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1994 				uint64_t item_flags,
1995 				const struct rte_flow_item *gre_item,
1996 				struct rte_flow_error *error)
1997 {
1998 	const rte_be32_t *mask = item->mask;
1999 	int ret = 0;
2000 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2001 	const struct rte_flow_item_gre *gre_spec;
2002 	const struct rte_flow_item_gre *gre_mask;
2003 
2004 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2005 		return rte_flow_error_set(error, ENOTSUP,
2006 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 					  "Multiple GRE key not support");
2008 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2009 		return rte_flow_error_set(error, ENOTSUP,
2010 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2011 					  "No preceding GRE header");
2012 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2013 		return rte_flow_error_set(error, ENOTSUP,
2014 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2015 					  "GRE key following a wrong item");
2016 	gre_mask = gre_item->mask;
2017 	if (!gre_mask)
2018 		gre_mask = &rte_flow_item_gre_mask;
2019 	gre_spec = gre_item->spec;
2020 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2021 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2022 		return rte_flow_error_set(error, EINVAL,
2023 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2024 					  "Key bit must be on");
2025 
2026 	if (!mask)
2027 		mask = &gre_key_default_mask;
2028 	ret = mlx5_flow_item_acceptable
2029 		(item, (const uint8_t *)mask,
2030 		 (const uint8_t *)&gre_key_default_mask,
2031 		 sizeof(rte_be32_t), error);
2032 	return ret;
2033 }
2034 
2035 /**
2036  * Validate GRE item.
2037  *
2038  * @param[in] item
2039  *   Item specification.
2040  * @param[in] item_flags
2041  *   Bit flags to mark detected items.
2042  * @param[in] target_protocol
2043  *   The next protocol in the previous item.
2044  * @param[out] error
2045  *   Pointer to error structure.
2046  *
2047  * @return
2048  *   0 on success, a negative errno value otherwise and rte_errno is set.
2049  */
2050 int
2051 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2052 			    uint64_t item_flags,
2053 			    uint8_t target_protocol,
2054 			    struct rte_flow_error *error)
2055 {
2056 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2057 	const struct rte_flow_item_gre *mask = item->mask;
2058 	int ret;
2059 	const struct rte_flow_item_gre nic_mask = {
2060 		.c_rsvd0_ver = RTE_BE16(0xB000),
2061 		.protocol = RTE_BE16(UINT16_MAX),
2062 	};
2063 
2064 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2065 		return rte_flow_error_set(error, EINVAL,
2066 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2067 					  "protocol filtering not compatible"
2068 					  " with this GRE layer");
2069 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2070 		return rte_flow_error_set(error, ENOTSUP,
2071 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2072 					  "multiple tunnel layers not"
2073 					  " supported");
2074 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2075 		return rte_flow_error_set(error, ENOTSUP,
2076 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2077 					  "L3 Layer is missing");
2078 	if (!mask)
2079 		mask = &rte_flow_item_gre_mask;
2080 	ret = mlx5_flow_item_acceptable
2081 		(item, (const uint8_t *)mask,
2082 		 (const uint8_t *)&nic_mask,
2083 		 sizeof(struct rte_flow_item_gre), error);
2084 	if (ret < 0)
2085 		return ret;
2086 #ifndef HAVE_MLX5DV_DR
2087 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2088 	if (spec && (spec->protocol & mask->protocol))
2089 		return rte_flow_error_set(error, ENOTSUP,
2090 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2091 					  "without MPLS support the"
2092 					  " specification cannot be used for"
2093 					  " filtering");
2094 #endif
2095 #endif
2096 	return 0;
2097 }
2098 
2099 /**
2100  * Validate Geneve item.
2101  *
2102  * @param[in] item
2103  *   Item specification.
2104  * @param[in] itemFlags
2105  *   Bit-fields that holds the items detected until now.
2106  * @param[in] enPriv
2107  *   Pointer to the private data structure.
2108  * @param[out] error
2109  *   Pointer to error structure.
2110  *
2111  * @return
2112  *   0 on success, a negative errno value otherwise and rte_errno is set.
2113  */
2114 
2115 int
2116 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2117 			       uint64_t item_flags,
2118 			       struct rte_eth_dev *dev,
2119 			       struct rte_flow_error *error)
2120 {
2121 	struct mlx5_priv *priv = dev->data->dev_private;
2122 	const struct rte_flow_item_geneve *spec = item->spec;
2123 	const struct rte_flow_item_geneve *mask = item->mask;
2124 	int ret;
2125 	uint16_t gbhdr;
2126 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2127 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2128 	const struct rte_flow_item_geneve nic_mask = {
2129 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2130 		.vni = "\xff\xff\xff",
2131 		.protocol = RTE_BE16(UINT16_MAX),
2132 	};
2133 
2134 	if (!(priv->config.hca_attr.flex_parser_protocols &
2135 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2136 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2137 		return rte_flow_error_set(error, ENOTSUP,
2138 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2139 					  "L3 Geneve is not enabled by device"
2140 					  " parameter and/or not configured in"
2141 					  " firmware");
2142 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2143 		return rte_flow_error_set(error, ENOTSUP,
2144 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2145 					  "multiple tunnel layers not"
2146 					  " supported");
2147 	/*
2148 	 * Verify only UDPv4 is present as defined in
2149 	 * https://tools.ietf.org/html/rfc7348
2150 	 */
2151 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2152 		return rte_flow_error_set(error, EINVAL,
2153 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2154 					  "no outer UDP layer found");
2155 	if (!mask)
2156 		mask = &rte_flow_item_geneve_mask;
2157 	ret = mlx5_flow_item_acceptable
2158 				  (item, (const uint8_t *)mask,
2159 				   (const uint8_t *)&nic_mask,
2160 				   sizeof(struct rte_flow_item_geneve), error);
2161 	if (ret)
2162 		return ret;
2163 	if (spec) {
2164 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2165 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2166 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2167 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2168 			return rte_flow_error_set(error, ENOTSUP,
2169 						  RTE_FLOW_ERROR_TYPE_ITEM,
2170 						  item,
2171 						  "Geneve protocol unsupported"
2172 						  " fields are being used");
2173 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2174 			return rte_flow_error_set
2175 					(error, ENOTSUP,
2176 					 RTE_FLOW_ERROR_TYPE_ITEM,
2177 					 item,
2178 					 "Unsupported Geneve options length");
2179 	}
2180 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2181 		return rte_flow_error_set
2182 				    (error, ENOTSUP,
2183 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2184 				     "Geneve tunnel must be fully defined");
2185 	return 0;
2186 }
2187 
2188 /**
2189  * Validate MPLS item.
2190  *
2191  * @param[in] dev
2192  *   Pointer to the rte_eth_dev structure.
2193  * @param[in] item
2194  *   Item specification.
2195  * @param[in] item_flags
2196  *   Bit-fields that holds the items detected until now.
2197  * @param[in] prev_layer
2198  *   The protocol layer indicated in previous item.
2199  * @param[out] error
2200  *   Pointer to error structure.
2201  *
2202  * @return
2203  *   0 on success, a negative errno value otherwise and rte_errno is set.
2204  */
2205 int
2206 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2207 			     const struct rte_flow_item *item __rte_unused,
2208 			     uint64_t item_flags __rte_unused,
2209 			     uint64_t prev_layer __rte_unused,
2210 			     struct rte_flow_error *error)
2211 {
2212 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2213 	const struct rte_flow_item_mpls *mask = item->mask;
2214 	struct mlx5_priv *priv = dev->data->dev_private;
2215 	int ret;
2216 
2217 	if (!priv->config.mpls_en)
2218 		return rte_flow_error_set(error, ENOTSUP,
2219 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2220 					  "MPLS not supported or"
2221 					  " disabled in firmware"
2222 					  " configuration.");
2223 	/* MPLS over IP, UDP, GRE is allowed */
2224 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2225 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2226 			    MLX5_FLOW_LAYER_GRE)))
2227 		return rte_flow_error_set(error, EINVAL,
2228 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 					  "protocol filtering not compatible"
2230 					  " with MPLS layer");
2231 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2232 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2233 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2234 		return rte_flow_error_set(error, ENOTSUP,
2235 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2236 					  "multiple tunnel layers not"
2237 					  " supported");
2238 	if (!mask)
2239 		mask = &rte_flow_item_mpls_mask;
2240 	ret = mlx5_flow_item_acceptable
2241 		(item, (const uint8_t *)mask,
2242 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2243 		 sizeof(struct rte_flow_item_mpls), error);
2244 	if (ret < 0)
2245 		return ret;
2246 	return 0;
2247 #endif
2248 	return rte_flow_error_set(error, ENOTSUP,
2249 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2250 				  "MPLS is not supported by Verbs, please"
2251 				  " update.");
2252 }
2253 
2254 /**
2255  * Validate NVGRE item.
2256  *
2257  * @param[in] item
2258  *   Item specification.
2259  * @param[in] item_flags
2260  *   Bit flags to mark detected items.
2261  * @param[in] target_protocol
2262  *   The next protocol in the previous item.
2263  * @param[out] error
2264  *   Pointer to error structure.
2265  *
2266  * @return
2267  *   0 on success, a negative errno value otherwise and rte_errno is set.
2268  */
2269 int
2270 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2271 			      uint64_t item_flags,
2272 			      uint8_t target_protocol,
2273 			      struct rte_flow_error *error)
2274 {
2275 	const struct rte_flow_item_nvgre *mask = item->mask;
2276 	int ret;
2277 
2278 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2279 		return rte_flow_error_set(error, EINVAL,
2280 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2281 					  "protocol filtering not compatible"
2282 					  " with this GRE layer");
2283 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2284 		return rte_flow_error_set(error, ENOTSUP,
2285 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2286 					  "multiple tunnel layers not"
2287 					  " supported");
2288 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2289 		return rte_flow_error_set(error, ENOTSUP,
2290 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2291 					  "L3 Layer is missing");
2292 	if (!mask)
2293 		mask = &rte_flow_item_nvgre_mask;
2294 	ret = mlx5_flow_item_acceptable
2295 		(item, (const uint8_t *)mask,
2296 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2297 		 sizeof(struct rte_flow_item_nvgre), error);
2298 	if (ret < 0)
2299 		return ret;
2300 	return 0;
2301 }
2302 
2303 /* Allocate unique ID for the split Q/RSS subflows. */
2304 static uint32_t
2305 flow_qrss_get_id(struct rte_eth_dev *dev)
2306 {
2307 	struct mlx5_priv *priv = dev->data->dev_private;
2308 	uint32_t qrss_id, ret;
2309 
2310 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2311 	if (ret)
2312 		return 0;
2313 	assert(qrss_id);
2314 	return qrss_id;
2315 }
2316 
2317 /* Free unique ID for the split Q/RSS subflows. */
2318 static void
2319 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2320 {
2321 	struct mlx5_priv *priv = dev->data->dev_private;
2322 
2323 	if (qrss_id)
2324 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2325 }
2326 
2327 /**
2328  * Release resource related QUEUE/RSS action split.
2329  *
2330  * @param dev
2331  *   Pointer to Ethernet device.
2332  * @param flow
2333  *   Flow to release id's from.
2334  */
2335 static void
2336 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2337 			     struct rte_flow *flow)
2338 {
2339 	struct mlx5_flow *dev_flow;
2340 
2341 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2342 		if (dev_flow->qrss_id)
2343 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2344 }
2345 
2346 static int
2347 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2348 		   const struct rte_flow_attr *attr __rte_unused,
2349 		   const struct rte_flow_item items[] __rte_unused,
2350 		   const struct rte_flow_action actions[] __rte_unused,
2351 		   bool external __rte_unused,
2352 		   struct rte_flow_error *error)
2353 {
2354 	return rte_flow_error_set(error, ENOTSUP,
2355 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2356 }
2357 
2358 static struct mlx5_flow *
2359 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2360 		  const struct rte_flow_item items[] __rte_unused,
2361 		  const struct rte_flow_action actions[] __rte_unused,
2362 		  struct rte_flow_error *error)
2363 {
2364 	rte_flow_error_set(error, ENOTSUP,
2365 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2366 	return NULL;
2367 }
2368 
2369 static int
2370 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2371 		    struct mlx5_flow *dev_flow __rte_unused,
2372 		    const struct rte_flow_attr *attr __rte_unused,
2373 		    const struct rte_flow_item items[] __rte_unused,
2374 		    const struct rte_flow_action actions[] __rte_unused,
2375 		    struct rte_flow_error *error)
2376 {
2377 	return rte_flow_error_set(error, ENOTSUP,
2378 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2379 }
2380 
2381 static int
2382 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2383 		struct rte_flow *flow __rte_unused,
2384 		struct rte_flow_error *error)
2385 {
2386 	return rte_flow_error_set(error, ENOTSUP,
2387 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2388 }
2389 
2390 static void
2391 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2392 		 struct rte_flow *flow __rte_unused)
2393 {
2394 }
2395 
2396 static void
2397 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2398 		  struct rte_flow *flow __rte_unused)
2399 {
2400 }
2401 
2402 static int
2403 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2404 		struct rte_flow *flow __rte_unused,
2405 		const struct rte_flow_action *actions __rte_unused,
2406 		void *data __rte_unused,
2407 		struct rte_flow_error *error)
2408 {
2409 	return rte_flow_error_set(error, ENOTSUP,
2410 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2411 }
2412 
2413 /* Void driver to protect from null pointer reference. */
2414 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2415 	.validate = flow_null_validate,
2416 	.prepare = flow_null_prepare,
2417 	.translate = flow_null_translate,
2418 	.apply = flow_null_apply,
2419 	.remove = flow_null_remove,
2420 	.destroy = flow_null_destroy,
2421 	.query = flow_null_query,
2422 };
2423 
2424 /**
2425  * Select flow driver type according to flow attributes and device
2426  * configuration.
2427  *
2428  * @param[in] dev
2429  *   Pointer to the dev structure.
2430  * @param[in] attr
2431  *   Pointer to the flow attributes.
2432  *
2433  * @return
2434  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2435  */
2436 static enum mlx5_flow_drv_type
2437 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2438 {
2439 	struct mlx5_priv *priv = dev->data->dev_private;
2440 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2441 
2442 	if (attr->transfer && priv->config.dv_esw_en)
2443 		type = MLX5_FLOW_TYPE_DV;
2444 	if (!attr->transfer)
2445 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2446 						 MLX5_FLOW_TYPE_VERBS;
2447 	return type;
2448 }
2449 
2450 #define flow_get_drv_ops(type) flow_drv_ops[type]
2451 
2452 /**
2453  * Flow driver validation API. This abstracts calling driver specific functions.
2454  * The type of flow driver is determined according to flow attributes.
2455  *
2456  * @param[in] dev
2457  *   Pointer to the dev structure.
2458  * @param[in] attr
2459  *   Pointer to the flow attributes.
2460  * @param[in] items
2461  *   Pointer to the list of items.
2462  * @param[in] actions
2463  *   Pointer to the list of actions.
2464  * @param[in] external
2465  *   This flow rule is created by request external to PMD.
2466  * @param[out] error
2467  *   Pointer to the error structure.
2468  *
2469  * @return
2470  *   0 on success, a negative errno value otherwise and rte_errno is set.
2471  */
2472 static inline int
2473 flow_drv_validate(struct rte_eth_dev *dev,
2474 		  const struct rte_flow_attr *attr,
2475 		  const struct rte_flow_item items[],
2476 		  const struct rte_flow_action actions[],
2477 		  bool external, struct rte_flow_error *error)
2478 {
2479 	const struct mlx5_flow_driver_ops *fops;
2480 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2481 
2482 	fops = flow_get_drv_ops(type);
2483 	return fops->validate(dev, attr, items, actions, external, error);
2484 }
2485 
2486 /**
2487  * Flow driver preparation API. This abstracts calling driver specific
2488  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2489  * calculates the size of memory required for device flow, allocates the memory,
2490  * initializes the device flow and returns the pointer.
2491  *
2492  * @note
2493  *   This function initializes device flow structure such as dv or verbs in
2494  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2495  *   rest. For example, adding returning device flow to flow->dev_flow list and
2496  *   setting backward reference to the flow should be done out of this function.
2497  *   layers field is not filled either.
2498  *
2499  * @param[in] attr
2500  *   Pointer to the flow attributes.
2501  * @param[in] items
2502  *   Pointer to the list of items.
2503  * @param[in] actions
2504  *   Pointer to the list of actions.
2505  * @param[out] error
2506  *   Pointer to the error structure.
2507  *
2508  * @return
2509  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2510  */
2511 static inline struct mlx5_flow *
2512 flow_drv_prepare(const struct rte_flow *flow,
2513 		 const struct rte_flow_attr *attr,
2514 		 const struct rte_flow_item items[],
2515 		 const struct rte_flow_action actions[],
2516 		 struct rte_flow_error *error)
2517 {
2518 	const struct mlx5_flow_driver_ops *fops;
2519 	enum mlx5_flow_drv_type type = flow->drv_type;
2520 
2521 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2522 	fops = flow_get_drv_ops(type);
2523 	return fops->prepare(attr, items, actions, error);
2524 }
2525 
2526 /**
2527  * Flow driver translation API. This abstracts calling driver specific
2528  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2529  * translates a generic flow into a driver flow. flow_drv_prepare() must
2530  * precede.
2531  *
2532  * @note
2533  *   dev_flow->layers could be filled as a result of parsing during translation
2534  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2535  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2536  *   flow->actions could be overwritten even though all the expanded dev_flows
2537  *   have the same actions.
2538  *
2539  * @param[in] dev
2540  *   Pointer to the rte dev structure.
2541  * @param[in, out] dev_flow
2542  *   Pointer to the mlx5 flow.
2543  * @param[in] attr
2544  *   Pointer to the flow attributes.
2545  * @param[in] items
2546  *   Pointer to the list of items.
2547  * @param[in] actions
2548  *   Pointer to the list of actions.
2549  * @param[out] error
2550  *   Pointer to the error structure.
2551  *
2552  * @return
2553  *   0 on success, a negative errno value otherwise and rte_errno is set.
2554  */
2555 static inline int
2556 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2557 		   const struct rte_flow_attr *attr,
2558 		   const struct rte_flow_item items[],
2559 		   const struct rte_flow_action actions[],
2560 		   struct rte_flow_error *error)
2561 {
2562 	const struct mlx5_flow_driver_ops *fops;
2563 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2564 
2565 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2566 	fops = flow_get_drv_ops(type);
2567 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2568 }
2569 
2570 /**
2571  * Flow driver apply API. This abstracts calling driver specific functions.
2572  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2573  * translated driver flows on to device. flow_drv_translate() must precede.
2574  *
2575  * @param[in] dev
2576  *   Pointer to Ethernet device structure.
2577  * @param[in, out] flow
2578  *   Pointer to flow structure.
2579  * @param[out] error
2580  *   Pointer to error structure.
2581  *
2582  * @return
2583  *   0 on success, a negative errno value otherwise and rte_errno is set.
2584  */
2585 static inline int
2586 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2587 	       struct rte_flow_error *error)
2588 {
2589 	const struct mlx5_flow_driver_ops *fops;
2590 	enum mlx5_flow_drv_type type = flow->drv_type;
2591 
2592 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2593 	fops = flow_get_drv_ops(type);
2594 	return fops->apply(dev, flow, error);
2595 }
2596 
2597 /**
2598  * Flow driver remove API. This abstracts calling driver specific functions.
2599  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2600  * on device. All the resources of the flow should be freed by calling
2601  * flow_drv_destroy().
2602  *
2603  * @param[in] dev
2604  *   Pointer to Ethernet device.
2605  * @param[in, out] flow
2606  *   Pointer to flow structure.
2607  */
2608 static inline void
2609 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2610 {
2611 	const struct mlx5_flow_driver_ops *fops;
2612 	enum mlx5_flow_drv_type type = flow->drv_type;
2613 
2614 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2615 	fops = flow_get_drv_ops(type);
2616 	fops->remove(dev, flow);
2617 }
2618 
2619 /**
2620  * Flow driver destroy API. This abstracts calling driver specific functions.
2621  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2622  * on device and releases resources of the flow.
2623  *
2624  * @param[in] dev
2625  *   Pointer to Ethernet device.
2626  * @param[in, out] flow
2627  *   Pointer to flow structure.
2628  */
2629 static inline void
2630 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2631 {
2632 	const struct mlx5_flow_driver_ops *fops;
2633 	enum mlx5_flow_drv_type type = flow->drv_type;
2634 
2635 	flow_mreg_split_qrss_release(dev, flow);
2636 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2637 	fops = flow_get_drv_ops(type);
2638 	fops->destroy(dev, flow);
2639 }
2640 
2641 /**
2642  * Validate a flow supported by the NIC.
2643  *
2644  * @see rte_flow_validate()
2645  * @see rte_flow_ops
2646  */
2647 int
2648 mlx5_flow_validate(struct rte_eth_dev *dev,
2649 		   const struct rte_flow_attr *attr,
2650 		   const struct rte_flow_item items[],
2651 		   const struct rte_flow_action actions[],
2652 		   struct rte_flow_error *error)
2653 {
2654 	int ret;
2655 
2656 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2657 	if (ret < 0)
2658 		return ret;
2659 	return 0;
2660 }
2661 
2662 /**
2663  * Get port id item from the item list.
2664  *
2665  * @param[in] item
2666  *   Pointer to the list of items.
2667  *
2668  * @return
2669  *   Pointer to the port id item if exist, else return NULL.
2670  */
2671 static const struct rte_flow_item *
2672 find_port_id_item(const struct rte_flow_item *item)
2673 {
2674 	assert(item);
2675 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2676 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2677 			return item;
2678 	}
2679 	return NULL;
2680 }
2681 
2682 /**
2683  * Get RSS action from the action list.
2684  *
2685  * @param[in] actions
2686  *   Pointer to the list of actions.
2687  *
2688  * @return
2689  *   Pointer to the RSS action if exist, else return NULL.
2690  */
2691 static const struct rte_flow_action_rss*
2692 flow_get_rss_action(const struct rte_flow_action actions[])
2693 {
2694 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2695 		switch (actions->type) {
2696 		case RTE_FLOW_ACTION_TYPE_RSS:
2697 			return (const struct rte_flow_action_rss *)
2698 			       actions->conf;
2699 		default:
2700 			break;
2701 		}
2702 	}
2703 	return NULL;
2704 }
2705 
2706 static unsigned int
2707 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2708 {
2709 	const struct rte_flow_item *item;
2710 	unsigned int has_vlan = 0;
2711 
2712 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2713 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2714 			has_vlan = 1;
2715 			break;
2716 		}
2717 	}
2718 	if (has_vlan)
2719 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2720 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2721 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2722 			       MLX5_EXPANSION_ROOT_OUTER;
2723 }
2724 
2725 /**
2726  * Get QUEUE/RSS action from the action list.
2727  *
2728  * @param[in] actions
2729  *   Pointer to the list of actions.
2730  * @param[out] qrss
2731  *   Pointer to the return pointer.
2732  * @param[out] qrss_type
2733  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2734  *   if no QUEUE/RSS is found.
2735  *
2736  * @return
2737  *   Total number of actions.
2738  */
2739 static int
2740 flow_parse_qrss_action(const struct rte_flow_action actions[],
2741 		       const struct rte_flow_action **qrss)
2742 {
2743 	int actions_n = 0;
2744 
2745 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2746 		switch (actions->type) {
2747 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2748 		case RTE_FLOW_ACTION_TYPE_RSS:
2749 			*qrss = actions;
2750 			break;
2751 		default:
2752 			break;
2753 		}
2754 		actions_n++;
2755 	}
2756 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2757 	return actions_n + 1;
2758 }
2759 
2760 /**
2761  * Check meter action from the action list.
2762  *
2763  * @param[in] actions
2764  *   Pointer to the list of actions.
2765  * @param[out] mtr
2766  *   Pointer to the meter exist flag.
2767  *
2768  * @return
2769  *   Total number of actions.
2770  */
2771 static int
2772 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2773 {
2774 	int actions_n = 0;
2775 
2776 	assert(mtr);
2777 	*mtr = 0;
2778 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2779 		switch (actions->type) {
2780 		case RTE_FLOW_ACTION_TYPE_METER:
2781 			*mtr = 1;
2782 			break;
2783 		default:
2784 			break;
2785 		}
2786 		actions_n++;
2787 	}
2788 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2789 	return actions_n + 1;
2790 }
2791 
2792 /**
2793  * Check if the flow should be splited due to hairpin.
2794  * The reason for the split is that in current HW we can't
2795  * support encap on Rx, so if a flow have encap we move it
2796  * to Tx.
2797  *
2798  * @param dev
2799  *   Pointer to Ethernet device.
2800  * @param[in] attr
2801  *   Flow rule attributes.
2802  * @param[in] actions
2803  *   Associated actions (list terminated by the END action).
2804  *
2805  * @return
2806  *   > 0 the number of actions and the flow should be split,
2807  *   0 when no split required.
2808  */
2809 static int
2810 flow_check_hairpin_split(struct rte_eth_dev *dev,
2811 			 const struct rte_flow_attr *attr,
2812 			 const struct rte_flow_action actions[])
2813 {
2814 	int queue_action = 0;
2815 	int action_n = 0;
2816 	int encap = 0;
2817 	const struct rte_flow_action_queue *queue;
2818 	const struct rte_flow_action_rss *rss;
2819 	const struct rte_flow_action_raw_encap *raw_encap;
2820 
2821 	if (!attr->ingress)
2822 		return 0;
2823 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2824 		switch (actions->type) {
2825 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2826 			queue = actions->conf;
2827 			if (queue == NULL)
2828 				return 0;
2829 			if (mlx5_rxq_get_type(dev, queue->index) !=
2830 			    MLX5_RXQ_TYPE_HAIRPIN)
2831 				return 0;
2832 			queue_action = 1;
2833 			action_n++;
2834 			break;
2835 		case RTE_FLOW_ACTION_TYPE_RSS:
2836 			rss = actions->conf;
2837 			if (rss == NULL || rss->queue_num == 0)
2838 				return 0;
2839 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2840 			    MLX5_RXQ_TYPE_HAIRPIN)
2841 				return 0;
2842 			queue_action = 1;
2843 			action_n++;
2844 			break;
2845 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2846 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2847 			encap = 1;
2848 			action_n++;
2849 			break;
2850 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2851 			raw_encap = actions->conf;
2852 			if (raw_encap->size >
2853 			    (sizeof(struct rte_flow_item_eth) +
2854 			     sizeof(struct rte_flow_item_ipv4)))
2855 				encap = 1;
2856 			action_n++;
2857 			break;
2858 		default:
2859 			action_n++;
2860 			break;
2861 		}
2862 	}
2863 	if (encap == 1 && queue_action)
2864 		return action_n;
2865 	return 0;
2866 }
2867 
2868 /* Declare flow create/destroy prototype in advance. */
2869 static struct rte_flow *
2870 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2871 		 const struct rte_flow_attr *attr,
2872 		 const struct rte_flow_item items[],
2873 		 const struct rte_flow_action actions[],
2874 		 bool external, struct rte_flow_error *error);
2875 
2876 static void
2877 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2878 		  struct rte_flow *flow);
2879 
2880 /**
2881  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2882  *
2883  * As mark_id is unique, if there's already a registered flow for the mark_id,
2884  * return by increasing the reference counter of the resource. Otherwise, create
2885  * the resource (mcp_res) and flow.
2886  *
2887  * Flow looks like,
2888  *   - If ingress port is ANY and reg_c[1] is mark_id,
2889  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2890  *
2891  * For default flow (zero mark_id), flow is like,
2892  *   - If ingress port is ANY,
2893  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2894  *
2895  * @param dev
2896  *   Pointer to Ethernet device.
2897  * @param mark_id
2898  *   ID of MARK action, zero means default flow for META.
2899  * @param[out] error
2900  *   Perform verbose error reporting if not NULL.
2901  *
2902  * @return
2903  *   Associated resource on success, NULL otherwise and rte_errno is set.
2904  */
2905 static struct mlx5_flow_mreg_copy_resource *
2906 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2907 			  struct rte_flow_error *error)
2908 {
2909 	struct mlx5_priv *priv = dev->data->dev_private;
2910 	struct rte_flow_attr attr = {
2911 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2912 		.ingress = 1,
2913 	};
2914 	struct mlx5_rte_flow_item_tag tag_spec = {
2915 		.data = mark_id,
2916 	};
2917 	struct rte_flow_item items[] = {
2918 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2919 	};
2920 	struct rte_flow_action_mark ftag = {
2921 		.id = mark_id,
2922 	};
2923 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2924 		.dst = REG_B,
2925 		.src = 0,
2926 	};
2927 	struct rte_flow_action_jump jump = {
2928 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2929 	};
2930 	struct rte_flow_action actions[] = {
2931 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2932 	};
2933 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2934 	int ret;
2935 
2936 	/* Fill the register fileds in the flow. */
2937 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2938 	if (ret < 0)
2939 		return NULL;
2940 	tag_spec.id = ret;
2941 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2942 	if (ret < 0)
2943 		return NULL;
2944 	cp_mreg.src = ret;
2945 	/* Check if already registered. */
2946 	assert(priv->mreg_cp_tbl);
2947 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2948 	if (mcp_res) {
2949 		/* For non-default rule. */
2950 		if (mark_id)
2951 			mcp_res->refcnt++;
2952 		assert(mark_id || mcp_res->refcnt == 1);
2953 		return mcp_res;
2954 	}
2955 	/* Provide the full width of FLAG specific value. */
2956 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2957 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2958 	/* Build a new flow. */
2959 	if (mark_id) {
2960 		items[0] = (struct rte_flow_item){
2961 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2962 			.spec = &tag_spec,
2963 		};
2964 		items[1] = (struct rte_flow_item){
2965 			.type = RTE_FLOW_ITEM_TYPE_END,
2966 		};
2967 		actions[0] = (struct rte_flow_action){
2968 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2969 			.conf = &ftag,
2970 		};
2971 		actions[1] = (struct rte_flow_action){
2972 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2973 			.conf = &cp_mreg,
2974 		};
2975 		actions[2] = (struct rte_flow_action){
2976 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2977 			.conf = &jump,
2978 		};
2979 		actions[3] = (struct rte_flow_action){
2980 			.type = RTE_FLOW_ACTION_TYPE_END,
2981 		};
2982 	} else {
2983 		/* Default rule, wildcard match. */
2984 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2985 		items[0] = (struct rte_flow_item){
2986 			.type = RTE_FLOW_ITEM_TYPE_END,
2987 		};
2988 		actions[0] = (struct rte_flow_action){
2989 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2990 			.conf = &cp_mreg,
2991 		};
2992 		actions[1] = (struct rte_flow_action){
2993 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2994 			.conf = &jump,
2995 		};
2996 		actions[2] = (struct rte_flow_action){
2997 			.type = RTE_FLOW_ACTION_TYPE_END,
2998 		};
2999 	}
3000 	/* Build a new entry. */
3001 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3002 	if (!mcp_res) {
3003 		rte_errno = ENOMEM;
3004 		return NULL;
3005 	}
3006 	/*
3007 	 * The copy Flows are not included in any list. There
3008 	 * ones are referenced from other Flows and can not
3009 	 * be applied, removed, deleted in ardbitrary order
3010 	 * by list traversing.
3011 	 */
3012 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3013 					 actions, false, error);
3014 	if (!mcp_res->flow)
3015 		goto error;
3016 	mcp_res->refcnt++;
3017 	mcp_res->hlist_ent.key = mark_id;
3018 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3019 				&mcp_res->hlist_ent);
3020 	assert(!ret);
3021 	if (ret)
3022 		goto error;
3023 	return mcp_res;
3024 error:
3025 	if (mcp_res->flow)
3026 		flow_list_destroy(dev, NULL, mcp_res->flow);
3027 	rte_free(mcp_res);
3028 	return NULL;
3029 }
3030 
3031 /**
3032  * Release flow in RX_CP_TBL.
3033  *
3034  * @param dev
3035  *   Pointer to Ethernet device.
3036  * @flow
3037  *   Parent flow for wich copying is provided.
3038  */
3039 static void
3040 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3041 			  struct rte_flow *flow)
3042 {
3043 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3044 	struct mlx5_priv *priv = dev->data->dev_private;
3045 
3046 	if (!mcp_res || !priv->mreg_cp_tbl)
3047 		return;
3048 	if (flow->copy_applied) {
3049 		assert(mcp_res->appcnt);
3050 		flow->copy_applied = 0;
3051 		--mcp_res->appcnt;
3052 		if (!mcp_res->appcnt)
3053 			flow_drv_remove(dev, mcp_res->flow);
3054 	}
3055 	/*
3056 	 * We do not check availability of metadata registers here,
3057 	 * because copy resources are allocated in this case.
3058 	 */
3059 	if (--mcp_res->refcnt)
3060 		return;
3061 	assert(mcp_res->flow);
3062 	flow_list_destroy(dev, NULL, mcp_res->flow);
3063 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3064 	rte_free(mcp_res);
3065 	flow->mreg_copy = NULL;
3066 }
3067 
3068 /**
3069  * Start flow in RX_CP_TBL.
3070  *
3071  * @param dev
3072  *   Pointer to Ethernet device.
3073  * @flow
3074  *   Parent flow for wich copying is provided.
3075  *
3076  * @return
3077  *   0 on success, a negative errno value otherwise and rte_errno is set.
3078  */
3079 static int
3080 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3081 			    struct rte_flow *flow)
3082 {
3083 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3084 	int ret;
3085 
3086 	if (!mcp_res || flow->copy_applied)
3087 		return 0;
3088 	if (!mcp_res->appcnt) {
3089 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3090 		if (ret)
3091 			return ret;
3092 	}
3093 	++mcp_res->appcnt;
3094 	flow->copy_applied = 1;
3095 	return 0;
3096 }
3097 
3098 /**
3099  * Stop flow in RX_CP_TBL.
3100  *
3101  * @param dev
3102  *   Pointer to Ethernet device.
3103  * @flow
3104  *   Parent flow for wich copying is provided.
3105  */
3106 static void
3107 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3108 			   struct rte_flow *flow)
3109 {
3110 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3111 
3112 	if (!mcp_res || !flow->copy_applied)
3113 		return;
3114 	assert(mcp_res->appcnt);
3115 	--mcp_res->appcnt;
3116 	flow->copy_applied = 0;
3117 	if (!mcp_res->appcnt)
3118 		flow_drv_remove(dev, mcp_res->flow);
3119 }
3120 
3121 /**
3122  * Remove the default copy action from RX_CP_TBL.
3123  *
3124  * @param dev
3125  *   Pointer to Ethernet device.
3126  */
3127 static void
3128 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3129 {
3130 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3131 	struct mlx5_priv *priv = dev->data->dev_private;
3132 
3133 	/* Check if default flow is registered. */
3134 	if (!priv->mreg_cp_tbl)
3135 		return;
3136 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, 0ULL);
3137 	if (!mcp_res)
3138 		return;
3139 	assert(mcp_res->flow);
3140 	flow_list_destroy(dev, NULL, mcp_res->flow);
3141 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3142 	rte_free(mcp_res);
3143 }
3144 
3145 /**
3146  * Add the default copy action in in RX_CP_TBL.
3147  *
3148  * @param dev
3149  *   Pointer to Ethernet device.
3150  * @param[out] error
3151  *   Perform verbose error reporting if not NULL.
3152  *
3153  * @return
3154  *   0 for success, negative value otherwise and rte_errno is set.
3155  */
3156 static int
3157 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3158 				  struct rte_flow_error *error)
3159 {
3160 	struct mlx5_priv *priv = dev->data->dev_private;
3161 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3162 
3163 	/* Check whether extensive metadata feature is engaged. */
3164 	if (!priv->config.dv_flow_en ||
3165 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3166 	    !mlx5_flow_ext_mreg_supported(dev) ||
3167 	    !priv->sh->dv_regc0_mask)
3168 		return 0;
3169 	mcp_res = flow_mreg_add_copy_action(dev, 0, error);
3170 	if (!mcp_res)
3171 		return -rte_errno;
3172 	return 0;
3173 }
3174 
3175 /**
3176  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3177  *
3178  * All the flow having Q/RSS action should be split by
3179  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3180  * performs the following,
3181  *   - CQE->flow_tag := reg_c[1] (MARK)
3182  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3183  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3184  * but there should be a flow per each MARK ID set by MARK action.
3185  *
3186  * For the aforementioned reason, if there's a MARK action in flow's action
3187  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3188  * the MARK ID to CQE's flow_tag like,
3189  *   - If reg_c[1] is mark_id,
3190  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3191  *
3192  * For SET_META action which stores value in reg_c[0], as the destination is
3193  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3194  * MARK ID means the default flow. The default flow looks like,
3195  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3196  *
3197  * @param dev
3198  *   Pointer to Ethernet device.
3199  * @param flow
3200  *   Pointer to flow structure.
3201  * @param[in] actions
3202  *   Pointer to the list of actions.
3203  * @param[out] error
3204  *   Perform verbose error reporting if not NULL.
3205  *
3206  * @return
3207  *   0 on success, negative value otherwise and rte_errno is set.
3208  */
3209 static int
3210 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3211 			    struct rte_flow *flow,
3212 			    const struct rte_flow_action *actions,
3213 			    struct rte_flow_error *error)
3214 {
3215 	struct mlx5_priv *priv = dev->data->dev_private;
3216 	struct mlx5_dev_config *config = &priv->config;
3217 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3218 	const struct rte_flow_action_mark *mark;
3219 
3220 	/* Check whether extensive metadata feature is engaged. */
3221 	if (!config->dv_flow_en ||
3222 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3223 	    !mlx5_flow_ext_mreg_supported(dev) ||
3224 	    !priv->sh->dv_regc0_mask)
3225 		return 0;
3226 	/* Find MARK action. */
3227 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3228 		switch (actions->type) {
3229 		case RTE_FLOW_ACTION_TYPE_FLAG:
3230 			mcp_res = flow_mreg_add_copy_action
3231 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3232 			if (!mcp_res)
3233 				return -rte_errno;
3234 			flow->mreg_copy = mcp_res;
3235 			if (dev->data->dev_started) {
3236 				mcp_res->appcnt++;
3237 				flow->copy_applied = 1;
3238 			}
3239 			return 0;
3240 		case RTE_FLOW_ACTION_TYPE_MARK:
3241 			mark = (const struct rte_flow_action_mark *)
3242 				actions->conf;
3243 			mcp_res =
3244 				flow_mreg_add_copy_action(dev, mark->id, error);
3245 			if (!mcp_res)
3246 				return -rte_errno;
3247 			flow->mreg_copy = mcp_res;
3248 			if (dev->data->dev_started) {
3249 				mcp_res->appcnt++;
3250 				flow->copy_applied = 1;
3251 			}
3252 			return 0;
3253 		default:
3254 			break;
3255 		}
3256 	}
3257 	return 0;
3258 }
3259 
3260 #define MLX5_MAX_SPLIT_ACTIONS 24
3261 #define MLX5_MAX_SPLIT_ITEMS 24
3262 
3263 /**
3264  * Split the hairpin flow.
3265  * Since HW can't support encap on Rx we move the encap to Tx.
3266  * If the count action is after the encap then we also
3267  * move the count action. in this case the count will also measure
3268  * the outer bytes.
3269  *
3270  * @param dev
3271  *   Pointer to Ethernet device.
3272  * @param[in] actions
3273  *   Associated actions (list terminated by the END action).
3274  * @param[out] actions_rx
3275  *   Rx flow actions.
3276  * @param[out] actions_tx
3277  *   Tx flow actions..
3278  * @param[out] pattern_tx
3279  *   The pattern items for the Tx flow.
3280  * @param[out] flow_id
3281  *   The flow ID connected to this flow.
3282  *
3283  * @return
3284  *   0 on success.
3285  */
3286 static int
3287 flow_hairpin_split(struct rte_eth_dev *dev,
3288 		   const struct rte_flow_action actions[],
3289 		   struct rte_flow_action actions_rx[],
3290 		   struct rte_flow_action actions_tx[],
3291 		   struct rte_flow_item pattern_tx[],
3292 		   uint32_t *flow_id)
3293 {
3294 	struct mlx5_priv *priv = dev->data->dev_private;
3295 	const struct rte_flow_action_raw_encap *raw_encap;
3296 	const struct rte_flow_action_raw_decap *raw_decap;
3297 	struct mlx5_rte_flow_action_set_tag *set_tag;
3298 	struct rte_flow_action *tag_action;
3299 	struct mlx5_rte_flow_item_tag *tag_item;
3300 	struct rte_flow_item *item;
3301 	char *addr;
3302 	int encap = 0;
3303 
3304 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3305 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3306 		switch (actions->type) {
3307 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3308 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3309 			rte_memcpy(actions_tx, actions,
3310 			       sizeof(struct rte_flow_action));
3311 			actions_tx++;
3312 			break;
3313 		case RTE_FLOW_ACTION_TYPE_COUNT:
3314 			if (encap) {
3315 				rte_memcpy(actions_tx, actions,
3316 					   sizeof(struct rte_flow_action));
3317 				actions_tx++;
3318 			} else {
3319 				rte_memcpy(actions_rx, actions,
3320 					   sizeof(struct rte_flow_action));
3321 				actions_rx++;
3322 			}
3323 			break;
3324 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3325 			raw_encap = actions->conf;
3326 			if (raw_encap->size >
3327 			    (sizeof(struct rte_flow_item_eth) +
3328 			     sizeof(struct rte_flow_item_ipv4))) {
3329 				memcpy(actions_tx, actions,
3330 				       sizeof(struct rte_flow_action));
3331 				actions_tx++;
3332 				encap = 1;
3333 			} else {
3334 				rte_memcpy(actions_rx, actions,
3335 					   sizeof(struct rte_flow_action));
3336 				actions_rx++;
3337 			}
3338 			break;
3339 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3340 			raw_decap = actions->conf;
3341 			if (raw_decap->size <
3342 			    (sizeof(struct rte_flow_item_eth) +
3343 			     sizeof(struct rte_flow_item_ipv4))) {
3344 				memcpy(actions_tx, actions,
3345 				       sizeof(struct rte_flow_action));
3346 				actions_tx++;
3347 			} else {
3348 				rte_memcpy(actions_rx, actions,
3349 					   sizeof(struct rte_flow_action));
3350 				actions_rx++;
3351 			}
3352 			break;
3353 		default:
3354 			rte_memcpy(actions_rx, actions,
3355 				   sizeof(struct rte_flow_action));
3356 			actions_rx++;
3357 			break;
3358 		}
3359 	}
3360 	/* Add set meta action and end action for the Rx flow. */
3361 	tag_action = actions_rx;
3362 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3363 	actions_rx++;
3364 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3365 	actions_rx++;
3366 	set_tag = (void *)actions_rx;
3367 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3368 	assert(set_tag->id > REG_NONE);
3369 	set_tag->data = *flow_id;
3370 	tag_action->conf = set_tag;
3371 	/* Create Tx item list. */
3372 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3373 	addr = (void *)&pattern_tx[2];
3374 	item = pattern_tx;
3375 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3376 	tag_item = (void *)addr;
3377 	tag_item->data = *flow_id;
3378 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3379 	assert(set_tag->id > REG_NONE);
3380 	item->spec = tag_item;
3381 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3382 	tag_item = (void *)addr;
3383 	tag_item->data = UINT32_MAX;
3384 	tag_item->id = UINT16_MAX;
3385 	item->mask = tag_item;
3386 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3387 	item->last = NULL;
3388 	item++;
3389 	item->type = RTE_FLOW_ITEM_TYPE_END;
3390 	return 0;
3391 }
3392 
3393 /**
3394  * The last stage of splitting chain, just creates the subflow
3395  * without any modification.
3396  *
3397  * @param dev
3398  *   Pointer to Ethernet device.
3399  * @param[in] flow
3400  *   Parent flow structure pointer.
3401  * @param[in, out] sub_flow
3402  *   Pointer to return the created subflow, may be NULL.
3403  * @param[in] attr
3404  *   Flow rule attributes.
3405  * @param[in] items
3406  *   Pattern specification (list terminated by the END pattern item).
3407  * @param[in] actions
3408  *   Associated actions (list terminated by the END action).
3409  * @param[in] external
3410  *   This flow rule is created by request external to PMD.
3411  * @param[out] error
3412  *   Perform verbose error reporting if not NULL.
3413  * @return
3414  *   0 on success, negative value otherwise
3415  */
3416 static int
3417 flow_create_split_inner(struct rte_eth_dev *dev,
3418 			struct rte_flow *flow,
3419 			struct mlx5_flow **sub_flow,
3420 			const struct rte_flow_attr *attr,
3421 			const struct rte_flow_item items[],
3422 			const struct rte_flow_action actions[],
3423 			bool external, struct rte_flow_error *error)
3424 {
3425 	struct mlx5_flow *dev_flow;
3426 
3427 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3428 	if (!dev_flow)
3429 		return -rte_errno;
3430 	dev_flow->flow = flow;
3431 	dev_flow->external = external;
3432 	/* Subflow object was created, we must include one in the list. */
3433 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3434 	if (sub_flow)
3435 		*sub_flow = dev_flow;
3436 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3437 }
3438 
3439 /**
3440  * Split the meter flow.
3441  *
3442  * As meter flow will split to three sub flow, other than meter
3443  * action, the other actions make sense to only meter accepts
3444  * the packet. If it need to be dropped, no other additional
3445  * actions should be take.
3446  *
3447  * One kind of special action which decapsulates the L3 tunnel
3448  * header will be in the prefix sub flow, as not to take the
3449  * L3 tunnel header into account.
3450  *
3451  * @param dev
3452  *   Pointer to Ethernet device.
3453  * @param[in] actions
3454  *   Associated actions (list terminated by the END action).
3455  * @param[out] actions_sfx
3456  *   Suffix flow actions.
3457  * @param[out] actions_pre
3458  *   Prefix flow actions.
3459  * @param[out] pattern_sfx
3460  *   The pattern items for the suffix flow.
3461  * @param[out] tag_sfx
3462  *   Pointer to suffix flow tag.
3463  *
3464  * @return
3465  *   0 on success.
3466  */
3467 static int
3468 flow_meter_split_prep(struct rte_eth_dev *dev,
3469 		 const struct rte_flow_action actions[],
3470 		 struct rte_flow_action actions_sfx[],
3471 		 struct rte_flow_action actions_pre[])
3472 {
3473 	struct rte_flow_action *tag_action;
3474 	struct mlx5_rte_flow_action_set_tag *set_tag;
3475 	struct rte_flow_error error;
3476 	const struct rte_flow_action_raw_encap *raw_encap;
3477 	const struct rte_flow_action_raw_decap *raw_decap;
3478 	uint32_t tag_id;
3479 
3480 	/* Add the extra tag action first. */
3481 	tag_action = actions_pre;
3482 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3483 	actions_pre++;
3484 	/* Prepare the actions for prefix and suffix flow. */
3485 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3486 		switch (actions->type) {
3487 		case RTE_FLOW_ACTION_TYPE_METER:
3488 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3489 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3490 			memcpy(actions_pre, actions,
3491 			       sizeof(struct rte_flow_action));
3492 			actions_pre++;
3493 			break;
3494 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3495 			raw_encap = actions->conf;
3496 			if (raw_encap->size >
3497 			    (sizeof(struct rte_flow_item_eth) +
3498 			     sizeof(struct rte_flow_item_ipv4))) {
3499 				memcpy(actions_sfx, actions,
3500 				       sizeof(struct rte_flow_action));
3501 				actions_sfx++;
3502 			} else {
3503 				rte_memcpy(actions_pre, actions,
3504 					   sizeof(struct rte_flow_action));
3505 				actions_pre++;
3506 			}
3507 			break;
3508 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3509 			raw_decap = actions->conf;
3510 			/* Size 0 decap means 50 bytes as vxlan decap. */
3511 			if (raw_decap->size && (raw_decap->size <
3512 			    (sizeof(struct rte_flow_item_eth) +
3513 			     sizeof(struct rte_flow_item_ipv4)))) {
3514 				memcpy(actions_sfx, actions,
3515 				       sizeof(struct rte_flow_action));
3516 				actions_sfx++;
3517 			} else {
3518 				rte_memcpy(actions_pre, actions,
3519 					   sizeof(struct rte_flow_action));
3520 				actions_pre++;
3521 			}
3522 			break;
3523 		default:
3524 			memcpy(actions_sfx, actions,
3525 				sizeof(struct rte_flow_action));
3526 			actions_sfx++;
3527 			break;
3528 		}
3529 	}
3530 	/* Add end action to the actions. */
3531 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3532 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3533 	actions_pre++;
3534 	/* Set the tag. */
3535 	set_tag = (void *)actions_pre;
3536 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3537 	/*
3538 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3539 	 */
3540 	tag_id = flow_qrss_get_id(dev);
3541 	set_tag->data = rte_cpu_to_be_32(tag_id);
3542 	tag_action->conf = set_tag;
3543 	return tag_id;
3544 }
3545 
3546 /**
3547  * Split action list having QUEUE/RSS for metadata register copy.
3548  *
3549  * Once Q/RSS action is detected in user's action list, the flow action
3550  * should be split in order to copy metadata registers, which will happen in
3551  * RX_CP_TBL like,
3552  *   - CQE->flow_tag := reg_c[1] (MARK)
3553  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3554  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3555  * This is because the last action of each flow must be a terminal action
3556  * (QUEUE, RSS or DROP).
3557  *
3558  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3559  * stored and kept in the mlx5_flow structure per each sub_flow.
3560  *
3561  * The Q/RSS action is replaced with,
3562  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3563  * And the following JUMP action is added at the end,
3564  *   - JUMP, to RX_CP_TBL.
3565  *
3566  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3567  * flow_create_split_metadata() routine. The flow will look like,
3568  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3569  *
3570  * @param dev
3571  *   Pointer to Ethernet device.
3572  * @param[out] split_actions
3573  *   Pointer to store split actions to jump to CP_TBL.
3574  * @param[in] actions
3575  *   Pointer to the list of original flow actions.
3576  * @param[in] qrss
3577  *   Pointer to the Q/RSS action.
3578  * @param[in] actions_n
3579  *   Number of original actions.
3580  * @param[out] error
3581  *   Perform verbose error reporting if not NULL.
3582  *
3583  * @return
3584  *   non-zero unique flow_id on success, otherwise 0 and
3585  *   error/rte_error are set.
3586  */
3587 static uint32_t
3588 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3589 			  struct rte_flow_action *split_actions,
3590 			  const struct rte_flow_action *actions,
3591 			  const struct rte_flow_action *qrss,
3592 			  int actions_n, struct rte_flow_error *error)
3593 {
3594 	struct mlx5_rte_flow_action_set_tag *set_tag;
3595 	struct rte_flow_action_jump *jump;
3596 	const int qrss_idx = qrss - actions;
3597 	uint32_t flow_id = 0;
3598 	int ret = 0;
3599 
3600 	/*
3601 	 * Given actions will be split
3602 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3603 	 * - Add jump to mreg CP_TBL.
3604 	 * As a result, there will be one more action.
3605 	 */
3606 	++actions_n;
3607 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3608 	set_tag = (void *)(split_actions + actions_n);
3609 	/*
3610 	 * If tag action is not set to void(it means we are not the meter
3611 	 * suffix flow), add the tag action. Since meter suffix flow already
3612 	 * has the tag added.
3613 	 */
3614 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3615 		/*
3616 		 * Allocate the new subflow ID. This one is unique within
3617 		 * device and not shared with representors. Otherwise,
3618 		 * we would have to resolve multi-thread access synch
3619 		 * issue. Each flow on the shared device is appended
3620 		 * with source vport identifier, so the resulting
3621 		 * flows will be unique in the shared (by master and
3622 		 * representors) domain even if they have coinciding
3623 		 * IDs.
3624 		 */
3625 		flow_id = flow_qrss_get_id(dev);
3626 		if (!flow_id)
3627 			return rte_flow_error_set(error, ENOMEM,
3628 						  RTE_FLOW_ERROR_TYPE_ACTION,
3629 						  NULL, "can't allocate id "
3630 						  "for split Q/RSS subflow");
3631 		/* Internal SET_TAG action to set flow ID. */
3632 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3633 			.data = flow_id,
3634 		};
3635 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3636 		if (ret < 0)
3637 			return ret;
3638 		set_tag->id = ret;
3639 		/* Construct new actions array. */
3640 		/* Replace QUEUE/RSS action. */
3641 		split_actions[qrss_idx] = (struct rte_flow_action){
3642 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3643 			.conf = set_tag,
3644 		};
3645 	}
3646 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3647 	jump = (void *)(set_tag + 1);
3648 	*jump = (struct rte_flow_action_jump){
3649 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3650 	};
3651 	split_actions[actions_n - 2] = (struct rte_flow_action){
3652 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3653 		.conf = jump,
3654 	};
3655 	split_actions[actions_n - 1] = (struct rte_flow_action){
3656 		.type = RTE_FLOW_ACTION_TYPE_END,
3657 	};
3658 	return flow_id;
3659 }
3660 
3661 /**
3662  * Extend the given action list for Tx metadata copy.
3663  *
3664  * Copy the given action list to the ext_actions and add flow metadata register
3665  * copy action in order to copy reg_a set by WQE to reg_c[0].
3666  *
3667  * @param[out] ext_actions
3668  *   Pointer to the extended action list.
3669  * @param[in] actions
3670  *   Pointer to the list of actions.
3671  * @param[in] actions_n
3672  *   Number of actions in the list.
3673  * @param[out] error
3674  *   Perform verbose error reporting if not NULL.
3675  *
3676  * @return
3677  *   0 on success, negative value otherwise
3678  */
3679 static int
3680 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3681 		       struct rte_flow_action *ext_actions,
3682 		       const struct rte_flow_action *actions,
3683 		       int actions_n, struct rte_flow_error *error)
3684 {
3685 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3686 		(struct mlx5_flow_action_copy_mreg *)
3687 			(ext_actions + actions_n + 1);
3688 	int ret;
3689 
3690 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3691 	if (ret < 0)
3692 		return ret;
3693 	cp_mreg->dst = ret;
3694 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3695 	if (ret < 0)
3696 		return ret;
3697 	cp_mreg->src = ret;
3698 	memcpy(ext_actions, actions,
3699 			sizeof(*ext_actions) * actions_n);
3700 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3701 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3702 		.conf = cp_mreg,
3703 	};
3704 	ext_actions[actions_n] = (struct rte_flow_action){
3705 		.type = RTE_FLOW_ACTION_TYPE_END,
3706 	};
3707 	return 0;
3708 }
3709 
3710 /**
3711  * The splitting for metadata feature.
3712  *
3713  * - Q/RSS action on NIC Rx should be split in order to pass by
3714  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3715  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3716  *
3717  * - All the actions on NIC Tx should have a mreg copy action to
3718  *   copy reg_a from WQE to reg_c[0].
3719  *
3720  * @param dev
3721  *   Pointer to Ethernet device.
3722  * @param[in] flow
3723  *   Parent flow structure pointer.
3724  * @param[in] attr
3725  *   Flow rule attributes.
3726  * @param[in] items
3727  *   Pattern specification (list terminated by the END pattern item).
3728  * @param[in] actions
3729  *   Associated actions (list terminated by the END action).
3730  * @param[in] external
3731  *   This flow rule is created by request external to PMD.
3732  * @param[out] error
3733  *   Perform verbose error reporting if not NULL.
3734  * @return
3735  *   0 on success, negative value otherwise
3736  */
3737 static int
3738 flow_create_split_metadata(struct rte_eth_dev *dev,
3739 			   struct rte_flow *flow,
3740 			   const struct rte_flow_attr *attr,
3741 			   const struct rte_flow_item items[],
3742 			   const struct rte_flow_action actions[],
3743 			   bool external, struct rte_flow_error *error)
3744 {
3745 	struct mlx5_priv *priv = dev->data->dev_private;
3746 	struct mlx5_dev_config *config = &priv->config;
3747 	const struct rte_flow_action *qrss = NULL;
3748 	struct rte_flow_action *ext_actions = NULL;
3749 	struct mlx5_flow *dev_flow = NULL;
3750 	uint32_t qrss_id = 0;
3751 	int mtr_sfx = 0;
3752 	size_t act_size;
3753 	int actions_n;
3754 	int ret;
3755 
3756 	/* Check whether extensive metadata feature is engaged. */
3757 	if (!config->dv_flow_en ||
3758 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3759 	    !mlx5_flow_ext_mreg_supported(dev))
3760 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3761 					       actions, external, error);
3762 	actions_n = flow_parse_qrss_action(actions, &qrss);
3763 	if (qrss) {
3764 		/* Exclude hairpin flows from splitting. */
3765 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3766 			const struct rte_flow_action_queue *queue;
3767 
3768 			queue = qrss->conf;
3769 			if (mlx5_rxq_get_type(dev, queue->index) ==
3770 			    MLX5_RXQ_TYPE_HAIRPIN)
3771 				qrss = NULL;
3772 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3773 			const struct rte_flow_action_rss *rss;
3774 
3775 			rss = qrss->conf;
3776 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3777 			    MLX5_RXQ_TYPE_HAIRPIN)
3778 				qrss = NULL;
3779 		}
3780 	}
3781 	if (qrss) {
3782 		/* Check if it is in meter suffix table. */
3783 		mtr_sfx = attr->group == (attr->transfer ?
3784 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3785 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3786 		/*
3787 		 * Q/RSS action on NIC Rx should be split in order to pass by
3788 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3789 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3790 		 */
3791 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3792 			   sizeof(struct rte_flow_action_set_tag) +
3793 			   sizeof(struct rte_flow_action_jump);
3794 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3795 		if (!ext_actions)
3796 			return rte_flow_error_set(error, ENOMEM,
3797 						  RTE_FLOW_ERROR_TYPE_ACTION,
3798 						  NULL, "no memory to split "
3799 						  "metadata flow");
3800 		/*
3801 		 * If we are the suffix flow of meter, tag already exist.
3802 		 * Set the tag action to void.
3803 		 */
3804 		if (mtr_sfx)
3805 			ext_actions[qrss - actions].type =
3806 						RTE_FLOW_ACTION_TYPE_VOID;
3807 		else
3808 			ext_actions[qrss - actions].type =
3809 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3810 		/*
3811 		 * Create the new actions list with removed Q/RSS action
3812 		 * and appended set tag and jump to register copy table
3813 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3814 		 * in advance, because it is needed for set tag action.
3815 		 */
3816 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3817 						    qrss, actions_n, error);
3818 		if (!mtr_sfx && !qrss_id) {
3819 			ret = -rte_errno;
3820 			goto exit;
3821 		}
3822 	} else if (attr->egress && !attr->transfer) {
3823 		/*
3824 		 * All the actions on NIC Tx should have a metadata register
3825 		 * copy action to copy reg_a from WQE to reg_c[meta]
3826 		 */
3827 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3828 			   sizeof(struct mlx5_flow_action_copy_mreg);
3829 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3830 		if (!ext_actions)
3831 			return rte_flow_error_set(error, ENOMEM,
3832 						  RTE_FLOW_ERROR_TYPE_ACTION,
3833 						  NULL, "no memory to split "
3834 						  "metadata flow");
3835 		/* Create the action list appended with copy register. */
3836 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3837 					     actions_n, error);
3838 		if (ret < 0)
3839 			goto exit;
3840 	}
3841 	/* Add the unmodified original or prefix subflow. */
3842 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3843 				      ext_actions ? ext_actions : actions,
3844 				      external, error);
3845 	if (ret < 0)
3846 		goto exit;
3847 	assert(dev_flow);
3848 	if (qrss) {
3849 		const struct rte_flow_attr q_attr = {
3850 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3851 			.ingress = 1,
3852 		};
3853 		/* Internal PMD action to set register. */
3854 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3855 			.data = qrss_id,
3856 			.id = 0,
3857 		};
3858 		struct rte_flow_item q_items[] = {
3859 			{
3860 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3861 				.spec = &q_tag_spec,
3862 				.last = NULL,
3863 				.mask = NULL,
3864 			},
3865 			{
3866 				.type = RTE_FLOW_ITEM_TYPE_END,
3867 			},
3868 		};
3869 		struct rte_flow_action q_actions[] = {
3870 			{
3871 				.type = qrss->type,
3872 				.conf = qrss->conf,
3873 			},
3874 			{
3875 				.type = RTE_FLOW_ACTION_TYPE_END,
3876 			},
3877 		};
3878 		uint64_t hash_fields = dev_flow->hash_fields;
3879 
3880 		/*
3881 		 * Configure the tag item only if there is no meter subflow.
3882 		 * Since tag is already marked in the meter suffix subflow
3883 		 * we can just use the meter suffix items as is.
3884 		 */
3885 		if (qrss_id) {
3886 			/* Not meter subflow. */
3887 			assert(!mtr_sfx);
3888 			/*
3889 			 * Put unique id in prefix flow due to it is destroyed
3890 			 * after suffix flow and id will be freed after there
3891 			 * is no actual flows with this id and identifier
3892 			 * reallocation becomes possible (for example, for
3893 			 * other flows in other threads).
3894 			 */
3895 			dev_flow->qrss_id = qrss_id;
3896 			qrss_id = 0;
3897 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3898 						   error);
3899 			if (ret < 0)
3900 				goto exit;
3901 			q_tag_spec.id = ret;
3902 		}
3903 		dev_flow = NULL;
3904 		/* Add suffix subflow to execute Q/RSS. */
3905 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3906 					      &q_attr, mtr_sfx ? items :
3907 					      q_items, q_actions,
3908 					      external, error);
3909 		if (ret < 0)
3910 			goto exit;
3911 		assert(dev_flow);
3912 		dev_flow->hash_fields = hash_fields;
3913 	}
3914 
3915 exit:
3916 	/*
3917 	 * We do not destroy the partially created sub_flows in case of error.
3918 	 * These ones are included into parent flow list and will be destroyed
3919 	 * by flow_drv_destroy.
3920 	 */
3921 	flow_qrss_free_id(dev, qrss_id);
3922 	rte_free(ext_actions);
3923 	return ret;
3924 }
3925 
3926 /**
3927  * The splitting for meter feature.
3928  *
3929  * - The meter flow will be split to two flows as prefix and
3930  *   suffix flow. The packets make sense only it pass the prefix
3931  *   meter action.
3932  *
3933  * - Reg_C_5 is used for the packet to match betweend prefix and
3934  *   suffix flow.
3935  *
3936  * @param dev
3937  *   Pointer to Ethernet device.
3938  * @param[in] flow
3939  *   Parent flow structure pointer.
3940  * @param[in] attr
3941  *   Flow rule attributes.
3942  * @param[in] items
3943  *   Pattern specification (list terminated by the END pattern item).
3944  * @param[in] actions
3945  *   Associated actions (list terminated by the END action).
3946  * @param[in] external
3947  *   This flow rule is created by request external to PMD.
3948  * @param[out] error
3949  *   Perform verbose error reporting if not NULL.
3950  * @return
3951  *   0 on success, negative value otherwise
3952  */
3953 static int
3954 flow_create_split_meter(struct rte_eth_dev *dev,
3955 			   struct rte_flow *flow,
3956 			   const struct rte_flow_attr *attr,
3957 			   const struct rte_flow_item items[],
3958 			   const struct rte_flow_action actions[],
3959 			   bool external, struct rte_flow_error *error)
3960 {
3961 	struct mlx5_priv *priv = dev->data->dev_private;
3962 	struct rte_flow_action *sfx_actions = NULL;
3963 	struct rte_flow_action *pre_actions = NULL;
3964 	struct rte_flow_item *sfx_items = NULL;
3965 	const  struct rte_flow_item *sfx_port_id_item;
3966 	struct mlx5_flow *dev_flow = NULL;
3967 	struct rte_flow_attr sfx_attr = *attr;
3968 	uint32_t mtr = 0;
3969 	uint32_t mtr_tag_id = 0;
3970 	size_t act_size;
3971 	size_t item_size;
3972 	int actions_n = 0;
3973 	int ret;
3974 
3975 	if (priv->mtr_en)
3976 		actions_n = flow_check_meter_action(actions, &mtr);
3977 	if (mtr) {
3978 		struct mlx5_rte_flow_item_tag *tag_spec;
3979 		/* The five prefix actions: meter, decap, encap, tag, end. */
3980 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3981 			   sizeof(struct rte_flow_action_set_tag);
3982 		/* tag, end. */
3983 #define METER_SUFFIX_ITEM 3
3984 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3985 			    sizeof(struct mlx5_rte_flow_item_tag);
3986 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3987 		if (!sfx_actions)
3988 			return rte_flow_error_set(error, ENOMEM,
3989 						  RTE_FLOW_ERROR_TYPE_ACTION,
3990 						  NULL, "no memory to split "
3991 						  "meter flow");
3992 		pre_actions = sfx_actions + actions_n;
3993 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
3994 						     pre_actions);
3995 		if (!mtr_tag_id) {
3996 			ret = -rte_errno;
3997 			goto exit;
3998 		}
3999 		/* Add the prefix subflow. */
4000 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4001 						  pre_actions, external, error);
4002 		if (ret) {
4003 			ret = -rte_errno;
4004 			goto exit;
4005 		}
4006 		dev_flow->mtr_flow_id = mtr_tag_id;
4007 		/* Prepare the suffix flow match pattern. */
4008 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4009 			     act_size);
4010 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4011 			    METER_SUFFIX_ITEM);
4012 		tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
4013 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4014 						    error);
4015 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4016 		sfx_items->spec = tag_spec;
4017 		sfx_items->last = NULL;
4018 		sfx_items->mask = NULL;
4019 		sfx_items++;
4020 		sfx_port_id_item = find_port_id_item(items);
4021 		if (sfx_port_id_item) {
4022 			memcpy(sfx_items, sfx_port_id_item,
4023 			       sizeof(*sfx_items));
4024 			sfx_items++;
4025 		}
4026 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4027 		sfx_items -= METER_SUFFIX_ITEM;
4028 		/* Setting the sfx group atrr. */
4029 		sfx_attr.group = sfx_attr.transfer ?
4030 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4031 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4032 	}
4033 	/* Add the prefix subflow. */
4034 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4035 					 sfx_items ? sfx_items : items,
4036 					 sfx_actions ? sfx_actions : actions,
4037 					 external, error);
4038 exit:
4039 	if (sfx_actions)
4040 		rte_free(sfx_actions);
4041 	return ret;
4042 }
4043 
4044 /**
4045  * Split the flow to subflow set. The splitters might be linked
4046  * in the chain, like this:
4047  * flow_create_split_outer() calls:
4048  *   flow_create_split_meter() calls:
4049  *     flow_create_split_metadata(meter_subflow_0) calls:
4050  *       flow_create_split_inner(metadata_subflow_0)
4051  *       flow_create_split_inner(metadata_subflow_1)
4052  *       flow_create_split_inner(metadata_subflow_2)
4053  *     flow_create_split_metadata(meter_subflow_1) calls:
4054  *       flow_create_split_inner(metadata_subflow_0)
4055  *       flow_create_split_inner(metadata_subflow_1)
4056  *       flow_create_split_inner(metadata_subflow_2)
4057  *
4058  * This provide flexible way to add new levels of flow splitting.
4059  * The all of successfully created subflows are included to the
4060  * parent flow dev_flow list.
4061  *
4062  * @param dev
4063  *   Pointer to Ethernet device.
4064  * @param[in] flow
4065  *   Parent flow structure pointer.
4066  * @param[in] attr
4067  *   Flow rule attributes.
4068  * @param[in] items
4069  *   Pattern specification (list terminated by the END pattern item).
4070  * @param[in] actions
4071  *   Associated actions (list terminated by the END action).
4072  * @param[in] external
4073  *   This flow rule is created by request external to PMD.
4074  * @param[out] error
4075  *   Perform verbose error reporting if not NULL.
4076  * @return
4077  *   0 on success, negative value otherwise
4078  */
4079 static int
4080 flow_create_split_outer(struct rte_eth_dev *dev,
4081 			struct rte_flow *flow,
4082 			const struct rte_flow_attr *attr,
4083 			const struct rte_flow_item items[],
4084 			const struct rte_flow_action actions[],
4085 			bool external, struct rte_flow_error *error)
4086 {
4087 	int ret;
4088 
4089 	ret = flow_create_split_meter(dev, flow, attr, items,
4090 					 actions, external, error);
4091 	assert(ret <= 0);
4092 	return ret;
4093 }
4094 
4095 /**
4096  * Create a flow and add it to @p list.
4097  *
4098  * @param dev
4099  *   Pointer to Ethernet device.
4100  * @param list
4101  *   Pointer to a TAILQ flow list. If this parameter NULL,
4102  *   no list insertion occurred, flow is just created,
4103  *   this is caller's responsibility to track the
4104  *   created flow.
4105  * @param[in] attr
4106  *   Flow rule attributes.
4107  * @param[in] items
4108  *   Pattern specification (list terminated by the END pattern item).
4109  * @param[in] actions
4110  *   Associated actions (list terminated by the END action).
4111  * @param[in] external
4112  *   This flow rule is created by request external to PMD.
4113  * @param[out] error
4114  *   Perform verbose error reporting if not NULL.
4115  *
4116  * @return
4117  *   A flow on success, NULL otherwise and rte_errno is set.
4118  */
4119 static struct rte_flow *
4120 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4121 		 const struct rte_flow_attr *attr,
4122 		 const struct rte_flow_item items[],
4123 		 const struct rte_flow_action actions[],
4124 		 bool external, struct rte_flow_error *error)
4125 {
4126 	struct mlx5_priv *priv = dev->data->dev_private;
4127 	struct rte_flow *flow = NULL;
4128 	struct mlx5_flow *dev_flow;
4129 	const struct rte_flow_action_rss *rss;
4130 	union {
4131 		struct rte_flow_expand_rss buf;
4132 		uint8_t buffer[2048];
4133 	} expand_buffer;
4134 	union {
4135 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4136 		uint8_t buffer[2048];
4137 	} actions_rx;
4138 	union {
4139 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4140 		uint8_t buffer[2048];
4141 	} actions_hairpin_tx;
4142 	union {
4143 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4144 		uint8_t buffer[2048];
4145 	} items_tx;
4146 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4147 	const struct rte_flow_action *p_actions_rx = actions;
4148 	int ret;
4149 	uint32_t i;
4150 	uint32_t flow_size;
4151 	int hairpin_flow = 0;
4152 	uint32_t hairpin_id = 0;
4153 	struct rte_flow_attr attr_tx = { .priority = 0 };
4154 
4155 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4156 	if (hairpin_flow > 0) {
4157 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4158 			rte_errno = EINVAL;
4159 			return NULL;
4160 		}
4161 		flow_hairpin_split(dev, actions, actions_rx.actions,
4162 				   actions_hairpin_tx.actions, items_tx.items,
4163 				   &hairpin_id);
4164 		p_actions_rx = actions_rx.actions;
4165 	}
4166 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4167 				error);
4168 	if (ret < 0)
4169 		goto error_before_flow;
4170 	flow_size = sizeof(struct rte_flow);
4171 	rss = flow_get_rss_action(p_actions_rx);
4172 	if (rss)
4173 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4174 					    sizeof(void *));
4175 	else
4176 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4177 	flow = rte_calloc(__func__, 1, flow_size, 0);
4178 	if (!flow) {
4179 		rte_errno = ENOMEM;
4180 		goto error_before_flow;
4181 	}
4182 	flow->drv_type = flow_get_drv_type(dev, attr);
4183 	if (hairpin_id != 0)
4184 		flow->hairpin_flow_id = hairpin_id;
4185 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4186 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
4187 	flow->rss.queue = (void *)(flow + 1);
4188 	if (rss) {
4189 		/*
4190 		 * The following information is required by
4191 		 * mlx5_flow_hashfields_adjust() in advance.
4192 		 */
4193 		flow->rss.level = rss->level;
4194 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4195 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4196 	}
4197 	LIST_INIT(&flow->dev_flows);
4198 	if (rss && rss->types) {
4199 		unsigned int graph_root;
4200 
4201 		graph_root = find_graph_root(items, rss->level);
4202 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4203 					  items, rss->types,
4204 					  mlx5_support_expansion,
4205 					  graph_root);
4206 		assert(ret > 0 &&
4207 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4208 	} else {
4209 		buf->entries = 1;
4210 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4211 	}
4212 	for (i = 0; i < buf->entries; ++i) {
4213 		/*
4214 		 * The splitter may create multiple dev_flows,
4215 		 * depending on configuration. In the simplest
4216 		 * case it just creates unmodified original flow.
4217 		 */
4218 		ret = flow_create_split_outer(dev, flow, attr,
4219 					      buf->entry[i].pattern,
4220 					      p_actions_rx, external,
4221 					      error);
4222 		if (ret < 0)
4223 			goto error;
4224 	}
4225 	/* Create the tx flow. */
4226 	if (hairpin_flow) {
4227 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4228 		attr_tx.ingress = 0;
4229 		attr_tx.egress = 1;
4230 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4231 					    actions_hairpin_tx.actions, error);
4232 		if (!dev_flow)
4233 			goto error;
4234 		dev_flow->flow = flow;
4235 		dev_flow->external = 0;
4236 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4237 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4238 					 items_tx.items,
4239 					 actions_hairpin_tx.actions, error);
4240 		if (ret < 0)
4241 			goto error;
4242 	}
4243 	/*
4244 	 * Update the metadata register copy table. If extensive
4245 	 * metadata feature is enabled and registers are supported
4246 	 * we might create the extra rte_flow for each unique
4247 	 * MARK/FLAG action ID.
4248 	 *
4249 	 * The table is updated for ingress Flows only, because
4250 	 * the egress Flows belong to the different device and
4251 	 * copy table should be updated in peer NIC Rx domain.
4252 	 */
4253 	if (attr->ingress &&
4254 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4255 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4256 		if (ret)
4257 			goto error;
4258 	}
4259 	if (dev->data->dev_started) {
4260 		ret = flow_drv_apply(dev, flow, error);
4261 		if (ret < 0)
4262 			goto error;
4263 	}
4264 	if (list)
4265 		TAILQ_INSERT_TAIL(list, flow, next);
4266 	flow_rxq_flags_set(dev, flow);
4267 	return flow;
4268 error_before_flow:
4269 	if (hairpin_id)
4270 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4271 				     hairpin_id);
4272 	return NULL;
4273 error:
4274 	assert(flow);
4275 	flow_mreg_del_copy_action(dev, flow);
4276 	ret = rte_errno; /* Save rte_errno before cleanup. */
4277 	if (flow->hairpin_flow_id)
4278 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4279 				     flow->hairpin_flow_id);
4280 	assert(flow);
4281 	flow_drv_destroy(dev, flow);
4282 	rte_free(flow);
4283 	rte_errno = ret; /* Restore rte_errno. */
4284 	return NULL;
4285 }
4286 
4287 /**
4288  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4289  * incoming packets to table 1.
4290  *
4291  * Other flow rules, requested for group n, will be created in
4292  * e-switch table n+1.
4293  * Jump action to e-switch group n will be created to group n+1.
4294  *
4295  * Used when working in switchdev mode, to utilise advantages of table 1
4296  * and above.
4297  *
4298  * @param dev
4299  *   Pointer to Ethernet device.
4300  *
4301  * @return
4302  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4303  */
4304 struct rte_flow *
4305 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4306 {
4307 	const struct rte_flow_attr attr = {
4308 		.group = 0,
4309 		.priority = 0,
4310 		.ingress = 1,
4311 		.egress = 0,
4312 		.transfer = 1,
4313 	};
4314 	const struct rte_flow_item pattern = {
4315 		.type = RTE_FLOW_ITEM_TYPE_END,
4316 	};
4317 	struct rte_flow_action_jump jump = {
4318 		.group = 1,
4319 	};
4320 	const struct rte_flow_action actions[] = {
4321 		{
4322 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4323 			.conf = &jump,
4324 		},
4325 		{
4326 			.type = RTE_FLOW_ACTION_TYPE_END,
4327 		},
4328 	};
4329 	struct mlx5_priv *priv = dev->data->dev_private;
4330 	struct rte_flow_error error;
4331 
4332 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4333 				actions, false, &error);
4334 }
4335 
4336 /**
4337  * Create a flow.
4338  *
4339  * @see rte_flow_create()
4340  * @see rte_flow_ops
4341  */
4342 struct rte_flow *
4343 mlx5_flow_create(struct rte_eth_dev *dev,
4344 		 const struct rte_flow_attr *attr,
4345 		 const struct rte_flow_item items[],
4346 		 const struct rte_flow_action actions[],
4347 		 struct rte_flow_error *error)
4348 {
4349 	struct mlx5_priv *priv = dev->data->dev_private;
4350 
4351 	return flow_list_create(dev, &priv->flows,
4352 				attr, items, actions, true, error);
4353 }
4354 
4355 /**
4356  * Destroy a flow in a list.
4357  *
4358  * @param dev
4359  *   Pointer to Ethernet device.
4360  * @param list
4361  *   Pointer to a TAILQ flow list. If this parameter NULL,
4362  *   there is no flow removal from the list.
4363  * @param[in] flow
4364  *   Flow to destroy.
4365  */
4366 static void
4367 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4368 		  struct rte_flow *flow)
4369 {
4370 	struct mlx5_priv *priv = dev->data->dev_private;
4371 
4372 	/*
4373 	 * Update RX queue flags only if port is started, otherwise it is
4374 	 * already clean.
4375 	 */
4376 	if (dev->data->dev_started)
4377 		flow_rxq_flags_trim(dev, flow);
4378 	if (flow->hairpin_flow_id)
4379 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4380 				     flow->hairpin_flow_id);
4381 	flow_drv_destroy(dev, flow);
4382 	if (list)
4383 		TAILQ_REMOVE(list, flow, next);
4384 	flow_mreg_del_copy_action(dev, flow);
4385 	rte_free(flow->fdir);
4386 	rte_free(flow);
4387 }
4388 
4389 /**
4390  * Destroy all flows.
4391  *
4392  * @param dev
4393  *   Pointer to Ethernet device.
4394  * @param list
4395  *   Pointer to a TAILQ flow list.
4396  */
4397 void
4398 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4399 {
4400 	while (!TAILQ_EMPTY(list)) {
4401 		struct rte_flow *flow;
4402 
4403 		flow = TAILQ_FIRST(list);
4404 		flow_list_destroy(dev, list, flow);
4405 	}
4406 }
4407 
4408 /**
4409  * Remove all flows.
4410  *
4411  * @param dev
4412  *   Pointer to Ethernet device.
4413  * @param list
4414  *   Pointer to a TAILQ flow list.
4415  */
4416 void
4417 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4418 {
4419 	struct rte_flow *flow;
4420 
4421 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4422 		flow_drv_remove(dev, flow);
4423 		flow_mreg_stop_copy_action(dev, flow);
4424 	}
4425 	flow_mreg_del_default_copy_action(dev);
4426 	flow_rxq_flags_clear(dev);
4427 }
4428 
4429 /**
4430  * Add all flows.
4431  *
4432  * @param dev
4433  *   Pointer to Ethernet device.
4434  * @param list
4435  *   Pointer to a TAILQ flow list.
4436  *
4437  * @return
4438  *   0 on success, a negative errno value otherwise and rte_errno is set.
4439  */
4440 int
4441 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4442 {
4443 	struct rte_flow *flow;
4444 	struct rte_flow_error error;
4445 	int ret = 0;
4446 
4447 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4448 	ret = flow_mreg_add_default_copy_action(dev, &error);
4449 	if (ret < 0)
4450 		return -rte_errno;
4451 	/* Apply Flows created by application. */
4452 	TAILQ_FOREACH(flow, list, next) {
4453 		ret = flow_mreg_start_copy_action(dev, flow);
4454 		if (ret < 0)
4455 			goto error;
4456 		ret = flow_drv_apply(dev, flow, &error);
4457 		if (ret < 0)
4458 			goto error;
4459 		flow_rxq_flags_set(dev, flow);
4460 	}
4461 	return 0;
4462 error:
4463 	ret = rte_errno; /* Save rte_errno before cleanup. */
4464 	mlx5_flow_stop(dev, list);
4465 	rte_errno = ret; /* Restore rte_errno. */
4466 	return -rte_errno;
4467 }
4468 
4469 /**
4470  * Verify the flow list is empty
4471  *
4472  * @param dev
4473  *  Pointer to Ethernet device.
4474  *
4475  * @return the number of flows not released.
4476  */
4477 int
4478 mlx5_flow_verify(struct rte_eth_dev *dev)
4479 {
4480 	struct mlx5_priv *priv = dev->data->dev_private;
4481 	struct rte_flow *flow;
4482 	int ret = 0;
4483 
4484 	TAILQ_FOREACH(flow, &priv->flows, next) {
4485 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4486 			dev->data->port_id, (void *)flow);
4487 		++ret;
4488 	}
4489 	return ret;
4490 }
4491 
4492 /**
4493  * Enable default hairpin egress flow.
4494  *
4495  * @param dev
4496  *   Pointer to Ethernet device.
4497  * @param queue
4498  *   The queue index.
4499  *
4500  * @return
4501  *   0 on success, a negative errno value otherwise and rte_errno is set.
4502  */
4503 int
4504 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4505 			    uint32_t queue)
4506 {
4507 	struct mlx5_priv *priv = dev->data->dev_private;
4508 	const struct rte_flow_attr attr = {
4509 		.egress = 1,
4510 		.priority = 0,
4511 	};
4512 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4513 		.queue = queue,
4514 	};
4515 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4516 		.queue = UINT32_MAX,
4517 	};
4518 	struct rte_flow_item items[] = {
4519 		{
4520 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4521 			.spec = &queue_spec,
4522 			.last = NULL,
4523 			.mask = &queue_mask,
4524 		},
4525 		{
4526 			.type = RTE_FLOW_ITEM_TYPE_END,
4527 		},
4528 	};
4529 	struct rte_flow_action_jump jump = {
4530 		.group = MLX5_HAIRPIN_TX_TABLE,
4531 	};
4532 	struct rte_flow_action actions[2];
4533 	struct rte_flow *flow;
4534 	struct rte_flow_error error;
4535 
4536 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4537 	actions[0].conf = &jump;
4538 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4539 	flow = flow_list_create(dev, &priv->ctrl_flows,
4540 				&attr, items, actions, false, &error);
4541 	if (!flow) {
4542 		DRV_LOG(DEBUG,
4543 			"Failed to create ctrl flow: rte_errno(%d),"
4544 			" type(%d), message(%s)",
4545 			rte_errno, error.type,
4546 			error.message ? error.message : " (no stated reason)");
4547 		return -rte_errno;
4548 	}
4549 	return 0;
4550 }
4551 
4552 /**
4553  * Enable a control flow configured from the control plane.
4554  *
4555  * @param dev
4556  *   Pointer to Ethernet device.
4557  * @param eth_spec
4558  *   An Ethernet flow spec to apply.
4559  * @param eth_mask
4560  *   An Ethernet flow mask to apply.
4561  * @param vlan_spec
4562  *   A VLAN flow spec to apply.
4563  * @param vlan_mask
4564  *   A VLAN flow mask to apply.
4565  *
4566  * @return
4567  *   0 on success, a negative errno value otherwise and rte_errno is set.
4568  */
4569 int
4570 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4571 		    struct rte_flow_item_eth *eth_spec,
4572 		    struct rte_flow_item_eth *eth_mask,
4573 		    struct rte_flow_item_vlan *vlan_spec,
4574 		    struct rte_flow_item_vlan *vlan_mask)
4575 {
4576 	struct mlx5_priv *priv = dev->data->dev_private;
4577 	const struct rte_flow_attr attr = {
4578 		.ingress = 1,
4579 		.priority = MLX5_FLOW_PRIO_RSVD,
4580 	};
4581 	struct rte_flow_item items[] = {
4582 		{
4583 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4584 			.spec = eth_spec,
4585 			.last = NULL,
4586 			.mask = eth_mask,
4587 		},
4588 		{
4589 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4590 					      RTE_FLOW_ITEM_TYPE_END,
4591 			.spec = vlan_spec,
4592 			.last = NULL,
4593 			.mask = vlan_mask,
4594 		},
4595 		{
4596 			.type = RTE_FLOW_ITEM_TYPE_END,
4597 		},
4598 	};
4599 	uint16_t queue[priv->reta_idx_n];
4600 	struct rte_flow_action_rss action_rss = {
4601 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4602 		.level = 0,
4603 		.types = priv->rss_conf.rss_hf,
4604 		.key_len = priv->rss_conf.rss_key_len,
4605 		.queue_num = priv->reta_idx_n,
4606 		.key = priv->rss_conf.rss_key,
4607 		.queue = queue,
4608 	};
4609 	struct rte_flow_action actions[] = {
4610 		{
4611 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4612 			.conf = &action_rss,
4613 		},
4614 		{
4615 			.type = RTE_FLOW_ACTION_TYPE_END,
4616 		},
4617 	};
4618 	struct rte_flow *flow;
4619 	struct rte_flow_error error;
4620 	unsigned int i;
4621 
4622 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4623 		return 0;
4624 	}
4625 	for (i = 0; i != priv->reta_idx_n; ++i)
4626 		queue[i] = (*priv->reta_idx)[i];
4627 	flow = flow_list_create(dev, &priv->ctrl_flows,
4628 				&attr, items, actions, false, &error);
4629 	if (!flow)
4630 		return -rte_errno;
4631 	return 0;
4632 }
4633 
4634 /**
4635  * Enable a flow control configured from the control plane.
4636  *
4637  * @param dev
4638  *   Pointer to Ethernet device.
4639  * @param eth_spec
4640  *   An Ethernet flow spec to apply.
4641  * @param eth_mask
4642  *   An Ethernet flow mask to apply.
4643  *
4644  * @return
4645  *   0 on success, a negative errno value otherwise and rte_errno is set.
4646  */
4647 int
4648 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4649 	       struct rte_flow_item_eth *eth_spec,
4650 	       struct rte_flow_item_eth *eth_mask)
4651 {
4652 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4653 }
4654 
4655 /**
4656  * Destroy a flow.
4657  *
4658  * @see rte_flow_destroy()
4659  * @see rte_flow_ops
4660  */
4661 int
4662 mlx5_flow_destroy(struct rte_eth_dev *dev,
4663 		  struct rte_flow *flow,
4664 		  struct rte_flow_error *error __rte_unused)
4665 {
4666 	struct mlx5_priv *priv = dev->data->dev_private;
4667 
4668 	flow_list_destroy(dev, &priv->flows, flow);
4669 	return 0;
4670 }
4671 
4672 /**
4673  * Destroy all flows.
4674  *
4675  * @see rte_flow_flush()
4676  * @see rte_flow_ops
4677  */
4678 int
4679 mlx5_flow_flush(struct rte_eth_dev *dev,
4680 		struct rte_flow_error *error __rte_unused)
4681 {
4682 	struct mlx5_priv *priv = dev->data->dev_private;
4683 
4684 	mlx5_flow_list_flush(dev, &priv->flows);
4685 	return 0;
4686 }
4687 
4688 /**
4689  * Isolated mode.
4690  *
4691  * @see rte_flow_isolate()
4692  * @see rte_flow_ops
4693  */
4694 int
4695 mlx5_flow_isolate(struct rte_eth_dev *dev,
4696 		  int enable,
4697 		  struct rte_flow_error *error)
4698 {
4699 	struct mlx5_priv *priv = dev->data->dev_private;
4700 
4701 	if (dev->data->dev_started) {
4702 		rte_flow_error_set(error, EBUSY,
4703 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4704 				   NULL,
4705 				   "port must be stopped first");
4706 		return -rte_errno;
4707 	}
4708 	priv->isolated = !!enable;
4709 	if (enable)
4710 		dev->dev_ops = &mlx5_dev_ops_isolate;
4711 	else
4712 		dev->dev_ops = &mlx5_dev_ops;
4713 	return 0;
4714 }
4715 
4716 /**
4717  * Query a flow.
4718  *
4719  * @see rte_flow_query()
4720  * @see rte_flow_ops
4721  */
4722 static int
4723 flow_drv_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 	const struct mlx5_flow_driver_ops *fops;
4730 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4731 
4732 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4733 	fops = flow_get_drv_ops(ftype);
4734 
4735 	return fops->query(dev, flow, actions, data, error);
4736 }
4737 
4738 /**
4739  * Query a flow.
4740  *
4741  * @see rte_flow_query()
4742  * @see rte_flow_ops
4743  */
4744 int
4745 mlx5_flow_query(struct rte_eth_dev *dev,
4746 		struct rte_flow *flow,
4747 		const struct rte_flow_action *actions,
4748 		void *data,
4749 		struct rte_flow_error *error)
4750 {
4751 	int ret;
4752 
4753 	ret = flow_drv_query(dev, flow, actions, data, error);
4754 	if (ret < 0)
4755 		return ret;
4756 	return 0;
4757 }
4758 
4759 /**
4760  * Convert a flow director filter to a generic flow.
4761  *
4762  * @param dev
4763  *   Pointer to Ethernet device.
4764  * @param fdir_filter
4765  *   Flow director filter to add.
4766  * @param attributes
4767  *   Generic flow parameters structure.
4768  *
4769  * @return
4770  *   0 on success, a negative errno value otherwise and rte_errno is set.
4771  */
4772 static int
4773 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4774 			 const struct rte_eth_fdir_filter *fdir_filter,
4775 			 struct mlx5_fdir *attributes)
4776 {
4777 	struct mlx5_priv *priv = dev->data->dev_private;
4778 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4779 	const struct rte_eth_fdir_masks *mask =
4780 		&dev->data->dev_conf.fdir_conf.mask;
4781 
4782 	/* Validate queue number. */
4783 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4784 		DRV_LOG(ERR, "port %u invalid queue number %d",
4785 			dev->data->port_id, fdir_filter->action.rx_queue);
4786 		rte_errno = EINVAL;
4787 		return -rte_errno;
4788 	}
4789 	attributes->attr.ingress = 1;
4790 	attributes->items[0] = (struct rte_flow_item) {
4791 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4792 		.spec = &attributes->l2,
4793 		.mask = &attributes->l2_mask,
4794 	};
4795 	switch (fdir_filter->action.behavior) {
4796 	case RTE_ETH_FDIR_ACCEPT:
4797 		attributes->actions[0] = (struct rte_flow_action){
4798 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4799 			.conf = &attributes->queue,
4800 		};
4801 		break;
4802 	case RTE_ETH_FDIR_REJECT:
4803 		attributes->actions[0] = (struct rte_flow_action){
4804 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4805 		};
4806 		break;
4807 	default:
4808 		DRV_LOG(ERR, "port %u invalid behavior %d",
4809 			dev->data->port_id,
4810 			fdir_filter->action.behavior);
4811 		rte_errno = ENOTSUP;
4812 		return -rte_errno;
4813 	}
4814 	attributes->queue.index = fdir_filter->action.rx_queue;
4815 	/* Handle L3. */
4816 	switch (fdir_filter->input.flow_type) {
4817 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4818 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4819 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4820 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4821 			.src_addr = input->flow.ip4_flow.src_ip,
4822 			.dst_addr = input->flow.ip4_flow.dst_ip,
4823 			.time_to_live = input->flow.ip4_flow.ttl,
4824 			.type_of_service = input->flow.ip4_flow.tos,
4825 		};
4826 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4827 			.src_addr = mask->ipv4_mask.src_ip,
4828 			.dst_addr = mask->ipv4_mask.dst_ip,
4829 			.time_to_live = mask->ipv4_mask.ttl,
4830 			.type_of_service = mask->ipv4_mask.tos,
4831 			.next_proto_id = mask->ipv4_mask.proto,
4832 		};
4833 		attributes->items[1] = (struct rte_flow_item){
4834 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4835 			.spec = &attributes->l3,
4836 			.mask = &attributes->l3_mask,
4837 		};
4838 		break;
4839 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4840 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4841 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4842 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4843 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4844 			.proto = input->flow.ipv6_flow.proto,
4845 		};
4846 
4847 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4848 		       input->flow.ipv6_flow.src_ip,
4849 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4850 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4851 		       input->flow.ipv6_flow.dst_ip,
4852 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4853 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4854 		       mask->ipv6_mask.src_ip,
4855 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4856 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4857 		       mask->ipv6_mask.dst_ip,
4858 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4859 		attributes->items[1] = (struct rte_flow_item){
4860 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4861 			.spec = &attributes->l3,
4862 			.mask = &attributes->l3_mask,
4863 		};
4864 		break;
4865 	default:
4866 		DRV_LOG(ERR, "port %u invalid flow type%d",
4867 			dev->data->port_id, fdir_filter->input.flow_type);
4868 		rte_errno = ENOTSUP;
4869 		return -rte_errno;
4870 	}
4871 	/* Handle L4. */
4872 	switch (fdir_filter->input.flow_type) {
4873 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4874 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4875 			.src_port = input->flow.udp4_flow.src_port,
4876 			.dst_port = input->flow.udp4_flow.dst_port,
4877 		};
4878 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4879 			.src_port = mask->src_port_mask,
4880 			.dst_port = mask->dst_port_mask,
4881 		};
4882 		attributes->items[2] = (struct rte_flow_item){
4883 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4884 			.spec = &attributes->l4,
4885 			.mask = &attributes->l4_mask,
4886 		};
4887 		break;
4888 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4889 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4890 			.src_port = input->flow.tcp4_flow.src_port,
4891 			.dst_port = input->flow.tcp4_flow.dst_port,
4892 		};
4893 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4894 			.src_port = mask->src_port_mask,
4895 			.dst_port = mask->dst_port_mask,
4896 		};
4897 		attributes->items[2] = (struct rte_flow_item){
4898 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4899 			.spec = &attributes->l4,
4900 			.mask = &attributes->l4_mask,
4901 		};
4902 		break;
4903 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4904 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4905 			.src_port = input->flow.udp6_flow.src_port,
4906 			.dst_port = input->flow.udp6_flow.dst_port,
4907 		};
4908 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4909 			.src_port = mask->src_port_mask,
4910 			.dst_port = mask->dst_port_mask,
4911 		};
4912 		attributes->items[2] = (struct rte_flow_item){
4913 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4914 			.spec = &attributes->l4,
4915 			.mask = &attributes->l4_mask,
4916 		};
4917 		break;
4918 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4919 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4920 			.src_port = input->flow.tcp6_flow.src_port,
4921 			.dst_port = input->flow.tcp6_flow.dst_port,
4922 		};
4923 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4924 			.src_port = mask->src_port_mask,
4925 			.dst_port = mask->dst_port_mask,
4926 		};
4927 		attributes->items[2] = (struct rte_flow_item){
4928 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4929 			.spec = &attributes->l4,
4930 			.mask = &attributes->l4_mask,
4931 		};
4932 		break;
4933 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4934 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4935 		break;
4936 	default:
4937 		DRV_LOG(ERR, "port %u invalid flow type%d",
4938 			dev->data->port_id, fdir_filter->input.flow_type);
4939 		rte_errno = ENOTSUP;
4940 		return -rte_errno;
4941 	}
4942 	return 0;
4943 }
4944 
4945 #define FLOW_FDIR_CMP(f1, f2, fld) \
4946 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4947 
4948 /**
4949  * Compare two FDIR flows. If items and actions are identical, the two flows are
4950  * regarded as same.
4951  *
4952  * @param dev
4953  *   Pointer to Ethernet device.
4954  * @param f1
4955  *   FDIR flow to compare.
4956  * @param f2
4957  *   FDIR flow to compare.
4958  *
4959  * @return
4960  *   Zero on match, 1 otherwise.
4961  */
4962 static int
4963 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4964 {
4965 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4966 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4967 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4968 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4969 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4970 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4971 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4972 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4973 		return 1;
4974 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4975 	    FLOW_FDIR_CMP(f1, f2, queue))
4976 		return 1;
4977 	return 0;
4978 }
4979 
4980 /**
4981  * Search device flow list to find out a matched FDIR flow.
4982  *
4983  * @param dev
4984  *   Pointer to Ethernet device.
4985  * @param fdir_flow
4986  *   FDIR flow to lookup.
4987  *
4988  * @return
4989  *   Pointer of flow if found, NULL otherwise.
4990  */
4991 static struct rte_flow *
4992 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
4993 {
4994 	struct mlx5_priv *priv = dev->data->dev_private;
4995 	struct rte_flow *flow = NULL;
4996 
4997 	assert(fdir_flow);
4998 	TAILQ_FOREACH(flow, &priv->flows, next) {
4999 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5000 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5001 				dev->data->port_id, (void *)flow);
5002 			break;
5003 		}
5004 	}
5005 	return flow;
5006 }
5007 
5008 /**
5009  * Add new flow director filter and store it in list.
5010  *
5011  * @param dev
5012  *   Pointer to Ethernet device.
5013  * @param fdir_filter
5014  *   Flow director filter to add.
5015  *
5016  * @return
5017  *   0 on success, a negative errno value otherwise and rte_errno is set.
5018  */
5019 static int
5020 flow_fdir_filter_add(struct rte_eth_dev *dev,
5021 		     const struct rte_eth_fdir_filter *fdir_filter)
5022 {
5023 	struct mlx5_priv *priv = dev->data->dev_private;
5024 	struct mlx5_fdir *fdir_flow;
5025 	struct rte_flow *flow;
5026 	int ret;
5027 
5028 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5029 	if (!fdir_flow) {
5030 		rte_errno = ENOMEM;
5031 		return -rte_errno;
5032 	}
5033 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5034 	if (ret)
5035 		goto error;
5036 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5037 	if (flow) {
5038 		rte_errno = EEXIST;
5039 		goto error;
5040 	}
5041 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5042 				fdir_flow->items, fdir_flow->actions, true,
5043 				NULL);
5044 	if (!flow)
5045 		goto error;
5046 	assert(!flow->fdir);
5047 	flow->fdir = fdir_flow;
5048 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5049 		dev->data->port_id, (void *)flow);
5050 	return 0;
5051 error:
5052 	rte_free(fdir_flow);
5053 	return -rte_errno;
5054 }
5055 
5056 /**
5057  * Delete specific filter.
5058  *
5059  * @param dev
5060  *   Pointer to Ethernet device.
5061  * @param fdir_filter
5062  *   Filter to be deleted.
5063  *
5064  * @return
5065  *   0 on success, a negative errno value otherwise and rte_errno is set.
5066  */
5067 static int
5068 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5069 			const struct rte_eth_fdir_filter *fdir_filter)
5070 {
5071 	struct mlx5_priv *priv = dev->data->dev_private;
5072 	struct rte_flow *flow;
5073 	struct mlx5_fdir fdir_flow = {
5074 		.attr.group = 0,
5075 	};
5076 	int ret;
5077 
5078 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5079 	if (ret)
5080 		return -rte_errno;
5081 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5082 	if (!flow) {
5083 		rte_errno = ENOENT;
5084 		return -rte_errno;
5085 	}
5086 	flow_list_destroy(dev, &priv->flows, flow);
5087 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5088 		dev->data->port_id, (void *)flow);
5089 	return 0;
5090 }
5091 
5092 /**
5093  * Update queue for specific filter.
5094  *
5095  * @param dev
5096  *   Pointer to Ethernet device.
5097  * @param fdir_filter
5098  *   Filter to be updated.
5099  *
5100  * @return
5101  *   0 on success, a negative errno value otherwise and rte_errno is set.
5102  */
5103 static int
5104 flow_fdir_filter_update(struct rte_eth_dev *dev,
5105 			const struct rte_eth_fdir_filter *fdir_filter)
5106 {
5107 	int ret;
5108 
5109 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5110 	if (ret)
5111 		return ret;
5112 	return flow_fdir_filter_add(dev, fdir_filter);
5113 }
5114 
5115 /**
5116  * Flush all filters.
5117  *
5118  * @param dev
5119  *   Pointer to Ethernet device.
5120  */
5121 static void
5122 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5123 {
5124 	struct mlx5_priv *priv = dev->data->dev_private;
5125 
5126 	mlx5_flow_list_flush(dev, &priv->flows);
5127 }
5128 
5129 /**
5130  * Get flow director information.
5131  *
5132  * @param dev
5133  *   Pointer to Ethernet device.
5134  * @param[out] fdir_info
5135  *   Resulting flow director information.
5136  */
5137 static void
5138 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5139 {
5140 	struct rte_eth_fdir_masks *mask =
5141 		&dev->data->dev_conf.fdir_conf.mask;
5142 
5143 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5144 	fdir_info->guarant_spc = 0;
5145 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5146 	fdir_info->max_flexpayload = 0;
5147 	fdir_info->flow_types_mask[0] = 0;
5148 	fdir_info->flex_payload_unit = 0;
5149 	fdir_info->max_flex_payload_segment_num = 0;
5150 	fdir_info->flex_payload_limit = 0;
5151 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5152 }
5153 
5154 /**
5155  * Deal with flow director operations.
5156  *
5157  * @param dev
5158  *   Pointer to Ethernet device.
5159  * @param filter_op
5160  *   Operation to perform.
5161  * @param arg
5162  *   Pointer to operation-specific structure.
5163  *
5164  * @return
5165  *   0 on success, a negative errno value otherwise and rte_errno is set.
5166  */
5167 static int
5168 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5169 		    void *arg)
5170 {
5171 	enum rte_fdir_mode fdir_mode =
5172 		dev->data->dev_conf.fdir_conf.mode;
5173 
5174 	if (filter_op == RTE_ETH_FILTER_NOP)
5175 		return 0;
5176 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5177 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5178 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5179 			dev->data->port_id, fdir_mode);
5180 		rte_errno = EINVAL;
5181 		return -rte_errno;
5182 	}
5183 	switch (filter_op) {
5184 	case RTE_ETH_FILTER_ADD:
5185 		return flow_fdir_filter_add(dev, arg);
5186 	case RTE_ETH_FILTER_UPDATE:
5187 		return flow_fdir_filter_update(dev, arg);
5188 	case RTE_ETH_FILTER_DELETE:
5189 		return flow_fdir_filter_delete(dev, arg);
5190 	case RTE_ETH_FILTER_FLUSH:
5191 		flow_fdir_filter_flush(dev);
5192 		break;
5193 	case RTE_ETH_FILTER_INFO:
5194 		flow_fdir_info_get(dev, arg);
5195 		break;
5196 	default:
5197 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5198 			dev->data->port_id, filter_op);
5199 		rte_errno = EINVAL;
5200 		return -rte_errno;
5201 	}
5202 	return 0;
5203 }
5204 
5205 /**
5206  * Manage filter operations.
5207  *
5208  * @param dev
5209  *   Pointer to Ethernet device structure.
5210  * @param filter_type
5211  *   Filter type.
5212  * @param filter_op
5213  *   Operation to perform.
5214  * @param arg
5215  *   Pointer to operation-specific structure.
5216  *
5217  * @return
5218  *   0 on success, a negative errno value otherwise and rte_errno is set.
5219  */
5220 int
5221 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5222 		     enum rte_filter_type filter_type,
5223 		     enum rte_filter_op filter_op,
5224 		     void *arg)
5225 {
5226 	switch (filter_type) {
5227 	case RTE_ETH_FILTER_GENERIC:
5228 		if (filter_op != RTE_ETH_FILTER_GET) {
5229 			rte_errno = EINVAL;
5230 			return -rte_errno;
5231 		}
5232 		*(const void **)arg = &mlx5_flow_ops;
5233 		return 0;
5234 	case RTE_ETH_FILTER_FDIR:
5235 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5236 	default:
5237 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5238 			dev->data->port_id, filter_type);
5239 		rte_errno = ENOTSUP;
5240 		return -rte_errno;
5241 	}
5242 	return 0;
5243 }
5244 
5245 /**
5246  * Create the needed meter and suffix tables.
5247  *
5248  * @param[in] dev
5249  *   Pointer to Ethernet device.
5250  * @param[in] fm
5251  *   Pointer to the flow meter.
5252  *
5253  * @return
5254  *   Pointer to table set on success, NULL otherwise.
5255  */
5256 struct mlx5_meter_domains_infos *
5257 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5258 			  const struct mlx5_flow_meter *fm)
5259 {
5260 	const struct mlx5_flow_driver_ops *fops;
5261 
5262 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5263 	return fops->create_mtr_tbls(dev, fm);
5264 }
5265 
5266 /**
5267  * Destroy the meter table set.
5268  *
5269  * @param[in] dev
5270  *   Pointer to Ethernet device.
5271  * @param[in] tbl
5272  *   Pointer to the meter table set.
5273  *
5274  * @return
5275  *   0 on success.
5276  */
5277 int
5278 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5279 			   struct mlx5_meter_domains_infos *tbls)
5280 {
5281 	const struct mlx5_flow_driver_ops *fops;
5282 
5283 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5284 	return fops->destroy_mtr_tbls(dev, tbls);
5285 }
5286 
5287 /**
5288  * Create policer rules.
5289  *
5290  * @param[in] dev
5291  *   Pointer to Ethernet device.
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_create_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->create_policer_rules(dev, fm, attr);
5309 }
5310 
5311 /**
5312  * Destroy policer rules.
5313  *
5314  * @param[in] fm
5315  *   Pointer to flow meter structure.
5316  * @param[in] attr
5317  *   Pointer to flow attributes.
5318  *
5319  * @return
5320  *   0 on success, -1 otherwise.
5321  */
5322 int
5323 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5324 				struct mlx5_flow_meter *fm,
5325 				const struct rte_flow_attr *attr)
5326 {
5327 	const struct mlx5_flow_driver_ops *fops;
5328 
5329 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5330 	return fops->destroy_policer_rules(dev, fm, attr);
5331 }
5332 
5333 /**
5334  * Allocate a counter.
5335  *
5336  * @param[in] dev
5337  *   Pointer to Ethernet device structure.
5338  *
5339  * @return
5340  *   Pointer to allocated counter  on success, NULL otherwise.
5341  */
5342 struct mlx5_flow_counter *
5343 mlx5_counter_alloc(struct rte_eth_dev *dev)
5344 {
5345 	const struct mlx5_flow_driver_ops *fops;
5346 	struct rte_flow_attr attr = { .transfer = 0 };
5347 
5348 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5349 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5350 		return fops->counter_alloc(dev);
5351 	}
5352 	DRV_LOG(ERR,
5353 		"port %u counter allocate is not supported.",
5354 		 dev->data->port_id);
5355 	return NULL;
5356 }
5357 
5358 /**
5359  * Free a counter.
5360  *
5361  * @param[in] dev
5362  *   Pointer to Ethernet device structure.
5363  * @param[in] cnt
5364  *   Pointer to counter to be free.
5365  */
5366 void
5367 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5368 {
5369 	const struct mlx5_flow_driver_ops *fops;
5370 	struct rte_flow_attr attr = { .transfer = 0 };
5371 
5372 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5373 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5374 		fops->counter_free(dev, cnt);
5375 		return;
5376 	}
5377 	DRV_LOG(ERR,
5378 		"port %u counter free is not supported.",
5379 		 dev->data->port_id);
5380 }
5381 
5382 /**
5383  * Query counter statistics.
5384  *
5385  * @param[in] dev
5386  *   Pointer to Ethernet device structure.
5387  * @param[in] cnt
5388  *   Pointer to counter to query.
5389  * @param[in] clear
5390  *   Set to clear counter statistics.
5391  * @param[out] pkts
5392  *   The counter hits packets number to save.
5393  * @param[out] bytes
5394  *   The counter hits bytes number to save.
5395  *
5396  * @return
5397  *   0 on success, a negative errno value otherwise.
5398  */
5399 int
5400 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5401 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5402 {
5403 	const struct mlx5_flow_driver_ops *fops;
5404 	struct rte_flow_attr attr = { .transfer = 0 };
5405 
5406 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5407 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5408 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5409 	}
5410 	DRV_LOG(ERR,
5411 		"port %u counter query is not supported.",
5412 		 dev->data->port_id);
5413 	return -ENOTSUP;
5414 }
5415 
5416 #define MLX5_POOL_QUERY_FREQ_US 1000000
5417 
5418 /**
5419  * Set the periodic procedure for triggering asynchronous batch queries for all
5420  * the counter pools.
5421  *
5422  * @param[in] sh
5423  *   Pointer to mlx5_ibv_shared object.
5424  */
5425 void
5426 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5427 {
5428 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5429 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5430 	uint32_t us;
5431 
5432 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5433 	pools_n += rte_atomic16_read(&cont->n_valid);
5434 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5435 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5436 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5437 		sh->cmng.query_thread_on = 0;
5438 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5439 	} else {
5440 		sh->cmng.query_thread_on = 1;
5441 	}
5442 }
5443 
5444 /**
5445  * The periodic procedure for triggering asynchronous batch queries for all the
5446  * counter pools. This function is probably called by the host thread.
5447  *
5448  * @param[in] arg
5449  *   The parameter for the alarm process.
5450  */
5451 void
5452 mlx5_flow_query_alarm(void *arg)
5453 {
5454 	struct mlx5_ibv_shared *sh = arg;
5455 	struct mlx5_devx_obj *dcs;
5456 	uint16_t offset;
5457 	int ret;
5458 	uint8_t batch = sh->cmng.batch;
5459 	uint16_t pool_index = sh->cmng.pool_index;
5460 	struct mlx5_pools_container *cont;
5461 	struct mlx5_pools_container *mcont;
5462 	struct mlx5_flow_counter_pool *pool;
5463 
5464 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5465 		goto set_alarm;
5466 next_container:
5467 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5468 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5469 	/* Check if resize was done and need to flip a container. */
5470 	if (cont != mcont) {
5471 		if (cont->pools) {
5472 			/* Clean the old container. */
5473 			rte_free(cont->pools);
5474 			memset(cont, 0, sizeof(*cont));
5475 		}
5476 		rte_cio_wmb();
5477 		 /* Flip the host container. */
5478 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5479 		cont = mcont;
5480 	}
5481 	if (!cont->pools) {
5482 		/* 2 empty containers case is unexpected. */
5483 		if (unlikely(batch != sh->cmng.batch))
5484 			goto set_alarm;
5485 		batch ^= 0x1;
5486 		pool_index = 0;
5487 		goto next_container;
5488 	}
5489 	pool = cont->pools[pool_index];
5490 	if (pool->raw_hw)
5491 		/* There is a pool query in progress. */
5492 		goto set_alarm;
5493 	pool->raw_hw =
5494 		LIST_FIRST(&sh->cmng.free_stat_raws);
5495 	if (!pool->raw_hw)
5496 		/* No free counter statistics raw memory. */
5497 		goto set_alarm;
5498 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5499 							      (&pool->a64_dcs);
5500 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5501 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5502 					       offset, NULL, NULL,
5503 					       pool->raw_hw->mem_mng->dm->id,
5504 					       (void *)(uintptr_t)
5505 					       (pool->raw_hw->data + offset),
5506 					       sh->devx_comp,
5507 					       (uint64_t)(uintptr_t)pool);
5508 	if (ret) {
5509 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5510 			" %d", pool->min_dcs->id);
5511 		pool->raw_hw = NULL;
5512 		goto set_alarm;
5513 	}
5514 	pool->raw_hw->min_dcs_id = dcs->id;
5515 	LIST_REMOVE(pool->raw_hw, next);
5516 	sh->cmng.pending_queries++;
5517 	pool_index++;
5518 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5519 		batch ^= 0x1;
5520 		pool_index = 0;
5521 	}
5522 set_alarm:
5523 	sh->cmng.batch = batch;
5524 	sh->cmng.pool_index = pool_index;
5525 	mlx5_set_query_alarm(sh);
5526 }
5527 
5528 /**
5529  * Handler for the HW respond about ready values from an asynchronous batch
5530  * query. This function is probably called by the host thread.
5531  *
5532  * @param[in] sh
5533  *   The pointer to the shared IB device context.
5534  * @param[in] async_id
5535  *   The Devx async ID.
5536  * @param[in] status
5537  *   The status of the completion.
5538  */
5539 void
5540 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5541 				  uint64_t async_id, int status)
5542 {
5543 	struct mlx5_flow_counter_pool *pool =
5544 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5545 	struct mlx5_counter_stats_raw *raw_to_free;
5546 
5547 	if (unlikely(status)) {
5548 		raw_to_free = pool->raw_hw;
5549 	} else {
5550 		raw_to_free = pool->raw;
5551 		rte_spinlock_lock(&pool->sl);
5552 		pool->raw = pool->raw_hw;
5553 		rte_spinlock_unlock(&pool->sl);
5554 		rte_atomic64_add(&pool->query_gen, 1);
5555 		/* Be sure the new raw counters data is updated in memory. */
5556 		rte_cio_wmb();
5557 	}
5558 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5559 	pool->raw_hw = NULL;
5560 	sh->cmng.pending_queries--;
5561 }
5562 
5563 /**
5564  * Translate the rte_flow group index to HW table value.
5565  *
5566  * @param[in] attributes
5567  *   Pointer to flow attributes
5568  * @param[in] external
5569  *   Value is part of flow rule created by request external to PMD.
5570  * @param[in] group
5571  *   rte_flow group index value.
5572  * @param[out] table
5573  *   HW table value.
5574  * @param[out] error
5575  *   Pointer to error structure.
5576  *
5577  * @return
5578  *   0 on success, a negative errno value otherwise and rte_errno is set.
5579  */
5580 int
5581 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5582 			 uint32_t group, uint32_t *table,
5583 			 struct rte_flow_error *error)
5584 {
5585 	if (attributes->transfer && external) {
5586 		if (group == UINT32_MAX)
5587 			return rte_flow_error_set
5588 						(error, EINVAL,
5589 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5590 						 NULL,
5591 						 "group index not supported");
5592 		*table = group + 1;
5593 	} else {
5594 		*table = group;
5595 	}
5596 	return 0;
5597 }
5598 
5599 /**
5600  * Discover availability of metadata reg_c's.
5601  *
5602  * Iteratively use test flows to check availability.
5603  *
5604  * @param[in] dev
5605  *   Pointer to the Ethernet device structure.
5606  *
5607  * @return
5608  *   0 on success, a negative errno value otherwise and rte_errno is set.
5609  */
5610 int
5611 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5612 {
5613 	struct mlx5_priv *priv = dev->data->dev_private;
5614 	struct mlx5_dev_config *config = &priv->config;
5615 	enum modify_reg idx;
5616 	int n = 0;
5617 
5618 	/* reg_c[0] and reg_c[1] are reserved. */
5619 	config->flow_mreg_c[n++] = REG_C_0;
5620 	config->flow_mreg_c[n++] = REG_C_1;
5621 	/* Discover availability of other reg_c's. */
5622 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5623 		struct rte_flow_attr attr = {
5624 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5625 			.priority = MLX5_FLOW_PRIO_RSVD,
5626 			.ingress = 1,
5627 		};
5628 		struct rte_flow_item items[] = {
5629 			[0] = {
5630 				.type = RTE_FLOW_ITEM_TYPE_END,
5631 			},
5632 		};
5633 		struct rte_flow_action actions[] = {
5634 			[0] = {
5635 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5636 				.conf = &(struct mlx5_flow_action_copy_mreg){
5637 					.src = REG_C_1,
5638 					.dst = idx,
5639 				},
5640 			},
5641 			[1] = {
5642 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5643 				.conf = &(struct rte_flow_action_jump){
5644 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5645 				},
5646 			},
5647 			[2] = {
5648 				.type = RTE_FLOW_ACTION_TYPE_END,
5649 			},
5650 		};
5651 		struct rte_flow *flow;
5652 		struct rte_flow_error error;
5653 
5654 		if (!config->dv_flow_en)
5655 			break;
5656 		/* Create internal flow, validation skips copy action. */
5657 		flow = flow_list_create(dev, NULL, &attr, items,
5658 					actions, false, &error);
5659 		if (!flow)
5660 			continue;
5661 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5662 			config->flow_mreg_c[n++] = idx;
5663 		flow_list_destroy(dev, NULL, flow);
5664 	}
5665 	for (; n < MLX5_MREG_C_NUM; ++n)
5666 		config->flow_mreg_c[n] = REG_NONE;
5667 	return 0;
5668 }
5669