xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 72a944dba1639fff11f74c5ae1ab7d622bcc39fd)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 /* Verbs header. */
14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
15 #ifdef PEDANTIC
16 #pragma GCC diagnostic ignored "-Wpedantic"
17 #endif
18 #include <infiniband/verbs.h>
19 #ifdef PEDANTIC
20 #pragma GCC diagnostic error "-Wpedantic"
21 #endif
22 
23 #include <rte_common.h>
24 #include <rte_ether.h>
25 #include <rte_ethdev_driver.h>
26 #include <rte_flow.h>
27 #include <rte_flow_driver.h>
28 #include <rte_malloc.h>
29 #include <rte_ip.h>
30 
31 #include <mlx5_glue.h>
32 #include <mlx5_devx_cmds.h>
33 #include <mlx5_prm.h>
34 
35 #include "mlx5_defs.h"
36 #include "mlx5.h"
37 #include "mlx5_flow.h"
38 #include "mlx5_rxtx.h"
39 
40 /* Dev ops structure defined in mlx5.c */
41 extern const struct eth_dev_ops mlx5_dev_ops;
42 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
43 
44 /** Device flow drivers. */
45 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
46 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
47 #endif
48 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
49 
50 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
51 
52 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
53 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
54 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
55 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
56 #endif
57 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
58 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
59 };
60 
61 enum mlx5_expansion {
62 	MLX5_EXPANSION_ROOT,
63 	MLX5_EXPANSION_ROOT_OUTER,
64 	MLX5_EXPANSION_ROOT_ETH_VLAN,
65 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
66 	MLX5_EXPANSION_OUTER_ETH,
67 	MLX5_EXPANSION_OUTER_ETH_VLAN,
68 	MLX5_EXPANSION_OUTER_VLAN,
69 	MLX5_EXPANSION_OUTER_IPV4,
70 	MLX5_EXPANSION_OUTER_IPV4_UDP,
71 	MLX5_EXPANSION_OUTER_IPV4_TCP,
72 	MLX5_EXPANSION_OUTER_IPV6,
73 	MLX5_EXPANSION_OUTER_IPV6_UDP,
74 	MLX5_EXPANSION_OUTER_IPV6_TCP,
75 	MLX5_EXPANSION_VXLAN,
76 	MLX5_EXPANSION_VXLAN_GPE,
77 	MLX5_EXPANSION_GRE,
78 	MLX5_EXPANSION_MPLS,
79 	MLX5_EXPANSION_ETH,
80 	MLX5_EXPANSION_ETH_VLAN,
81 	MLX5_EXPANSION_VLAN,
82 	MLX5_EXPANSION_IPV4,
83 	MLX5_EXPANSION_IPV4_UDP,
84 	MLX5_EXPANSION_IPV4_TCP,
85 	MLX5_EXPANSION_IPV6,
86 	MLX5_EXPANSION_IPV6_UDP,
87 	MLX5_EXPANSION_IPV6_TCP,
88 };
89 
90 /** Supported expansion of items. */
91 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
92 	[MLX5_EXPANSION_ROOT] = {
93 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
94 						 MLX5_EXPANSION_IPV4,
95 						 MLX5_EXPANSION_IPV6),
96 		.type = RTE_FLOW_ITEM_TYPE_END,
97 	},
98 	[MLX5_EXPANSION_ROOT_OUTER] = {
99 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
100 						 MLX5_EXPANSION_OUTER_IPV4,
101 						 MLX5_EXPANSION_OUTER_IPV6),
102 		.type = RTE_FLOW_ITEM_TYPE_END,
103 	},
104 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
105 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
106 		.type = RTE_FLOW_ITEM_TYPE_END,
107 	},
108 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
109 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
110 		.type = RTE_FLOW_ITEM_TYPE_END,
111 	},
112 	[MLX5_EXPANSION_OUTER_ETH] = {
113 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
114 						 MLX5_EXPANSION_OUTER_IPV6,
115 						 MLX5_EXPANSION_MPLS),
116 		.type = RTE_FLOW_ITEM_TYPE_ETH,
117 		.rss_types = 0,
118 	},
119 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
120 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
121 		.type = RTE_FLOW_ITEM_TYPE_ETH,
122 		.rss_types = 0,
123 	},
124 	[MLX5_EXPANSION_OUTER_VLAN] = {
125 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
126 						 MLX5_EXPANSION_OUTER_IPV6),
127 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
128 	},
129 	[MLX5_EXPANSION_OUTER_IPV4] = {
130 		.next = RTE_FLOW_EXPAND_RSS_NEXT
131 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
132 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
133 			 MLX5_EXPANSION_GRE,
134 			 MLX5_EXPANSION_IPV4,
135 			 MLX5_EXPANSION_IPV6),
136 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
137 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
138 			ETH_RSS_NONFRAG_IPV4_OTHER,
139 	},
140 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
141 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
142 						 MLX5_EXPANSION_VXLAN_GPE),
143 		.type = RTE_FLOW_ITEM_TYPE_UDP,
144 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
145 	},
146 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
147 		.type = RTE_FLOW_ITEM_TYPE_TCP,
148 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
149 	},
150 	[MLX5_EXPANSION_OUTER_IPV6] = {
151 		.next = RTE_FLOW_EXPAND_RSS_NEXT
152 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
153 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
154 			 MLX5_EXPANSION_IPV4,
155 			 MLX5_EXPANSION_IPV6),
156 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
157 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
158 			ETH_RSS_NONFRAG_IPV6_OTHER,
159 	},
160 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
161 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
162 						 MLX5_EXPANSION_VXLAN_GPE),
163 		.type = RTE_FLOW_ITEM_TYPE_UDP,
164 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
165 	},
166 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
167 		.type = RTE_FLOW_ITEM_TYPE_TCP,
168 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
169 	},
170 	[MLX5_EXPANSION_VXLAN] = {
171 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
172 						 MLX5_EXPANSION_IPV4,
173 						 MLX5_EXPANSION_IPV6),
174 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
175 	},
176 	[MLX5_EXPANSION_VXLAN_GPE] = {
177 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
178 						 MLX5_EXPANSION_IPV4,
179 						 MLX5_EXPANSION_IPV6),
180 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
181 	},
182 	[MLX5_EXPANSION_GRE] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
184 		.type = RTE_FLOW_ITEM_TYPE_GRE,
185 	},
186 	[MLX5_EXPANSION_MPLS] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
190 	},
191 	[MLX5_EXPANSION_ETH] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
193 						 MLX5_EXPANSION_IPV6),
194 		.type = RTE_FLOW_ITEM_TYPE_ETH,
195 	},
196 	[MLX5_EXPANSION_ETH_VLAN] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
198 		.type = RTE_FLOW_ITEM_TYPE_ETH,
199 	},
200 	[MLX5_EXPANSION_VLAN] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
202 						 MLX5_EXPANSION_IPV6),
203 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
204 	},
205 	[MLX5_EXPANSION_IPV4] = {
206 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
207 						 MLX5_EXPANSION_IPV4_TCP),
208 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
209 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
210 			ETH_RSS_NONFRAG_IPV4_OTHER,
211 	},
212 	[MLX5_EXPANSION_IPV4_UDP] = {
213 		.type = RTE_FLOW_ITEM_TYPE_UDP,
214 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
215 	},
216 	[MLX5_EXPANSION_IPV4_TCP] = {
217 		.type = RTE_FLOW_ITEM_TYPE_TCP,
218 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
219 	},
220 	[MLX5_EXPANSION_IPV6] = {
221 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
222 						 MLX5_EXPANSION_IPV6_TCP),
223 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
224 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
225 			ETH_RSS_NONFRAG_IPV6_OTHER,
226 	},
227 	[MLX5_EXPANSION_IPV6_UDP] = {
228 		.type = RTE_FLOW_ITEM_TYPE_UDP,
229 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
230 	},
231 	[MLX5_EXPANSION_IPV6_TCP] = {
232 		.type = RTE_FLOW_ITEM_TYPE_TCP,
233 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
234 	},
235 };
236 
237 static const struct rte_flow_ops mlx5_flow_ops = {
238 	.validate = mlx5_flow_validate,
239 	.create = mlx5_flow_create,
240 	.destroy = mlx5_flow_destroy,
241 	.flush = mlx5_flow_flush,
242 	.isolate = mlx5_flow_isolate,
243 	.query = mlx5_flow_query,
244 	.dev_dump = mlx5_flow_dev_dump,
245 };
246 
247 /* Convert FDIR request to Generic flow. */
248 struct mlx5_fdir {
249 	struct rte_flow_attr attr;
250 	struct rte_flow_item items[4];
251 	struct rte_flow_item_eth l2;
252 	struct rte_flow_item_eth l2_mask;
253 	union {
254 		struct rte_flow_item_ipv4 ipv4;
255 		struct rte_flow_item_ipv6 ipv6;
256 	} l3;
257 	union {
258 		struct rte_flow_item_ipv4 ipv4;
259 		struct rte_flow_item_ipv6 ipv6;
260 	} l3_mask;
261 	union {
262 		struct rte_flow_item_udp udp;
263 		struct rte_flow_item_tcp tcp;
264 	} l4;
265 	union {
266 		struct rte_flow_item_udp udp;
267 		struct rte_flow_item_tcp tcp;
268 	} l4_mask;
269 	struct rte_flow_action actions[2];
270 	struct rte_flow_action_queue queue;
271 };
272 
273 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
274 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
275 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
276 };
277 
278 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
279 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
280 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
281 	{ 9, 10, 11 }, { 12, 13, 14 },
282 };
283 
284 /* Tunnel information. */
285 struct mlx5_flow_tunnel_info {
286 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
287 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
288 };
289 
290 static struct mlx5_flow_tunnel_info tunnels_info[] = {
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
293 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
297 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
301 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
302 	},
303 	{
304 		.tunnel = MLX5_FLOW_LAYER_GRE,
305 		.ptype = RTE_PTYPE_TUNNEL_GRE,
306 	},
307 	{
308 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
309 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
310 	},
311 	{
312 		.tunnel = MLX5_FLOW_LAYER_MPLS,
313 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
314 	},
315 	{
316 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
317 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
318 	},
319 	{
320 		.tunnel = MLX5_FLOW_LAYER_IPIP,
321 		.ptype = RTE_PTYPE_TUNNEL_IP,
322 	},
323 	{
324 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
325 		.ptype = RTE_PTYPE_TUNNEL_IP,
326 	},
327 	{
328 		.tunnel = MLX5_FLOW_LAYER_GTP,
329 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
330 	},
331 };
332 
333 /**
334  * Translate tag ID to register.
335  *
336  * @param[in] dev
337  *   Pointer to the Ethernet device structure.
338  * @param[in] feature
339  *   The feature that request the register.
340  * @param[in] id
341  *   The request register ID.
342  * @param[out] error
343  *   Error description in case of any.
344  *
345  * @return
346  *   The request register on success, a negative errno
347  *   value otherwise and rte_errno is set.
348  */
349 int
350 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
351 		     enum mlx5_feature_name feature,
352 		     uint32_t id,
353 		     struct rte_flow_error *error)
354 {
355 	struct mlx5_priv *priv = dev->data->dev_private;
356 	struct mlx5_dev_config *config = &priv->config;
357 	enum modify_reg start_reg;
358 	bool skip_mtr_reg = false;
359 
360 	switch (feature) {
361 	case MLX5_HAIRPIN_RX:
362 		return REG_B;
363 	case MLX5_HAIRPIN_TX:
364 		return REG_A;
365 	case MLX5_METADATA_RX:
366 		switch (config->dv_xmeta_en) {
367 		case MLX5_XMETA_MODE_LEGACY:
368 			return REG_B;
369 		case MLX5_XMETA_MODE_META16:
370 			return REG_C_0;
371 		case MLX5_XMETA_MODE_META32:
372 			return REG_C_1;
373 		}
374 		break;
375 	case MLX5_METADATA_TX:
376 		return REG_A;
377 	case MLX5_METADATA_FDB:
378 		switch (config->dv_xmeta_en) {
379 		case MLX5_XMETA_MODE_LEGACY:
380 			return REG_NONE;
381 		case MLX5_XMETA_MODE_META16:
382 			return REG_C_0;
383 		case MLX5_XMETA_MODE_META32:
384 			return REG_C_1;
385 		}
386 		break;
387 	case MLX5_FLOW_MARK:
388 		switch (config->dv_xmeta_en) {
389 		case MLX5_XMETA_MODE_LEGACY:
390 			return REG_NONE;
391 		case MLX5_XMETA_MODE_META16:
392 			return REG_C_1;
393 		case MLX5_XMETA_MODE_META32:
394 			return REG_C_0;
395 		}
396 		break;
397 	case MLX5_MTR_SFX:
398 		/*
399 		 * If meter color and flow match share one register, flow match
400 		 * should use the meter color register for match.
401 		 */
402 		if (priv->mtr_reg_share)
403 			return priv->mtr_color_reg;
404 		else
405 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
406 			       REG_C_3;
407 	case MLX5_MTR_COLOR:
408 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
409 		return priv->mtr_color_reg;
410 	case MLX5_COPY_MARK:
411 		/*
412 		 * Metadata COPY_MARK register using is in meter suffix sub
413 		 * flow while with meter. It's safe to share the same register.
414 		 */
415 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
416 	case MLX5_APP_TAG:
417 		/*
418 		 * If meter is enable, it will engage the register for color
419 		 * match and flow match. If meter color match is not using the
420 		 * REG_C_2, need to skip the REG_C_x be used by meter color
421 		 * match.
422 		 * If meter is disable, free to use all available registers.
423 		 */
424 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
425 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
426 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
427 		if (id > (REG_C_7 - start_reg))
428 			return rte_flow_error_set(error, EINVAL,
429 						  RTE_FLOW_ERROR_TYPE_ITEM,
430 						  NULL, "invalid tag id");
431 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
432 			return rte_flow_error_set(error, ENOTSUP,
433 						  RTE_FLOW_ERROR_TYPE_ITEM,
434 						  NULL, "unsupported tag id");
435 		/*
436 		 * This case means meter is using the REG_C_x great than 2.
437 		 * Take care not to conflict with meter color REG_C_x.
438 		 * If the available index REG_C_y >= REG_C_x, skip the
439 		 * color register.
440 		 */
441 		if (skip_mtr_reg && config->flow_mreg_c
442 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
443 			if (config->flow_mreg_c
444 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
445 				return config->flow_mreg_c
446 					       [id + 1 + start_reg - REG_C_0];
447 			return rte_flow_error_set(error, ENOTSUP,
448 						  RTE_FLOW_ERROR_TYPE_ITEM,
449 						  NULL, "unsupported tag id");
450 		}
451 		return config->flow_mreg_c[id + start_reg - REG_C_0];
452 	}
453 	MLX5_ASSERT(false);
454 	return rte_flow_error_set(error, EINVAL,
455 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
456 				  NULL, "invalid feature name");
457 }
458 
459 /**
460  * Check extensive flow metadata register support.
461  *
462  * @param dev
463  *   Pointer to rte_eth_dev structure.
464  *
465  * @return
466  *   True if device supports extensive flow metadata register, otherwise false.
467  */
468 bool
469 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
470 {
471 	struct mlx5_priv *priv = dev->data->dev_private;
472 	struct mlx5_dev_config *config = &priv->config;
473 
474 	/*
475 	 * Having available reg_c can be regarded inclusively as supporting
476 	 * extensive flow metadata register, which could mean,
477 	 * - metadata register copy action by modify header.
478 	 * - 16 modify header actions is supported.
479 	 * - reg_c's are preserved across different domain (FDB and NIC) on
480 	 *   packet loopback by flow lookup miss.
481 	 */
482 	return config->flow_mreg_c[2] != REG_NONE;
483 }
484 
485 /**
486  * Discover the maximum number of priority available.
487  *
488  * @param[in] dev
489  *   Pointer to the Ethernet device structure.
490  *
491  * @return
492  *   number of supported flow priority on success, a negative errno
493  *   value otherwise and rte_errno is set.
494  */
495 int
496 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
497 {
498 	struct mlx5_priv *priv = dev->data->dev_private;
499 	struct {
500 		struct ibv_flow_attr attr;
501 		struct ibv_flow_spec_eth eth;
502 		struct ibv_flow_spec_action_drop drop;
503 	} flow_attr = {
504 		.attr = {
505 			.num_of_specs = 2,
506 			.port = (uint8_t)priv->ibv_port,
507 		},
508 		.eth = {
509 			.type = IBV_FLOW_SPEC_ETH,
510 			.size = sizeof(struct ibv_flow_spec_eth),
511 		},
512 		.drop = {
513 			.size = sizeof(struct ibv_flow_spec_action_drop),
514 			.type = IBV_FLOW_SPEC_ACTION_DROP,
515 		},
516 	};
517 	struct ibv_flow *flow;
518 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
519 	uint16_t vprio[] = { 8, 16 };
520 	int i;
521 	int priority = 0;
522 
523 	if (!drop) {
524 		rte_errno = ENOTSUP;
525 		return -rte_errno;
526 	}
527 	for (i = 0; i != RTE_DIM(vprio); i++) {
528 		flow_attr.attr.priority = vprio[i] - 1;
529 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
530 		if (!flow)
531 			break;
532 		claim_zero(mlx5_glue->destroy_flow(flow));
533 		priority = vprio[i];
534 	}
535 	mlx5_hrxq_drop_release(dev);
536 	switch (priority) {
537 	case 8:
538 		priority = RTE_DIM(priority_map_3);
539 		break;
540 	case 16:
541 		priority = RTE_DIM(priority_map_5);
542 		break;
543 	default:
544 		rte_errno = ENOTSUP;
545 		DRV_LOG(ERR,
546 			"port %u verbs maximum priority: %d expected 8/16",
547 			dev->data->port_id, priority);
548 		return -rte_errno;
549 	}
550 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
551 		dev->data->port_id, priority);
552 	return priority;
553 }
554 
555 /**
556  * Adjust flow priority based on the highest layer and the request priority.
557  *
558  * @param[in] dev
559  *   Pointer to the Ethernet device structure.
560  * @param[in] priority
561  *   The rule base priority.
562  * @param[in] subpriority
563  *   The priority based on the items.
564  *
565  * @return
566  *   The new priority.
567  */
568 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
569 				   uint32_t subpriority)
570 {
571 	uint32_t res = 0;
572 	struct mlx5_priv *priv = dev->data->dev_private;
573 
574 	switch (priv->config.flow_prio) {
575 	case RTE_DIM(priority_map_3):
576 		res = priority_map_3[priority][subpriority];
577 		break;
578 	case RTE_DIM(priority_map_5):
579 		res = priority_map_5[priority][subpriority];
580 		break;
581 	}
582 	return  res;
583 }
584 
585 /**
586  * Verify the @p item specifications (spec, last, mask) are compatible with the
587  * NIC capabilities.
588  *
589  * @param[in] item
590  *   Item specification.
591  * @param[in] mask
592  *   @p item->mask or flow default bit-masks.
593  * @param[in] nic_mask
594  *   Bit-masks covering supported fields by the NIC to compare with user mask.
595  * @param[in] size
596  *   Bit-masks size in bytes.
597  * @param[out] error
598  *   Pointer to error structure.
599  *
600  * @return
601  *   0 on success, a negative errno value otherwise and rte_errno is set.
602  */
603 int
604 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
605 			  const uint8_t *mask,
606 			  const uint8_t *nic_mask,
607 			  unsigned int size,
608 			  struct rte_flow_error *error)
609 {
610 	unsigned int i;
611 
612 	MLX5_ASSERT(nic_mask);
613 	for (i = 0; i < size; ++i)
614 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
615 			return rte_flow_error_set(error, ENOTSUP,
616 						  RTE_FLOW_ERROR_TYPE_ITEM,
617 						  item,
618 						  "mask enables non supported"
619 						  " bits");
620 	if (!item->spec && (item->mask || item->last))
621 		return rte_flow_error_set(error, EINVAL,
622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
623 					  "mask/last without a spec is not"
624 					  " supported");
625 	if (item->spec && item->last) {
626 		uint8_t spec[size];
627 		uint8_t last[size];
628 		unsigned int i;
629 		int ret;
630 
631 		for (i = 0; i < size; ++i) {
632 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
633 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
634 		}
635 		ret = memcmp(spec, last, size);
636 		if (ret != 0)
637 			return rte_flow_error_set(error, EINVAL,
638 						  RTE_FLOW_ERROR_TYPE_ITEM,
639 						  item,
640 						  "range is not valid");
641 	}
642 	return 0;
643 }
644 
645 /**
646  * Adjust the hash fields according to the @p flow information.
647  *
648  * @param[in] dev_flow.
649  *   Pointer to the mlx5_flow.
650  * @param[in] tunnel
651  *   1 when the hash field is for a tunnel item.
652  * @param[in] layer_types
653  *   ETH_RSS_* types.
654  * @param[in] hash_fields
655  *   Item hash fields.
656  *
657  * @return
658  *   The hash fields that should be used.
659  */
660 uint64_t
661 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
662 			    int tunnel __rte_unused, uint64_t layer_types,
663 			    uint64_t hash_fields)
664 {
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 	int rss_request_inner = rss_desc->level >= 2;
667 
668 	/* Check RSS hash level for tunnel. */
669 	if (tunnel && rss_request_inner)
670 		hash_fields |= IBV_RX_HASH_INNER;
671 	else if (tunnel || rss_request_inner)
672 		return 0;
673 #endif
674 	/* Check if requested layer matches RSS hash fields. */
675 	if (!(rss_desc->types & layer_types))
676 		return 0;
677 	return hash_fields;
678 }
679 
680 /**
681  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
682  * if several tunnel rules are used on this queue, the tunnel ptype will be
683  * cleared.
684  *
685  * @param rxq_ctrl
686  *   Rx queue to update.
687  */
688 static void
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
690 {
691 	unsigned int i;
692 	uint32_t tunnel_ptype = 0;
693 
694 	/* Look up for the ptype to use. */
695 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 		if (!rxq_ctrl->flow_tunnels_n[i])
697 			continue;
698 		if (!tunnel_ptype) {
699 			tunnel_ptype = tunnels_info[i].ptype;
700 		} else {
701 			tunnel_ptype = 0;
702 			break;
703 		}
704 	}
705 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
706 }
707 
708 /**
709  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
710  * flow.
711  *
712  * @param[in] dev
713  *   Pointer to the Ethernet device structure.
714  * @param[in] dev_handle
715  *   Pointer to device flow handle structure.
716  */
717 static void
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
719 		       struct mlx5_flow_handle *dev_handle)
720 {
721 	struct mlx5_priv *priv = dev->data->dev_private;
722 	const int mark = dev_handle->mark;
723 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
724 	struct mlx5_hrxq *hrxq;
725 	unsigned int i;
726 
727 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
728 		return;
729 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
730 			      dev_handle->rix_hrxq);
731 	if (!hrxq)
732 		return;
733 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
734 		int idx = hrxq->ind_table->queues[i];
735 		struct mlx5_rxq_ctrl *rxq_ctrl =
736 			container_of((*priv->rxqs)[idx],
737 				     struct mlx5_rxq_ctrl, rxq);
738 
739 		/*
740 		 * To support metadata register copy on Tx loopback,
741 		 * this must be always enabled (metadata may arive
742 		 * from other port - not from local flows only.
743 		 */
744 		if (priv->config.dv_flow_en &&
745 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
746 		    mlx5_flow_ext_mreg_supported(dev)) {
747 			rxq_ctrl->rxq.mark = 1;
748 			rxq_ctrl->flow_mark_n = 1;
749 		} else if (mark) {
750 			rxq_ctrl->rxq.mark = 1;
751 			rxq_ctrl->flow_mark_n++;
752 		}
753 		if (tunnel) {
754 			unsigned int j;
755 
756 			/* Increase the counter matching the flow. */
757 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
758 				if ((tunnels_info[j].tunnel &
759 				     dev_handle->layers) ==
760 				    tunnels_info[j].tunnel) {
761 					rxq_ctrl->flow_tunnels_n[j]++;
762 					break;
763 				}
764 			}
765 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
766 		}
767 	}
768 }
769 
770 /**
771  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
772  *
773  * @param[in] dev
774  *   Pointer to the Ethernet device structure.
775  * @param[in] flow
776  *   Pointer to flow structure.
777  */
778 static void
779 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
780 {
781 	struct mlx5_priv *priv = dev->data->dev_private;
782 	uint32_t handle_idx;
783 	struct mlx5_flow_handle *dev_handle;
784 
785 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
786 		       handle_idx, dev_handle, next)
787 		flow_drv_rxq_flags_set(dev, dev_handle);
788 }
789 
790 /**
791  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
792  * device flow if no other flow uses it with the same kind of request.
793  *
794  * @param dev
795  *   Pointer to Ethernet device.
796  * @param[in] dev_handle
797  *   Pointer to the device flow handle structure.
798  */
799 static void
800 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
801 			struct mlx5_flow_handle *dev_handle)
802 {
803 	struct mlx5_priv *priv = dev->data->dev_private;
804 	const int mark = dev_handle->mark;
805 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
806 	struct mlx5_hrxq *hrxq;
807 	unsigned int i;
808 
809 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
810 		return;
811 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
812 			      dev_handle->rix_hrxq);
813 	if (!hrxq)
814 		return;
815 	MLX5_ASSERT(dev->data->dev_started);
816 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
817 		int idx = hrxq->ind_table->queues[i];
818 		struct mlx5_rxq_ctrl *rxq_ctrl =
819 			container_of((*priv->rxqs)[idx],
820 				     struct mlx5_rxq_ctrl, rxq);
821 
822 		if (priv->config.dv_flow_en &&
823 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
824 		    mlx5_flow_ext_mreg_supported(dev)) {
825 			rxq_ctrl->rxq.mark = 1;
826 			rxq_ctrl->flow_mark_n = 1;
827 		} else if (mark) {
828 			rxq_ctrl->flow_mark_n--;
829 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
830 		}
831 		if (tunnel) {
832 			unsigned int j;
833 
834 			/* Decrease the counter matching the flow. */
835 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
836 				if ((tunnels_info[j].tunnel &
837 				     dev_handle->layers) ==
838 				    tunnels_info[j].tunnel) {
839 					rxq_ctrl->flow_tunnels_n[j]--;
840 					break;
841 				}
842 			}
843 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
844 		}
845 	}
846 }
847 
848 /**
849  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
850  * @p flow if no other flow uses it with the same kind of request.
851  *
852  * @param dev
853  *   Pointer to Ethernet device.
854  * @param[in] flow
855  *   Pointer to the flow.
856  */
857 static void
858 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
859 {
860 	struct mlx5_priv *priv = dev->data->dev_private;
861 	uint32_t handle_idx;
862 	struct mlx5_flow_handle *dev_handle;
863 
864 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
865 		       handle_idx, dev_handle, next)
866 		flow_drv_rxq_flags_trim(dev, dev_handle);
867 }
868 
869 /**
870  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
871  *
872  * @param dev
873  *   Pointer to Ethernet device.
874  */
875 static void
876 flow_rxq_flags_clear(struct rte_eth_dev *dev)
877 {
878 	struct mlx5_priv *priv = dev->data->dev_private;
879 	unsigned int i;
880 
881 	for (i = 0; i != priv->rxqs_n; ++i) {
882 		struct mlx5_rxq_ctrl *rxq_ctrl;
883 		unsigned int j;
884 
885 		if (!(*priv->rxqs)[i])
886 			continue;
887 		rxq_ctrl = container_of((*priv->rxqs)[i],
888 					struct mlx5_rxq_ctrl, rxq);
889 		rxq_ctrl->flow_mark_n = 0;
890 		rxq_ctrl->rxq.mark = 0;
891 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
892 			rxq_ctrl->flow_tunnels_n[j] = 0;
893 		rxq_ctrl->rxq.tunnel = 0;
894 	}
895 }
896 
897 /*
898  * return a pointer to the desired action in the list of actions.
899  *
900  * @param[in] actions
901  *   The list of actions to search the action in.
902  * @param[in] action
903  *   The action to find.
904  *
905  * @return
906  *   Pointer to the action in the list, if found. NULL otherwise.
907  */
908 const struct rte_flow_action *
909 mlx5_flow_find_action(const struct rte_flow_action *actions,
910 		      enum rte_flow_action_type action)
911 {
912 	if (actions == NULL)
913 		return NULL;
914 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
915 		if (actions->type == action)
916 			return actions;
917 	return NULL;
918 }
919 
920 /*
921  * Validate the flag action.
922  *
923  * @param[in] action_flags
924  *   Bit-fields that holds the actions detected until now.
925  * @param[in] attr
926  *   Attributes of flow that includes this action.
927  * @param[out] error
928  *   Pointer to error structure.
929  *
930  * @return
931  *   0 on success, a negative errno value otherwise and rte_errno is set.
932  */
933 int
934 mlx5_flow_validate_action_flag(uint64_t action_flags,
935 			       const struct rte_flow_attr *attr,
936 			       struct rte_flow_error *error)
937 {
938 	if (action_flags & MLX5_FLOW_ACTION_MARK)
939 		return rte_flow_error_set(error, EINVAL,
940 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
941 					  "can't mark and flag in same flow");
942 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
943 		return rte_flow_error_set(error, EINVAL,
944 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
945 					  "can't have 2 flag"
946 					  " actions in same flow");
947 	if (attr->egress)
948 		return rte_flow_error_set(error, ENOTSUP,
949 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
950 					  "flag action not supported for "
951 					  "egress");
952 	return 0;
953 }
954 
955 /*
956  * Validate the mark action.
957  *
958  * @param[in] action
959  *   Pointer to the queue action.
960  * @param[in] action_flags
961  *   Bit-fields that holds the actions detected until now.
962  * @param[in] attr
963  *   Attributes of flow that includes this action.
964  * @param[out] error
965  *   Pointer to error structure.
966  *
967  * @return
968  *   0 on success, a negative errno value otherwise and rte_errno is set.
969  */
970 int
971 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
972 			       uint64_t action_flags,
973 			       const struct rte_flow_attr *attr,
974 			       struct rte_flow_error *error)
975 {
976 	const struct rte_flow_action_mark *mark = action->conf;
977 
978 	if (!mark)
979 		return rte_flow_error_set(error, EINVAL,
980 					  RTE_FLOW_ERROR_TYPE_ACTION,
981 					  action,
982 					  "configuration cannot be null");
983 	if (mark->id >= MLX5_FLOW_MARK_MAX)
984 		return rte_flow_error_set(error, EINVAL,
985 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
986 					  &mark->id,
987 					  "mark id must in 0 <= id < "
988 					  RTE_STR(MLX5_FLOW_MARK_MAX));
989 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
990 		return rte_flow_error_set(error, EINVAL,
991 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
992 					  "can't flag and mark in same flow");
993 	if (action_flags & MLX5_FLOW_ACTION_MARK)
994 		return rte_flow_error_set(error, EINVAL,
995 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
996 					  "can't have 2 mark actions in same"
997 					  " flow");
998 	if (attr->egress)
999 		return rte_flow_error_set(error, ENOTSUP,
1000 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1001 					  "mark action not supported for "
1002 					  "egress");
1003 	return 0;
1004 }
1005 
1006 /*
1007  * Validate the drop action.
1008  *
1009  * @param[in] action_flags
1010  *   Bit-fields that holds the actions detected until now.
1011  * @param[in] attr
1012  *   Attributes of flow that includes this action.
1013  * @param[out] error
1014  *   Pointer to error structure.
1015  *
1016  * @return
1017  *   0 on success, a negative errno value otherwise and rte_errno is set.
1018  */
1019 int
1020 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1021 			       const struct rte_flow_attr *attr,
1022 			       struct rte_flow_error *error)
1023 {
1024 	if (attr->egress)
1025 		return rte_flow_error_set(error, ENOTSUP,
1026 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1027 					  "drop action not supported for "
1028 					  "egress");
1029 	return 0;
1030 }
1031 
1032 /*
1033  * Validate the queue action.
1034  *
1035  * @param[in] action
1036  *   Pointer to the queue action.
1037  * @param[in] action_flags
1038  *   Bit-fields that holds the actions detected until now.
1039  * @param[in] dev
1040  *   Pointer to the Ethernet device structure.
1041  * @param[in] attr
1042  *   Attributes of flow that includes this action.
1043  * @param[out] error
1044  *   Pointer to error structure.
1045  *
1046  * @return
1047  *   0 on success, a negative errno value otherwise and rte_errno is set.
1048  */
1049 int
1050 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1051 				uint64_t action_flags,
1052 				struct rte_eth_dev *dev,
1053 				const struct rte_flow_attr *attr,
1054 				struct rte_flow_error *error)
1055 {
1056 	struct mlx5_priv *priv = dev->data->dev_private;
1057 	const struct rte_flow_action_queue *queue = action->conf;
1058 
1059 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1060 		return rte_flow_error_set(error, EINVAL,
1061 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1062 					  "can't have 2 fate actions in"
1063 					  " same flow");
1064 	if (!priv->rxqs_n)
1065 		return rte_flow_error_set(error, EINVAL,
1066 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1067 					  NULL, "No Rx queues configured");
1068 	if (queue->index >= priv->rxqs_n)
1069 		return rte_flow_error_set(error, EINVAL,
1070 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1071 					  &queue->index,
1072 					  "queue index out of range");
1073 	if (!(*priv->rxqs)[queue->index])
1074 		return rte_flow_error_set(error, EINVAL,
1075 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1076 					  &queue->index,
1077 					  "queue is not configured");
1078 	if (attr->egress)
1079 		return rte_flow_error_set(error, ENOTSUP,
1080 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1081 					  "queue action not supported for "
1082 					  "egress");
1083 	return 0;
1084 }
1085 
1086 /*
1087  * Validate the rss action.
1088  *
1089  * @param[in] action
1090  *   Pointer to the queue action.
1091  * @param[in] action_flags
1092  *   Bit-fields that holds the actions detected until now.
1093  * @param[in] dev
1094  *   Pointer to the Ethernet device structure.
1095  * @param[in] attr
1096  *   Attributes of flow that includes this action.
1097  * @param[in] item_flags
1098  *   Items that were detected.
1099  * @param[out] error
1100  *   Pointer to error structure.
1101  *
1102  * @return
1103  *   0 on success, a negative errno value otherwise and rte_errno is set.
1104  */
1105 int
1106 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1107 			      uint64_t action_flags,
1108 			      struct rte_eth_dev *dev,
1109 			      const struct rte_flow_attr *attr,
1110 			      uint64_t item_flags,
1111 			      struct rte_flow_error *error)
1112 {
1113 	struct mlx5_priv *priv = dev->data->dev_private;
1114 	const struct rte_flow_action_rss *rss = action->conf;
1115 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1116 	unsigned int i;
1117 
1118 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1119 		return rte_flow_error_set(error, EINVAL,
1120 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1121 					  "can't have 2 fate actions"
1122 					  " in same flow");
1123 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1124 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127 					  &rss->func,
1128 					  "RSS hash function not supported");
1129 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1130 	if (rss->level > 2)
1131 #else
1132 	if (rss->level > 1)
1133 #endif
1134 		return rte_flow_error_set(error, ENOTSUP,
1135 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136 					  &rss->level,
1137 					  "tunnel RSS is not supported");
1138 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1139 	if (rss->key_len == 0 && rss->key != NULL)
1140 		return rte_flow_error_set(error, ENOTSUP,
1141 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142 					  &rss->key_len,
1143 					  "RSS hash key length 0");
1144 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1145 		return rte_flow_error_set(error, ENOTSUP,
1146 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1147 					  &rss->key_len,
1148 					  "RSS hash key too small");
1149 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1150 		return rte_flow_error_set(error, ENOTSUP,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1152 					  &rss->key_len,
1153 					  "RSS hash key too large");
1154 	if (rss->queue_num > priv->config.ind_table_max_size)
1155 		return rte_flow_error_set(error, ENOTSUP,
1156 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1157 					  &rss->queue_num,
1158 					  "number of queues too large");
1159 	if (rss->types & MLX5_RSS_HF_MASK)
1160 		return rte_flow_error_set(error, ENOTSUP,
1161 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 					  &rss->types,
1163 					  "some RSS protocols are not"
1164 					  " supported");
1165 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1166 	    !(rss->types & ETH_RSS_IP))
1167 		return rte_flow_error_set(error, EINVAL,
1168 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1169 					  "L3 partial RSS requested but L3 RSS"
1170 					  " type not specified");
1171 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1172 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1173 		return rte_flow_error_set(error, EINVAL,
1174 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1175 					  "L4 partial RSS requested but L4 RSS"
1176 					  " type not specified");
1177 	if (!priv->rxqs_n)
1178 		return rte_flow_error_set(error, EINVAL,
1179 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1180 					  NULL, "No Rx queues configured");
1181 	if (!rss->queue_num)
1182 		return rte_flow_error_set(error, EINVAL,
1183 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1184 					  NULL, "No queues configured");
1185 	for (i = 0; i != rss->queue_num; ++i) {
1186 		if (rss->queue[i] >= priv->rxqs_n)
1187 			return rte_flow_error_set
1188 				(error, EINVAL,
1189 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1190 				 &rss->queue[i], "queue index out of range");
1191 		if (!(*priv->rxqs)[rss->queue[i]])
1192 			return rte_flow_error_set
1193 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1194 				 &rss->queue[i], "queue is not configured");
1195 	}
1196 	if (attr->egress)
1197 		return rte_flow_error_set(error, ENOTSUP,
1198 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1199 					  "rss action not supported for "
1200 					  "egress");
1201 	if (rss->level > 1 &&  !tunnel)
1202 		return rte_flow_error_set(error, EINVAL,
1203 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204 					  "inner RSS is not supported for "
1205 					  "non-tunnel flows");
1206 	return 0;
1207 }
1208 
1209 /*
1210  * Validate the count action.
1211  *
1212  * @param[in] dev
1213  *   Pointer to the Ethernet device structure.
1214  * @param[in] attr
1215  *   Attributes of flow that includes this action.
1216  * @param[out] error
1217  *   Pointer to error structure.
1218  *
1219  * @return
1220  *   0 on success, a negative errno value otherwise and rte_errno is set.
1221  */
1222 int
1223 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1224 				const struct rte_flow_attr *attr,
1225 				struct rte_flow_error *error)
1226 {
1227 	if (attr->egress)
1228 		return rte_flow_error_set(error, ENOTSUP,
1229 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230 					  "count action not supported for "
1231 					  "egress");
1232 	return 0;
1233 }
1234 
1235 /**
1236  * Verify the @p attributes will be correctly understood by the NIC and store
1237  * them in the @p flow if everything is correct.
1238  *
1239  * @param[in] dev
1240  *   Pointer to the Ethernet device structure.
1241  * @param[in] attributes
1242  *   Pointer to flow attributes
1243  * @param[out] error
1244  *   Pointer to error structure.
1245  *
1246  * @return
1247  *   0 on success, a negative errno value otherwise and rte_errno is set.
1248  */
1249 int
1250 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1251 			      const struct rte_flow_attr *attributes,
1252 			      struct rte_flow_error *error)
1253 {
1254 	struct mlx5_priv *priv = dev->data->dev_private;
1255 	uint32_t priority_max = priv->config.flow_prio - 1;
1256 
1257 	if (attributes->group)
1258 		return rte_flow_error_set(error, ENOTSUP,
1259 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1260 					  NULL, "groups is not supported");
1261 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1262 	    attributes->priority >= priority_max)
1263 		return rte_flow_error_set(error, ENOTSUP,
1264 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1265 					  NULL, "priority out of range");
1266 	if (attributes->egress)
1267 		return rte_flow_error_set(error, ENOTSUP,
1268 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1269 					  "egress is not supported");
1270 	if (attributes->transfer && !priv->config.dv_esw_en)
1271 		return rte_flow_error_set(error, ENOTSUP,
1272 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1273 					  NULL, "transfer is not supported");
1274 	if (!attributes->ingress)
1275 		return rte_flow_error_set(error, EINVAL,
1276 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1277 					  NULL,
1278 					  "ingress attribute is mandatory");
1279 	return 0;
1280 }
1281 
1282 /**
1283  * Validate ICMP6 item.
1284  *
1285  * @param[in] item
1286  *   Item specification.
1287  * @param[in] item_flags
1288  *   Bit-fields that holds the items detected until now.
1289  * @param[out] error
1290  *   Pointer to error structure.
1291  *
1292  * @return
1293  *   0 on success, a negative errno value otherwise and rte_errno is set.
1294  */
1295 int
1296 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1297 			       uint64_t item_flags,
1298 			       uint8_t target_protocol,
1299 			       struct rte_flow_error *error)
1300 {
1301 	const struct rte_flow_item_icmp6 *mask = item->mask;
1302 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1303 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1304 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1305 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1306 				      MLX5_FLOW_LAYER_OUTER_L4;
1307 	int ret;
1308 
1309 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1310 		return rte_flow_error_set(error, EINVAL,
1311 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1312 					  "protocol filtering not compatible"
1313 					  " with ICMP6 layer");
1314 	if (!(item_flags & l3m))
1315 		return rte_flow_error_set(error, EINVAL,
1316 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1317 					  "IPv6 is mandatory to filter on"
1318 					  " ICMP6");
1319 	if (item_flags & l4m)
1320 		return rte_flow_error_set(error, EINVAL,
1321 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1322 					  "multiple L4 layers not supported");
1323 	if (!mask)
1324 		mask = &rte_flow_item_icmp6_mask;
1325 	ret = mlx5_flow_item_acceptable
1326 		(item, (const uint8_t *)mask,
1327 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1328 		 sizeof(struct rte_flow_item_icmp6), error);
1329 	if (ret < 0)
1330 		return ret;
1331 	return 0;
1332 }
1333 
1334 /**
1335  * Validate ICMP item.
1336  *
1337  * @param[in] item
1338  *   Item specification.
1339  * @param[in] item_flags
1340  *   Bit-fields that holds the items detected until now.
1341  * @param[out] error
1342  *   Pointer to error structure.
1343  *
1344  * @return
1345  *   0 on success, a negative errno value otherwise and rte_errno is set.
1346  */
1347 int
1348 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1349 			     uint64_t item_flags,
1350 			     uint8_t target_protocol,
1351 			     struct rte_flow_error *error)
1352 {
1353 	const struct rte_flow_item_icmp *mask = item->mask;
1354 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1355 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1356 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1357 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1358 				      MLX5_FLOW_LAYER_OUTER_L4;
1359 	int ret;
1360 
1361 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1362 		return rte_flow_error_set(error, EINVAL,
1363 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1364 					  "protocol filtering not compatible"
1365 					  " with ICMP layer");
1366 	if (!(item_flags & l3m))
1367 		return rte_flow_error_set(error, EINVAL,
1368 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1369 					  "IPv4 is mandatory to filter"
1370 					  " on ICMP");
1371 	if (item_flags & l4m)
1372 		return rte_flow_error_set(error, EINVAL,
1373 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1374 					  "multiple L4 layers not supported");
1375 	if (!mask)
1376 		mask = &rte_flow_item_icmp_mask;
1377 	ret = mlx5_flow_item_acceptable
1378 		(item, (const uint8_t *)mask,
1379 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1380 		 sizeof(struct rte_flow_item_icmp), error);
1381 	if (ret < 0)
1382 		return ret;
1383 	return 0;
1384 }
1385 
1386 /**
1387  * Validate Ethernet item.
1388  *
1389  * @param[in] item
1390  *   Item specification.
1391  * @param[in] item_flags
1392  *   Bit-fields that holds the items detected until now.
1393  * @param[out] error
1394  *   Pointer to error structure.
1395  *
1396  * @return
1397  *   0 on success, a negative errno value otherwise and rte_errno is set.
1398  */
1399 int
1400 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1401 			    uint64_t item_flags,
1402 			    struct rte_flow_error *error)
1403 {
1404 	const struct rte_flow_item_eth *mask = item->mask;
1405 	const struct rte_flow_item_eth nic_mask = {
1406 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1407 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1408 		.type = RTE_BE16(0xffff),
1409 	};
1410 	int ret;
1411 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1412 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1413 				       MLX5_FLOW_LAYER_OUTER_L2;
1414 
1415 	if (item_flags & ethm)
1416 		return rte_flow_error_set(error, ENOTSUP,
1417 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1418 					  "multiple L2 layers not supported");
1419 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1420 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1421 		return rte_flow_error_set(error, EINVAL,
1422 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1423 					  "L2 layer should not follow "
1424 					  "L3 layers");
1425 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1426 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1427 		return rte_flow_error_set(error, EINVAL,
1428 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1429 					  "L2 layer should not follow VLAN");
1430 	if (!mask)
1431 		mask = &rte_flow_item_eth_mask;
1432 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1433 					(const uint8_t *)&nic_mask,
1434 					sizeof(struct rte_flow_item_eth),
1435 					error);
1436 	return ret;
1437 }
1438 
1439 /**
1440  * Validate VLAN item.
1441  *
1442  * @param[in] item
1443  *   Item specification.
1444  * @param[in] item_flags
1445  *   Bit-fields that holds the items detected until now.
1446  * @param[in] dev
1447  *   Ethernet device flow is being created on.
1448  * @param[out] error
1449  *   Pointer to error structure.
1450  *
1451  * @return
1452  *   0 on success, a negative errno value otherwise and rte_errno is set.
1453  */
1454 int
1455 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1456 			     uint64_t item_flags,
1457 			     struct rte_eth_dev *dev,
1458 			     struct rte_flow_error *error)
1459 {
1460 	const struct rte_flow_item_vlan *spec = item->spec;
1461 	const struct rte_flow_item_vlan *mask = item->mask;
1462 	const struct rte_flow_item_vlan nic_mask = {
1463 		.tci = RTE_BE16(UINT16_MAX),
1464 		.inner_type = RTE_BE16(UINT16_MAX),
1465 	};
1466 	uint16_t vlan_tag = 0;
1467 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1468 	int ret;
1469 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1470 					MLX5_FLOW_LAYER_INNER_L4) :
1471 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1472 					MLX5_FLOW_LAYER_OUTER_L4);
1473 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1474 					MLX5_FLOW_LAYER_OUTER_VLAN;
1475 
1476 	if (item_flags & vlanm)
1477 		return rte_flow_error_set(error, EINVAL,
1478 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1479 					  "multiple VLAN layers not supported");
1480 	else if ((item_flags & l34m) != 0)
1481 		return rte_flow_error_set(error, EINVAL,
1482 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 					  "VLAN cannot follow L3/L4 layer");
1484 	if (!mask)
1485 		mask = &rte_flow_item_vlan_mask;
1486 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1487 					(const uint8_t *)&nic_mask,
1488 					sizeof(struct rte_flow_item_vlan),
1489 					error);
1490 	if (ret)
1491 		return ret;
1492 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1493 		struct mlx5_priv *priv = dev->data->dev_private;
1494 
1495 		if (priv->vmwa_context) {
1496 			/*
1497 			 * Non-NULL context means we have a virtual machine
1498 			 * and SR-IOV enabled, we have to create VLAN interface
1499 			 * to make hypervisor to setup E-Switch vport
1500 			 * context correctly. We avoid creating the multiple
1501 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1502 			 */
1503 			return rte_flow_error_set(error, EINVAL,
1504 						  RTE_FLOW_ERROR_TYPE_ITEM,
1505 						  item,
1506 						  "VLAN tag mask is not"
1507 						  " supported in virtual"
1508 						  " environment");
1509 		}
1510 	}
1511 	if (spec) {
1512 		vlan_tag = spec->tci;
1513 		vlan_tag &= mask->tci;
1514 	}
1515 	/*
1516 	 * From verbs perspective an empty VLAN is equivalent
1517 	 * to a packet without VLAN layer.
1518 	 */
1519 	if (!vlan_tag)
1520 		return rte_flow_error_set(error, EINVAL,
1521 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1522 					  item->spec,
1523 					  "VLAN cannot be empty");
1524 	return 0;
1525 }
1526 
1527 /**
1528  * Validate IPV4 item.
1529  *
1530  * @param[in] item
1531  *   Item specification.
1532  * @param[in] item_flags
1533  *   Bit-fields that holds the items detected until now.
1534  * @param[in] acc_mask
1535  *   Acceptable mask, if NULL default internal default mask
1536  *   will be used to check whether item fields are supported.
1537  * @param[out] error
1538  *   Pointer to error structure.
1539  *
1540  * @return
1541  *   0 on success, a negative errno value otherwise and rte_errno is set.
1542  */
1543 int
1544 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1545 			     uint64_t item_flags,
1546 			     uint64_t last_item,
1547 			     uint16_t ether_type,
1548 			     const struct rte_flow_item_ipv4 *acc_mask,
1549 			     struct rte_flow_error *error)
1550 {
1551 	const struct rte_flow_item_ipv4 *mask = item->mask;
1552 	const struct rte_flow_item_ipv4 *spec = item->spec;
1553 	const struct rte_flow_item_ipv4 nic_mask = {
1554 		.hdr = {
1555 			.src_addr = RTE_BE32(0xffffffff),
1556 			.dst_addr = RTE_BE32(0xffffffff),
1557 			.type_of_service = 0xff,
1558 			.next_proto_id = 0xff,
1559 		},
1560 	};
1561 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1562 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1563 				      MLX5_FLOW_LAYER_OUTER_L3;
1564 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1565 				      MLX5_FLOW_LAYER_OUTER_L4;
1566 	int ret;
1567 	uint8_t next_proto = 0xFF;
1568 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1569 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1570 				  MLX5_FLOW_LAYER_INNER_VLAN);
1571 
1572 	if ((last_item & l2_vlan) && ether_type &&
1573 	    ether_type != RTE_ETHER_TYPE_IPV4)
1574 		return rte_flow_error_set(error, EINVAL,
1575 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 					  "IPv4 cannot follow L2/VLAN layer "
1577 					  "which ether type is not IPv4");
1578 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1579 		if (mask && spec)
1580 			next_proto = mask->hdr.next_proto_id &
1581 				     spec->hdr.next_proto_id;
1582 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1583 			return rte_flow_error_set(error, EINVAL,
1584 						  RTE_FLOW_ERROR_TYPE_ITEM,
1585 						  item,
1586 						  "multiple tunnel "
1587 						  "not supported");
1588 	}
1589 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1592 					  "wrong tunnel type - IPv6 specified "
1593 					  "but IPv4 item provided");
1594 	if (item_flags & l3m)
1595 		return rte_flow_error_set(error, ENOTSUP,
1596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1597 					  "multiple L3 layers not supported");
1598 	else if (item_flags & l4m)
1599 		return rte_flow_error_set(error, EINVAL,
1600 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1601 					  "L3 cannot follow an L4 layer.");
1602 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1603 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1604 		return rte_flow_error_set(error, EINVAL,
1605 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1606 					  "L3 cannot follow an NVGRE layer.");
1607 	if (!mask)
1608 		mask = &rte_flow_item_ipv4_mask;
1609 	else if (mask->hdr.next_proto_id != 0 &&
1610 		 mask->hdr.next_proto_id != 0xff)
1611 		return rte_flow_error_set(error, EINVAL,
1612 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1613 					  "partial mask is not supported"
1614 					  " for protocol");
1615 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1616 					acc_mask ? (const uint8_t *)acc_mask
1617 						 : (const uint8_t *)&nic_mask,
1618 					sizeof(struct rte_flow_item_ipv4),
1619 					error);
1620 	if (ret < 0)
1621 		return ret;
1622 	return 0;
1623 }
1624 
1625 /**
1626  * Validate IPV6 item.
1627  *
1628  * @param[in] item
1629  *   Item specification.
1630  * @param[in] item_flags
1631  *   Bit-fields that holds the items detected until now.
1632  * @param[in] acc_mask
1633  *   Acceptable mask, if NULL default internal default mask
1634  *   will be used to check whether item fields are supported.
1635  * @param[out] error
1636  *   Pointer to error structure.
1637  *
1638  * @return
1639  *   0 on success, a negative errno value otherwise and rte_errno is set.
1640  */
1641 int
1642 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1643 			     uint64_t item_flags,
1644 			     uint64_t last_item,
1645 			     uint16_t ether_type,
1646 			     const struct rte_flow_item_ipv6 *acc_mask,
1647 			     struct rte_flow_error *error)
1648 {
1649 	const struct rte_flow_item_ipv6 *mask = item->mask;
1650 	const struct rte_flow_item_ipv6 *spec = item->spec;
1651 	const struct rte_flow_item_ipv6 nic_mask = {
1652 		.hdr = {
1653 			.src_addr =
1654 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1655 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1656 			.dst_addr =
1657 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1658 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1659 			.vtc_flow = RTE_BE32(0xffffffff),
1660 			.proto = 0xff,
1661 		},
1662 	};
1663 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1664 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1665 				      MLX5_FLOW_LAYER_OUTER_L3;
1666 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1667 				      MLX5_FLOW_LAYER_OUTER_L4;
1668 	int ret;
1669 	uint8_t next_proto = 0xFF;
1670 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1671 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1672 				  MLX5_FLOW_LAYER_INNER_VLAN);
1673 
1674 	if ((last_item & l2_vlan) && ether_type &&
1675 	    ether_type != RTE_ETHER_TYPE_IPV6)
1676 		return rte_flow_error_set(error, EINVAL,
1677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1678 					  "IPv6 cannot follow L2/VLAN layer "
1679 					  "which ether type is not IPv6");
1680 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1681 		if (mask && spec)
1682 			next_proto = mask->hdr.proto & spec->hdr.proto;
1683 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1684 			return rte_flow_error_set(error, EINVAL,
1685 						  RTE_FLOW_ERROR_TYPE_ITEM,
1686 						  item,
1687 						  "multiple tunnel "
1688 						  "not supported");
1689 	}
1690 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1691 		return rte_flow_error_set(error, EINVAL,
1692 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1693 					  "wrong tunnel type - IPv4 specified "
1694 					  "but IPv6 item provided");
1695 	if (item_flags & l3m)
1696 		return rte_flow_error_set(error, ENOTSUP,
1697 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1698 					  "multiple L3 layers not supported");
1699 	else if (item_flags & l4m)
1700 		return rte_flow_error_set(error, EINVAL,
1701 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1702 					  "L3 cannot follow an L4 layer.");
1703 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1704 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1705 		return rte_flow_error_set(error, EINVAL,
1706 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 					  "L3 cannot follow an NVGRE layer.");
1708 	if (!mask)
1709 		mask = &rte_flow_item_ipv6_mask;
1710 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1711 					acc_mask ? (const uint8_t *)acc_mask
1712 						 : (const uint8_t *)&nic_mask,
1713 					sizeof(struct rte_flow_item_ipv6),
1714 					error);
1715 	if (ret < 0)
1716 		return ret;
1717 	return 0;
1718 }
1719 
1720 /**
1721  * Validate UDP item.
1722  *
1723  * @param[in] item
1724  *   Item specification.
1725  * @param[in] item_flags
1726  *   Bit-fields that holds the items detected until now.
1727  * @param[in] target_protocol
1728  *   The next protocol in the previous item.
1729  * @param[in] flow_mask
1730  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1731  * @param[out] error
1732  *   Pointer to error structure.
1733  *
1734  * @return
1735  *   0 on success, a negative errno value otherwise and rte_errno is set.
1736  */
1737 int
1738 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1739 			    uint64_t item_flags,
1740 			    uint8_t target_protocol,
1741 			    struct rte_flow_error *error)
1742 {
1743 	const struct rte_flow_item_udp *mask = item->mask;
1744 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1745 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1746 				      MLX5_FLOW_LAYER_OUTER_L3;
1747 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1748 				      MLX5_FLOW_LAYER_OUTER_L4;
1749 	int ret;
1750 
1751 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1752 		return rte_flow_error_set(error, EINVAL,
1753 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1754 					  "protocol filtering not compatible"
1755 					  " with UDP layer");
1756 	if (!(item_flags & l3m))
1757 		return rte_flow_error_set(error, EINVAL,
1758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1759 					  "L3 is mandatory to filter on L4");
1760 	if (item_flags & l4m)
1761 		return rte_flow_error_set(error, EINVAL,
1762 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1763 					  "multiple L4 layers not supported");
1764 	if (!mask)
1765 		mask = &rte_flow_item_udp_mask;
1766 	ret = mlx5_flow_item_acceptable
1767 		(item, (const uint8_t *)mask,
1768 		 (const uint8_t *)&rte_flow_item_udp_mask,
1769 		 sizeof(struct rte_flow_item_udp), error);
1770 	if (ret < 0)
1771 		return ret;
1772 	return 0;
1773 }
1774 
1775 /**
1776  * Validate TCP item.
1777  *
1778  * @param[in] item
1779  *   Item specification.
1780  * @param[in] item_flags
1781  *   Bit-fields that holds the items detected until now.
1782  * @param[in] target_protocol
1783  *   The next protocol in the previous item.
1784  * @param[out] error
1785  *   Pointer to error structure.
1786  *
1787  * @return
1788  *   0 on success, a negative errno value otherwise and rte_errno is set.
1789  */
1790 int
1791 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1792 			    uint64_t item_flags,
1793 			    uint8_t target_protocol,
1794 			    const struct rte_flow_item_tcp *flow_mask,
1795 			    struct rte_flow_error *error)
1796 {
1797 	const struct rte_flow_item_tcp *mask = item->mask;
1798 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1799 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1800 				      MLX5_FLOW_LAYER_OUTER_L3;
1801 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1802 				      MLX5_FLOW_LAYER_OUTER_L4;
1803 	int ret;
1804 
1805 	MLX5_ASSERT(flow_mask);
1806 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1807 		return rte_flow_error_set(error, EINVAL,
1808 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1809 					  "protocol filtering not compatible"
1810 					  " with TCP layer");
1811 	if (!(item_flags & l3m))
1812 		return rte_flow_error_set(error, EINVAL,
1813 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1814 					  "L3 is mandatory to filter on L4");
1815 	if (item_flags & l4m)
1816 		return rte_flow_error_set(error, EINVAL,
1817 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1818 					  "multiple L4 layers not supported");
1819 	if (!mask)
1820 		mask = &rte_flow_item_tcp_mask;
1821 	ret = mlx5_flow_item_acceptable
1822 		(item, (const uint8_t *)mask,
1823 		 (const uint8_t *)flow_mask,
1824 		 sizeof(struct rte_flow_item_tcp), error);
1825 	if (ret < 0)
1826 		return ret;
1827 	return 0;
1828 }
1829 
1830 /**
1831  * Validate VXLAN item.
1832  *
1833  * @param[in] item
1834  *   Item specification.
1835  * @param[in] item_flags
1836  *   Bit-fields that holds the items detected until now.
1837  * @param[in] target_protocol
1838  *   The next protocol in the previous item.
1839  * @param[out] error
1840  *   Pointer to error structure.
1841  *
1842  * @return
1843  *   0 on success, a negative errno value otherwise and rte_errno is set.
1844  */
1845 int
1846 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1847 			      uint64_t item_flags,
1848 			      struct rte_flow_error *error)
1849 {
1850 	const struct rte_flow_item_vxlan *spec = item->spec;
1851 	const struct rte_flow_item_vxlan *mask = item->mask;
1852 	int ret;
1853 	union vni {
1854 		uint32_t vlan_id;
1855 		uint8_t vni[4];
1856 	} id = { .vlan_id = 0, };
1857 
1858 
1859 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1860 		return rte_flow_error_set(error, ENOTSUP,
1861 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1862 					  "multiple tunnel layers not"
1863 					  " supported");
1864 	/*
1865 	 * Verify only UDPv4 is present as defined in
1866 	 * https://tools.ietf.org/html/rfc7348
1867 	 */
1868 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1869 		return rte_flow_error_set(error, EINVAL,
1870 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1871 					  "no outer UDP layer found");
1872 	if (!mask)
1873 		mask = &rte_flow_item_vxlan_mask;
1874 	ret = mlx5_flow_item_acceptable
1875 		(item, (const uint8_t *)mask,
1876 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1877 		 sizeof(struct rte_flow_item_vxlan),
1878 		 error);
1879 	if (ret < 0)
1880 		return ret;
1881 	if (spec) {
1882 		memcpy(&id.vni[1], spec->vni, 3);
1883 		memcpy(&id.vni[1], mask->vni, 3);
1884 	}
1885 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1886 		return rte_flow_error_set(error, ENOTSUP,
1887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 					  "VXLAN tunnel must be fully defined");
1889 	return 0;
1890 }
1891 
1892 /**
1893  * Validate VXLAN_GPE item.
1894  *
1895  * @param[in] item
1896  *   Item specification.
1897  * @param[in] item_flags
1898  *   Bit-fields that holds the items detected until now.
1899  * @param[in] priv
1900  *   Pointer to the private data structure.
1901  * @param[in] target_protocol
1902  *   The next protocol in the previous item.
1903  * @param[out] error
1904  *   Pointer to error structure.
1905  *
1906  * @return
1907  *   0 on success, a negative errno value otherwise and rte_errno is set.
1908  */
1909 int
1910 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1911 				  uint64_t item_flags,
1912 				  struct rte_eth_dev *dev,
1913 				  struct rte_flow_error *error)
1914 {
1915 	struct mlx5_priv *priv = dev->data->dev_private;
1916 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1917 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1918 	int ret;
1919 	union vni {
1920 		uint32_t vlan_id;
1921 		uint8_t vni[4];
1922 	} id = { .vlan_id = 0, };
1923 
1924 	if (!priv->config.l3_vxlan_en)
1925 		return rte_flow_error_set(error, ENOTSUP,
1926 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 					  "L3 VXLAN is not enabled by device"
1928 					  " parameter and/or not configured in"
1929 					  " firmware");
1930 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1931 		return rte_flow_error_set(error, ENOTSUP,
1932 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 					  "multiple tunnel layers not"
1934 					  " supported");
1935 	/*
1936 	 * Verify only UDPv4 is present as defined in
1937 	 * https://tools.ietf.org/html/rfc7348
1938 	 */
1939 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1940 		return rte_flow_error_set(error, EINVAL,
1941 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1942 					  "no outer UDP layer found");
1943 	if (!mask)
1944 		mask = &rte_flow_item_vxlan_gpe_mask;
1945 	ret = mlx5_flow_item_acceptable
1946 		(item, (const uint8_t *)mask,
1947 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1948 		 sizeof(struct rte_flow_item_vxlan_gpe),
1949 		 error);
1950 	if (ret < 0)
1951 		return ret;
1952 	if (spec) {
1953 		if (spec->protocol)
1954 			return rte_flow_error_set(error, ENOTSUP,
1955 						  RTE_FLOW_ERROR_TYPE_ITEM,
1956 						  item,
1957 						  "VxLAN-GPE protocol"
1958 						  " not supported");
1959 		memcpy(&id.vni[1], spec->vni, 3);
1960 		memcpy(&id.vni[1], mask->vni, 3);
1961 	}
1962 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1963 		return rte_flow_error_set(error, ENOTSUP,
1964 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1965 					  "VXLAN-GPE tunnel must be fully"
1966 					  " defined");
1967 	return 0;
1968 }
1969 /**
1970  * Validate GRE Key item.
1971  *
1972  * @param[in] item
1973  *   Item specification.
1974  * @param[in] item_flags
1975  *   Bit flags to mark detected items.
1976  * @param[in] gre_item
1977  *   Pointer to gre_item
1978  * @param[out] error
1979  *   Pointer to error structure.
1980  *
1981  * @return
1982  *   0 on success, a negative errno value otherwise and rte_errno is set.
1983  */
1984 int
1985 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1986 				uint64_t item_flags,
1987 				const struct rte_flow_item *gre_item,
1988 				struct rte_flow_error *error)
1989 {
1990 	const rte_be32_t *mask = item->mask;
1991 	int ret = 0;
1992 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1993 	const struct rte_flow_item_gre *gre_spec;
1994 	const struct rte_flow_item_gre *gre_mask;
1995 
1996 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1997 		return rte_flow_error_set(error, ENOTSUP,
1998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 					  "Multiple GRE key not support");
2000 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2001 		return rte_flow_error_set(error, ENOTSUP,
2002 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2003 					  "No preceding GRE header");
2004 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2005 		return rte_flow_error_set(error, ENOTSUP,
2006 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 					  "GRE key following a wrong item");
2008 	gre_mask = gre_item->mask;
2009 	if (!gre_mask)
2010 		gre_mask = &rte_flow_item_gre_mask;
2011 	gre_spec = gre_item->spec;
2012 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2013 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2014 		return rte_flow_error_set(error, EINVAL,
2015 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2016 					  "Key bit must be on");
2017 
2018 	if (!mask)
2019 		mask = &gre_key_default_mask;
2020 	ret = mlx5_flow_item_acceptable
2021 		(item, (const uint8_t *)mask,
2022 		 (const uint8_t *)&gre_key_default_mask,
2023 		 sizeof(rte_be32_t), error);
2024 	return ret;
2025 }
2026 
2027 /**
2028  * Validate GRE item.
2029  *
2030  * @param[in] item
2031  *   Item specification.
2032  * @param[in] item_flags
2033  *   Bit flags to mark detected items.
2034  * @param[in] target_protocol
2035  *   The next protocol in the previous item.
2036  * @param[out] error
2037  *   Pointer to error structure.
2038  *
2039  * @return
2040  *   0 on success, a negative errno value otherwise and rte_errno is set.
2041  */
2042 int
2043 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2044 			    uint64_t item_flags,
2045 			    uint8_t target_protocol,
2046 			    struct rte_flow_error *error)
2047 {
2048 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2049 	const struct rte_flow_item_gre *mask = item->mask;
2050 	int ret;
2051 	const struct rte_flow_item_gre nic_mask = {
2052 		.c_rsvd0_ver = RTE_BE16(0xB000),
2053 		.protocol = RTE_BE16(UINT16_MAX),
2054 	};
2055 
2056 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2057 		return rte_flow_error_set(error, EINVAL,
2058 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2059 					  "protocol filtering not compatible"
2060 					  " with this GRE layer");
2061 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2062 		return rte_flow_error_set(error, ENOTSUP,
2063 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2064 					  "multiple tunnel layers not"
2065 					  " supported");
2066 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2067 		return rte_flow_error_set(error, ENOTSUP,
2068 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2069 					  "L3 Layer is missing");
2070 	if (!mask)
2071 		mask = &rte_flow_item_gre_mask;
2072 	ret = mlx5_flow_item_acceptable
2073 		(item, (const uint8_t *)mask,
2074 		 (const uint8_t *)&nic_mask,
2075 		 sizeof(struct rte_flow_item_gre), error);
2076 	if (ret < 0)
2077 		return ret;
2078 #ifndef HAVE_MLX5DV_DR
2079 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2080 	if (spec && (spec->protocol & mask->protocol))
2081 		return rte_flow_error_set(error, ENOTSUP,
2082 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2083 					  "without MPLS support the"
2084 					  " specification cannot be used for"
2085 					  " filtering");
2086 #endif
2087 #endif
2088 	return 0;
2089 }
2090 
2091 /**
2092  * Validate Geneve item.
2093  *
2094  * @param[in] item
2095  *   Item specification.
2096  * @param[in] itemFlags
2097  *   Bit-fields that holds the items detected until now.
2098  * @param[in] enPriv
2099  *   Pointer to the private data structure.
2100  * @param[out] error
2101  *   Pointer to error structure.
2102  *
2103  * @return
2104  *   0 on success, a negative errno value otherwise and rte_errno is set.
2105  */
2106 
2107 int
2108 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2109 			       uint64_t item_flags,
2110 			       struct rte_eth_dev *dev,
2111 			       struct rte_flow_error *error)
2112 {
2113 	struct mlx5_priv *priv = dev->data->dev_private;
2114 	const struct rte_flow_item_geneve *spec = item->spec;
2115 	const struct rte_flow_item_geneve *mask = item->mask;
2116 	int ret;
2117 	uint16_t gbhdr;
2118 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2119 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2120 	const struct rte_flow_item_geneve nic_mask = {
2121 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2122 		.vni = "\xff\xff\xff",
2123 		.protocol = RTE_BE16(UINT16_MAX),
2124 	};
2125 
2126 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2127 		return rte_flow_error_set(error, ENOTSUP,
2128 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2129 					  "L3 Geneve is not enabled by device"
2130 					  " parameter and/or not configured in"
2131 					  " firmware");
2132 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2133 		return rte_flow_error_set(error, ENOTSUP,
2134 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2135 					  "multiple tunnel layers not"
2136 					  " supported");
2137 	/*
2138 	 * Verify only UDPv4 is present as defined in
2139 	 * https://tools.ietf.org/html/rfc7348
2140 	 */
2141 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2142 		return rte_flow_error_set(error, EINVAL,
2143 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2144 					  "no outer UDP layer found");
2145 	if (!mask)
2146 		mask = &rte_flow_item_geneve_mask;
2147 	ret = mlx5_flow_item_acceptable
2148 				  (item, (const uint8_t *)mask,
2149 				   (const uint8_t *)&nic_mask,
2150 				   sizeof(struct rte_flow_item_geneve), error);
2151 	if (ret)
2152 		return ret;
2153 	if (spec) {
2154 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2155 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2156 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2157 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2158 			return rte_flow_error_set(error, ENOTSUP,
2159 						  RTE_FLOW_ERROR_TYPE_ITEM,
2160 						  item,
2161 						  "Geneve protocol unsupported"
2162 						  " fields are being used");
2163 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2164 			return rte_flow_error_set
2165 					(error, ENOTSUP,
2166 					 RTE_FLOW_ERROR_TYPE_ITEM,
2167 					 item,
2168 					 "Unsupported Geneve options length");
2169 	}
2170 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2171 		return rte_flow_error_set
2172 				    (error, ENOTSUP,
2173 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2174 				     "Geneve tunnel must be fully defined");
2175 	return 0;
2176 }
2177 
2178 /**
2179  * Validate MPLS item.
2180  *
2181  * @param[in] dev
2182  *   Pointer to the rte_eth_dev structure.
2183  * @param[in] item
2184  *   Item specification.
2185  * @param[in] item_flags
2186  *   Bit-fields that holds the items detected until now.
2187  * @param[in] prev_layer
2188  *   The protocol layer indicated in previous item.
2189  * @param[out] error
2190  *   Pointer to error structure.
2191  *
2192  * @return
2193  *   0 on success, a negative errno value otherwise and rte_errno is set.
2194  */
2195 int
2196 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2197 			     const struct rte_flow_item *item __rte_unused,
2198 			     uint64_t item_flags __rte_unused,
2199 			     uint64_t prev_layer __rte_unused,
2200 			     struct rte_flow_error *error)
2201 {
2202 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2203 	const struct rte_flow_item_mpls *mask = item->mask;
2204 	struct mlx5_priv *priv = dev->data->dev_private;
2205 	int ret;
2206 
2207 	if (!priv->config.mpls_en)
2208 		return rte_flow_error_set(error, ENOTSUP,
2209 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2210 					  "MPLS not supported or"
2211 					  " disabled in firmware"
2212 					  " configuration.");
2213 	/* MPLS over IP, UDP, GRE is allowed */
2214 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2215 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2216 			    MLX5_FLOW_LAYER_GRE)))
2217 		return rte_flow_error_set(error, EINVAL,
2218 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 					  "protocol filtering not compatible"
2220 					  " with MPLS layer");
2221 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2222 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2223 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2224 		return rte_flow_error_set(error, ENOTSUP,
2225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 					  "multiple tunnel layers not"
2227 					  " supported");
2228 	if (!mask)
2229 		mask = &rte_flow_item_mpls_mask;
2230 	ret = mlx5_flow_item_acceptable
2231 		(item, (const uint8_t *)mask,
2232 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2233 		 sizeof(struct rte_flow_item_mpls), error);
2234 	if (ret < 0)
2235 		return ret;
2236 	return 0;
2237 #endif
2238 	return rte_flow_error_set(error, ENOTSUP,
2239 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 				  "MPLS is not supported by Verbs, please"
2241 				  " update.");
2242 }
2243 
2244 /**
2245  * Validate NVGRE item.
2246  *
2247  * @param[in] item
2248  *   Item specification.
2249  * @param[in] item_flags
2250  *   Bit flags to mark detected items.
2251  * @param[in] target_protocol
2252  *   The next protocol in the previous item.
2253  * @param[out] error
2254  *   Pointer to error structure.
2255  *
2256  * @return
2257  *   0 on success, a negative errno value otherwise and rte_errno is set.
2258  */
2259 int
2260 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2261 			      uint64_t item_flags,
2262 			      uint8_t target_protocol,
2263 			      struct rte_flow_error *error)
2264 {
2265 	const struct rte_flow_item_nvgre *mask = item->mask;
2266 	int ret;
2267 
2268 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2269 		return rte_flow_error_set(error, EINVAL,
2270 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 					  "protocol filtering not compatible"
2272 					  " with this GRE layer");
2273 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2274 		return rte_flow_error_set(error, ENOTSUP,
2275 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 					  "multiple tunnel layers not"
2277 					  " supported");
2278 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2279 		return rte_flow_error_set(error, ENOTSUP,
2280 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2281 					  "L3 Layer is missing");
2282 	if (!mask)
2283 		mask = &rte_flow_item_nvgre_mask;
2284 	ret = mlx5_flow_item_acceptable
2285 		(item, (const uint8_t *)mask,
2286 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2287 		 sizeof(struct rte_flow_item_nvgre), error);
2288 	if (ret < 0)
2289 		return ret;
2290 	return 0;
2291 }
2292 
2293 /* Allocate unique ID for the split Q/RSS subflows. */
2294 static uint32_t
2295 flow_qrss_get_id(struct rte_eth_dev *dev)
2296 {
2297 	struct mlx5_priv *priv = dev->data->dev_private;
2298 	uint32_t qrss_id, ret;
2299 
2300 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2301 	if (ret)
2302 		return 0;
2303 	MLX5_ASSERT(qrss_id);
2304 	return qrss_id;
2305 }
2306 
2307 /* Free unique ID for the split Q/RSS subflows. */
2308 static void
2309 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2310 {
2311 	struct mlx5_priv *priv = dev->data->dev_private;
2312 
2313 	if (qrss_id)
2314 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2315 }
2316 
2317 /**
2318  * Release resource related QUEUE/RSS action split.
2319  *
2320  * @param dev
2321  *   Pointer to Ethernet device.
2322  * @param flow
2323  *   Flow to release id's from.
2324  */
2325 static void
2326 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2327 			     struct rte_flow *flow)
2328 {
2329 	struct mlx5_priv *priv = dev->data->dev_private;
2330 	uint32_t handle_idx;
2331 	struct mlx5_flow_handle *dev_handle;
2332 
2333 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2334 		       handle_idx, dev_handle, next)
2335 		if (dev_handle->split_flow_id)
2336 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2337 }
2338 
2339 static int
2340 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2341 		   const struct rte_flow_attr *attr __rte_unused,
2342 		   const struct rte_flow_item items[] __rte_unused,
2343 		   const struct rte_flow_action actions[] __rte_unused,
2344 		   bool external __rte_unused,
2345 		   int hairpin __rte_unused,
2346 		   struct rte_flow_error *error)
2347 {
2348 	return rte_flow_error_set(error, ENOTSUP,
2349 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2350 }
2351 
2352 static struct mlx5_flow *
2353 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2354 		  const struct rte_flow_attr *attr __rte_unused,
2355 		  const struct rte_flow_item items[] __rte_unused,
2356 		  const struct rte_flow_action actions[] __rte_unused,
2357 		  struct rte_flow_error *error)
2358 {
2359 	rte_flow_error_set(error, ENOTSUP,
2360 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2361 	return NULL;
2362 }
2363 
2364 static int
2365 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2366 		    struct mlx5_flow *dev_flow __rte_unused,
2367 		    const struct rte_flow_attr *attr __rte_unused,
2368 		    const struct rte_flow_item items[] __rte_unused,
2369 		    const struct rte_flow_action actions[] __rte_unused,
2370 		    struct rte_flow_error *error)
2371 {
2372 	return rte_flow_error_set(error, ENOTSUP,
2373 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2374 }
2375 
2376 static int
2377 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2378 		struct rte_flow *flow __rte_unused,
2379 		struct rte_flow_error *error)
2380 {
2381 	return rte_flow_error_set(error, ENOTSUP,
2382 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2383 }
2384 
2385 static void
2386 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2387 		 struct rte_flow *flow __rte_unused)
2388 {
2389 }
2390 
2391 static void
2392 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2393 		  struct rte_flow *flow __rte_unused)
2394 {
2395 }
2396 
2397 static int
2398 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2399 		struct rte_flow *flow __rte_unused,
2400 		const struct rte_flow_action *actions __rte_unused,
2401 		void *data __rte_unused,
2402 		struct rte_flow_error *error)
2403 {
2404 	return rte_flow_error_set(error, ENOTSUP,
2405 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2406 }
2407 
2408 /* Void driver to protect from null pointer reference. */
2409 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2410 	.validate = flow_null_validate,
2411 	.prepare = flow_null_prepare,
2412 	.translate = flow_null_translate,
2413 	.apply = flow_null_apply,
2414 	.remove = flow_null_remove,
2415 	.destroy = flow_null_destroy,
2416 	.query = flow_null_query,
2417 };
2418 
2419 /**
2420  * Select flow driver type according to flow attributes and device
2421  * configuration.
2422  *
2423  * @param[in] dev
2424  *   Pointer to the dev structure.
2425  * @param[in] attr
2426  *   Pointer to the flow attributes.
2427  *
2428  * @return
2429  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2430  */
2431 static enum mlx5_flow_drv_type
2432 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2433 {
2434 	struct mlx5_priv *priv = dev->data->dev_private;
2435 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2436 
2437 	if (attr->transfer && priv->config.dv_esw_en)
2438 		type = MLX5_FLOW_TYPE_DV;
2439 	if (!attr->transfer)
2440 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2441 						 MLX5_FLOW_TYPE_VERBS;
2442 	return type;
2443 }
2444 
2445 #define flow_get_drv_ops(type) flow_drv_ops[type]
2446 
2447 /**
2448  * Flow driver validation API. This abstracts calling driver specific functions.
2449  * The type of flow driver is determined according to flow attributes.
2450  *
2451  * @param[in] dev
2452  *   Pointer to the dev structure.
2453  * @param[in] attr
2454  *   Pointer to the flow attributes.
2455  * @param[in] items
2456  *   Pointer to the list of items.
2457  * @param[in] actions
2458  *   Pointer to the list of actions.
2459  * @param[in] external
2460  *   This flow rule is created by request external to PMD.
2461  * @param[in] hairpin
2462  *   Number of hairpin TX actions, 0 means classic flow.
2463  * @param[out] error
2464  *   Pointer to the error structure.
2465  *
2466  * @return
2467  *   0 on success, a negative errno value otherwise and rte_errno is set.
2468  */
2469 static inline int
2470 flow_drv_validate(struct rte_eth_dev *dev,
2471 		  const struct rte_flow_attr *attr,
2472 		  const struct rte_flow_item items[],
2473 		  const struct rte_flow_action actions[],
2474 		  bool external, int hairpin, struct rte_flow_error *error)
2475 {
2476 	const struct mlx5_flow_driver_ops *fops;
2477 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2478 
2479 	fops = flow_get_drv_ops(type);
2480 	return fops->validate(dev, attr, items, actions, external,
2481 			      hairpin, error);
2482 }
2483 
2484 /**
2485  * Flow driver preparation API. This abstracts calling driver specific
2486  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2487  * calculates the size of memory required for device flow, allocates the memory,
2488  * initializes the device flow and returns the pointer.
2489  *
2490  * @note
2491  *   This function initializes device flow structure such as dv or verbs in
2492  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2493  *   rest. For example, adding returning device flow to flow->dev_flow list and
2494  *   setting backward reference to the flow should be done out of this function.
2495  *   layers field is not filled either.
2496  *
2497  * @param[in] dev
2498  *   Pointer to the dev structure.
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(struct rte_eth_dev *dev,
2513 		 const struct rte_flow *flow,
2514 		 const struct rte_flow_attr *attr,
2515 		 const struct rte_flow_item items[],
2516 		 const struct rte_flow_action actions[],
2517 		 struct rte_flow_error *error)
2518 {
2519 	const struct mlx5_flow_driver_ops *fops;
2520 	enum mlx5_flow_drv_type type = flow->drv_type;
2521 
2522 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2523 	fops = flow_get_drv_ops(type);
2524 	return fops->prepare(dev, attr, items, actions, error);
2525 }
2526 
2527 /**
2528  * Flow driver translation API. This abstracts calling driver specific
2529  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2530  * translates a generic flow into a driver flow. flow_drv_prepare() must
2531  * precede.
2532  *
2533  * @note
2534  *   dev_flow->layers could be filled as a result of parsing during translation
2535  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2536  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2537  *   flow->actions could be overwritten even though all the expanded dev_flows
2538  *   have the same actions.
2539  *
2540  * @param[in] dev
2541  *   Pointer to the rte dev structure.
2542  * @param[in, out] dev_flow
2543  *   Pointer to the mlx5 flow.
2544  * @param[in] attr
2545  *   Pointer to the flow attributes.
2546  * @param[in] items
2547  *   Pointer to the list of items.
2548  * @param[in] actions
2549  *   Pointer to the list of actions.
2550  * @param[out] error
2551  *   Pointer to the error structure.
2552  *
2553  * @return
2554  *   0 on success, a negative errno value otherwise and rte_errno is set.
2555  */
2556 static inline int
2557 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2558 		   const struct rte_flow_attr *attr,
2559 		   const struct rte_flow_item items[],
2560 		   const struct rte_flow_action actions[],
2561 		   struct rte_flow_error *error)
2562 {
2563 	const struct mlx5_flow_driver_ops *fops;
2564 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2565 
2566 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2567 	fops = flow_get_drv_ops(type);
2568 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2569 }
2570 
2571 /**
2572  * Flow driver apply API. This abstracts calling driver specific functions.
2573  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2574  * translated driver flows on to device. flow_drv_translate() must precede.
2575  *
2576  * @param[in] dev
2577  *   Pointer to Ethernet device structure.
2578  * @param[in, out] flow
2579  *   Pointer to flow structure.
2580  * @param[out] error
2581  *   Pointer to error structure.
2582  *
2583  * @return
2584  *   0 on success, a negative errno value otherwise and rte_errno is set.
2585  */
2586 static inline int
2587 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2588 	       struct rte_flow_error *error)
2589 {
2590 	const struct mlx5_flow_driver_ops *fops;
2591 	enum mlx5_flow_drv_type type = flow->drv_type;
2592 
2593 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2594 	fops = flow_get_drv_ops(type);
2595 	return fops->apply(dev, flow, error);
2596 }
2597 
2598 /**
2599  * Flow driver remove API. This abstracts calling driver specific functions.
2600  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2601  * on device. All the resources of the flow should be freed by calling
2602  * flow_drv_destroy().
2603  *
2604  * @param[in] dev
2605  *   Pointer to Ethernet device.
2606  * @param[in, out] flow
2607  *   Pointer to flow structure.
2608  */
2609 static inline void
2610 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2611 {
2612 	const struct mlx5_flow_driver_ops *fops;
2613 	enum mlx5_flow_drv_type type = flow->drv_type;
2614 
2615 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2616 	fops = flow_get_drv_ops(type);
2617 	fops->remove(dev, flow);
2618 }
2619 
2620 /**
2621  * Flow driver destroy API. This abstracts calling driver specific functions.
2622  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2623  * on device and releases resources of the flow.
2624  *
2625  * @param[in] dev
2626  *   Pointer to Ethernet device.
2627  * @param[in, out] flow
2628  *   Pointer to flow structure.
2629  */
2630 static inline void
2631 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2632 {
2633 	const struct mlx5_flow_driver_ops *fops;
2634 	enum mlx5_flow_drv_type type = flow->drv_type;
2635 
2636 	flow_mreg_split_qrss_release(dev, flow);
2637 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2638 	fops = flow_get_drv_ops(type);
2639 	fops->destroy(dev, flow);
2640 }
2641 
2642 /**
2643  * Get RSS action from the action list.
2644  *
2645  * @param[in] actions
2646  *   Pointer to the list of actions.
2647  *
2648  * @return
2649  *   Pointer to the RSS action if exist, else return NULL.
2650  */
2651 static const struct rte_flow_action_rss*
2652 flow_get_rss_action(const struct rte_flow_action actions[])
2653 {
2654 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2655 		switch (actions->type) {
2656 		case RTE_FLOW_ACTION_TYPE_RSS:
2657 			return (const struct rte_flow_action_rss *)
2658 			       actions->conf;
2659 		default:
2660 			break;
2661 		}
2662 	}
2663 	return NULL;
2664 }
2665 
2666 static unsigned int
2667 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2668 {
2669 	const struct rte_flow_item *item;
2670 	unsigned int has_vlan = 0;
2671 
2672 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2673 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2674 			has_vlan = 1;
2675 			break;
2676 		}
2677 	}
2678 	if (has_vlan)
2679 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2680 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2681 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2682 			       MLX5_EXPANSION_ROOT_OUTER;
2683 }
2684 
2685 /**
2686  *  Get layer flags from the prefix flow.
2687  *
2688  *  Some flows may be split to several subflows, the prefix subflow gets the
2689  *  match items and the suffix sub flow gets the actions.
2690  *  Some actions need the user defined match item flags to get the detail for
2691  *  the action.
2692  *  This function helps the suffix flow to get the item layer flags from prefix
2693  *  subflow.
2694  *
2695  * @param[in] dev_flow
2696  *   Pointer the created preifx subflow.
2697  *
2698  * @return
2699  *   The layers get from prefix subflow.
2700  */
2701 static inline uint64_t
2702 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2703 {
2704 	uint64_t layers = 0;
2705 
2706 	/*
2707 	 * Layers bits could be localization, but usually the compiler will
2708 	 * help to do the optimization work for source code.
2709 	 * If no decap actions, use the layers directly.
2710 	 */
2711 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2712 		return dev_flow->handle->layers;
2713 	/* Convert L3 layers with decap action. */
2714 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2715 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2716 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2717 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2718 	/* Convert L4 layers with decap action.  */
2719 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2720 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2721 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2722 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2723 	return layers;
2724 }
2725 
2726 /**
2727  * Get metadata split action information.
2728  *
2729  * @param[in] actions
2730  *   Pointer to the list of actions.
2731  * @param[out] qrss
2732  *   Pointer to the return pointer.
2733  * @param[out] qrss_type
2734  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2735  *   if no QUEUE/RSS is found.
2736  * @param[out] encap_idx
2737  *   Pointer to the index of the encap action if exists, otherwise the last
2738  *   action index.
2739  *
2740  * @return
2741  *   Total number of actions.
2742  */
2743 static int
2744 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2745 				       const struct rte_flow_action **qrss,
2746 				       int *encap_idx)
2747 {
2748 	const struct rte_flow_action_raw_encap *raw_encap;
2749 	int actions_n = 0;
2750 	int raw_decap_idx = -1;
2751 
2752 	*encap_idx = -1;
2753 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2754 		switch (actions->type) {
2755 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2756 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2757 			*encap_idx = actions_n;
2758 			break;
2759 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2760 			raw_decap_idx = actions_n;
2761 			break;
2762 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2763 			raw_encap = actions->conf;
2764 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2765 				*encap_idx = raw_decap_idx != -1 ?
2766 						      raw_decap_idx : actions_n;
2767 			break;
2768 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2769 		case RTE_FLOW_ACTION_TYPE_RSS:
2770 			*qrss = actions;
2771 			break;
2772 		default:
2773 			break;
2774 		}
2775 		actions_n++;
2776 	}
2777 	if (*encap_idx == -1)
2778 		*encap_idx = actions_n;
2779 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2780 	return actions_n + 1;
2781 }
2782 
2783 /**
2784  * Check meter action from the action list.
2785  *
2786  * @param[in] actions
2787  *   Pointer to the list of actions.
2788  * @param[out] mtr
2789  *   Pointer to the meter exist flag.
2790  *
2791  * @return
2792  *   Total number of actions.
2793  */
2794 static int
2795 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2796 {
2797 	int actions_n = 0;
2798 
2799 	MLX5_ASSERT(mtr);
2800 	*mtr = 0;
2801 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2802 		switch (actions->type) {
2803 		case RTE_FLOW_ACTION_TYPE_METER:
2804 			*mtr = 1;
2805 			break;
2806 		default:
2807 			break;
2808 		}
2809 		actions_n++;
2810 	}
2811 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2812 	return actions_n + 1;
2813 }
2814 
2815 /**
2816  * Check if the flow should be splited due to hairpin.
2817  * The reason for the split is that in current HW we can't
2818  * support encap on Rx, so if a flow have encap we move it
2819  * to Tx.
2820  *
2821  * @param dev
2822  *   Pointer to Ethernet device.
2823  * @param[in] attr
2824  *   Flow rule attributes.
2825  * @param[in] actions
2826  *   Associated actions (list terminated by the END action).
2827  *
2828  * @return
2829  *   > 0 the number of actions and the flow should be split,
2830  *   0 when no split required.
2831  */
2832 static int
2833 flow_check_hairpin_split(struct rte_eth_dev *dev,
2834 			 const struct rte_flow_attr *attr,
2835 			 const struct rte_flow_action actions[])
2836 {
2837 	int queue_action = 0;
2838 	int action_n = 0;
2839 	int encap = 0;
2840 	const struct rte_flow_action_queue *queue;
2841 	const struct rte_flow_action_rss *rss;
2842 	const struct rte_flow_action_raw_encap *raw_encap;
2843 
2844 	if (!attr->ingress)
2845 		return 0;
2846 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2847 		switch (actions->type) {
2848 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2849 			queue = actions->conf;
2850 			if (queue == NULL)
2851 				return 0;
2852 			if (mlx5_rxq_get_type(dev, queue->index) !=
2853 			    MLX5_RXQ_TYPE_HAIRPIN)
2854 				return 0;
2855 			queue_action = 1;
2856 			action_n++;
2857 			break;
2858 		case RTE_FLOW_ACTION_TYPE_RSS:
2859 			rss = actions->conf;
2860 			if (rss == NULL || rss->queue_num == 0)
2861 				return 0;
2862 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2863 			    MLX5_RXQ_TYPE_HAIRPIN)
2864 				return 0;
2865 			queue_action = 1;
2866 			action_n++;
2867 			break;
2868 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2869 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2870 			encap = 1;
2871 			action_n++;
2872 			break;
2873 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2874 			raw_encap = actions->conf;
2875 			if (raw_encap->size >
2876 			    (sizeof(struct rte_flow_item_eth) +
2877 			     sizeof(struct rte_flow_item_ipv4)))
2878 				encap = 1;
2879 			action_n++;
2880 			break;
2881 		default:
2882 			action_n++;
2883 			break;
2884 		}
2885 	}
2886 	if (encap == 1 && queue_action)
2887 		return action_n;
2888 	return 0;
2889 }
2890 
2891 /* Declare flow create/destroy prototype in advance. */
2892 static uint32_t
2893 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2894 		 const struct rte_flow_attr *attr,
2895 		 const struct rte_flow_item items[],
2896 		 const struct rte_flow_action actions[],
2897 		 bool external, struct rte_flow_error *error);
2898 
2899 static void
2900 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2901 		  uint32_t flow_idx);
2902 
2903 /**
2904  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2905  *
2906  * As mark_id is unique, if there's already a registered flow for the mark_id,
2907  * return by increasing the reference counter of the resource. Otherwise, create
2908  * the resource (mcp_res) and flow.
2909  *
2910  * Flow looks like,
2911  *   - If ingress port is ANY and reg_c[1] is mark_id,
2912  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2913  *
2914  * For default flow (zero mark_id), flow is like,
2915  *   - If ingress port is ANY,
2916  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2917  *
2918  * @param dev
2919  *   Pointer to Ethernet device.
2920  * @param mark_id
2921  *   ID of MARK action, zero means default flow for META.
2922  * @param[out] error
2923  *   Perform verbose error reporting if not NULL.
2924  *
2925  * @return
2926  *   Associated resource on success, NULL otherwise and rte_errno is set.
2927  */
2928 static struct mlx5_flow_mreg_copy_resource *
2929 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2930 			  struct rte_flow_error *error)
2931 {
2932 	struct mlx5_priv *priv = dev->data->dev_private;
2933 	struct rte_flow_attr attr = {
2934 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2935 		.ingress = 1,
2936 	};
2937 	struct mlx5_rte_flow_item_tag tag_spec = {
2938 		.data = mark_id,
2939 	};
2940 	struct rte_flow_item items[] = {
2941 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2942 	};
2943 	struct rte_flow_action_mark ftag = {
2944 		.id = mark_id,
2945 	};
2946 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2947 		.dst = REG_B,
2948 		.src = 0,
2949 	};
2950 	struct rte_flow_action_jump jump = {
2951 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2952 	};
2953 	struct rte_flow_action actions[] = {
2954 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2955 	};
2956 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2957 	uint32_t idx = 0;
2958 	int ret;
2959 
2960 	/* Fill the register fileds in the flow. */
2961 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2962 	if (ret < 0)
2963 		return NULL;
2964 	tag_spec.id = ret;
2965 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2966 	if (ret < 0)
2967 		return NULL;
2968 	cp_mreg.src = ret;
2969 	/* Check if already registered. */
2970 	MLX5_ASSERT(priv->mreg_cp_tbl);
2971 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2972 	if (mcp_res) {
2973 		/* For non-default rule. */
2974 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2975 			mcp_res->refcnt++;
2976 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2977 			    mcp_res->refcnt == 1);
2978 		return mcp_res;
2979 	}
2980 	/* Provide the full width of FLAG specific value. */
2981 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2982 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2983 	/* Build a new flow. */
2984 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2985 		items[0] = (struct rte_flow_item){
2986 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2987 			.spec = &tag_spec,
2988 		};
2989 		items[1] = (struct rte_flow_item){
2990 			.type = RTE_FLOW_ITEM_TYPE_END,
2991 		};
2992 		actions[0] = (struct rte_flow_action){
2993 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2994 			.conf = &ftag,
2995 		};
2996 		actions[1] = (struct rte_flow_action){
2997 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2998 			.conf = &cp_mreg,
2999 		};
3000 		actions[2] = (struct rte_flow_action){
3001 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3002 			.conf = &jump,
3003 		};
3004 		actions[3] = (struct rte_flow_action){
3005 			.type = RTE_FLOW_ACTION_TYPE_END,
3006 		};
3007 	} else {
3008 		/* Default rule, wildcard match. */
3009 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3010 		items[0] = (struct rte_flow_item){
3011 			.type = RTE_FLOW_ITEM_TYPE_END,
3012 		};
3013 		actions[0] = (struct rte_flow_action){
3014 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3015 			.conf = &cp_mreg,
3016 		};
3017 		actions[1] = (struct rte_flow_action){
3018 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3019 			.conf = &jump,
3020 		};
3021 		actions[2] = (struct rte_flow_action){
3022 			.type = RTE_FLOW_ACTION_TYPE_END,
3023 		};
3024 	}
3025 	/* Build a new entry. */
3026 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3027 	if (!mcp_res) {
3028 		rte_errno = ENOMEM;
3029 		return NULL;
3030 	}
3031 	mcp_res->idx = idx;
3032 	/*
3033 	 * The copy Flows are not included in any list. There
3034 	 * ones are referenced from other Flows and can not
3035 	 * be applied, removed, deleted in ardbitrary order
3036 	 * by list traversing.
3037 	 */
3038 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3039 					 actions, false, error);
3040 	if (!mcp_res->rix_flow)
3041 		goto error;
3042 	mcp_res->refcnt++;
3043 	mcp_res->hlist_ent.key = mark_id;
3044 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3045 				&mcp_res->hlist_ent);
3046 	MLX5_ASSERT(!ret);
3047 	if (ret)
3048 		goto error;
3049 	return mcp_res;
3050 error:
3051 	if (mcp_res->rix_flow)
3052 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3053 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3054 	return NULL;
3055 }
3056 
3057 /**
3058  * Release flow in RX_CP_TBL.
3059  *
3060  * @param dev
3061  *   Pointer to Ethernet device.
3062  * @flow
3063  *   Parent flow for wich copying is provided.
3064  */
3065 static void
3066 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3067 			  struct rte_flow *flow)
3068 {
3069 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3070 	struct mlx5_priv *priv = dev->data->dev_private;
3071 
3072 	if (!flow->rix_mreg_copy)
3073 		return;
3074 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3075 				 flow->rix_mreg_copy);
3076 	if (!mcp_res || !priv->mreg_cp_tbl)
3077 		return;
3078 	if (flow->copy_applied) {
3079 		MLX5_ASSERT(mcp_res->appcnt);
3080 		flow->copy_applied = 0;
3081 		--mcp_res->appcnt;
3082 		if (!mcp_res->appcnt) {
3083 			struct rte_flow *mcp_flow = mlx5_ipool_get
3084 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3085 					mcp_res->rix_flow);
3086 
3087 			if (mcp_flow)
3088 				flow_drv_remove(dev, mcp_flow);
3089 		}
3090 	}
3091 	/*
3092 	 * We do not check availability of metadata registers here,
3093 	 * because copy resources are not allocated in this case.
3094 	 */
3095 	if (--mcp_res->refcnt)
3096 		return;
3097 	MLX5_ASSERT(mcp_res->rix_flow);
3098 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3099 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3100 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3101 	flow->rix_mreg_copy = 0;
3102 }
3103 
3104 /**
3105  * Start flow in RX_CP_TBL.
3106  *
3107  * @param dev
3108  *   Pointer to Ethernet device.
3109  * @flow
3110  *   Parent flow for wich copying is provided.
3111  *
3112  * @return
3113  *   0 on success, a negative errno value otherwise and rte_errno is set.
3114  */
3115 static int
3116 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3117 			    struct rte_flow *flow)
3118 {
3119 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3120 	struct mlx5_priv *priv = dev->data->dev_private;
3121 	int ret;
3122 
3123 	if (!flow->rix_mreg_copy || flow->copy_applied)
3124 		return 0;
3125 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3126 				 flow->rix_mreg_copy);
3127 	if (!mcp_res)
3128 		return 0;
3129 	if (!mcp_res->appcnt) {
3130 		struct rte_flow *mcp_flow = mlx5_ipool_get
3131 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3132 				mcp_res->rix_flow);
3133 
3134 		if (mcp_flow) {
3135 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3136 			if (ret)
3137 				return ret;
3138 		}
3139 	}
3140 	++mcp_res->appcnt;
3141 	flow->copy_applied = 1;
3142 	return 0;
3143 }
3144 
3145 /**
3146  * Stop flow in RX_CP_TBL.
3147  *
3148  * @param dev
3149  *   Pointer to Ethernet device.
3150  * @flow
3151  *   Parent flow for wich copying is provided.
3152  */
3153 static void
3154 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3155 			   struct rte_flow *flow)
3156 {
3157 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3158 	struct mlx5_priv *priv = dev->data->dev_private;
3159 
3160 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3161 		return;
3162 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3163 				 flow->rix_mreg_copy);
3164 	if (!mcp_res)
3165 		return;
3166 	MLX5_ASSERT(mcp_res->appcnt);
3167 	--mcp_res->appcnt;
3168 	flow->copy_applied = 0;
3169 	if (!mcp_res->appcnt) {
3170 		struct rte_flow *mcp_flow = mlx5_ipool_get
3171 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3172 				mcp_res->rix_flow);
3173 
3174 		if (mcp_flow)
3175 			flow_drv_remove(dev, mcp_flow);
3176 	}
3177 }
3178 
3179 /**
3180  * Remove the default copy action from RX_CP_TBL.
3181  *
3182  * @param dev
3183  *   Pointer to Ethernet device.
3184  */
3185 static void
3186 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3187 {
3188 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3189 	struct mlx5_priv *priv = dev->data->dev_private;
3190 
3191 	/* Check if default flow is registered. */
3192 	if (!priv->mreg_cp_tbl)
3193 		return;
3194 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3195 					    MLX5_DEFAULT_COPY_ID);
3196 	if (!mcp_res)
3197 		return;
3198 	MLX5_ASSERT(mcp_res->rix_flow);
3199 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3200 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3201 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3202 }
3203 
3204 /**
3205  * Add the default copy action in in RX_CP_TBL.
3206  *
3207  * @param dev
3208  *   Pointer to Ethernet device.
3209  * @param[out] error
3210  *   Perform verbose error reporting if not NULL.
3211  *
3212  * @return
3213  *   0 for success, negative value otherwise and rte_errno is set.
3214  */
3215 static int
3216 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3217 				  struct rte_flow_error *error)
3218 {
3219 	struct mlx5_priv *priv = dev->data->dev_private;
3220 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3221 
3222 	/* Check whether extensive metadata feature is engaged. */
3223 	if (!priv->config.dv_flow_en ||
3224 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3225 	    !mlx5_flow_ext_mreg_supported(dev) ||
3226 	    !priv->sh->dv_regc0_mask)
3227 		return 0;
3228 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3229 	if (!mcp_res)
3230 		return -rte_errno;
3231 	return 0;
3232 }
3233 
3234 /**
3235  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3236  *
3237  * All the flow having Q/RSS action should be split by
3238  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3239  * performs the following,
3240  *   - CQE->flow_tag := reg_c[1] (MARK)
3241  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3242  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3243  * but there should be a flow per each MARK ID set by MARK action.
3244  *
3245  * For the aforementioned reason, if there's a MARK action in flow's action
3246  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3247  * the MARK ID to CQE's flow_tag like,
3248  *   - If reg_c[1] is mark_id,
3249  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3250  *
3251  * For SET_META action which stores value in reg_c[0], as the destination is
3252  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3253  * MARK ID means the default flow. The default flow looks like,
3254  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3255  *
3256  * @param dev
3257  *   Pointer to Ethernet device.
3258  * @param flow
3259  *   Pointer to flow structure.
3260  * @param[in] actions
3261  *   Pointer to the list of actions.
3262  * @param[out] error
3263  *   Perform verbose error reporting if not NULL.
3264  *
3265  * @return
3266  *   0 on success, negative value otherwise and rte_errno is set.
3267  */
3268 static int
3269 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3270 			    struct rte_flow *flow,
3271 			    const struct rte_flow_action *actions,
3272 			    struct rte_flow_error *error)
3273 {
3274 	struct mlx5_priv *priv = dev->data->dev_private;
3275 	struct mlx5_dev_config *config = &priv->config;
3276 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3277 	const struct rte_flow_action_mark *mark;
3278 
3279 	/* Check whether extensive metadata feature is engaged. */
3280 	if (!config->dv_flow_en ||
3281 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3282 	    !mlx5_flow_ext_mreg_supported(dev) ||
3283 	    !priv->sh->dv_regc0_mask)
3284 		return 0;
3285 	/* Find MARK action. */
3286 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3287 		switch (actions->type) {
3288 		case RTE_FLOW_ACTION_TYPE_FLAG:
3289 			mcp_res = flow_mreg_add_copy_action
3290 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3291 			if (!mcp_res)
3292 				return -rte_errno;
3293 			flow->rix_mreg_copy = mcp_res->idx;
3294 			if (dev->data->dev_started) {
3295 				mcp_res->appcnt++;
3296 				flow->copy_applied = 1;
3297 			}
3298 			return 0;
3299 		case RTE_FLOW_ACTION_TYPE_MARK:
3300 			mark = (const struct rte_flow_action_mark *)
3301 				actions->conf;
3302 			mcp_res =
3303 				flow_mreg_add_copy_action(dev, mark->id, error);
3304 			if (!mcp_res)
3305 				return -rte_errno;
3306 			flow->rix_mreg_copy = mcp_res->idx;
3307 			if (dev->data->dev_started) {
3308 				mcp_res->appcnt++;
3309 				flow->copy_applied = 1;
3310 			}
3311 			return 0;
3312 		default:
3313 			break;
3314 		}
3315 	}
3316 	return 0;
3317 }
3318 
3319 #define MLX5_MAX_SPLIT_ACTIONS 24
3320 #define MLX5_MAX_SPLIT_ITEMS 24
3321 
3322 /**
3323  * Split the hairpin flow.
3324  * Since HW can't support encap on Rx we move the encap to Tx.
3325  * If the count action is after the encap then we also
3326  * move the count action. in this case the count will also measure
3327  * the outer bytes.
3328  *
3329  * @param dev
3330  *   Pointer to Ethernet device.
3331  * @param[in] actions
3332  *   Associated actions (list terminated by the END action).
3333  * @param[out] actions_rx
3334  *   Rx flow actions.
3335  * @param[out] actions_tx
3336  *   Tx flow actions..
3337  * @param[out] pattern_tx
3338  *   The pattern items for the Tx flow.
3339  * @param[out] flow_id
3340  *   The flow ID connected to this flow.
3341  *
3342  * @return
3343  *   0 on success.
3344  */
3345 static int
3346 flow_hairpin_split(struct rte_eth_dev *dev,
3347 		   const struct rte_flow_action actions[],
3348 		   struct rte_flow_action actions_rx[],
3349 		   struct rte_flow_action actions_tx[],
3350 		   struct rte_flow_item pattern_tx[],
3351 		   uint32_t *flow_id)
3352 {
3353 	struct mlx5_priv *priv = dev->data->dev_private;
3354 	const struct rte_flow_action_raw_encap *raw_encap;
3355 	const struct rte_flow_action_raw_decap *raw_decap;
3356 	struct mlx5_rte_flow_action_set_tag *set_tag;
3357 	struct rte_flow_action *tag_action;
3358 	struct mlx5_rte_flow_item_tag *tag_item;
3359 	struct rte_flow_item *item;
3360 	char *addr;
3361 	int encap = 0;
3362 
3363 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3364 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3365 		switch (actions->type) {
3366 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3367 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3368 			rte_memcpy(actions_tx, actions,
3369 			       sizeof(struct rte_flow_action));
3370 			actions_tx++;
3371 			break;
3372 		case RTE_FLOW_ACTION_TYPE_COUNT:
3373 			if (encap) {
3374 				rte_memcpy(actions_tx, actions,
3375 					   sizeof(struct rte_flow_action));
3376 				actions_tx++;
3377 			} else {
3378 				rte_memcpy(actions_rx, actions,
3379 					   sizeof(struct rte_flow_action));
3380 				actions_rx++;
3381 			}
3382 			break;
3383 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3384 			raw_encap = actions->conf;
3385 			if (raw_encap->size >
3386 			    (sizeof(struct rte_flow_item_eth) +
3387 			     sizeof(struct rte_flow_item_ipv4))) {
3388 				memcpy(actions_tx, actions,
3389 				       sizeof(struct rte_flow_action));
3390 				actions_tx++;
3391 				encap = 1;
3392 			} else {
3393 				rte_memcpy(actions_rx, actions,
3394 					   sizeof(struct rte_flow_action));
3395 				actions_rx++;
3396 			}
3397 			break;
3398 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3399 			raw_decap = actions->conf;
3400 			if (raw_decap->size <
3401 			    (sizeof(struct rte_flow_item_eth) +
3402 			     sizeof(struct rte_flow_item_ipv4))) {
3403 				memcpy(actions_tx, actions,
3404 				       sizeof(struct rte_flow_action));
3405 				actions_tx++;
3406 			} else {
3407 				rte_memcpy(actions_rx, actions,
3408 					   sizeof(struct rte_flow_action));
3409 				actions_rx++;
3410 			}
3411 			break;
3412 		default:
3413 			rte_memcpy(actions_rx, actions,
3414 				   sizeof(struct rte_flow_action));
3415 			actions_rx++;
3416 			break;
3417 		}
3418 	}
3419 	/* Add set meta action and end action for the Rx flow. */
3420 	tag_action = actions_rx;
3421 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3422 	actions_rx++;
3423 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3424 	actions_rx++;
3425 	set_tag = (void *)actions_rx;
3426 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3427 	MLX5_ASSERT(set_tag->id > REG_NONE);
3428 	set_tag->data = *flow_id;
3429 	tag_action->conf = set_tag;
3430 	/* Create Tx item list. */
3431 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3432 	addr = (void *)&pattern_tx[2];
3433 	item = pattern_tx;
3434 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3435 	tag_item = (void *)addr;
3436 	tag_item->data = *flow_id;
3437 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3438 	MLX5_ASSERT(set_tag->id > REG_NONE);
3439 	item->spec = tag_item;
3440 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3441 	tag_item = (void *)addr;
3442 	tag_item->data = UINT32_MAX;
3443 	tag_item->id = UINT16_MAX;
3444 	item->mask = tag_item;
3445 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3446 	item->last = NULL;
3447 	item++;
3448 	item->type = RTE_FLOW_ITEM_TYPE_END;
3449 	return 0;
3450 }
3451 
3452 /**
3453  * The last stage of splitting chain, just creates the subflow
3454  * without any modification.
3455  *
3456  * @param[in] dev
3457  *   Pointer to Ethernet device.
3458  * @param[in] flow
3459  *   Parent flow structure pointer.
3460  * @param[in, out] sub_flow
3461  *   Pointer to return the created subflow, may be NULL.
3462  * @param[in] prefix_layers
3463  *   Prefix subflow layers, may be 0.
3464  * @param[in] attr
3465  *   Flow rule attributes.
3466  * @param[in] items
3467  *   Pattern specification (list terminated by the END pattern item).
3468  * @param[in] actions
3469  *   Associated actions (list terminated by the END action).
3470  * @param[in] external
3471  *   This flow rule is created by request external to PMD.
3472  * @param[out] error
3473  *   Perform verbose error reporting if not NULL.
3474  * @return
3475  *   0 on success, negative value otherwise
3476  */
3477 static int
3478 flow_create_split_inner(struct rte_eth_dev *dev,
3479 			struct rte_flow *flow,
3480 			struct mlx5_flow **sub_flow,
3481 			uint64_t prefix_layers,
3482 			const struct rte_flow_attr *attr,
3483 			const struct rte_flow_item items[],
3484 			const struct rte_flow_action actions[],
3485 			bool external, struct rte_flow_error *error)
3486 {
3487 	struct mlx5_flow *dev_flow;
3488 
3489 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3490 	if (!dev_flow)
3491 		return -rte_errno;
3492 	dev_flow->flow = flow;
3493 	dev_flow->external = external;
3494 	/* Subflow object was created, we must include one in the list. */
3495 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3496 		      dev_flow->handle, next);
3497 	/*
3498 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3499 	 * flow may need some user defined item layer flags.
3500 	 */
3501 	if (prefix_layers)
3502 		dev_flow->handle->layers = prefix_layers;
3503 	if (sub_flow)
3504 		*sub_flow = dev_flow;
3505 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3506 }
3507 
3508 /**
3509  * Split the meter flow.
3510  *
3511  * As meter flow will split to three sub flow, other than meter
3512  * action, the other actions make sense to only meter accepts
3513  * the packet. If it need to be dropped, no other additional
3514  * actions should be take.
3515  *
3516  * One kind of special action which decapsulates the L3 tunnel
3517  * header will be in the prefix sub flow, as not to take the
3518  * L3 tunnel header into account.
3519  *
3520  * @param dev
3521  *   Pointer to Ethernet device.
3522  * @param[in] items
3523  *   Pattern specification (list terminated by the END pattern item).
3524  * @param[out] sfx_items
3525  *   Suffix flow match items (list terminated by the END pattern item).
3526  * @param[in] actions
3527  *   Associated actions (list terminated by the END action).
3528  * @param[out] actions_sfx
3529  *   Suffix flow actions.
3530  * @param[out] actions_pre
3531  *   Prefix flow actions.
3532  * @param[out] pattern_sfx
3533  *   The pattern items for the suffix flow.
3534  * @param[out] tag_sfx
3535  *   Pointer to suffix flow tag.
3536  *
3537  * @return
3538  *   0 on success.
3539  */
3540 static int
3541 flow_meter_split_prep(struct rte_eth_dev *dev,
3542 		 const struct rte_flow_item items[],
3543 		 struct rte_flow_item sfx_items[],
3544 		 const struct rte_flow_action actions[],
3545 		 struct rte_flow_action actions_sfx[],
3546 		 struct rte_flow_action actions_pre[])
3547 {
3548 	struct rte_flow_action *tag_action = NULL;
3549 	struct rte_flow_item *tag_item;
3550 	struct mlx5_rte_flow_action_set_tag *set_tag;
3551 	struct rte_flow_error error;
3552 	const struct rte_flow_action_raw_encap *raw_encap;
3553 	const struct rte_flow_action_raw_decap *raw_decap;
3554 	struct mlx5_rte_flow_item_tag *tag_spec;
3555 	struct mlx5_rte_flow_item_tag *tag_mask;
3556 	uint32_t tag_id;
3557 	bool copy_vlan = false;
3558 
3559 	/* Prepare the actions for prefix and suffix flow. */
3560 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3561 		struct rte_flow_action **action_cur = NULL;
3562 
3563 		switch (actions->type) {
3564 		case RTE_FLOW_ACTION_TYPE_METER:
3565 			/* Add the extra tag action first. */
3566 			tag_action = actions_pre;
3567 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3568 			actions_pre++;
3569 			action_cur = &actions_pre;
3570 			break;
3571 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3572 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3573 			action_cur = &actions_pre;
3574 			break;
3575 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3576 			raw_encap = actions->conf;
3577 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3578 				action_cur = &actions_pre;
3579 			break;
3580 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3581 			raw_decap = actions->conf;
3582 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3583 				action_cur = &actions_pre;
3584 			break;
3585 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3586 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3587 			copy_vlan = true;
3588 			break;
3589 		default:
3590 			break;
3591 		}
3592 		if (!action_cur)
3593 			action_cur = &actions_sfx;
3594 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3595 		(*action_cur)++;
3596 	}
3597 	/* Add end action to the actions. */
3598 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3599 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3600 	actions_pre++;
3601 	/* Set the tag. */
3602 	set_tag = (void *)actions_pre;
3603 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3604 	/*
3605 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3606 	 */
3607 	tag_id = flow_qrss_get_id(dev);
3608 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3609 	assert(tag_action);
3610 	tag_action->conf = set_tag;
3611 	/* Prepare the suffix subflow items. */
3612 	tag_item = sfx_items++;
3613 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3614 		int item_type = items->type;
3615 
3616 		switch (item_type) {
3617 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3618 			memcpy(sfx_items, items, sizeof(*sfx_items));
3619 			sfx_items++;
3620 			break;
3621 		case RTE_FLOW_ITEM_TYPE_VLAN:
3622 			if (copy_vlan) {
3623 				memcpy(sfx_items, items, sizeof(*sfx_items));
3624 				/*
3625 				 * Convert to internal match item, it is used
3626 				 * for vlan push and set vid.
3627 				 */
3628 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3629 				sfx_items++;
3630 			}
3631 			break;
3632 		default:
3633 			break;
3634 		}
3635 	}
3636 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3637 	sfx_items++;
3638 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3639 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3640 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3641 	tag_mask = tag_spec + 1;
3642 	tag_mask->data = 0xffffff00;
3643 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3644 	tag_item->spec = tag_spec;
3645 	tag_item->last = NULL;
3646 	tag_item->mask = tag_mask;
3647 	return tag_id;
3648 }
3649 
3650 /**
3651  * Split action list having QUEUE/RSS for metadata register copy.
3652  *
3653  * Once Q/RSS action is detected in user's action list, the flow action
3654  * should be split in order to copy metadata registers, which will happen in
3655  * RX_CP_TBL like,
3656  *   - CQE->flow_tag := reg_c[1] (MARK)
3657  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3658  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3659  * This is because the last action of each flow must be a terminal action
3660  * (QUEUE, RSS or DROP).
3661  *
3662  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3663  * stored and kept in the mlx5_flow structure per each sub_flow.
3664  *
3665  * The Q/RSS action is replaced with,
3666  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3667  * And the following JUMP action is added at the end,
3668  *   - JUMP, to RX_CP_TBL.
3669  *
3670  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3671  * flow_create_split_metadata() routine. The flow will look like,
3672  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3673  *
3674  * @param dev
3675  *   Pointer to Ethernet device.
3676  * @param[out] split_actions
3677  *   Pointer to store split actions to jump to CP_TBL.
3678  * @param[in] actions
3679  *   Pointer to the list of original flow actions.
3680  * @param[in] qrss
3681  *   Pointer to the Q/RSS action.
3682  * @param[in] actions_n
3683  *   Number of original actions.
3684  * @param[out] error
3685  *   Perform verbose error reporting if not NULL.
3686  *
3687  * @return
3688  *   non-zero unique flow_id on success, otherwise 0 and
3689  *   error/rte_error are set.
3690  */
3691 static uint32_t
3692 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3693 			  struct rte_flow_action *split_actions,
3694 			  const struct rte_flow_action *actions,
3695 			  const struct rte_flow_action *qrss,
3696 			  int actions_n, struct rte_flow_error *error)
3697 {
3698 	struct mlx5_rte_flow_action_set_tag *set_tag;
3699 	struct rte_flow_action_jump *jump;
3700 	const int qrss_idx = qrss - actions;
3701 	uint32_t flow_id = 0;
3702 	int ret = 0;
3703 
3704 	/*
3705 	 * Given actions will be split
3706 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3707 	 * - Add jump to mreg CP_TBL.
3708 	 * As a result, there will be one more action.
3709 	 */
3710 	++actions_n;
3711 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3712 	set_tag = (void *)(split_actions + actions_n);
3713 	/*
3714 	 * If tag action is not set to void(it means we are not the meter
3715 	 * suffix flow), add the tag action. Since meter suffix flow already
3716 	 * has the tag added.
3717 	 */
3718 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3719 		/*
3720 		 * Allocate the new subflow ID. This one is unique within
3721 		 * device and not shared with representors. Otherwise,
3722 		 * we would have to resolve multi-thread access synch
3723 		 * issue. Each flow on the shared device is appended
3724 		 * with source vport identifier, so the resulting
3725 		 * flows will be unique in the shared (by master and
3726 		 * representors) domain even if they have coinciding
3727 		 * IDs.
3728 		 */
3729 		flow_id = flow_qrss_get_id(dev);
3730 		if (!flow_id)
3731 			return rte_flow_error_set(error, ENOMEM,
3732 						  RTE_FLOW_ERROR_TYPE_ACTION,
3733 						  NULL, "can't allocate id "
3734 						  "for split Q/RSS subflow");
3735 		/* Internal SET_TAG action to set flow ID. */
3736 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3737 			.data = flow_id,
3738 		};
3739 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3740 		if (ret < 0)
3741 			return ret;
3742 		set_tag->id = ret;
3743 		/* Construct new actions array. */
3744 		/* Replace QUEUE/RSS action. */
3745 		split_actions[qrss_idx] = (struct rte_flow_action){
3746 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3747 			.conf = set_tag,
3748 		};
3749 	}
3750 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3751 	jump = (void *)(set_tag + 1);
3752 	*jump = (struct rte_flow_action_jump){
3753 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3754 	};
3755 	split_actions[actions_n - 2] = (struct rte_flow_action){
3756 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3757 		.conf = jump,
3758 	};
3759 	split_actions[actions_n - 1] = (struct rte_flow_action){
3760 		.type = RTE_FLOW_ACTION_TYPE_END,
3761 	};
3762 	return flow_id;
3763 }
3764 
3765 /**
3766  * Extend the given action list for Tx metadata copy.
3767  *
3768  * Copy the given action list to the ext_actions and add flow metadata register
3769  * copy action in order to copy reg_a set by WQE to reg_c[0].
3770  *
3771  * @param[out] ext_actions
3772  *   Pointer to the extended action list.
3773  * @param[in] actions
3774  *   Pointer to the list of actions.
3775  * @param[in] actions_n
3776  *   Number of actions in the list.
3777  * @param[out] error
3778  *   Perform verbose error reporting if not NULL.
3779  * @param[in] encap_idx
3780  *   The encap action inndex.
3781  *
3782  * @return
3783  *   0 on success, negative value otherwise
3784  */
3785 static int
3786 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3787 		       struct rte_flow_action *ext_actions,
3788 		       const struct rte_flow_action *actions,
3789 		       int actions_n, struct rte_flow_error *error,
3790 		       int encap_idx)
3791 {
3792 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3793 		(struct mlx5_flow_action_copy_mreg *)
3794 			(ext_actions + actions_n + 1);
3795 	int ret;
3796 
3797 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3798 	if (ret < 0)
3799 		return ret;
3800 	cp_mreg->dst = ret;
3801 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3802 	if (ret < 0)
3803 		return ret;
3804 	cp_mreg->src = ret;
3805 	if (encap_idx != 0)
3806 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3807 	if (encap_idx == actions_n - 1) {
3808 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3809 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3810 			.conf = cp_mreg,
3811 		};
3812 		ext_actions[actions_n] = (struct rte_flow_action){
3813 			.type = RTE_FLOW_ACTION_TYPE_END,
3814 		};
3815 	} else {
3816 		ext_actions[encap_idx] = (struct rte_flow_action){
3817 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3818 			.conf = cp_mreg,
3819 		};
3820 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3821 				sizeof(*ext_actions) * (actions_n - encap_idx));
3822 	}
3823 	return 0;
3824 }
3825 
3826 /**
3827  * The splitting for metadata feature.
3828  *
3829  * - Q/RSS action on NIC Rx should be split in order to pass by
3830  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3831  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3832  *
3833  * - All the actions on NIC Tx should have a mreg copy action to
3834  *   copy reg_a from WQE to reg_c[0].
3835  *
3836  * @param dev
3837  *   Pointer to Ethernet device.
3838  * @param[in] flow
3839  *   Parent flow structure pointer.
3840  * @param[in] prefix_layers
3841  *   Prefix flow layer flags.
3842  * @param[in] attr
3843  *   Flow rule attributes.
3844  * @param[in] items
3845  *   Pattern specification (list terminated by the END pattern item).
3846  * @param[in] actions
3847  *   Associated actions (list terminated by the END action).
3848  * @param[in] external
3849  *   This flow rule is created by request external to PMD.
3850  * @param[out] error
3851  *   Perform verbose error reporting if not NULL.
3852  * @return
3853  *   0 on success, negative value otherwise
3854  */
3855 static int
3856 flow_create_split_metadata(struct rte_eth_dev *dev,
3857 			   struct rte_flow *flow,
3858 			   uint64_t prefix_layers,
3859 			   const struct rte_flow_attr *attr,
3860 			   const struct rte_flow_item items[],
3861 			   const struct rte_flow_action actions[],
3862 			   bool external, struct rte_flow_error *error)
3863 {
3864 	struct mlx5_priv *priv = dev->data->dev_private;
3865 	struct mlx5_dev_config *config = &priv->config;
3866 	const struct rte_flow_action *qrss = NULL;
3867 	struct rte_flow_action *ext_actions = NULL;
3868 	struct mlx5_flow *dev_flow = NULL;
3869 	uint32_t qrss_id = 0;
3870 	int mtr_sfx = 0;
3871 	size_t act_size;
3872 	int actions_n;
3873 	int encap_idx;
3874 	int ret;
3875 
3876 	/* Check whether extensive metadata feature is engaged. */
3877 	if (!config->dv_flow_en ||
3878 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3879 	    !mlx5_flow_ext_mreg_supported(dev))
3880 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3881 					       attr, items, actions, external,
3882 					       error);
3883 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3884 							   &encap_idx);
3885 	if (qrss) {
3886 		/* Exclude hairpin flows from splitting. */
3887 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3888 			const struct rte_flow_action_queue *queue;
3889 
3890 			queue = qrss->conf;
3891 			if (mlx5_rxq_get_type(dev, queue->index) ==
3892 			    MLX5_RXQ_TYPE_HAIRPIN)
3893 				qrss = NULL;
3894 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3895 			const struct rte_flow_action_rss *rss;
3896 
3897 			rss = qrss->conf;
3898 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3899 			    MLX5_RXQ_TYPE_HAIRPIN)
3900 				qrss = NULL;
3901 		}
3902 	}
3903 	if (qrss) {
3904 		/* Check if it is in meter suffix table. */
3905 		mtr_sfx = attr->group == (attr->transfer ?
3906 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3907 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3908 		/*
3909 		 * Q/RSS action on NIC Rx should be split in order to pass by
3910 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3911 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3912 		 */
3913 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3914 			   sizeof(struct rte_flow_action_set_tag) +
3915 			   sizeof(struct rte_flow_action_jump);
3916 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3917 		if (!ext_actions)
3918 			return rte_flow_error_set(error, ENOMEM,
3919 						  RTE_FLOW_ERROR_TYPE_ACTION,
3920 						  NULL, "no memory to split "
3921 						  "metadata flow");
3922 		/*
3923 		 * If we are the suffix flow of meter, tag already exist.
3924 		 * Set the tag action to void.
3925 		 */
3926 		if (mtr_sfx)
3927 			ext_actions[qrss - actions].type =
3928 						RTE_FLOW_ACTION_TYPE_VOID;
3929 		else
3930 			ext_actions[qrss - actions].type =
3931 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3932 		/*
3933 		 * Create the new actions list with removed Q/RSS action
3934 		 * and appended set tag and jump to register copy table
3935 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3936 		 * in advance, because it is needed for set tag action.
3937 		 */
3938 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3939 						    qrss, actions_n, error);
3940 		if (!mtr_sfx && !qrss_id) {
3941 			ret = -rte_errno;
3942 			goto exit;
3943 		}
3944 	} else if (attr->egress && !attr->transfer) {
3945 		/*
3946 		 * All the actions on NIC Tx should have a metadata register
3947 		 * copy action to copy reg_a from WQE to reg_c[meta]
3948 		 */
3949 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3950 			   sizeof(struct mlx5_flow_action_copy_mreg);
3951 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3952 		if (!ext_actions)
3953 			return rte_flow_error_set(error, ENOMEM,
3954 						  RTE_FLOW_ERROR_TYPE_ACTION,
3955 						  NULL, "no memory to split "
3956 						  "metadata flow");
3957 		/* Create the action list appended with copy register. */
3958 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3959 					     actions_n, error, encap_idx);
3960 		if (ret < 0)
3961 			goto exit;
3962 	}
3963 	/* Add the unmodified original or prefix subflow. */
3964 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3965 				      items, ext_actions ? ext_actions :
3966 				      actions, external, error);
3967 	if (ret < 0)
3968 		goto exit;
3969 	MLX5_ASSERT(dev_flow);
3970 	if (qrss) {
3971 		const struct rte_flow_attr q_attr = {
3972 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3973 			.ingress = 1,
3974 		};
3975 		/* Internal PMD action to set register. */
3976 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3977 			.data = qrss_id,
3978 			.id = 0,
3979 		};
3980 		struct rte_flow_item q_items[] = {
3981 			{
3982 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3983 				.spec = &q_tag_spec,
3984 				.last = NULL,
3985 				.mask = NULL,
3986 			},
3987 			{
3988 				.type = RTE_FLOW_ITEM_TYPE_END,
3989 			},
3990 		};
3991 		struct rte_flow_action q_actions[] = {
3992 			{
3993 				.type = qrss->type,
3994 				.conf = qrss->conf,
3995 			},
3996 			{
3997 				.type = RTE_FLOW_ACTION_TYPE_END,
3998 			},
3999 		};
4000 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4001 
4002 		/*
4003 		 * Configure the tag item only if there is no meter subflow.
4004 		 * Since tag is already marked in the meter suffix subflow
4005 		 * we can just use the meter suffix items as is.
4006 		 */
4007 		if (qrss_id) {
4008 			/* Not meter subflow. */
4009 			MLX5_ASSERT(!mtr_sfx);
4010 			/*
4011 			 * Put unique id in prefix flow due to it is destroyed
4012 			 * after suffix flow and id will be freed after there
4013 			 * is no actual flows with this id and identifier
4014 			 * reallocation becomes possible (for example, for
4015 			 * other flows in other threads).
4016 			 */
4017 			dev_flow->handle->split_flow_id = qrss_id;
4018 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4019 						   error);
4020 			if (ret < 0)
4021 				goto exit;
4022 			q_tag_spec.id = ret;
4023 		}
4024 		dev_flow = NULL;
4025 		/* Add suffix subflow to execute Q/RSS. */
4026 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4027 					      &q_attr, mtr_sfx ? items :
4028 					      q_items, q_actions,
4029 					      external, error);
4030 		if (ret < 0)
4031 			goto exit;
4032 		/* qrss ID should be freed if failed. */
4033 		qrss_id = 0;
4034 		MLX5_ASSERT(dev_flow);
4035 	}
4036 
4037 exit:
4038 	/*
4039 	 * We do not destroy the partially created sub_flows in case of error.
4040 	 * These ones are included into parent flow list and will be destroyed
4041 	 * by flow_drv_destroy.
4042 	 */
4043 	flow_qrss_free_id(dev, qrss_id);
4044 	rte_free(ext_actions);
4045 	return ret;
4046 }
4047 
4048 /**
4049  * The splitting for meter feature.
4050  *
4051  * - The meter flow will be split to two flows as prefix and
4052  *   suffix flow. The packets make sense only it pass the prefix
4053  *   meter action.
4054  *
4055  * - Reg_C_5 is used for the packet to match betweend prefix and
4056  *   suffix flow.
4057  *
4058  * @param dev
4059  *   Pointer to Ethernet device.
4060  * @param[in] flow
4061  *   Parent flow structure pointer.
4062  * @param[in] attr
4063  *   Flow rule attributes.
4064  * @param[in] items
4065  *   Pattern specification (list terminated by the END pattern item).
4066  * @param[in] actions
4067  *   Associated actions (list terminated by the END action).
4068  * @param[in] external
4069  *   This flow rule is created by request external to PMD.
4070  * @param[out] error
4071  *   Perform verbose error reporting if not NULL.
4072  * @return
4073  *   0 on success, negative value otherwise
4074  */
4075 static int
4076 flow_create_split_meter(struct rte_eth_dev *dev,
4077 			   struct rte_flow *flow,
4078 			   const struct rte_flow_attr *attr,
4079 			   const struct rte_flow_item items[],
4080 			   const struct rte_flow_action actions[],
4081 			   bool external, struct rte_flow_error *error)
4082 {
4083 	struct mlx5_priv *priv = dev->data->dev_private;
4084 	struct rte_flow_action *sfx_actions = NULL;
4085 	struct rte_flow_action *pre_actions = NULL;
4086 	struct rte_flow_item *sfx_items = NULL;
4087 	struct mlx5_flow *dev_flow = NULL;
4088 	struct rte_flow_attr sfx_attr = *attr;
4089 	uint32_t mtr = 0;
4090 	uint32_t mtr_tag_id = 0;
4091 	size_t act_size;
4092 	size_t item_size;
4093 	int actions_n = 0;
4094 	int ret;
4095 
4096 	if (priv->mtr_en)
4097 		actions_n = flow_check_meter_action(actions, &mtr);
4098 	if (mtr) {
4099 		/* The five prefix actions: meter, decap, encap, tag, end. */
4100 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4101 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4102 		/* tag, vlan, port id, end. */
4103 #define METER_SUFFIX_ITEM 4
4104 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4105 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4106 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4107 		if (!sfx_actions)
4108 			return rte_flow_error_set(error, ENOMEM,
4109 						  RTE_FLOW_ERROR_TYPE_ACTION,
4110 						  NULL, "no memory to split "
4111 						  "meter flow");
4112 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4113 			     act_size);
4114 		pre_actions = sfx_actions + actions_n;
4115 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4116 						   actions, sfx_actions,
4117 						   pre_actions);
4118 		if (!mtr_tag_id) {
4119 			ret = -rte_errno;
4120 			goto exit;
4121 		}
4122 		/* Add the prefix subflow. */
4123 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4124 					      items, pre_actions, external,
4125 					      error);
4126 		if (ret) {
4127 			ret = -rte_errno;
4128 			goto exit;
4129 		}
4130 		dev_flow->handle->split_flow_id = mtr_tag_id;
4131 		/* Setting the sfx group atrr. */
4132 		sfx_attr.group = sfx_attr.transfer ?
4133 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4134 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4135 	}
4136 	/* Add the prefix subflow. */
4137 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4138 					 flow_get_prefix_layer_flags(dev_flow) :
4139 					 0, &sfx_attr,
4140 					 sfx_items ? sfx_items : items,
4141 					 sfx_actions ? sfx_actions : actions,
4142 					 external, error);
4143 exit:
4144 	if (sfx_actions)
4145 		rte_free(sfx_actions);
4146 	return ret;
4147 }
4148 
4149 /**
4150  * Split the flow to subflow set. The splitters might be linked
4151  * in the chain, like this:
4152  * flow_create_split_outer() calls:
4153  *   flow_create_split_meter() calls:
4154  *     flow_create_split_metadata(meter_subflow_0) calls:
4155  *       flow_create_split_inner(metadata_subflow_0)
4156  *       flow_create_split_inner(metadata_subflow_1)
4157  *       flow_create_split_inner(metadata_subflow_2)
4158  *     flow_create_split_metadata(meter_subflow_1) calls:
4159  *       flow_create_split_inner(metadata_subflow_0)
4160  *       flow_create_split_inner(metadata_subflow_1)
4161  *       flow_create_split_inner(metadata_subflow_2)
4162  *
4163  * This provide flexible way to add new levels of flow splitting.
4164  * The all of successfully created subflows are included to the
4165  * parent flow dev_flow list.
4166  *
4167  * @param dev
4168  *   Pointer to Ethernet device.
4169  * @param[in] flow
4170  *   Parent flow structure pointer.
4171  * @param[in] attr
4172  *   Flow rule attributes.
4173  * @param[in] items
4174  *   Pattern specification (list terminated by the END pattern item).
4175  * @param[in] actions
4176  *   Associated actions (list terminated by the END action).
4177  * @param[in] external
4178  *   This flow rule is created by request external to PMD.
4179  * @param[out] error
4180  *   Perform verbose error reporting if not NULL.
4181  * @return
4182  *   0 on success, negative value otherwise
4183  */
4184 static int
4185 flow_create_split_outer(struct rte_eth_dev *dev,
4186 			struct rte_flow *flow,
4187 			const struct rte_flow_attr *attr,
4188 			const struct rte_flow_item items[],
4189 			const struct rte_flow_action actions[],
4190 			bool external, struct rte_flow_error *error)
4191 {
4192 	int ret;
4193 
4194 	ret = flow_create_split_meter(dev, flow, attr, items,
4195 					 actions, external, error);
4196 	MLX5_ASSERT(ret <= 0);
4197 	return ret;
4198 }
4199 
4200 /**
4201  * Create a flow and add it to @p list.
4202  *
4203  * @param dev
4204  *   Pointer to Ethernet device.
4205  * @param list
4206  *   Pointer to a TAILQ flow list. If this parameter NULL,
4207  *   no list insertion occurred, flow is just created,
4208  *   this is caller's responsibility to track the
4209  *   created flow.
4210  * @param[in] attr
4211  *   Flow rule attributes.
4212  * @param[in] items
4213  *   Pattern specification (list terminated by the END pattern item).
4214  * @param[in] actions
4215  *   Associated actions (list terminated by the END action).
4216  * @param[in] external
4217  *   This flow rule is created by request external to PMD.
4218  * @param[out] error
4219  *   Perform verbose error reporting if not NULL.
4220  *
4221  * @return
4222  *   A flow index on success, 0 otherwise and rte_errno is set.
4223  */
4224 static uint32_t
4225 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4226 		 const struct rte_flow_attr *attr,
4227 		 const struct rte_flow_item items[],
4228 		 const struct rte_flow_action actions[],
4229 		 bool external, struct rte_flow_error *error)
4230 {
4231 	struct mlx5_priv *priv = dev->data->dev_private;
4232 	struct rte_flow *flow = NULL;
4233 	struct mlx5_flow *dev_flow;
4234 	const struct rte_flow_action_rss *rss;
4235 	union {
4236 		struct rte_flow_expand_rss buf;
4237 		uint8_t buffer[2048];
4238 	} expand_buffer;
4239 	union {
4240 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4241 		uint8_t buffer[2048];
4242 	} actions_rx;
4243 	union {
4244 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4245 		uint8_t buffer[2048];
4246 	} actions_hairpin_tx;
4247 	union {
4248 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4249 		uint8_t buffer[2048];
4250 	} items_tx;
4251 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4252 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4253 					      priv->rss_desc)[!!priv->flow_idx];
4254 	const struct rte_flow_action *p_actions_rx = actions;
4255 	uint32_t i;
4256 	uint32_t idx = 0;
4257 	int hairpin_flow;
4258 	uint32_t hairpin_id = 0;
4259 	struct rte_flow_attr attr_tx = { .priority = 0 };
4260 	int ret;
4261 
4262 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4263 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4264 				external, hairpin_flow, error);
4265 	if (ret < 0)
4266 		return 0;
4267 	if (hairpin_flow > 0) {
4268 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4269 			rte_errno = EINVAL;
4270 			return 0;
4271 		}
4272 		flow_hairpin_split(dev, actions, actions_rx.actions,
4273 				   actions_hairpin_tx.actions, items_tx.items,
4274 				   &hairpin_id);
4275 		p_actions_rx = actions_rx.actions;
4276 	}
4277 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4278 	if (!flow) {
4279 		rte_errno = ENOMEM;
4280 		goto error_before_flow;
4281 	}
4282 	flow->drv_type = flow_get_drv_type(dev, attr);
4283 	if (hairpin_id != 0)
4284 		flow->hairpin_flow_id = hairpin_id;
4285 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4286 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4287 	memset(rss_desc, 0, sizeof(*rss_desc));
4288 	rss = flow_get_rss_action(p_actions_rx);
4289 	if (rss) {
4290 		/*
4291 		 * The following information is required by
4292 		 * mlx5_flow_hashfields_adjust() in advance.
4293 		 */
4294 		rss_desc->level = rss->level;
4295 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4296 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4297 	}
4298 	flow->dev_handles = 0;
4299 	if (rss && rss->types) {
4300 		unsigned int graph_root;
4301 
4302 		graph_root = find_graph_root(items, rss->level);
4303 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4304 					  items, rss->types,
4305 					  mlx5_support_expansion,
4306 					  graph_root);
4307 		MLX5_ASSERT(ret > 0 &&
4308 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4309 	} else {
4310 		buf->entries = 1;
4311 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4312 	}
4313 	/*
4314 	 * Record the start index when there is a nested call. All sub-flows
4315 	 * need to be translated before another calling.
4316 	 * No need to use ping-pong buffer to save memory here.
4317 	 */
4318 	if (priv->flow_idx) {
4319 		MLX5_ASSERT(!priv->flow_nested_idx);
4320 		priv->flow_nested_idx = priv->flow_idx;
4321 	}
4322 	for (i = 0; i < buf->entries; ++i) {
4323 		/*
4324 		 * The splitter may create multiple dev_flows,
4325 		 * depending on configuration. In the simplest
4326 		 * case it just creates unmodified original flow.
4327 		 */
4328 		ret = flow_create_split_outer(dev, flow, attr,
4329 					      buf->entry[i].pattern,
4330 					      p_actions_rx, external,
4331 					      error);
4332 		if (ret < 0)
4333 			goto error;
4334 	}
4335 	/* Create the tx flow. */
4336 	if (hairpin_flow) {
4337 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4338 		attr_tx.ingress = 0;
4339 		attr_tx.egress = 1;
4340 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4341 					    actions_hairpin_tx.actions, error);
4342 		if (!dev_flow)
4343 			goto error;
4344 		dev_flow->flow = flow;
4345 		dev_flow->external = 0;
4346 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4347 			      dev_flow->handle, next);
4348 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4349 					 items_tx.items,
4350 					 actions_hairpin_tx.actions, error);
4351 		if (ret < 0)
4352 			goto error;
4353 	}
4354 	/*
4355 	 * Update the metadata register copy table. If extensive
4356 	 * metadata feature is enabled and registers are supported
4357 	 * we might create the extra rte_flow for each unique
4358 	 * MARK/FLAG action ID.
4359 	 *
4360 	 * The table is updated for ingress Flows only, because
4361 	 * the egress Flows belong to the different device and
4362 	 * copy table should be updated in peer NIC Rx domain.
4363 	 */
4364 	if (attr->ingress &&
4365 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4366 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4367 		if (ret)
4368 			goto error;
4369 	}
4370 	/*
4371 	 * If the flow is external (from application) OR device is started, then
4372 	 * the flow will be applied immediately.
4373 	 */
4374 	if (external || dev->data->dev_started) {
4375 		ret = flow_drv_apply(dev, flow, error);
4376 		if (ret < 0)
4377 			goto error;
4378 	}
4379 	if (list)
4380 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4381 			     flow, next);
4382 	flow_rxq_flags_set(dev, flow);
4383 	/* Nested flow creation index recovery. */
4384 	priv->flow_idx = priv->flow_nested_idx;
4385 	if (priv->flow_nested_idx)
4386 		priv->flow_nested_idx = 0;
4387 	return idx;
4388 error:
4389 	MLX5_ASSERT(flow);
4390 	ret = rte_errno; /* Save rte_errno before cleanup. */
4391 	flow_mreg_del_copy_action(dev, flow);
4392 	flow_drv_destroy(dev, flow);
4393 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4394 	rte_errno = ret; /* Restore rte_errno. */
4395 error_before_flow:
4396 	ret = rte_errno;
4397 	if (hairpin_id)
4398 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4399 				     hairpin_id);
4400 	rte_errno = ret;
4401 	priv->flow_idx = priv->flow_nested_idx;
4402 	if (priv->flow_nested_idx)
4403 		priv->flow_nested_idx = 0;
4404 	return 0;
4405 }
4406 
4407 /**
4408  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4409  * incoming packets to table 1.
4410  *
4411  * Other flow rules, requested for group n, will be created in
4412  * e-switch table n+1.
4413  * Jump action to e-switch group n will be created to group n+1.
4414  *
4415  * Used when working in switchdev mode, to utilise advantages of table 1
4416  * and above.
4417  *
4418  * @param dev
4419  *   Pointer to Ethernet device.
4420  *
4421  * @return
4422  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4423  */
4424 struct rte_flow *
4425 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4426 {
4427 	const struct rte_flow_attr attr = {
4428 		.group = 0,
4429 		.priority = 0,
4430 		.ingress = 1,
4431 		.egress = 0,
4432 		.transfer = 1,
4433 	};
4434 	const struct rte_flow_item pattern = {
4435 		.type = RTE_FLOW_ITEM_TYPE_END,
4436 	};
4437 	struct rte_flow_action_jump jump = {
4438 		.group = 1,
4439 	};
4440 	const struct rte_flow_action actions[] = {
4441 		{
4442 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4443 			.conf = &jump,
4444 		},
4445 		{
4446 			.type = RTE_FLOW_ACTION_TYPE_END,
4447 		},
4448 	};
4449 	struct mlx5_priv *priv = dev->data->dev_private;
4450 	struct rte_flow_error error;
4451 
4452 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4453 						   &attr, &pattern,
4454 						   actions, false, &error);
4455 }
4456 
4457 /**
4458  * Validate a flow supported by the NIC.
4459  *
4460  * @see rte_flow_validate()
4461  * @see rte_flow_ops
4462  */
4463 int
4464 mlx5_flow_validate(struct rte_eth_dev *dev,
4465 		   const struct rte_flow_attr *attr,
4466 		   const struct rte_flow_item items[],
4467 		   const struct rte_flow_action actions[],
4468 		   struct rte_flow_error *error)
4469 {
4470 	int hairpin_flow;
4471 
4472 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4473 	return flow_drv_validate(dev, attr, items, actions,
4474 				true, hairpin_flow, error);
4475 }
4476 
4477 /**
4478  * Create a flow.
4479  *
4480  * @see rte_flow_create()
4481  * @see rte_flow_ops
4482  */
4483 struct rte_flow *
4484 mlx5_flow_create(struct rte_eth_dev *dev,
4485 		 const struct rte_flow_attr *attr,
4486 		 const struct rte_flow_item items[],
4487 		 const struct rte_flow_action actions[],
4488 		 struct rte_flow_error *error)
4489 {
4490 	struct mlx5_priv *priv = dev->data->dev_private;
4491 
4492 	/*
4493 	 * If the device is not started yet, it is not allowed to created a
4494 	 * flow from application. PMD default flows and traffic control flows
4495 	 * are not affected.
4496 	 */
4497 	if (unlikely(!dev->data->dev_started)) {
4498 		rte_errno = ENODEV;
4499 		DRV_LOG(DEBUG, "port %u is not started when "
4500 			"inserting a flow", dev->data->port_id);
4501 		return NULL;
4502 	}
4503 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4504 				  attr, items, actions, true, error);
4505 }
4506 
4507 /**
4508  * Destroy a flow in a list.
4509  *
4510  * @param dev
4511  *   Pointer to Ethernet device.
4512  * @param list
4513  *   Pointer to the Indexed flow list. If this parameter NULL,
4514  *   there is no flow removal from the list. Be noted that as
4515  *   flow is add to the indexed list, memory of the indexed
4516  *   list points to maybe changed as flow destroyed.
4517  * @param[in] flow_idx
4518  *   Index of flow to destroy.
4519  */
4520 static void
4521 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4522 		  uint32_t flow_idx)
4523 {
4524 	struct mlx5_priv *priv = dev->data->dev_private;
4525 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4526 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4527 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4528 
4529 	if (!flow)
4530 		return;
4531 	/*
4532 	 * Update RX queue flags only if port is started, otherwise it is
4533 	 * already clean.
4534 	 */
4535 	if (dev->data->dev_started)
4536 		flow_rxq_flags_trim(dev, flow);
4537 	if (flow->hairpin_flow_id)
4538 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4539 				     flow->hairpin_flow_id);
4540 	flow_drv_destroy(dev, flow);
4541 	if (list)
4542 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4543 			     flow_idx, flow, next);
4544 	flow_mreg_del_copy_action(dev, flow);
4545 	if (flow->fdir) {
4546 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4547 			if (priv_fdir_flow->rix_flow == flow_idx)
4548 				break;
4549 		}
4550 		if (priv_fdir_flow) {
4551 			LIST_REMOVE(priv_fdir_flow, next);
4552 			rte_free(priv_fdir_flow->fdir);
4553 			rte_free(priv_fdir_flow);
4554 		}
4555 	}
4556 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4557 }
4558 
4559 /**
4560  * Destroy all flows.
4561  *
4562  * @param dev
4563  *   Pointer to Ethernet device.
4564  * @param list
4565  *   Pointer to the Indexed flow list.
4566  * @param active
4567  *   If flushing is called avtively.
4568  */
4569 void
4570 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4571 {
4572 	uint32_t num_flushed = 0;
4573 
4574 	while (*list) {
4575 		flow_list_destroy(dev, list, *list);
4576 		num_flushed++;
4577 	}
4578 	if (active) {
4579 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4580 			dev->data->port_id, num_flushed);
4581 	}
4582 }
4583 
4584 /**
4585  * Remove all flows.
4586  *
4587  * @param dev
4588  *   Pointer to Ethernet device.
4589  * @param list
4590  *   Pointer to the Indexed flow list.
4591  */
4592 void
4593 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4594 {
4595 	struct mlx5_priv *priv = dev->data->dev_private;
4596 	struct rte_flow *flow = NULL;
4597 	uint32_t idx;
4598 
4599 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4600 		      flow, next) {
4601 		flow_drv_remove(dev, flow);
4602 		flow_mreg_stop_copy_action(dev, flow);
4603 	}
4604 	flow_mreg_del_default_copy_action(dev);
4605 	flow_rxq_flags_clear(dev);
4606 }
4607 
4608 /**
4609  * Add all flows.
4610  *
4611  * @param dev
4612  *   Pointer to Ethernet device.
4613  * @param list
4614  *   Pointer to the Indexed flow list.
4615  *
4616  * @return
4617  *   0 on success, a negative errno value otherwise and rte_errno is set.
4618  */
4619 int
4620 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4621 {
4622 	struct mlx5_priv *priv = dev->data->dev_private;
4623 	struct rte_flow *flow = NULL;
4624 	struct rte_flow_error error;
4625 	uint32_t idx;
4626 	int ret = 0;
4627 
4628 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4629 	ret = flow_mreg_add_default_copy_action(dev, &error);
4630 	if (ret < 0)
4631 		return -rte_errno;
4632 	/* Apply Flows created by application. */
4633 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4634 		      flow, next) {
4635 		ret = flow_mreg_start_copy_action(dev, flow);
4636 		if (ret < 0)
4637 			goto error;
4638 		ret = flow_drv_apply(dev, flow, &error);
4639 		if (ret < 0)
4640 			goto error;
4641 		flow_rxq_flags_set(dev, flow);
4642 	}
4643 	return 0;
4644 error:
4645 	ret = rte_errno; /* Save rte_errno before cleanup. */
4646 	mlx5_flow_stop(dev, list);
4647 	rte_errno = ret; /* Restore rte_errno. */
4648 	return -rte_errno;
4649 }
4650 
4651 /**
4652  * Stop all default actions for flows.
4653  *
4654  * @param dev
4655  *   Pointer to Ethernet device.
4656  */
4657 void
4658 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4659 {
4660 	flow_mreg_del_default_copy_action(dev);
4661 }
4662 
4663 /**
4664  * Start all default actions for flows.
4665  *
4666  * @param dev
4667  *   Pointer to Ethernet device.
4668  * @return
4669  *   0 on success, a negative errno value otherwise and rte_errno is set.
4670  */
4671 int
4672 mlx5_flow_start_default(struct rte_eth_dev *dev)
4673 {
4674 	struct rte_flow_error error;
4675 
4676 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4677 	return flow_mreg_add_default_copy_action(dev, &error);
4678 }
4679 
4680 /**
4681  * Allocate intermediate resources for flow creation.
4682  *
4683  * @param dev
4684  *   Pointer to Ethernet device.
4685  */
4686 void
4687 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4688 {
4689 	struct mlx5_priv *priv = dev->data->dev_private;
4690 
4691 	if (!priv->inter_flows) {
4692 		priv->inter_flows = rte_calloc(__func__, 1,
4693 				    MLX5_NUM_MAX_DEV_FLOWS *
4694 				    sizeof(struct mlx5_flow) +
4695 				    (sizeof(struct mlx5_flow_rss_desc) +
4696 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4697 		if (!priv->inter_flows) {
4698 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4699 			return;
4700 		}
4701 	}
4702 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4703 			 [MLX5_NUM_MAX_DEV_FLOWS];
4704 	/* Reset the index. */
4705 	priv->flow_idx = 0;
4706 	priv->flow_nested_idx = 0;
4707 }
4708 
4709 /**
4710  * Free intermediate resources for flows.
4711  *
4712  * @param dev
4713  *   Pointer to Ethernet device.
4714  */
4715 void
4716 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4717 {
4718 	struct mlx5_priv *priv = dev->data->dev_private;
4719 
4720 	rte_free(priv->inter_flows);
4721 	priv->inter_flows = NULL;
4722 }
4723 
4724 /**
4725  * Verify the flow list is empty
4726  *
4727  * @param dev
4728  *  Pointer to Ethernet device.
4729  *
4730  * @return the number of flows not released.
4731  */
4732 int
4733 mlx5_flow_verify(struct rte_eth_dev *dev)
4734 {
4735 	struct mlx5_priv *priv = dev->data->dev_private;
4736 	struct rte_flow *flow;
4737 	uint32_t idx;
4738 	int ret = 0;
4739 
4740 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4741 		      flow, next) {
4742 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4743 			dev->data->port_id, (void *)flow);
4744 		++ret;
4745 	}
4746 	return ret;
4747 }
4748 
4749 /**
4750  * Enable default hairpin egress flow.
4751  *
4752  * @param dev
4753  *   Pointer to Ethernet device.
4754  * @param queue
4755  *   The queue index.
4756  *
4757  * @return
4758  *   0 on success, a negative errno value otherwise and rte_errno is set.
4759  */
4760 int
4761 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4762 			    uint32_t queue)
4763 {
4764 	struct mlx5_priv *priv = dev->data->dev_private;
4765 	const struct rte_flow_attr attr = {
4766 		.egress = 1,
4767 		.priority = 0,
4768 	};
4769 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4770 		.queue = queue,
4771 	};
4772 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4773 		.queue = UINT32_MAX,
4774 	};
4775 	struct rte_flow_item items[] = {
4776 		{
4777 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4778 			.spec = &queue_spec,
4779 			.last = NULL,
4780 			.mask = &queue_mask,
4781 		},
4782 		{
4783 			.type = RTE_FLOW_ITEM_TYPE_END,
4784 		},
4785 	};
4786 	struct rte_flow_action_jump jump = {
4787 		.group = MLX5_HAIRPIN_TX_TABLE,
4788 	};
4789 	struct rte_flow_action actions[2];
4790 	uint32_t flow_idx;
4791 	struct rte_flow_error error;
4792 
4793 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4794 	actions[0].conf = &jump;
4795 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4796 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4797 				&attr, items, actions, false, &error);
4798 	if (!flow_idx) {
4799 		DRV_LOG(DEBUG,
4800 			"Failed to create ctrl flow: rte_errno(%d),"
4801 			" type(%d), message(%s)",
4802 			rte_errno, error.type,
4803 			error.message ? error.message : " (no stated reason)");
4804 		return -rte_errno;
4805 	}
4806 	return 0;
4807 }
4808 
4809 /**
4810  * Enable a control flow configured from the control plane.
4811  *
4812  * @param dev
4813  *   Pointer to Ethernet device.
4814  * @param eth_spec
4815  *   An Ethernet flow spec to apply.
4816  * @param eth_mask
4817  *   An Ethernet flow mask to apply.
4818  * @param vlan_spec
4819  *   A VLAN flow spec to apply.
4820  * @param vlan_mask
4821  *   A VLAN flow mask to apply.
4822  *
4823  * @return
4824  *   0 on success, a negative errno value otherwise and rte_errno is set.
4825  */
4826 int
4827 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4828 		    struct rte_flow_item_eth *eth_spec,
4829 		    struct rte_flow_item_eth *eth_mask,
4830 		    struct rte_flow_item_vlan *vlan_spec,
4831 		    struct rte_flow_item_vlan *vlan_mask)
4832 {
4833 	struct mlx5_priv *priv = dev->data->dev_private;
4834 	const struct rte_flow_attr attr = {
4835 		.ingress = 1,
4836 		.priority = MLX5_FLOW_PRIO_RSVD,
4837 	};
4838 	struct rte_flow_item items[] = {
4839 		{
4840 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4841 			.spec = eth_spec,
4842 			.last = NULL,
4843 			.mask = eth_mask,
4844 		},
4845 		{
4846 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4847 					      RTE_FLOW_ITEM_TYPE_END,
4848 			.spec = vlan_spec,
4849 			.last = NULL,
4850 			.mask = vlan_mask,
4851 		},
4852 		{
4853 			.type = RTE_FLOW_ITEM_TYPE_END,
4854 		},
4855 	};
4856 	uint16_t queue[priv->reta_idx_n];
4857 	struct rte_flow_action_rss action_rss = {
4858 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4859 		.level = 0,
4860 		.types = priv->rss_conf.rss_hf,
4861 		.key_len = priv->rss_conf.rss_key_len,
4862 		.queue_num = priv->reta_idx_n,
4863 		.key = priv->rss_conf.rss_key,
4864 		.queue = queue,
4865 	};
4866 	struct rte_flow_action actions[] = {
4867 		{
4868 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4869 			.conf = &action_rss,
4870 		},
4871 		{
4872 			.type = RTE_FLOW_ACTION_TYPE_END,
4873 		},
4874 	};
4875 	uint32_t flow_idx;
4876 	struct rte_flow_error error;
4877 	unsigned int i;
4878 
4879 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4880 		return 0;
4881 	}
4882 	for (i = 0; i != priv->reta_idx_n; ++i)
4883 		queue[i] = (*priv->reta_idx)[i];
4884 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4885 				&attr, items, actions, false, &error);
4886 	if (!flow_idx)
4887 		return -rte_errno;
4888 	return 0;
4889 }
4890 
4891 /**
4892  * Enable a flow control configured from the control plane.
4893  *
4894  * @param dev
4895  *   Pointer to Ethernet device.
4896  * @param eth_spec
4897  *   An Ethernet flow spec to apply.
4898  * @param eth_mask
4899  *   An Ethernet flow mask to apply.
4900  *
4901  * @return
4902  *   0 on success, a negative errno value otherwise and rte_errno is set.
4903  */
4904 int
4905 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4906 	       struct rte_flow_item_eth *eth_spec,
4907 	       struct rte_flow_item_eth *eth_mask)
4908 {
4909 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4910 }
4911 
4912 /**
4913  * Destroy a flow.
4914  *
4915  * @see rte_flow_destroy()
4916  * @see rte_flow_ops
4917  */
4918 int
4919 mlx5_flow_destroy(struct rte_eth_dev *dev,
4920 		  struct rte_flow *flow,
4921 		  struct rte_flow_error *error __rte_unused)
4922 {
4923 	struct mlx5_priv *priv = dev->data->dev_private;
4924 
4925 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
4926 	return 0;
4927 }
4928 
4929 /**
4930  * Destroy all flows.
4931  *
4932  * @see rte_flow_flush()
4933  * @see rte_flow_ops
4934  */
4935 int
4936 mlx5_flow_flush(struct rte_eth_dev *dev,
4937 		struct rte_flow_error *error __rte_unused)
4938 {
4939 	struct mlx5_priv *priv = dev->data->dev_private;
4940 
4941 	mlx5_flow_list_flush(dev, &priv->flows, false);
4942 	return 0;
4943 }
4944 
4945 /**
4946  * Isolated mode.
4947  *
4948  * @see rte_flow_isolate()
4949  * @see rte_flow_ops
4950  */
4951 int
4952 mlx5_flow_isolate(struct rte_eth_dev *dev,
4953 		  int enable,
4954 		  struct rte_flow_error *error)
4955 {
4956 	struct mlx5_priv *priv = dev->data->dev_private;
4957 
4958 	if (dev->data->dev_started) {
4959 		rte_flow_error_set(error, EBUSY,
4960 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4961 				   NULL,
4962 				   "port must be stopped first");
4963 		return -rte_errno;
4964 	}
4965 	priv->isolated = !!enable;
4966 	if (enable)
4967 		dev->dev_ops = &mlx5_dev_ops_isolate;
4968 	else
4969 		dev->dev_ops = &mlx5_dev_ops;
4970 	return 0;
4971 }
4972 
4973 /**
4974  * Query a flow.
4975  *
4976  * @see rte_flow_query()
4977  * @see rte_flow_ops
4978  */
4979 static int
4980 flow_drv_query(struct rte_eth_dev *dev,
4981 	       uint32_t flow_idx,
4982 	       const struct rte_flow_action *actions,
4983 	       void *data,
4984 	       struct rte_flow_error *error)
4985 {
4986 	struct mlx5_priv *priv = dev->data->dev_private;
4987 	const struct mlx5_flow_driver_ops *fops;
4988 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4989 					       [MLX5_IPOOL_RTE_FLOW],
4990 					       flow_idx);
4991 	enum mlx5_flow_drv_type ftype;
4992 
4993 	if (!flow) {
4994 		return rte_flow_error_set(error, ENOENT,
4995 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4996 			  NULL,
4997 			  "invalid flow handle");
4998 	}
4999 	ftype = flow->drv_type;
5000 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5001 	fops = flow_get_drv_ops(ftype);
5002 
5003 	return fops->query(dev, flow, actions, data, error);
5004 }
5005 
5006 /**
5007  * Query a flow.
5008  *
5009  * @see rte_flow_query()
5010  * @see rte_flow_ops
5011  */
5012 int
5013 mlx5_flow_query(struct rte_eth_dev *dev,
5014 		struct rte_flow *flow,
5015 		const struct rte_flow_action *actions,
5016 		void *data,
5017 		struct rte_flow_error *error)
5018 {
5019 	int ret;
5020 
5021 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5022 			     error);
5023 	if (ret < 0)
5024 		return ret;
5025 	return 0;
5026 }
5027 
5028 /**
5029  * Convert a flow director filter to a generic flow.
5030  *
5031  * @param dev
5032  *   Pointer to Ethernet device.
5033  * @param fdir_filter
5034  *   Flow director filter to add.
5035  * @param attributes
5036  *   Generic flow parameters structure.
5037  *
5038  * @return
5039  *   0 on success, a negative errno value otherwise and rte_errno is set.
5040  */
5041 static int
5042 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5043 			 const struct rte_eth_fdir_filter *fdir_filter,
5044 			 struct mlx5_fdir *attributes)
5045 {
5046 	struct mlx5_priv *priv = dev->data->dev_private;
5047 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5048 	const struct rte_eth_fdir_masks *mask =
5049 		&dev->data->dev_conf.fdir_conf.mask;
5050 
5051 	/* Validate queue number. */
5052 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5053 		DRV_LOG(ERR, "port %u invalid queue number %d",
5054 			dev->data->port_id, fdir_filter->action.rx_queue);
5055 		rte_errno = EINVAL;
5056 		return -rte_errno;
5057 	}
5058 	attributes->attr.ingress = 1;
5059 	attributes->items[0] = (struct rte_flow_item) {
5060 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5061 		.spec = &attributes->l2,
5062 		.mask = &attributes->l2_mask,
5063 	};
5064 	switch (fdir_filter->action.behavior) {
5065 	case RTE_ETH_FDIR_ACCEPT:
5066 		attributes->actions[0] = (struct rte_flow_action){
5067 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5068 			.conf = &attributes->queue,
5069 		};
5070 		break;
5071 	case RTE_ETH_FDIR_REJECT:
5072 		attributes->actions[0] = (struct rte_flow_action){
5073 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5074 		};
5075 		break;
5076 	default:
5077 		DRV_LOG(ERR, "port %u invalid behavior %d",
5078 			dev->data->port_id,
5079 			fdir_filter->action.behavior);
5080 		rte_errno = ENOTSUP;
5081 		return -rte_errno;
5082 	}
5083 	attributes->queue.index = fdir_filter->action.rx_queue;
5084 	/* Handle L3. */
5085 	switch (fdir_filter->input.flow_type) {
5086 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5087 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5088 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5089 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5090 			.src_addr = input->flow.ip4_flow.src_ip,
5091 			.dst_addr = input->flow.ip4_flow.dst_ip,
5092 			.time_to_live = input->flow.ip4_flow.ttl,
5093 			.type_of_service = input->flow.ip4_flow.tos,
5094 		};
5095 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5096 			.src_addr = mask->ipv4_mask.src_ip,
5097 			.dst_addr = mask->ipv4_mask.dst_ip,
5098 			.time_to_live = mask->ipv4_mask.ttl,
5099 			.type_of_service = mask->ipv4_mask.tos,
5100 			.next_proto_id = mask->ipv4_mask.proto,
5101 		};
5102 		attributes->items[1] = (struct rte_flow_item){
5103 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5104 			.spec = &attributes->l3,
5105 			.mask = &attributes->l3_mask,
5106 		};
5107 		break;
5108 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5109 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5110 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5111 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5112 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5113 			.proto = input->flow.ipv6_flow.proto,
5114 		};
5115 
5116 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5117 		       input->flow.ipv6_flow.src_ip,
5118 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5119 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5120 		       input->flow.ipv6_flow.dst_ip,
5121 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5122 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5123 		       mask->ipv6_mask.src_ip,
5124 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5125 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5126 		       mask->ipv6_mask.dst_ip,
5127 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5128 		attributes->items[1] = (struct rte_flow_item){
5129 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5130 			.spec = &attributes->l3,
5131 			.mask = &attributes->l3_mask,
5132 		};
5133 		break;
5134 	default:
5135 		DRV_LOG(ERR, "port %u invalid flow type%d",
5136 			dev->data->port_id, fdir_filter->input.flow_type);
5137 		rte_errno = ENOTSUP;
5138 		return -rte_errno;
5139 	}
5140 	/* Handle L4. */
5141 	switch (fdir_filter->input.flow_type) {
5142 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5143 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5144 			.src_port = input->flow.udp4_flow.src_port,
5145 			.dst_port = input->flow.udp4_flow.dst_port,
5146 		};
5147 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5148 			.src_port = mask->src_port_mask,
5149 			.dst_port = mask->dst_port_mask,
5150 		};
5151 		attributes->items[2] = (struct rte_flow_item){
5152 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5153 			.spec = &attributes->l4,
5154 			.mask = &attributes->l4_mask,
5155 		};
5156 		break;
5157 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5158 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5159 			.src_port = input->flow.tcp4_flow.src_port,
5160 			.dst_port = input->flow.tcp4_flow.dst_port,
5161 		};
5162 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5163 			.src_port = mask->src_port_mask,
5164 			.dst_port = mask->dst_port_mask,
5165 		};
5166 		attributes->items[2] = (struct rte_flow_item){
5167 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5168 			.spec = &attributes->l4,
5169 			.mask = &attributes->l4_mask,
5170 		};
5171 		break;
5172 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5173 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5174 			.src_port = input->flow.udp6_flow.src_port,
5175 			.dst_port = input->flow.udp6_flow.dst_port,
5176 		};
5177 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5178 			.src_port = mask->src_port_mask,
5179 			.dst_port = mask->dst_port_mask,
5180 		};
5181 		attributes->items[2] = (struct rte_flow_item){
5182 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5183 			.spec = &attributes->l4,
5184 			.mask = &attributes->l4_mask,
5185 		};
5186 		break;
5187 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5188 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5189 			.src_port = input->flow.tcp6_flow.src_port,
5190 			.dst_port = input->flow.tcp6_flow.dst_port,
5191 		};
5192 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5193 			.src_port = mask->src_port_mask,
5194 			.dst_port = mask->dst_port_mask,
5195 		};
5196 		attributes->items[2] = (struct rte_flow_item){
5197 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5198 			.spec = &attributes->l4,
5199 			.mask = &attributes->l4_mask,
5200 		};
5201 		break;
5202 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5203 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5204 		break;
5205 	default:
5206 		DRV_LOG(ERR, "port %u invalid flow type%d",
5207 			dev->data->port_id, fdir_filter->input.flow_type);
5208 		rte_errno = ENOTSUP;
5209 		return -rte_errno;
5210 	}
5211 	return 0;
5212 }
5213 
5214 #define FLOW_FDIR_CMP(f1, f2, fld) \
5215 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5216 
5217 /**
5218  * Compare two FDIR flows. If items and actions are identical, the two flows are
5219  * regarded as same.
5220  *
5221  * @param dev
5222  *   Pointer to Ethernet device.
5223  * @param f1
5224  *   FDIR flow to compare.
5225  * @param f2
5226  *   FDIR flow to compare.
5227  *
5228  * @return
5229  *   Zero on match, 1 otherwise.
5230  */
5231 static int
5232 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5233 {
5234 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5235 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5236 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5237 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5238 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5239 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5240 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5241 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5242 		return 1;
5243 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5244 	    FLOW_FDIR_CMP(f1, f2, queue))
5245 		return 1;
5246 	return 0;
5247 }
5248 
5249 /**
5250  * Search device flow list to find out a matched FDIR flow.
5251  *
5252  * @param dev
5253  *   Pointer to Ethernet device.
5254  * @param fdir_flow
5255  *   FDIR flow to lookup.
5256  *
5257  * @return
5258  *   Index of flow if found, 0 otherwise.
5259  */
5260 static uint32_t
5261 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5262 {
5263 	struct mlx5_priv *priv = dev->data->dev_private;
5264 	uint32_t flow_idx = 0;
5265 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5266 
5267 	MLX5_ASSERT(fdir_flow);
5268 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5269 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5270 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5271 				dev->data->port_id, flow_idx);
5272 			flow_idx = priv_fdir_flow->rix_flow;
5273 			break;
5274 		}
5275 	}
5276 	return flow_idx;
5277 }
5278 
5279 /**
5280  * Add new flow director filter and store it in list.
5281  *
5282  * @param dev
5283  *   Pointer to Ethernet device.
5284  * @param fdir_filter
5285  *   Flow director filter to add.
5286  *
5287  * @return
5288  *   0 on success, a negative errno value otherwise and rte_errno is set.
5289  */
5290 static int
5291 flow_fdir_filter_add(struct rte_eth_dev *dev,
5292 		     const struct rte_eth_fdir_filter *fdir_filter)
5293 {
5294 	struct mlx5_priv *priv = dev->data->dev_private;
5295 	struct mlx5_fdir *fdir_flow;
5296 	struct rte_flow *flow;
5297 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5298 	uint32_t flow_idx;
5299 	int ret;
5300 
5301 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5302 	if (!fdir_flow) {
5303 		rte_errno = ENOMEM;
5304 		return -rte_errno;
5305 	}
5306 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5307 	if (ret)
5308 		goto error;
5309 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5310 	if (flow_idx) {
5311 		rte_errno = EEXIST;
5312 		goto error;
5313 	}
5314 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5315 				     0);
5316 	if (!priv_fdir_flow) {
5317 		rte_errno = ENOMEM;
5318 		goto error;
5319 	}
5320 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5321 				    fdir_flow->items, fdir_flow->actions, true,
5322 				    NULL);
5323 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5324 	if (!flow)
5325 		goto error;
5326 	flow->fdir = 1;
5327 	priv_fdir_flow->fdir = fdir_flow;
5328 	priv_fdir_flow->rix_flow = flow_idx;
5329 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5330 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5331 		dev->data->port_id, (void *)flow);
5332 	return 0;
5333 error:
5334 	rte_free(priv_fdir_flow);
5335 	rte_free(fdir_flow);
5336 	return -rte_errno;
5337 }
5338 
5339 /**
5340  * Delete specific filter.
5341  *
5342  * @param dev
5343  *   Pointer to Ethernet device.
5344  * @param fdir_filter
5345  *   Filter to be deleted.
5346  *
5347  * @return
5348  *   0 on success, a negative errno value otherwise and rte_errno is set.
5349  */
5350 static int
5351 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5352 			const struct rte_eth_fdir_filter *fdir_filter)
5353 {
5354 	struct mlx5_priv *priv = dev->data->dev_private;
5355 	uint32_t flow_idx;
5356 	struct mlx5_fdir fdir_flow = {
5357 		.attr.group = 0,
5358 	};
5359 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5360 	int ret;
5361 
5362 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5363 	if (ret)
5364 		return -rte_errno;
5365 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5366 		/* Find the fdir in priv list */
5367 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5368 			break;
5369 	}
5370 	if (!priv_fdir_flow)
5371 		return 0;
5372 	LIST_REMOVE(priv_fdir_flow, next);
5373 	flow_idx = priv_fdir_flow->rix_flow;
5374 	flow_list_destroy(dev, &priv->flows, flow_idx);
5375 	rte_free(priv_fdir_flow->fdir);
5376 	rte_free(priv_fdir_flow);
5377 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5378 		dev->data->port_id, flow_idx);
5379 	return 0;
5380 }
5381 
5382 /**
5383  * Update queue for specific filter.
5384  *
5385  * @param dev
5386  *   Pointer to Ethernet device.
5387  * @param fdir_filter
5388  *   Filter to be updated.
5389  *
5390  * @return
5391  *   0 on success, a negative errno value otherwise and rte_errno is set.
5392  */
5393 static int
5394 flow_fdir_filter_update(struct rte_eth_dev *dev,
5395 			const struct rte_eth_fdir_filter *fdir_filter)
5396 {
5397 	int ret;
5398 
5399 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5400 	if (ret)
5401 		return ret;
5402 	return flow_fdir_filter_add(dev, fdir_filter);
5403 }
5404 
5405 /**
5406  * Flush all filters.
5407  *
5408  * @param dev
5409  *   Pointer to Ethernet device.
5410  */
5411 static void
5412 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5413 {
5414 	struct mlx5_priv *priv = dev->data->dev_private;
5415 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5416 
5417 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5418 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5419 		LIST_REMOVE(priv_fdir_flow, next);
5420 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5421 		rte_free(priv_fdir_flow->fdir);
5422 		rte_free(priv_fdir_flow);
5423 	}
5424 }
5425 
5426 /**
5427  * Get flow director information.
5428  *
5429  * @param dev
5430  *   Pointer to Ethernet device.
5431  * @param[out] fdir_info
5432  *   Resulting flow director information.
5433  */
5434 static void
5435 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5436 {
5437 	struct rte_eth_fdir_masks *mask =
5438 		&dev->data->dev_conf.fdir_conf.mask;
5439 
5440 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5441 	fdir_info->guarant_spc = 0;
5442 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5443 	fdir_info->max_flexpayload = 0;
5444 	fdir_info->flow_types_mask[0] = 0;
5445 	fdir_info->flex_payload_unit = 0;
5446 	fdir_info->max_flex_payload_segment_num = 0;
5447 	fdir_info->flex_payload_limit = 0;
5448 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5449 }
5450 
5451 /**
5452  * Deal with flow director operations.
5453  *
5454  * @param dev
5455  *   Pointer to Ethernet device.
5456  * @param filter_op
5457  *   Operation to perform.
5458  * @param arg
5459  *   Pointer to operation-specific structure.
5460  *
5461  * @return
5462  *   0 on success, a negative errno value otherwise and rte_errno is set.
5463  */
5464 static int
5465 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5466 		    void *arg)
5467 {
5468 	enum rte_fdir_mode fdir_mode =
5469 		dev->data->dev_conf.fdir_conf.mode;
5470 
5471 	if (filter_op == RTE_ETH_FILTER_NOP)
5472 		return 0;
5473 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5474 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5475 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5476 			dev->data->port_id, fdir_mode);
5477 		rte_errno = EINVAL;
5478 		return -rte_errno;
5479 	}
5480 	switch (filter_op) {
5481 	case RTE_ETH_FILTER_ADD:
5482 		return flow_fdir_filter_add(dev, arg);
5483 	case RTE_ETH_FILTER_UPDATE:
5484 		return flow_fdir_filter_update(dev, arg);
5485 	case RTE_ETH_FILTER_DELETE:
5486 		return flow_fdir_filter_delete(dev, arg);
5487 	case RTE_ETH_FILTER_FLUSH:
5488 		flow_fdir_filter_flush(dev);
5489 		break;
5490 	case RTE_ETH_FILTER_INFO:
5491 		flow_fdir_info_get(dev, arg);
5492 		break;
5493 	default:
5494 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5495 			dev->data->port_id, filter_op);
5496 		rte_errno = EINVAL;
5497 		return -rte_errno;
5498 	}
5499 	return 0;
5500 }
5501 
5502 /**
5503  * Manage filter operations.
5504  *
5505  * @param dev
5506  *   Pointer to Ethernet device structure.
5507  * @param filter_type
5508  *   Filter type.
5509  * @param filter_op
5510  *   Operation to perform.
5511  * @param arg
5512  *   Pointer to operation-specific structure.
5513  *
5514  * @return
5515  *   0 on success, a negative errno value otherwise and rte_errno is set.
5516  */
5517 int
5518 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5519 		     enum rte_filter_type filter_type,
5520 		     enum rte_filter_op filter_op,
5521 		     void *arg)
5522 {
5523 	switch (filter_type) {
5524 	case RTE_ETH_FILTER_GENERIC:
5525 		if (filter_op != RTE_ETH_FILTER_GET) {
5526 			rte_errno = EINVAL;
5527 			return -rte_errno;
5528 		}
5529 		*(const void **)arg = &mlx5_flow_ops;
5530 		return 0;
5531 	case RTE_ETH_FILTER_FDIR:
5532 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5533 	default:
5534 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5535 			dev->data->port_id, filter_type);
5536 		rte_errno = ENOTSUP;
5537 		return -rte_errno;
5538 	}
5539 	return 0;
5540 }
5541 
5542 /**
5543  * Create the needed meter and suffix tables.
5544  *
5545  * @param[in] dev
5546  *   Pointer to Ethernet device.
5547  * @param[in] fm
5548  *   Pointer to the flow meter.
5549  *
5550  * @return
5551  *   Pointer to table set on success, NULL otherwise.
5552  */
5553 struct mlx5_meter_domains_infos *
5554 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5555 			  const struct mlx5_flow_meter *fm)
5556 {
5557 	const struct mlx5_flow_driver_ops *fops;
5558 
5559 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5560 	return fops->create_mtr_tbls(dev, fm);
5561 }
5562 
5563 /**
5564  * Destroy the meter table set.
5565  *
5566  * @param[in] dev
5567  *   Pointer to Ethernet device.
5568  * @param[in] tbl
5569  *   Pointer to the meter table set.
5570  *
5571  * @return
5572  *   0 on success.
5573  */
5574 int
5575 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5576 			   struct mlx5_meter_domains_infos *tbls)
5577 {
5578 	const struct mlx5_flow_driver_ops *fops;
5579 
5580 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5581 	return fops->destroy_mtr_tbls(dev, tbls);
5582 }
5583 
5584 /**
5585  * Create policer rules.
5586  *
5587  * @param[in] dev
5588  *   Pointer to Ethernet device.
5589  * @param[in] fm
5590  *   Pointer to flow meter structure.
5591  * @param[in] attr
5592  *   Pointer to flow attributes.
5593  *
5594  * @return
5595  *   0 on success, -1 otherwise.
5596  */
5597 int
5598 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5599 			       struct mlx5_flow_meter *fm,
5600 			       const struct rte_flow_attr *attr)
5601 {
5602 	const struct mlx5_flow_driver_ops *fops;
5603 
5604 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5605 	return fops->create_policer_rules(dev, fm, attr);
5606 }
5607 
5608 /**
5609  * Destroy policer rules.
5610  *
5611  * @param[in] fm
5612  *   Pointer to flow meter structure.
5613  * @param[in] attr
5614  *   Pointer to flow attributes.
5615  *
5616  * @return
5617  *   0 on success, -1 otherwise.
5618  */
5619 int
5620 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5621 				struct mlx5_flow_meter *fm,
5622 				const struct rte_flow_attr *attr)
5623 {
5624 	const struct mlx5_flow_driver_ops *fops;
5625 
5626 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5627 	return fops->destroy_policer_rules(dev, fm, attr);
5628 }
5629 
5630 /**
5631  * Allocate a counter.
5632  *
5633  * @param[in] dev
5634  *   Pointer to Ethernet device structure.
5635  *
5636  * @return
5637  *   Index to allocated counter  on success, 0 otherwise.
5638  */
5639 uint32_t
5640 mlx5_counter_alloc(struct rte_eth_dev *dev)
5641 {
5642 	const struct mlx5_flow_driver_ops *fops;
5643 	struct rte_flow_attr attr = { .transfer = 0 };
5644 
5645 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5646 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5647 		return fops->counter_alloc(dev);
5648 	}
5649 	DRV_LOG(ERR,
5650 		"port %u counter allocate is not supported.",
5651 		 dev->data->port_id);
5652 	return 0;
5653 }
5654 
5655 /**
5656  * Free a counter.
5657  *
5658  * @param[in] dev
5659  *   Pointer to Ethernet device structure.
5660  * @param[in] cnt
5661  *   Index to counter to be free.
5662  */
5663 void
5664 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5665 {
5666 	const struct mlx5_flow_driver_ops *fops;
5667 	struct rte_flow_attr attr = { .transfer = 0 };
5668 
5669 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5670 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5671 		fops->counter_free(dev, cnt);
5672 		return;
5673 	}
5674 	DRV_LOG(ERR,
5675 		"port %u counter free is not supported.",
5676 		 dev->data->port_id);
5677 }
5678 
5679 /**
5680  * Query counter statistics.
5681  *
5682  * @param[in] dev
5683  *   Pointer to Ethernet device structure.
5684  * @param[in] cnt
5685  *   Index to counter to query.
5686  * @param[in] clear
5687  *   Set to clear counter statistics.
5688  * @param[out] pkts
5689  *   The counter hits packets number to save.
5690  * @param[out] bytes
5691  *   The counter hits bytes number to save.
5692  *
5693  * @return
5694  *   0 on success, a negative errno value otherwise.
5695  */
5696 int
5697 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5698 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5699 {
5700 	const struct mlx5_flow_driver_ops *fops;
5701 	struct rte_flow_attr attr = { .transfer = 0 };
5702 
5703 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5704 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5705 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5706 	}
5707 	DRV_LOG(ERR,
5708 		"port %u counter query is not supported.",
5709 		 dev->data->port_id);
5710 	return -ENOTSUP;
5711 }
5712 
5713 #define MLX5_POOL_QUERY_FREQ_US 1000000
5714 
5715 /**
5716  * Set the periodic procedure for triggering asynchronous batch queries for all
5717  * the counter pools.
5718  *
5719  * @param[in] sh
5720  *   Pointer to mlx5_ibv_shared object.
5721  */
5722 void
5723 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5724 {
5725 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5726 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5727 	uint32_t us;
5728 
5729 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5730 	pools_n += rte_atomic16_read(&cont->n_valid);
5731 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5732 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5733 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5734 		sh->cmng.query_thread_on = 0;
5735 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5736 	} else {
5737 		sh->cmng.query_thread_on = 1;
5738 	}
5739 }
5740 
5741 /**
5742  * The periodic procedure for triggering asynchronous batch queries for all the
5743  * counter pools. This function is probably called by the host thread.
5744  *
5745  * @param[in] arg
5746  *   The parameter for the alarm process.
5747  */
5748 void
5749 mlx5_flow_query_alarm(void *arg)
5750 {
5751 	struct mlx5_ibv_shared *sh = arg;
5752 	struct mlx5_devx_obj *dcs;
5753 	uint16_t offset;
5754 	int ret;
5755 	uint8_t batch = sh->cmng.batch;
5756 	uint16_t pool_index = sh->cmng.pool_index;
5757 	struct mlx5_pools_container *cont;
5758 	struct mlx5_pools_container *mcont;
5759 	struct mlx5_flow_counter_pool *pool;
5760 
5761 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5762 		goto set_alarm;
5763 next_container:
5764 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5765 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5766 	/* Check if resize was done and need to flip a container. */
5767 	if (cont != mcont) {
5768 		if (cont->pools) {
5769 			/* Clean the old container. */
5770 			rte_free(cont->pools);
5771 			memset(cont, 0, sizeof(*cont));
5772 		}
5773 		rte_cio_wmb();
5774 		 /* Flip the host container. */
5775 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5776 		cont = mcont;
5777 	}
5778 	if (!cont->pools) {
5779 		/* 2 empty containers case is unexpected. */
5780 		if (unlikely(batch != sh->cmng.batch))
5781 			goto set_alarm;
5782 		batch ^= 0x1;
5783 		pool_index = 0;
5784 		goto next_container;
5785 	}
5786 	pool = cont->pools[pool_index];
5787 	if (pool->raw_hw)
5788 		/* There is a pool query in progress. */
5789 		goto set_alarm;
5790 	pool->raw_hw =
5791 		LIST_FIRST(&sh->cmng.free_stat_raws);
5792 	if (!pool->raw_hw)
5793 		/* No free counter statistics raw memory. */
5794 		goto set_alarm;
5795 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5796 							      (&pool->a64_dcs);
5797 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5798 	/*
5799 	 * Identify the counters released between query trigger and query
5800 	 * handle more effiecntly. The counter released in this gap period
5801 	 * should wait for a new round of query as the new arrived packets
5802 	 * will not be taken into account.
5803 	 */
5804 	rte_atomic64_add(&pool->start_query_gen, 1);
5805 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5806 					       offset, NULL, NULL,
5807 					       pool->raw_hw->mem_mng->dm->id,
5808 					       (void *)(uintptr_t)
5809 					       (pool->raw_hw->data + offset),
5810 					       sh->devx_comp,
5811 					       (uint64_t)(uintptr_t)pool);
5812 	if (ret) {
5813 		rte_atomic64_sub(&pool->start_query_gen, 1);
5814 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5815 			" %d", pool->min_dcs->id);
5816 		pool->raw_hw = NULL;
5817 		goto set_alarm;
5818 	}
5819 	pool->raw_hw->min_dcs_id = dcs->id;
5820 	LIST_REMOVE(pool->raw_hw, next);
5821 	sh->cmng.pending_queries++;
5822 	pool_index++;
5823 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5824 		batch ^= 0x1;
5825 		pool_index = 0;
5826 	}
5827 set_alarm:
5828 	sh->cmng.batch = batch;
5829 	sh->cmng.pool_index = pool_index;
5830 	mlx5_set_query_alarm(sh);
5831 }
5832 
5833 /**
5834  * Handler for the HW respond about ready values from an asynchronous batch
5835  * query. This function is probably called by the host thread.
5836  *
5837  * @param[in] sh
5838  *   The pointer to the shared IB device context.
5839  * @param[in] async_id
5840  *   The Devx async ID.
5841  * @param[in] status
5842  *   The status of the completion.
5843  */
5844 void
5845 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5846 				  uint64_t async_id, int status)
5847 {
5848 	struct mlx5_flow_counter_pool *pool =
5849 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5850 	struct mlx5_counter_stats_raw *raw_to_free;
5851 
5852 	if (unlikely(status)) {
5853 		rte_atomic64_sub(&pool->start_query_gen, 1);
5854 		raw_to_free = pool->raw_hw;
5855 	} else {
5856 		raw_to_free = pool->raw;
5857 		rte_spinlock_lock(&pool->sl);
5858 		pool->raw = pool->raw_hw;
5859 		rte_spinlock_unlock(&pool->sl);
5860 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5861 			    rte_atomic64_read(&pool->start_query_gen));
5862 		rte_atomic64_set(&pool->end_query_gen,
5863 				 rte_atomic64_read(&pool->start_query_gen));
5864 		/* Be sure the new raw counters data is updated in memory. */
5865 		rte_cio_wmb();
5866 	}
5867 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5868 	pool->raw_hw = NULL;
5869 	sh->cmng.pending_queries--;
5870 }
5871 
5872 /**
5873  * Translate the rte_flow group index to HW table value.
5874  *
5875  * @param[in] attributes
5876  *   Pointer to flow attributes
5877  * @param[in] external
5878  *   Value is part of flow rule created by request external to PMD.
5879  * @param[in] group
5880  *   rte_flow group index value.
5881  * @param[out] fdb_def_rule
5882  *   Whether fdb jump to table 1 is configured.
5883  * @param[out] table
5884  *   HW table value.
5885  * @param[out] error
5886  *   Pointer to error structure.
5887  *
5888  * @return
5889  *   0 on success, a negative errno value otherwise and rte_errno is set.
5890  */
5891 int
5892 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5893 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5894 			 struct rte_flow_error *error)
5895 {
5896 	if (attributes->transfer && external && fdb_def_rule) {
5897 		if (group == UINT32_MAX)
5898 			return rte_flow_error_set
5899 						(error, EINVAL,
5900 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5901 						 NULL,
5902 						 "group index not supported");
5903 		*table = group + 1;
5904 	} else {
5905 		*table = group;
5906 	}
5907 	return 0;
5908 }
5909 
5910 /**
5911  * Discover availability of metadata reg_c's.
5912  *
5913  * Iteratively use test flows to check availability.
5914  *
5915  * @param[in] dev
5916  *   Pointer to the Ethernet device structure.
5917  *
5918  * @return
5919  *   0 on success, a negative errno value otherwise and rte_errno is set.
5920  */
5921 int
5922 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5923 {
5924 	struct mlx5_priv *priv = dev->data->dev_private;
5925 	struct mlx5_dev_config *config = &priv->config;
5926 	enum modify_reg idx;
5927 	int n = 0;
5928 
5929 	/* reg_c[0] and reg_c[1] are reserved. */
5930 	config->flow_mreg_c[n++] = REG_C_0;
5931 	config->flow_mreg_c[n++] = REG_C_1;
5932 	/* Discover availability of other reg_c's. */
5933 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5934 		struct rte_flow_attr attr = {
5935 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5936 			.priority = MLX5_FLOW_PRIO_RSVD,
5937 			.ingress = 1,
5938 		};
5939 		struct rte_flow_item items[] = {
5940 			[0] = {
5941 				.type = RTE_FLOW_ITEM_TYPE_END,
5942 			},
5943 		};
5944 		struct rte_flow_action actions[] = {
5945 			[0] = {
5946 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5947 				.conf = &(struct mlx5_flow_action_copy_mreg){
5948 					.src = REG_C_1,
5949 					.dst = idx,
5950 				},
5951 			},
5952 			[1] = {
5953 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5954 				.conf = &(struct rte_flow_action_jump){
5955 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5956 				},
5957 			},
5958 			[2] = {
5959 				.type = RTE_FLOW_ACTION_TYPE_END,
5960 			},
5961 		};
5962 		uint32_t flow_idx;
5963 		struct rte_flow *flow;
5964 		struct rte_flow_error error;
5965 
5966 		if (!config->dv_flow_en)
5967 			break;
5968 		/* Create internal flow, validation skips copy action. */
5969 		flow_idx = flow_list_create(dev, NULL, &attr, items,
5970 					    actions, false, &error);
5971 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
5972 				      flow_idx);
5973 		if (!flow)
5974 			continue;
5975 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5976 			config->flow_mreg_c[n++] = idx;
5977 		flow_list_destroy(dev, NULL, flow_idx);
5978 	}
5979 	for (; n < MLX5_MREG_C_NUM; ++n)
5980 		config->flow_mreg_c[n] = REG_NONE;
5981 	return 0;
5982 }
5983 
5984 /**
5985  * Dump flow raw hw data to file
5986  *
5987  * @param[in] dev
5988  *    The pointer to Ethernet device.
5989  * @param[in] file
5990  *   A pointer to a file for output.
5991  * @param[out] error
5992  *   Perform verbose error reporting if not NULL. PMDs initialize this
5993  *   structure in case of error only.
5994  * @return
5995  *   0 on success, a nagative value otherwise.
5996  */
5997 int
5998 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5999 		   FILE *file,
6000 		   struct rte_flow_error *error __rte_unused)
6001 {
6002 	struct mlx5_priv *priv = dev->data->dev_private;
6003 	struct mlx5_ibv_shared *sh = priv->sh;
6004 
6005 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6006 				       sh->tx_domain, file);
6007 }
6008