xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision fa2d01c87d2b8b871e735114c76f637e7ee776bf)
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_cycles.h>
28 #include <rte_flow_driver.h>
29 #include <rte_malloc.h>
30 #include <rte_ip.h>
31 
32 #include <mlx5_glue.h>
33 #include <mlx5_devx_cmds.h>
34 #include <mlx5_prm.h>
35 
36 #include "mlx5_defs.h"
37 #include "mlx5.h"
38 #include "mlx5_flow.h"
39 #include "mlx5_rxtx.h"
40 
41 /* Dev ops structure defined in mlx5.c */
42 extern const struct eth_dev_ops mlx5_dev_ops;
43 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
44 
45 /** Device flow drivers. */
46 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
47 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
48 #endif
49 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
50 
51 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
52 
53 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
54 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
55 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
56 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
57 #endif
58 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
59 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
60 };
61 
62 enum mlx5_expansion {
63 	MLX5_EXPANSION_ROOT,
64 	MLX5_EXPANSION_ROOT_OUTER,
65 	MLX5_EXPANSION_ROOT_ETH_VLAN,
66 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
67 	MLX5_EXPANSION_OUTER_ETH,
68 	MLX5_EXPANSION_OUTER_ETH_VLAN,
69 	MLX5_EXPANSION_OUTER_VLAN,
70 	MLX5_EXPANSION_OUTER_IPV4,
71 	MLX5_EXPANSION_OUTER_IPV4_UDP,
72 	MLX5_EXPANSION_OUTER_IPV4_TCP,
73 	MLX5_EXPANSION_OUTER_IPV6,
74 	MLX5_EXPANSION_OUTER_IPV6_UDP,
75 	MLX5_EXPANSION_OUTER_IPV6_TCP,
76 	MLX5_EXPANSION_VXLAN,
77 	MLX5_EXPANSION_VXLAN_GPE,
78 	MLX5_EXPANSION_GRE,
79 	MLX5_EXPANSION_MPLS,
80 	MLX5_EXPANSION_ETH,
81 	MLX5_EXPANSION_ETH_VLAN,
82 	MLX5_EXPANSION_VLAN,
83 	MLX5_EXPANSION_IPV4,
84 	MLX5_EXPANSION_IPV4_UDP,
85 	MLX5_EXPANSION_IPV4_TCP,
86 	MLX5_EXPANSION_IPV6,
87 	MLX5_EXPANSION_IPV6_UDP,
88 	MLX5_EXPANSION_IPV6_TCP,
89 };
90 
91 /** Supported expansion of items. */
92 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
93 	[MLX5_EXPANSION_ROOT] = {
94 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
95 						 MLX5_EXPANSION_IPV4,
96 						 MLX5_EXPANSION_IPV6),
97 		.type = RTE_FLOW_ITEM_TYPE_END,
98 	},
99 	[MLX5_EXPANSION_ROOT_OUTER] = {
100 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
101 						 MLX5_EXPANSION_OUTER_IPV4,
102 						 MLX5_EXPANSION_OUTER_IPV6),
103 		.type = RTE_FLOW_ITEM_TYPE_END,
104 	},
105 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
106 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
107 		.type = RTE_FLOW_ITEM_TYPE_END,
108 	},
109 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
110 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
111 		.type = RTE_FLOW_ITEM_TYPE_END,
112 	},
113 	[MLX5_EXPANSION_OUTER_ETH] = {
114 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
115 						 MLX5_EXPANSION_OUTER_IPV6,
116 						 MLX5_EXPANSION_MPLS),
117 		.type = RTE_FLOW_ITEM_TYPE_ETH,
118 		.rss_types = 0,
119 	},
120 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
121 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
122 		.type = RTE_FLOW_ITEM_TYPE_ETH,
123 		.rss_types = 0,
124 	},
125 	[MLX5_EXPANSION_OUTER_VLAN] = {
126 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
127 						 MLX5_EXPANSION_OUTER_IPV6),
128 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
129 	},
130 	[MLX5_EXPANSION_OUTER_IPV4] = {
131 		.next = RTE_FLOW_EXPAND_RSS_NEXT
132 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
133 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
134 			 MLX5_EXPANSION_GRE,
135 			 MLX5_EXPANSION_IPV4,
136 			 MLX5_EXPANSION_IPV6),
137 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
138 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
139 			ETH_RSS_NONFRAG_IPV4_OTHER,
140 	},
141 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
142 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
143 						 MLX5_EXPANSION_VXLAN_GPE),
144 		.type = RTE_FLOW_ITEM_TYPE_UDP,
145 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
146 	},
147 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
148 		.type = RTE_FLOW_ITEM_TYPE_TCP,
149 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
150 	},
151 	[MLX5_EXPANSION_OUTER_IPV6] = {
152 		.next = RTE_FLOW_EXPAND_RSS_NEXT
153 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
154 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
155 			 MLX5_EXPANSION_IPV4,
156 			 MLX5_EXPANSION_IPV6),
157 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
158 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
159 			ETH_RSS_NONFRAG_IPV6_OTHER,
160 	},
161 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
162 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
163 						 MLX5_EXPANSION_VXLAN_GPE),
164 		.type = RTE_FLOW_ITEM_TYPE_UDP,
165 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
166 	},
167 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
168 		.type = RTE_FLOW_ITEM_TYPE_TCP,
169 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
170 	},
171 	[MLX5_EXPANSION_VXLAN] = {
172 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173 						 MLX5_EXPANSION_IPV4,
174 						 MLX5_EXPANSION_IPV6),
175 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
176 	},
177 	[MLX5_EXPANSION_VXLAN_GPE] = {
178 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
179 						 MLX5_EXPANSION_IPV4,
180 						 MLX5_EXPANSION_IPV6),
181 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
182 	},
183 	[MLX5_EXPANSION_GRE] = {
184 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
185 		.type = RTE_FLOW_ITEM_TYPE_GRE,
186 	},
187 	[MLX5_EXPANSION_MPLS] = {
188 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
189 						 MLX5_EXPANSION_IPV6),
190 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
191 	},
192 	[MLX5_EXPANSION_ETH] = {
193 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
194 						 MLX5_EXPANSION_IPV6),
195 		.type = RTE_FLOW_ITEM_TYPE_ETH,
196 	},
197 	[MLX5_EXPANSION_ETH_VLAN] = {
198 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
199 		.type = RTE_FLOW_ITEM_TYPE_ETH,
200 	},
201 	[MLX5_EXPANSION_VLAN] = {
202 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
203 						 MLX5_EXPANSION_IPV6),
204 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
205 	},
206 	[MLX5_EXPANSION_IPV4] = {
207 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
208 						 MLX5_EXPANSION_IPV4_TCP),
209 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
210 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
211 			ETH_RSS_NONFRAG_IPV4_OTHER,
212 	},
213 	[MLX5_EXPANSION_IPV4_UDP] = {
214 		.type = RTE_FLOW_ITEM_TYPE_UDP,
215 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
216 	},
217 	[MLX5_EXPANSION_IPV4_TCP] = {
218 		.type = RTE_FLOW_ITEM_TYPE_TCP,
219 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
220 	},
221 	[MLX5_EXPANSION_IPV6] = {
222 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
223 						 MLX5_EXPANSION_IPV6_TCP),
224 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
225 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
226 			ETH_RSS_NONFRAG_IPV6_OTHER,
227 	},
228 	[MLX5_EXPANSION_IPV6_UDP] = {
229 		.type = RTE_FLOW_ITEM_TYPE_UDP,
230 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
231 	},
232 	[MLX5_EXPANSION_IPV6_TCP] = {
233 		.type = RTE_FLOW_ITEM_TYPE_TCP,
234 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
235 	},
236 };
237 
238 static const struct rte_flow_ops mlx5_flow_ops = {
239 	.validate = mlx5_flow_validate,
240 	.create = mlx5_flow_create,
241 	.destroy = mlx5_flow_destroy,
242 	.flush = mlx5_flow_flush,
243 	.isolate = mlx5_flow_isolate,
244 	.query = mlx5_flow_query,
245 	.dev_dump = mlx5_flow_dev_dump,
246 	.get_aged_flows = mlx5_flow_get_aged_flows,
247 };
248 
249 /* Convert FDIR request to Generic flow. */
250 struct mlx5_fdir {
251 	struct rte_flow_attr attr;
252 	struct rte_flow_item items[4];
253 	struct rte_flow_item_eth l2;
254 	struct rte_flow_item_eth l2_mask;
255 	union {
256 		struct rte_flow_item_ipv4 ipv4;
257 		struct rte_flow_item_ipv6 ipv6;
258 	} l3;
259 	union {
260 		struct rte_flow_item_ipv4 ipv4;
261 		struct rte_flow_item_ipv6 ipv6;
262 	} l3_mask;
263 	union {
264 		struct rte_flow_item_udp udp;
265 		struct rte_flow_item_tcp tcp;
266 	} l4;
267 	union {
268 		struct rte_flow_item_udp udp;
269 		struct rte_flow_item_tcp tcp;
270 	} l4_mask;
271 	struct rte_flow_action actions[2];
272 	struct rte_flow_action_queue queue;
273 };
274 
275 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
276 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
277 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
278 };
279 
280 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
281 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
282 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
283 	{ 9, 10, 11 }, { 12, 13, 14 },
284 };
285 
286 /* Tunnel information. */
287 struct mlx5_flow_tunnel_info {
288 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
289 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
290 };
291 
292 static struct mlx5_flow_tunnel_info tunnels_info[] = {
293 	{
294 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
295 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
296 	},
297 	{
298 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
299 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
300 	},
301 	{
302 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
303 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
304 	},
305 	{
306 		.tunnel = MLX5_FLOW_LAYER_GRE,
307 		.ptype = RTE_PTYPE_TUNNEL_GRE,
308 	},
309 	{
310 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
311 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
312 	},
313 	{
314 		.tunnel = MLX5_FLOW_LAYER_MPLS,
315 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
316 	},
317 	{
318 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
319 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
320 	},
321 	{
322 		.tunnel = MLX5_FLOW_LAYER_IPIP,
323 		.ptype = RTE_PTYPE_TUNNEL_IP,
324 	},
325 	{
326 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
327 		.ptype = RTE_PTYPE_TUNNEL_IP,
328 	},
329 	{
330 		.tunnel = MLX5_FLOW_LAYER_GTP,
331 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
332 	},
333 };
334 
335 /**
336  * Translate tag ID to register.
337  *
338  * @param[in] dev
339  *   Pointer to the Ethernet device structure.
340  * @param[in] feature
341  *   The feature that request the register.
342  * @param[in] id
343  *   The request register ID.
344  * @param[out] error
345  *   Error description in case of any.
346  *
347  * @return
348  *   The request register on success, a negative errno
349  *   value otherwise and rte_errno is set.
350  */
351 int
352 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
353 		     enum mlx5_feature_name feature,
354 		     uint32_t id,
355 		     struct rte_flow_error *error)
356 {
357 	struct mlx5_priv *priv = dev->data->dev_private;
358 	struct mlx5_dev_config *config = &priv->config;
359 	enum modify_reg start_reg;
360 	bool skip_mtr_reg = false;
361 
362 	switch (feature) {
363 	case MLX5_HAIRPIN_RX:
364 		return REG_B;
365 	case MLX5_HAIRPIN_TX:
366 		return REG_A;
367 	case MLX5_METADATA_RX:
368 		switch (config->dv_xmeta_en) {
369 		case MLX5_XMETA_MODE_LEGACY:
370 			return REG_B;
371 		case MLX5_XMETA_MODE_META16:
372 			return REG_C_0;
373 		case MLX5_XMETA_MODE_META32:
374 			return REG_C_1;
375 		}
376 		break;
377 	case MLX5_METADATA_TX:
378 		return REG_A;
379 	case MLX5_METADATA_FDB:
380 		switch (config->dv_xmeta_en) {
381 		case MLX5_XMETA_MODE_LEGACY:
382 			return REG_NONE;
383 		case MLX5_XMETA_MODE_META16:
384 			return REG_C_0;
385 		case MLX5_XMETA_MODE_META32:
386 			return REG_C_1;
387 		}
388 		break;
389 	case MLX5_FLOW_MARK:
390 		switch (config->dv_xmeta_en) {
391 		case MLX5_XMETA_MODE_LEGACY:
392 			return REG_NONE;
393 		case MLX5_XMETA_MODE_META16:
394 			return REG_C_1;
395 		case MLX5_XMETA_MODE_META32:
396 			return REG_C_0;
397 		}
398 		break;
399 	case MLX5_MTR_SFX:
400 		/*
401 		 * If meter color and flow match share one register, flow match
402 		 * should use the meter color register for match.
403 		 */
404 		if (priv->mtr_reg_share)
405 			return priv->mtr_color_reg;
406 		else
407 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
408 			       REG_C_3;
409 	case MLX5_MTR_COLOR:
410 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
411 		return priv->mtr_color_reg;
412 	case MLX5_COPY_MARK:
413 		/*
414 		 * Metadata COPY_MARK register using is in meter suffix sub
415 		 * flow while with meter. It's safe to share the same register.
416 		 */
417 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
418 	case MLX5_APP_TAG:
419 		/*
420 		 * If meter is enable, it will engage the register for color
421 		 * match and flow match. If meter color match is not using the
422 		 * REG_C_2, need to skip the REG_C_x be used by meter color
423 		 * match.
424 		 * If meter is disable, free to use all available registers.
425 		 */
426 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
427 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
428 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
429 		if (id > (REG_C_7 - start_reg))
430 			return rte_flow_error_set(error, EINVAL,
431 						  RTE_FLOW_ERROR_TYPE_ITEM,
432 						  NULL, "invalid tag id");
433 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
434 			return rte_flow_error_set(error, ENOTSUP,
435 						  RTE_FLOW_ERROR_TYPE_ITEM,
436 						  NULL, "unsupported tag id");
437 		/*
438 		 * This case means meter is using the REG_C_x great than 2.
439 		 * Take care not to conflict with meter color REG_C_x.
440 		 * If the available index REG_C_y >= REG_C_x, skip the
441 		 * color register.
442 		 */
443 		if (skip_mtr_reg && config->flow_mreg_c
444 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
445 			if (config->flow_mreg_c
446 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
447 				return config->flow_mreg_c
448 					       [id + 1 + start_reg - REG_C_0];
449 			return rte_flow_error_set(error, ENOTSUP,
450 						  RTE_FLOW_ERROR_TYPE_ITEM,
451 						  NULL, "unsupported tag id");
452 		}
453 		return config->flow_mreg_c[id + start_reg - REG_C_0];
454 	}
455 	MLX5_ASSERT(false);
456 	return rte_flow_error_set(error, EINVAL,
457 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
458 				  NULL, "invalid feature name");
459 }
460 
461 /**
462  * Check extensive flow metadata register support.
463  *
464  * @param dev
465  *   Pointer to rte_eth_dev structure.
466  *
467  * @return
468  *   True if device supports extensive flow metadata register, otherwise false.
469  */
470 bool
471 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
472 {
473 	struct mlx5_priv *priv = dev->data->dev_private;
474 	struct mlx5_dev_config *config = &priv->config;
475 
476 	/*
477 	 * Having available reg_c can be regarded inclusively as supporting
478 	 * extensive flow metadata register, which could mean,
479 	 * - metadata register copy action by modify header.
480 	 * - 16 modify header actions is supported.
481 	 * - reg_c's are preserved across different domain (FDB and NIC) on
482 	 *   packet loopback by flow lookup miss.
483 	 */
484 	return config->flow_mreg_c[2] != REG_NONE;
485 }
486 
487 /**
488  * Discover the maximum number of priority available.
489  *
490  * @param[in] dev
491  *   Pointer to the Ethernet device structure.
492  *
493  * @return
494  *   number of supported flow priority on success, a negative errno
495  *   value otherwise and rte_errno is set.
496  */
497 int
498 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
499 {
500 	struct mlx5_priv *priv = dev->data->dev_private;
501 	struct {
502 		struct ibv_flow_attr attr;
503 		struct ibv_flow_spec_eth eth;
504 		struct ibv_flow_spec_action_drop drop;
505 	} flow_attr = {
506 		.attr = {
507 			.num_of_specs = 2,
508 			.port = (uint8_t)priv->ibv_port,
509 		},
510 		.eth = {
511 			.type = IBV_FLOW_SPEC_ETH,
512 			.size = sizeof(struct ibv_flow_spec_eth),
513 		},
514 		.drop = {
515 			.size = sizeof(struct ibv_flow_spec_action_drop),
516 			.type = IBV_FLOW_SPEC_ACTION_DROP,
517 		},
518 	};
519 	struct ibv_flow *flow;
520 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
521 	uint16_t vprio[] = { 8, 16 };
522 	int i;
523 	int priority = 0;
524 
525 	if (!drop) {
526 		rte_errno = ENOTSUP;
527 		return -rte_errno;
528 	}
529 	for (i = 0; i != RTE_DIM(vprio); i++) {
530 		flow_attr.attr.priority = vprio[i] - 1;
531 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
532 		if (!flow)
533 			break;
534 		claim_zero(mlx5_glue->destroy_flow(flow));
535 		priority = vprio[i];
536 	}
537 	mlx5_hrxq_drop_release(dev);
538 	switch (priority) {
539 	case 8:
540 		priority = RTE_DIM(priority_map_3);
541 		break;
542 	case 16:
543 		priority = RTE_DIM(priority_map_5);
544 		break;
545 	default:
546 		rte_errno = ENOTSUP;
547 		DRV_LOG(ERR,
548 			"port %u verbs maximum priority: %d expected 8/16",
549 			dev->data->port_id, priority);
550 		return -rte_errno;
551 	}
552 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
553 		dev->data->port_id, priority);
554 	return priority;
555 }
556 
557 /**
558  * Adjust flow priority based on the highest layer and the request priority.
559  *
560  * @param[in] dev
561  *   Pointer to the Ethernet device structure.
562  * @param[in] priority
563  *   The rule base priority.
564  * @param[in] subpriority
565  *   The priority based on the items.
566  *
567  * @return
568  *   The new priority.
569  */
570 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
571 				   uint32_t subpriority)
572 {
573 	uint32_t res = 0;
574 	struct mlx5_priv *priv = dev->data->dev_private;
575 
576 	switch (priv->config.flow_prio) {
577 	case RTE_DIM(priority_map_3):
578 		res = priority_map_3[priority][subpriority];
579 		break;
580 	case RTE_DIM(priority_map_5):
581 		res = priority_map_5[priority][subpriority];
582 		break;
583 	}
584 	return  res;
585 }
586 
587 /**
588  * Verify the @p item specifications (spec, last, mask) are compatible with the
589  * NIC capabilities.
590  *
591  * @param[in] item
592  *   Item specification.
593  * @param[in] mask
594  *   @p item->mask or flow default bit-masks.
595  * @param[in] nic_mask
596  *   Bit-masks covering supported fields by the NIC to compare with user mask.
597  * @param[in] size
598  *   Bit-masks size in bytes.
599  * @param[out] error
600  *   Pointer to error structure.
601  *
602  * @return
603  *   0 on success, a negative errno value otherwise and rte_errno is set.
604  */
605 int
606 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
607 			  const uint8_t *mask,
608 			  const uint8_t *nic_mask,
609 			  unsigned int size,
610 			  struct rte_flow_error *error)
611 {
612 	unsigned int i;
613 
614 	MLX5_ASSERT(nic_mask);
615 	for (i = 0; i < size; ++i)
616 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
617 			return rte_flow_error_set(error, ENOTSUP,
618 						  RTE_FLOW_ERROR_TYPE_ITEM,
619 						  item,
620 						  "mask enables non supported"
621 						  " bits");
622 	if (!item->spec && (item->mask || item->last))
623 		return rte_flow_error_set(error, EINVAL,
624 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
625 					  "mask/last without a spec is not"
626 					  " supported");
627 	if (item->spec && item->last) {
628 		uint8_t spec[size];
629 		uint8_t last[size];
630 		unsigned int i;
631 		int ret;
632 
633 		for (i = 0; i < size; ++i) {
634 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
635 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
636 		}
637 		ret = memcmp(spec, last, size);
638 		if (ret != 0)
639 			return rte_flow_error_set(error, EINVAL,
640 						  RTE_FLOW_ERROR_TYPE_ITEM,
641 						  item,
642 						  "range is not valid");
643 	}
644 	return 0;
645 }
646 
647 /**
648  * Adjust the hash fields according to the @p flow information.
649  *
650  * @param[in] dev_flow.
651  *   Pointer to the mlx5_flow.
652  * @param[in] tunnel
653  *   1 when the hash field is for a tunnel item.
654  * @param[in] layer_types
655  *   ETH_RSS_* types.
656  * @param[in] hash_fields
657  *   Item hash fields.
658  *
659  * @return
660  *   The hash fields that should be used.
661  */
662 uint64_t
663 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
664 			    int tunnel __rte_unused, uint64_t layer_types,
665 			    uint64_t hash_fields)
666 {
667 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
668 	int rss_request_inner = rss_desc->level >= 2;
669 
670 	/* Check RSS hash level for tunnel. */
671 	if (tunnel && rss_request_inner)
672 		hash_fields |= IBV_RX_HASH_INNER;
673 	else if (tunnel || rss_request_inner)
674 		return 0;
675 #endif
676 	/* Check if requested layer matches RSS hash fields. */
677 	if (!(rss_desc->types & layer_types))
678 		return 0;
679 	return hash_fields;
680 }
681 
682 /**
683  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
684  * if several tunnel rules are used on this queue, the tunnel ptype will be
685  * cleared.
686  *
687  * @param rxq_ctrl
688  *   Rx queue to update.
689  */
690 static void
691 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
692 {
693 	unsigned int i;
694 	uint32_t tunnel_ptype = 0;
695 
696 	/* Look up for the ptype to use. */
697 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
698 		if (!rxq_ctrl->flow_tunnels_n[i])
699 			continue;
700 		if (!tunnel_ptype) {
701 			tunnel_ptype = tunnels_info[i].ptype;
702 		} else {
703 			tunnel_ptype = 0;
704 			break;
705 		}
706 	}
707 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
708 }
709 
710 /**
711  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
712  * flow.
713  *
714  * @param[in] dev
715  *   Pointer to the Ethernet device structure.
716  * @param[in] dev_handle
717  *   Pointer to device flow handle structure.
718  */
719 static void
720 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
721 		       struct mlx5_flow_handle *dev_handle)
722 {
723 	struct mlx5_priv *priv = dev->data->dev_private;
724 	const int mark = dev_handle->mark;
725 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
726 	struct mlx5_hrxq *hrxq;
727 	unsigned int i;
728 
729 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
730 		return;
731 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
732 			      dev_handle->rix_hrxq);
733 	if (!hrxq)
734 		return;
735 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
736 		int idx = hrxq->ind_table->queues[i];
737 		struct mlx5_rxq_ctrl *rxq_ctrl =
738 			container_of((*priv->rxqs)[idx],
739 				     struct mlx5_rxq_ctrl, rxq);
740 
741 		/*
742 		 * To support metadata register copy on Tx loopback,
743 		 * this must be always enabled (metadata may arive
744 		 * from other port - not from local flows only.
745 		 */
746 		if (priv->config.dv_flow_en &&
747 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
748 		    mlx5_flow_ext_mreg_supported(dev)) {
749 			rxq_ctrl->rxq.mark = 1;
750 			rxq_ctrl->flow_mark_n = 1;
751 		} else if (mark) {
752 			rxq_ctrl->rxq.mark = 1;
753 			rxq_ctrl->flow_mark_n++;
754 		}
755 		if (tunnel) {
756 			unsigned int j;
757 
758 			/* Increase the counter matching the flow. */
759 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
760 				if ((tunnels_info[j].tunnel &
761 				     dev_handle->layers) ==
762 				    tunnels_info[j].tunnel) {
763 					rxq_ctrl->flow_tunnels_n[j]++;
764 					break;
765 				}
766 			}
767 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
768 		}
769 	}
770 }
771 
772 /**
773  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
774  *
775  * @param[in] dev
776  *   Pointer to the Ethernet device structure.
777  * @param[in] flow
778  *   Pointer to flow structure.
779  */
780 static void
781 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
782 {
783 	struct mlx5_priv *priv = dev->data->dev_private;
784 	uint32_t handle_idx;
785 	struct mlx5_flow_handle *dev_handle;
786 
787 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
788 		       handle_idx, dev_handle, next)
789 		flow_drv_rxq_flags_set(dev, dev_handle);
790 }
791 
792 /**
793  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
794  * device flow if no other flow uses it with the same kind of request.
795  *
796  * @param dev
797  *   Pointer to Ethernet device.
798  * @param[in] dev_handle
799  *   Pointer to the device flow handle structure.
800  */
801 static void
802 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
803 			struct mlx5_flow_handle *dev_handle)
804 {
805 	struct mlx5_priv *priv = dev->data->dev_private;
806 	const int mark = dev_handle->mark;
807 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
808 	struct mlx5_hrxq *hrxq;
809 	unsigned int i;
810 
811 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
812 		return;
813 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
814 			      dev_handle->rix_hrxq);
815 	if (!hrxq)
816 		return;
817 	MLX5_ASSERT(dev->data->dev_started);
818 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
819 		int idx = hrxq->ind_table->queues[i];
820 		struct mlx5_rxq_ctrl *rxq_ctrl =
821 			container_of((*priv->rxqs)[idx],
822 				     struct mlx5_rxq_ctrl, rxq);
823 
824 		if (priv->config.dv_flow_en &&
825 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
826 		    mlx5_flow_ext_mreg_supported(dev)) {
827 			rxq_ctrl->rxq.mark = 1;
828 			rxq_ctrl->flow_mark_n = 1;
829 		} else if (mark) {
830 			rxq_ctrl->flow_mark_n--;
831 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
832 		}
833 		if (tunnel) {
834 			unsigned int j;
835 
836 			/* Decrease the counter matching the flow. */
837 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
838 				if ((tunnels_info[j].tunnel &
839 				     dev_handle->layers) ==
840 				    tunnels_info[j].tunnel) {
841 					rxq_ctrl->flow_tunnels_n[j]--;
842 					break;
843 				}
844 			}
845 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
846 		}
847 	}
848 }
849 
850 /**
851  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
852  * @p flow if no other flow uses it with the same kind of request.
853  *
854  * @param dev
855  *   Pointer to Ethernet device.
856  * @param[in] flow
857  *   Pointer to the flow.
858  */
859 static void
860 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
861 {
862 	struct mlx5_priv *priv = dev->data->dev_private;
863 	uint32_t handle_idx;
864 	struct mlx5_flow_handle *dev_handle;
865 
866 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
867 		       handle_idx, dev_handle, next)
868 		flow_drv_rxq_flags_trim(dev, dev_handle);
869 }
870 
871 /**
872  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
873  *
874  * @param dev
875  *   Pointer to Ethernet device.
876  */
877 static void
878 flow_rxq_flags_clear(struct rte_eth_dev *dev)
879 {
880 	struct mlx5_priv *priv = dev->data->dev_private;
881 	unsigned int i;
882 
883 	for (i = 0; i != priv->rxqs_n; ++i) {
884 		struct mlx5_rxq_ctrl *rxq_ctrl;
885 		unsigned int j;
886 
887 		if (!(*priv->rxqs)[i])
888 			continue;
889 		rxq_ctrl = container_of((*priv->rxqs)[i],
890 					struct mlx5_rxq_ctrl, rxq);
891 		rxq_ctrl->flow_mark_n = 0;
892 		rxq_ctrl->rxq.mark = 0;
893 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
894 			rxq_ctrl->flow_tunnels_n[j] = 0;
895 		rxq_ctrl->rxq.tunnel = 0;
896 	}
897 }
898 
899 /**
900  * Set the Rx queue dynamic metadata (mask and offset) for a flow
901  *
902  * @param[in] dev
903  *   Pointer to the Ethernet device structure.
904  */
905 void
906 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
907 {
908 	struct mlx5_priv *priv = dev->data->dev_private;
909 	struct mlx5_rxq_data *data;
910 	unsigned int i;
911 
912 	for (i = 0; i != priv->rxqs_n; ++i) {
913 		if (!(*priv->rxqs)[i])
914 			continue;
915 		data = (*priv->rxqs)[i];
916 		if (!rte_flow_dynf_metadata_avail()) {
917 			data->dynf_meta = 0;
918 			data->flow_meta_mask = 0;
919 			data->flow_meta_offset = -1;
920 		} else {
921 			data->dynf_meta = 1;
922 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
923 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
924 		}
925 	}
926 }
927 
928 /*
929  * return a pointer to the desired action in the list of actions.
930  *
931  * @param[in] actions
932  *   The list of actions to search the action in.
933  * @param[in] action
934  *   The action to find.
935  *
936  * @return
937  *   Pointer to the action in the list, if found. NULL otherwise.
938  */
939 const struct rte_flow_action *
940 mlx5_flow_find_action(const struct rte_flow_action *actions,
941 		      enum rte_flow_action_type action)
942 {
943 	if (actions == NULL)
944 		return NULL;
945 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
946 		if (actions->type == action)
947 			return actions;
948 	return NULL;
949 }
950 
951 /*
952  * Validate the flag action.
953  *
954  * @param[in] action_flags
955  *   Bit-fields that holds the actions detected until now.
956  * @param[in] attr
957  *   Attributes of flow that includes this action.
958  * @param[out] error
959  *   Pointer to error structure.
960  *
961  * @return
962  *   0 on success, a negative errno value otherwise and rte_errno is set.
963  */
964 int
965 mlx5_flow_validate_action_flag(uint64_t action_flags,
966 			       const struct rte_flow_attr *attr,
967 			       struct rte_flow_error *error)
968 {
969 	if (action_flags & MLX5_FLOW_ACTION_MARK)
970 		return rte_flow_error_set(error, EINVAL,
971 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
972 					  "can't mark and flag in same flow");
973 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
974 		return rte_flow_error_set(error, EINVAL,
975 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 					  "can't have 2 flag"
977 					  " actions in same flow");
978 	if (attr->egress)
979 		return rte_flow_error_set(error, ENOTSUP,
980 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
981 					  "flag action not supported for "
982 					  "egress");
983 	return 0;
984 }
985 
986 /*
987  * Validate the mark action.
988  *
989  * @param[in] action
990  *   Pointer to the queue action.
991  * @param[in] action_flags
992  *   Bit-fields that holds the actions detected until now.
993  * @param[in] attr
994  *   Attributes of flow that includes this action.
995  * @param[out] error
996  *   Pointer to error structure.
997  *
998  * @return
999  *   0 on success, a negative errno value otherwise and rte_errno is set.
1000  */
1001 int
1002 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1003 			       uint64_t action_flags,
1004 			       const struct rte_flow_attr *attr,
1005 			       struct rte_flow_error *error)
1006 {
1007 	const struct rte_flow_action_mark *mark = action->conf;
1008 
1009 	if (!mark)
1010 		return rte_flow_error_set(error, EINVAL,
1011 					  RTE_FLOW_ERROR_TYPE_ACTION,
1012 					  action,
1013 					  "configuration cannot be null");
1014 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1015 		return rte_flow_error_set(error, EINVAL,
1016 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1017 					  &mark->id,
1018 					  "mark id must in 0 <= id < "
1019 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1020 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1021 		return rte_flow_error_set(error, EINVAL,
1022 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1023 					  "can't flag and mark in same flow");
1024 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1025 		return rte_flow_error_set(error, EINVAL,
1026 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1027 					  "can't have 2 mark actions in same"
1028 					  " flow");
1029 	if (attr->egress)
1030 		return rte_flow_error_set(error, ENOTSUP,
1031 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1032 					  "mark action not supported for "
1033 					  "egress");
1034 	return 0;
1035 }
1036 
1037 /*
1038  * Validate the drop action.
1039  *
1040  * @param[in] action_flags
1041  *   Bit-fields that holds the actions detected until now.
1042  * @param[in] attr
1043  *   Attributes of flow that includes this action.
1044  * @param[out] error
1045  *   Pointer to error structure.
1046  *
1047  * @return
1048  *   0 on success, a negative errno value otherwise and rte_errno is set.
1049  */
1050 int
1051 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1052 			       const struct rte_flow_attr *attr,
1053 			       struct rte_flow_error *error)
1054 {
1055 	if (attr->egress)
1056 		return rte_flow_error_set(error, ENOTSUP,
1057 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1058 					  "drop action not supported for "
1059 					  "egress");
1060 	return 0;
1061 }
1062 
1063 /*
1064  * Validate the queue action.
1065  *
1066  * @param[in] action
1067  *   Pointer to the queue action.
1068  * @param[in] action_flags
1069  *   Bit-fields that holds the actions detected until now.
1070  * @param[in] dev
1071  *   Pointer to the Ethernet device structure.
1072  * @param[in] attr
1073  *   Attributes of flow that includes this action.
1074  * @param[out] error
1075  *   Pointer to error structure.
1076  *
1077  * @return
1078  *   0 on success, a negative errno value otherwise and rte_errno is set.
1079  */
1080 int
1081 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1082 				uint64_t action_flags,
1083 				struct rte_eth_dev *dev,
1084 				const struct rte_flow_attr *attr,
1085 				struct rte_flow_error *error)
1086 {
1087 	struct mlx5_priv *priv = dev->data->dev_private;
1088 	const struct rte_flow_action_queue *queue = action->conf;
1089 
1090 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1091 		return rte_flow_error_set(error, EINVAL,
1092 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1093 					  "can't have 2 fate actions in"
1094 					  " same flow");
1095 	if (!priv->rxqs_n)
1096 		return rte_flow_error_set(error, EINVAL,
1097 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1098 					  NULL, "No Rx queues configured");
1099 	if (queue->index >= priv->rxqs_n)
1100 		return rte_flow_error_set(error, EINVAL,
1101 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1102 					  &queue->index,
1103 					  "queue index out of range");
1104 	if (!(*priv->rxqs)[queue->index])
1105 		return rte_flow_error_set(error, EINVAL,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &queue->index,
1108 					  "queue is not configured");
1109 	if (attr->egress)
1110 		return rte_flow_error_set(error, ENOTSUP,
1111 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1112 					  "queue action not supported for "
1113 					  "egress");
1114 	return 0;
1115 }
1116 
1117 /*
1118  * Validate the rss action.
1119  *
1120  * @param[in] action
1121  *   Pointer to the queue action.
1122  * @param[in] action_flags
1123  *   Bit-fields that holds the actions detected until now.
1124  * @param[in] dev
1125  *   Pointer to the Ethernet device structure.
1126  * @param[in] attr
1127  *   Attributes of flow that includes this action.
1128  * @param[in] item_flags
1129  *   Items that were detected.
1130  * @param[out] error
1131  *   Pointer to error structure.
1132  *
1133  * @return
1134  *   0 on success, a negative errno value otherwise and rte_errno is set.
1135  */
1136 int
1137 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1138 			      uint64_t action_flags,
1139 			      struct rte_eth_dev *dev,
1140 			      const struct rte_flow_attr *attr,
1141 			      uint64_t item_flags,
1142 			      struct rte_flow_error *error)
1143 {
1144 	struct mlx5_priv *priv = dev->data->dev_private;
1145 	const struct rte_flow_action_rss *rss = action->conf;
1146 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1147 	unsigned int i;
1148 
1149 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1152 					  "can't have 2 fate actions"
1153 					  " in same flow");
1154 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1155 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1156 		return rte_flow_error_set(error, ENOTSUP,
1157 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1158 					  &rss->func,
1159 					  "RSS hash function not supported");
1160 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1161 	if (rss->level > 2)
1162 #else
1163 	if (rss->level > 1)
1164 #endif
1165 		return rte_flow_error_set(error, ENOTSUP,
1166 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1167 					  &rss->level,
1168 					  "tunnel RSS is not supported");
1169 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1170 	if (rss->key_len == 0 && rss->key != NULL)
1171 		return rte_flow_error_set(error, ENOTSUP,
1172 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 					  &rss->key_len,
1174 					  "RSS hash key length 0");
1175 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1176 		return rte_flow_error_set(error, ENOTSUP,
1177 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1178 					  &rss->key_len,
1179 					  "RSS hash key too small");
1180 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1181 		return rte_flow_error_set(error, ENOTSUP,
1182 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183 					  &rss->key_len,
1184 					  "RSS hash key too large");
1185 	if (rss->queue_num > priv->config.ind_table_max_size)
1186 		return rte_flow_error_set(error, ENOTSUP,
1187 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1188 					  &rss->queue_num,
1189 					  "number of queues too large");
1190 	if (rss->types & MLX5_RSS_HF_MASK)
1191 		return rte_flow_error_set(error, ENOTSUP,
1192 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1193 					  &rss->types,
1194 					  "some RSS protocols are not"
1195 					  " supported");
1196 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1197 	    !(rss->types & ETH_RSS_IP))
1198 		return rte_flow_error_set(error, EINVAL,
1199 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1200 					  "L3 partial RSS requested but L3 RSS"
1201 					  " type not specified");
1202 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1203 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1204 		return rte_flow_error_set(error, EINVAL,
1205 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1206 					  "L4 partial RSS requested but L4 RSS"
1207 					  " type not specified");
1208 	if (!priv->rxqs_n)
1209 		return rte_flow_error_set(error, EINVAL,
1210 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1211 					  NULL, "No Rx queues configured");
1212 	if (!rss->queue_num)
1213 		return rte_flow_error_set(error, EINVAL,
1214 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1215 					  NULL, "No queues configured");
1216 	for (i = 0; i != rss->queue_num; ++i) {
1217 		if (rss->queue[i] >= priv->rxqs_n)
1218 			return rte_flow_error_set
1219 				(error, EINVAL,
1220 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1221 				 &rss->queue[i], "queue index out of range");
1222 		if (!(*priv->rxqs)[rss->queue[i]])
1223 			return rte_flow_error_set
1224 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1225 				 &rss->queue[i], "queue is not configured");
1226 	}
1227 	if (attr->egress)
1228 		return rte_flow_error_set(error, ENOTSUP,
1229 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230 					  "rss action not supported for "
1231 					  "egress");
1232 	if (rss->level > 1 &&  !tunnel)
1233 		return rte_flow_error_set(error, EINVAL,
1234 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1235 					  "inner RSS is not supported for "
1236 					  "non-tunnel flows");
1237 	return 0;
1238 }
1239 
1240 /*
1241  * Validate the count action.
1242  *
1243  * @param[in] dev
1244  *   Pointer to the Ethernet device structure.
1245  * @param[in] attr
1246  *   Attributes of flow that includes this action.
1247  * @param[out] error
1248  *   Pointer to error structure.
1249  *
1250  * @return
1251  *   0 on success, a negative errno value otherwise and rte_errno is set.
1252  */
1253 int
1254 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1255 				const struct rte_flow_attr *attr,
1256 				struct rte_flow_error *error)
1257 {
1258 	if (attr->egress)
1259 		return rte_flow_error_set(error, ENOTSUP,
1260 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1261 					  "count action not supported for "
1262 					  "egress");
1263 	return 0;
1264 }
1265 
1266 /**
1267  * Verify the @p attributes will be correctly understood by the NIC and store
1268  * them in the @p flow if everything is correct.
1269  *
1270  * @param[in] dev
1271  *   Pointer to the Ethernet device structure.
1272  * @param[in] attributes
1273  *   Pointer to flow attributes
1274  * @param[out] error
1275  *   Pointer to error structure.
1276  *
1277  * @return
1278  *   0 on success, a negative errno value otherwise and rte_errno is set.
1279  */
1280 int
1281 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1282 			      const struct rte_flow_attr *attributes,
1283 			      struct rte_flow_error *error)
1284 {
1285 	struct mlx5_priv *priv = dev->data->dev_private;
1286 	uint32_t priority_max = priv->config.flow_prio - 1;
1287 
1288 	if (attributes->group)
1289 		return rte_flow_error_set(error, ENOTSUP,
1290 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1291 					  NULL, "groups is not supported");
1292 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1293 	    attributes->priority >= priority_max)
1294 		return rte_flow_error_set(error, ENOTSUP,
1295 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1296 					  NULL, "priority out of range");
1297 	if (attributes->egress)
1298 		return rte_flow_error_set(error, ENOTSUP,
1299 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1300 					  "egress is not supported");
1301 	if (attributes->transfer && !priv->config.dv_esw_en)
1302 		return rte_flow_error_set(error, ENOTSUP,
1303 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1304 					  NULL, "transfer is not supported");
1305 	if (!attributes->ingress)
1306 		return rte_flow_error_set(error, EINVAL,
1307 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1308 					  NULL,
1309 					  "ingress attribute is mandatory");
1310 	return 0;
1311 }
1312 
1313 /**
1314  * Validate ICMP6 item.
1315  *
1316  * @param[in] item
1317  *   Item specification.
1318  * @param[in] item_flags
1319  *   Bit-fields that holds the items detected until now.
1320  * @param[out] error
1321  *   Pointer to error structure.
1322  *
1323  * @return
1324  *   0 on success, a negative errno value otherwise and rte_errno is set.
1325  */
1326 int
1327 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1328 			       uint64_t item_flags,
1329 			       uint8_t target_protocol,
1330 			       struct rte_flow_error *error)
1331 {
1332 	const struct rte_flow_item_icmp6 *mask = item->mask;
1333 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1334 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1335 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1336 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1337 				      MLX5_FLOW_LAYER_OUTER_L4;
1338 	int ret;
1339 
1340 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1341 		return rte_flow_error_set(error, EINVAL,
1342 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1343 					  "protocol filtering not compatible"
1344 					  " with ICMP6 layer");
1345 	if (!(item_flags & l3m))
1346 		return rte_flow_error_set(error, EINVAL,
1347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1348 					  "IPv6 is mandatory to filter on"
1349 					  " ICMP6");
1350 	if (item_flags & l4m)
1351 		return rte_flow_error_set(error, EINVAL,
1352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 					  "multiple L4 layers not supported");
1354 	if (!mask)
1355 		mask = &rte_flow_item_icmp6_mask;
1356 	ret = mlx5_flow_item_acceptable
1357 		(item, (const uint8_t *)mask,
1358 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1359 		 sizeof(struct rte_flow_item_icmp6), error);
1360 	if (ret < 0)
1361 		return ret;
1362 	return 0;
1363 }
1364 
1365 /**
1366  * Validate ICMP item.
1367  *
1368  * @param[in] item
1369  *   Item specification.
1370  * @param[in] item_flags
1371  *   Bit-fields that holds the items detected until now.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1380 			     uint64_t item_flags,
1381 			     uint8_t target_protocol,
1382 			     struct rte_flow_error *error)
1383 {
1384 	const struct rte_flow_item_icmp *mask = item->mask;
1385 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1386 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1387 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1388 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1389 				      MLX5_FLOW_LAYER_OUTER_L4;
1390 	int ret;
1391 
1392 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1393 		return rte_flow_error_set(error, EINVAL,
1394 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1395 					  "protocol filtering not compatible"
1396 					  " with ICMP layer");
1397 	if (!(item_flags & l3m))
1398 		return rte_flow_error_set(error, EINVAL,
1399 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1400 					  "IPv4 is mandatory to filter"
1401 					  " on ICMP");
1402 	if (item_flags & l4m)
1403 		return rte_flow_error_set(error, EINVAL,
1404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 					  "multiple L4 layers not supported");
1406 	if (!mask)
1407 		mask = &rte_flow_item_icmp_mask;
1408 	ret = mlx5_flow_item_acceptable
1409 		(item, (const uint8_t *)mask,
1410 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1411 		 sizeof(struct rte_flow_item_icmp), error);
1412 	if (ret < 0)
1413 		return ret;
1414 	return 0;
1415 }
1416 
1417 /**
1418  * Validate Ethernet item.
1419  *
1420  * @param[in] item
1421  *   Item specification.
1422  * @param[in] item_flags
1423  *   Bit-fields that holds the items detected until now.
1424  * @param[out] error
1425  *   Pointer to error structure.
1426  *
1427  * @return
1428  *   0 on success, a negative errno value otherwise and rte_errno is set.
1429  */
1430 int
1431 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1432 			    uint64_t item_flags,
1433 			    struct rte_flow_error *error)
1434 {
1435 	const struct rte_flow_item_eth *mask = item->mask;
1436 	const struct rte_flow_item_eth nic_mask = {
1437 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1438 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1439 		.type = RTE_BE16(0xffff),
1440 	};
1441 	int ret;
1442 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1443 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1444 				       MLX5_FLOW_LAYER_OUTER_L2;
1445 
1446 	if (item_flags & ethm)
1447 		return rte_flow_error_set(error, ENOTSUP,
1448 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1449 					  "multiple L2 layers not supported");
1450 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1451 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1452 		return rte_flow_error_set(error, EINVAL,
1453 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1454 					  "L2 layer should not follow "
1455 					  "L3 layers");
1456 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1457 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1458 		return rte_flow_error_set(error, EINVAL,
1459 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1460 					  "L2 layer should not follow VLAN");
1461 	if (!mask)
1462 		mask = &rte_flow_item_eth_mask;
1463 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1464 					(const uint8_t *)&nic_mask,
1465 					sizeof(struct rte_flow_item_eth),
1466 					error);
1467 	return ret;
1468 }
1469 
1470 /**
1471  * Validate VLAN item.
1472  *
1473  * @param[in] item
1474  *   Item specification.
1475  * @param[in] item_flags
1476  *   Bit-fields that holds the items detected until now.
1477  * @param[in] dev
1478  *   Ethernet device flow is being created on.
1479  * @param[out] error
1480  *   Pointer to error structure.
1481  *
1482  * @return
1483  *   0 on success, a negative errno value otherwise and rte_errno is set.
1484  */
1485 int
1486 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1487 			     uint64_t item_flags,
1488 			     struct rte_eth_dev *dev,
1489 			     struct rte_flow_error *error)
1490 {
1491 	const struct rte_flow_item_vlan *spec = item->spec;
1492 	const struct rte_flow_item_vlan *mask = item->mask;
1493 	const struct rte_flow_item_vlan nic_mask = {
1494 		.tci = RTE_BE16(UINT16_MAX),
1495 		.inner_type = RTE_BE16(UINT16_MAX),
1496 	};
1497 	uint16_t vlan_tag = 0;
1498 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1499 	int ret;
1500 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1501 					MLX5_FLOW_LAYER_INNER_L4) :
1502 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1503 					MLX5_FLOW_LAYER_OUTER_L4);
1504 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1505 					MLX5_FLOW_LAYER_OUTER_VLAN;
1506 
1507 	if (item_flags & vlanm)
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1510 					  "multiple VLAN layers not supported");
1511 	else if ((item_flags & l34m) != 0)
1512 		return rte_flow_error_set(error, EINVAL,
1513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1514 					  "VLAN cannot follow L3/L4 layer");
1515 	if (!mask)
1516 		mask = &rte_flow_item_vlan_mask;
1517 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1518 					(const uint8_t *)&nic_mask,
1519 					sizeof(struct rte_flow_item_vlan),
1520 					error);
1521 	if (ret)
1522 		return ret;
1523 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1524 		struct mlx5_priv *priv = dev->data->dev_private;
1525 
1526 		if (priv->vmwa_context) {
1527 			/*
1528 			 * Non-NULL context means we have a virtual machine
1529 			 * and SR-IOV enabled, we have to create VLAN interface
1530 			 * to make hypervisor to setup E-Switch vport
1531 			 * context correctly. We avoid creating the multiple
1532 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1533 			 */
1534 			return rte_flow_error_set(error, EINVAL,
1535 						  RTE_FLOW_ERROR_TYPE_ITEM,
1536 						  item,
1537 						  "VLAN tag mask is not"
1538 						  " supported in virtual"
1539 						  " environment");
1540 		}
1541 	}
1542 	if (spec) {
1543 		vlan_tag = spec->tci;
1544 		vlan_tag &= mask->tci;
1545 	}
1546 	/*
1547 	 * From verbs perspective an empty VLAN is equivalent
1548 	 * to a packet without VLAN layer.
1549 	 */
1550 	if (!vlan_tag)
1551 		return rte_flow_error_set(error, EINVAL,
1552 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1553 					  item->spec,
1554 					  "VLAN cannot be empty");
1555 	return 0;
1556 }
1557 
1558 /**
1559  * Validate IPV4 item.
1560  *
1561  * @param[in] item
1562  *   Item specification.
1563  * @param[in] item_flags
1564  *   Bit-fields that holds the items detected until now.
1565  * @param[in] acc_mask
1566  *   Acceptable mask, if NULL default internal default mask
1567  *   will be used to check whether item fields are supported.
1568  * @param[out] error
1569  *   Pointer to error structure.
1570  *
1571  * @return
1572  *   0 on success, a negative errno value otherwise and rte_errno is set.
1573  */
1574 int
1575 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1576 			     uint64_t item_flags,
1577 			     uint64_t last_item,
1578 			     uint16_t ether_type,
1579 			     const struct rte_flow_item_ipv4 *acc_mask,
1580 			     struct rte_flow_error *error)
1581 {
1582 	const struct rte_flow_item_ipv4 *mask = item->mask;
1583 	const struct rte_flow_item_ipv4 *spec = item->spec;
1584 	const struct rte_flow_item_ipv4 nic_mask = {
1585 		.hdr = {
1586 			.src_addr = RTE_BE32(0xffffffff),
1587 			.dst_addr = RTE_BE32(0xffffffff),
1588 			.type_of_service = 0xff,
1589 			.next_proto_id = 0xff,
1590 		},
1591 	};
1592 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1593 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1594 				      MLX5_FLOW_LAYER_OUTER_L3;
1595 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1596 				      MLX5_FLOW_LAYER_OUTER_L4;
1597 	int ret;
1598 	uint8_t next_proto = 0xFF;
1599 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1600 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1601 				  MLX5_FLOW_LAYER_INNER_VLAN);
1602 
1603 	if ((last_item & l2_vlan) && ether_type &&
1604 	    ether_type != RTE_ETHER_TYPE_IPV4)
1605 		return rte_flow_error_set(error, EINVAL,
1606 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1607 					  "IPv4 cannot follow L2/VLAN layer "
1608 					  "which ether type is not IPv4");
1609 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1610 		if (mask && spec)
1611 			next_proto = mask->hdr.next_proto_id &
1612 				     spec->hdr.next_proto_id;
1613 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1614 			return rte_flow_error_set(error, EINVAL,
1615 						  RTE_FLOW_ERROR_TYPE_ITEM,
1616 						  item,
1617 						  "multiple tunnel "
1618 						  "not supported");
1619 	}
1620 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1621 		return rte_flow_error_set(error, EINVAL,
1622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1623 					  "wrong tunnel type - IPv6 specified "
1624 					  "but IPv4 item provided");
1625 	if (item_flags & l3m)
1626 		return rte_flow_error_set(error, ENOTSUP,
1627 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1628 					  "multiple L3 layers not supported");
1629 	else if (item_flags & l4m)
1630 		return rte_flow_error_set(error, EINVAL,
1631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1632 					  "L3 cannot follow an L4 layer.");
1633 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1634 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1635 		return rte_flow_error_set(error, EINVAL,
1636 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1637 					  "L3 cannot follow an NVGRE layer.");
1638 	if (!mask)
1639 		mask = &rte_flow_item_ipv4_mask;
1640 	else if (mask->hdr.next_proto_id != 0 &&
1641 		 mask->hdr.next_proto_id != 0xff)
1642 		return rte_flow_error_set(error, EINVAL,
1643 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1644 					  "partial mask is not supported"
1645 					  " for protocol");
1646 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1647 					acc_mask ? (const uint8_t *)acc_mask
1648 						 : (const uint8_t *)&nic_mask,
1649 					sizeof(struct rte_flow_item_ipv4),
1650 					error);
1651 	if (ret < 0)
1652 		return ret;
1653 	return 0;
1654 }
1655 
1656 /**
1657  * Validate IPV6 item.
1658  *
1659  * @param[in] item
1660  *   Item specification.
1661  * @param[in] item_flags
1662  *   Bit-fields that holds the items detected until now.
1663  * @param[in] acc_mask
1664  *   Acceptable mask, if NULL default internal default mask
1665  *   will be used to check whether item fields are supported.
1666  * @param[out] error
1667  *   Pointer to error structure.
1668  *
1669  * @return
1670  *   0 on success, a negative errno value otherwise and rte_errno is set.
1671  */
1672 int
1673 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1674 			     uint64_t item_flags,
1675 			     uint64_t last_item,
1676 			     uint16_t ether_type,
1677 			     const struct rte_flow_item_ipv6 *acc_mask,
1678 			     struct rte_flow_error *error)
1679 {
1680 	const struct rte_flow_item_ipv6 *mask = item->mask;
1681 	const struct rte_flow_item_ipv6 *spec = item->spec;
1682 	const struct rte_flow_item_ipv6 nic_mask = {
1683 		.hdr = {
1684 			.src_addr =
1685 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1686 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1687 			.dst_addr =
1688 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1689 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1690 			.vtc_flow = RTE_BE32(0xffffffff),
1691 			.proto = 0xff,
1692 		},
1693 	};
1694 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1695 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1696 				      MLX5_FLOW_LAYER_OUTER_L3;
1697 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1698 				      MLX5_FLOW_LAYER_OUTER_L4;
1699 	int ret;
1700 	uint8_t next_proto = 0xFF;
1701 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1702 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1703 				  MLX5_FLOW_LAYER_INNER_VLAN);
1704 
1705 	if ((last_item & l2_vlan) && ether_type &&
1706 	    ether_type != RTE_ETHER_TYPE_IPV6)
1707 		return rte_flow_error_set(error, EINVAL,
1708 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1709 					  "IPv6 cannot follow L2/VLAN layer "
1710 					  "which ether type is not IPv6");
1711 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1712 		if (mask && spec)
1713 			next_proto = mask->hdr.proto & spec->hdr.proto;
1714 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1715 			return rte_flow_error_set(error, EINVAL,
1716 						  RTE_FLOW_ERROR_TYPE_ITEM,
1717 						  item,
1718 						  "multiple tunnel "
1719 						  "not supported");
1720 	}
1721 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1722 		return rte_flow_error_set(error, EINVAL,
1723 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1724 					  "wrong tunnel type - IPv4 specified "
1725 					  "but IPv6 item provided");
1726 	if (item_flags & l3m)
1727 		return rte_flow_error_set(error, ENOTSUP,
1728 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1729 					  "multiple L3 layers not supported");
1730 	else if (item_flags & l4m)
1731 		return rte_flow_error_set(error, EINVAL,
1732 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1733 					  "L3 cannot follow an L4 layer.");
1734 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1735 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1736 		return rte_flow_error_set(error, EINVAL,
1737 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1738 					  "L3 cannot follow an NVGRE layer.");
1739 	if (!mask)
1740 		mask = &rte_flow_item_ipv6_mask;
1741 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1742 					acc_mask ? (const uint8_t *)acc_mask
1743 						 : (const uint8_t *)&nic_mask,
1744 					sizeof(struct rte_flow_item_ipv6),
1745 					error);
1746 	if (ret < 0)
1747 		return ret;
1748 	return 0;
1749 }
1750 
1751 /**
1752  * Validate UDP item.
1753  *
1754  * @param[in] item
1755  *   Item specification.
1756  * @param[in] item_flags
1757  *   Bit-fields that holds the items detected until now.
1758  * @param[in] target_protocol
1759  *   The next protocol in the previous item.
1760  * @param[in] flow_mask
1761  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1762  * @param[out] error
1763  *   Pointer to error structure.
1764  *
1765  * @return
1766  *   0 on success, a negative errno value otherwise and rte_errno is set.
1767  */
1768 int
1769 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1770 			    uint64_t item_flags,
1771 			    uint8_t target_protocol,
1772 			    struct rte_flow_error *error)
1773 {
1774 	const struct rte_flow_item_udp *mask = item->mask;
1775 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1776 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1777 				      MLX5_FLOW_LAYER_OUTER_L3;
1778 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1779 				      MLX5_FLOW_LAYER_OUTER_L4;
1780 	int ret;
1781 
1782 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1783 		return rte_flow_error_set(error, EINVAL,
1784 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1785 					  "protocol filtering not compatible"
1786 					  " with UDP layer");
1787 	if (!(item_flags & l3m))
1788 		return rte_flow_error_set(error, EINVAL,
1789 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1790 					  "L3 is mandatory to filter on L4");
1791 	if (item_flags & l4m)
1792 		return rte_flow_error_set(error, EINVAL,
1793 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1794 					  "multiple L4 layers not supported");
1795 	if (!mask)
1796 		mask = &rte_flow_item_udp_mask;
1797 	ret = mlx5_flow_item_acceptable
1798 		(item, (const uint8_t *)mask,
1799 		 (const uint8_t *)&rte_flow_item_udp_mask,
1800 		 sizeof(struct rte_flow_item_udp), error);
1801 	if (ret < 0)
1802 		return ret;
1803 	return 0;
1804 }
1805 
1806 /**
1807  * Validate TCP item.
1808  *
1809  * @param[in] item
1810  *   Item specification.
1811  * @param[in] item_flags
1812  *   Bit-fields that holds the items detected until now.
1813  * @param[in] target_protocol
1814  *   The next protocol in the previous item.
1815  * @param[out] error
1816  *   Pointer to error structure.
1817  *
1818  * @return
1819  *   0 on success, a negative errno value otherwise and rte_errno is set.
1820  */
1821 int
1822 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1823 			    uint64_t item_flags,
1824 			    uint8_t target_protocol,
1825 			    const struct rte_flow_item_tcp *flow_mask,
1826 			    struct rte_flow_error *error)
1827 {
1828 	const struct rte_flow_item_tcp *mask = item->mask;
1829 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1830 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1831 				      MLX5_FLOW_LAYER_OUTER_L3;
1832 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1833 				      MLX5_FLOW_LAYER_OUTER_L4;
1834 	int ret;
1835 
1836 	MLX5_ASSERT(flow_mask);
1837 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1838 		return rte_flow_error_set(error, EINVAL,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "protocol filtering not compatible"
1841 					  " with TCP layer");
1842 	if (!(item_flags & l3m))
1843 		return rte_flow_error_set(error, EINVAL,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "L3 is mandatory to filter on L4");
1846 	if (item_flags & l4m)
1847 		return rte_flow_error_set(error, EINVAL,
1848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1849 					  "multiple L4 layers not supported");
1850 	if (!mask)
1851 		mask = &rte_flow_item_tcp_mask;
1852 	ret = mlx5_flow_item_acceptable
1853 		(item, (const uint8_t *)mask,
1854 		 (const uint8_t *)flow_mask,
1855 		 sizeof(struct rte_flow_item_tcp), error);
1856 	if (ret < 0)
1857 		return ret;
1858 	return 0;
1859 }
1860 
1861 /**
1862  * Validate VXLAN item.
1863  *
1864  * @param[in] item
1865  *   Item specification.
1866  * @param[in] item_flags
1867  *   Bit-fields that holds the items detected until now.
1868  * @param[in] target_protocol
1869  *   The next protocol in the previous item.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 int
1877 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1878 			      uint64_t item_flags,
1879 			      struct rte_flow_error *error)
1880 {
1881 	const struct rte_flow_item_vxlan *spec = item->spec;
1882 	const struct rte_flow_item_vxlan *mask = item->mask;
1883 	int ret;
1884 	union vni {
1885 		uint32_t vlan_id;
1886 		uint8_t vni[4];
1887 	} id = { .vlan_id = 0, };
1888 
1889 
1890 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1891 		return rte_flow_error_set(error, ENOTSUP,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "multiple tunnel layers not"
1894 					  " supported");
1895 	/*
1896 	 * Verify only UDPv4 is present as defined in
1897 	 * https://tools.ietf.org/html/rfc7348
1898 	 */
1899 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1900 		return rte_flow_error_set(error, EINVAL,
1901 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1902 					  "no outer UDP layer found");
1903 	if (!mask)
1904 		mask = &rte_flow_item_vxlan_mask;
1905 	ret = mlx5_flow_item_acceptable
1906 		(item, (const uint8_t *)mask,
1907 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1908 		 sizeof(struct rte_flow_item_vxlan),
1909 		 error);
1910 	if (ret < 0)
1911 		return ret;
1912 	if (spec) {
1913 		memcpy(&id.vni[1], spec->vni, 3);
1914 		memcpy(&id.vni[1], mask->vni, 3);
1915 	}
1916 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1917 		return rte_flow_error_set(error, ENOTSUP,
1918 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1919 					  "VXLAN tunnel must be fully defined");
1920 	return 0;
1921 }
1922 
1923 /**
1924  * Validate VXLAN_GPE item.
1925  *
1926  * @param[in] item
1927  *   Item specification.
1928  * @param[in] item_flags
1929  *   Bit-fields that holds the items detected until now.
1930  * @param[in] priv
1931  *   Pointer to the private data structure.
1932  * @param[in] target_protocol
1933  *   The next protocol in the previous item.
1934  * @param[out] error
1935  *   Pointer to error structure.
1936  *
1937  * @return
1938  *   0 on success, a negative errno value otherwise and rte_errno is set.
1939  */
1940 int
1941 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1942 				  uint64_t item_flags,
1943 				  struct rte_eth_dev *dev,
1944 				  struct rte_flow_error *error)
1945 {
1946 	struct mlx5_priv *priv = dev->data->dev_private;
1947 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1948 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1949 	int ret;
1950 	union vni {
1951 		uint32_t vlan_id;
1952 		uint8_t vni[4];
1953 	} id = { .vlan_id = 0, };
1954 
1955 	if (!priv->config.l3_vxlan_en)
1956 		return rte_flow_error_set(error, ENOTSUP,
1957 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1958 					  "L3 VXLAN is not enabled by device"
1959 					  " parameter and/or not configured in"
1960 					  " firmware");
1961 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1962 		return rte_flow_error_set(error, ENOTSUP,
1963 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1964 					  "multiple tunnel layers not"
1965 					  " supported");
1966 	/*
1967 	 * Verify only UDPv4 is present as defined in
1968 	 * https://tools.ietf.org/html/rfc7348
1969 	 */
1970 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1971 		return rte_flow_error_set(error, EINVAL,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "no outer UDP layer found");
1974 	if (!mask)
1975 		mask = &rte_flow_item_vxlan_gpe_mask;
1976 	ret = mlx5_flow_item_acceptable
1977 		(item, (const uint8_t *)mask,
1978 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1979 		 sizeof(struct rte_flow_item_vxlan_gpe),
1980 		 error);
1981 	if (ret < 0)
1982 		return ret;
1983 	if (spec) {
1984 		if (spec->protocol)
1985 			return rte_flow_error_set(error, ENOTSUP,
1986 						  RTE_FLOW_ERROR_TYPE_ITEM,
1987 						  item,
1988 						  "VxLAN-GPE protocol"
1989 						  " not supported");
1990 		memcpy(&id.vni[1], spec->vni, 3);
1991 		memcpy(&id.vni[1], mask->vni, 3);
1992 	}
1993 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1994 		return rte_flow_error_set(error, ENOTSUP,
1995 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1996 					  "VXLAN-GPE tunnel must be fully"
1997 					  " defined");
1998 	return 0;
1999 }
2000 /**
2001  * Validate GRE Key item.
2002  *
2003  * @param[in] item
2004  *   Item specification.
2005  * @param[in] item_flags
2006  *   Bit flags to mark detected items.
2007  * @param[in] gre_item
2008  *   Pointer to gre_item
2009  * @param[out] error
2010  *   Pointer to error structure.
2011  *
2012  * @return
2013  *   0 on success, a negative errno value otherwise and rte_errno is set.
2014  */
2015 int
2016 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2017 				uint64_t item_flags,
2018 				const struct rte_flow_item *gre_item,
2019 				struct rte_flow_error *error)
2020 {
2021 	const rte_be32_t *mask = item->mask;
2022 	int ret = 0;
2023 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2024 	const struct rte_flow_item_gre *gre_spec;
2025 	const struct rte_flow_item_gre *gre_mask;
2026 
2027 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2028 		return rte_flow_error_set(error, ENOTSUP,
2029 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2030 					  "Multiple GRE key not support");
2031 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2032 		return rte_flow_error_set(error, ENOTSUP,
2033 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2034 					  "No preceding GRE header");
2035 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2036 		return rte_flow_error_set(error, ENOTSUP,
2037 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2038 					  "GRE key following a wrong item");
2039 	gre_mask = gre_item->mask;
2040 	if (!gre_mask)
2041 		gre_mask = &rte_flow_item_gre_mask;
2042 	gre_spec = gre_item->spec;
2043 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2044 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "Key bit must be on");
2048 
2049 	if (!mask)
2050 		mask = &gre_key_default_mask;
2051 	ret = mlx5_flow_item_acceptable
2052 		(item, (const uint8_t *)mask,
2053 		 (const uint8_t *)&gre_key_default_mask,
2054 		 sizeof(rte_be32_t), error);
2055 	return ret;
2056 }
2057 
2058 /**
2059  * Validate GRE item.
2060  *
2061  * @param[in] item
2062  *   Item specification.
2063  * @param[in] item_flags
2064  *   Bit flags to mark detected items.
2065  * @param[in] target_protocol
2066  *   The next protocol in the previous item.
2067  * @param[out] error
2068  *   Pointer to error structure.
2069  *
2070  * @return
2071  *   0 on success, a negative errno value otherwise and rte_errno is set.
2072  */
2073 int
2074 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2075 			    uint64_t item_flags,
2076 			    uint8_t target_protocol,
2077 			    struct rte_flow_error *error)
2078 {
2079 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2080 	const struct rte_flow_item_gre *mask = item->mask;
2081 	int ret;
2082 	const struct rte_flow_item_gre nic_mask = {
2083 		.c_rsvd0_ver = RTE_BE16(0xB000),
2084 		.protocol = RTE_BE16(UINT16_MAX),
2085 	};
2086 
2087 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2088 		return rte_flow_error_set(error, EINVAL,
2089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2090 					  "protocol filtering not compatible"
2091 					  " with this GRE layer");
2092 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2093 		return rte_flow_error_set(error, ENOTSUP,
2094 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 					  "multiple tunnel layers not"
2096 					  " supported");
2097 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2098 		return rte_flow_error_set(error, ENOTSUP,
2099 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2100 					  "L3 Layer is missing");
2101 	if (!mask)
2102 		mask = &rte_flow_item_gre_mask;
2103 	ret = mlx5_flow_item_acceptable
2104 		(item, (const uint8_t *)mask,
2105 		 (const uint8_t *)&nic_mask,
2106 		 sizeof(struct rte_flow_item_gre), error);
2107 	if (ret < 0)
2108 		return ret;
2109 #ifndef HAVE_MLX5DV_DR
2110 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2111 	if (spec && (spec->protocol & mask->protocol))
2112 		return rte_flow_error_set(error, ENOTSUP,
2113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 					  "without MPLS support the"
2115 					  " specification cannot be used for"
2116 					  " filtering");
2117 #endif
2118 #endif
2119 	return 0;
2120 }
2121 
2122 /**
2123  * Validate Geneve item.
2124  *
2125  * @param[in] item
2126  *   Item specification.
2127  * @param[in] itemFlags
2128  *   Bit-fields that holds the items detected until now.
2129  * @param[in] enPriv
2130  *   Pointer to the private data structure.
2131  * @param[out] error
2132  *   Pointer to error structure.
2133  *
2134  * @return
2135  *   0 on success, a negative errno value otherwise and rte_errno is set.
2136  */
2137 
2138 int
2139 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2140 			       uint64_t item_flags,
2141 			       struct rte_eth_dev *dev,
2142 			       struct rte_flow_error *error)
2143 {
2144 	struct mlx5_priv *priv = dev->data->dev_private;
2145 	const struct rte_flow_item_geneve *spec = item->spec;
2146 	const struct rte_flow_item_geneve *mask = item->mask;
2147 	int ret;
2148 	uint16_t gbhdr;
2149 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2150 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2151 	const struct rte_flow_item_geneve nic_mask = {
2152 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2153 		.vni = "\xff\xff\xff",
2154 		.protocol = RTE_BE16(UINT16_MAX),
2155 	};
2156 
2157 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2158 		return rte_flow_error_set(error, ENOTSUP,
2159 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2160 					  "L3 Geneve is not enabled by device"
2161 					  " parameter and/or not configured in"
2162 					  " firmware");
2163 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2164 		return rte_flow_error_set(error, ENOTSUP,
2165 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2166 					  "multiple tunnel layers not"
2167 					  " supported");
2168 	/*
2169 	 * Verify only UDPv4 is present as defined in
2170 	 * https://tools.ietf.org/html/rfc7348
2171 	 */
2172 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2173 		return rte_flow_error_set(error, EINVAL,
2174 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2175 					  "no outer UDP layer found");
2176 	if (!mask)
2177 		mask = &rte_flow_item_geneve_mask;
2178 	ret = mlx5_flow_item_acceptable
2179 				  (item, (const uint8_t *)mask,
2180 				   (const uint8_t *)&nic_mask,
2181 				   sizeof(struct rte_flow_item_geneve), error);
2182 	if (ret)
2183 		return ret;
2184 	if (spec) {
2185 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2186 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2187 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2188 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2189 			return rte_flow_error_set(error, ENOTSUP,
2190 						  RTE_FLOW_ERROR_TYPE_ITEM,
2191 						  item,
2192 						  "Geneve protocol unsupported"
2193 						  " fields are being used");
2194 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2195 			return rte_flow_error_set
2196 					(error, ENOTSUP,
2197 					 RTE_FLOW_ERROR_TYPE_ITEM,
2198 					 item,
2199 					 "Unsupported Geneve options length");
2200 	}
2201 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2202 		return rte_flow_error_set
2203 				    (error, ENOTSUP,
2204 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2205 				     "Geneve tunnel must be fully defined");
2206 	return 0;
2207 }
2208 
2209 /**
2210  * Validate MPLS item.
2211  *
2212  * @param[in] dev
2213  *   Pointer to the rte_eth_dev structure.
2214  * @param[in] item
2215  *   Item specification.
2216  * @param[in] item_flags
2217  *   Bit-fields that holds the items detected until now.
2218  * @param[in] prev_layer
2219  *   The protocol layer indicated in previous item.
2220  * @param[out] error
2221  *   Pointer to error structure.
2222  *
2223  * @return
2224  *   0 on success, a negative errno value otherwise and rte_errno is set.
2225  */
2226 int
2227 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2228 			     const struct rte_flow_item *item __rte_unused,
2229 			     uint64_t item_flags __rte_unused,
2230 			     uint64_t prev_layer __rte_unused,
2231 			     struct rte_flow_error *error)
2232 {
2233 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2234 	const struct rte_flow_item_mpls *mask = item->mask;
2235 	struct mlx5_priv *priv = dev->data->dev_private;
2236 	int ret;
2237 
2238 	if (!priv->config.mpls_en)
2239 		return rte_flow_error_set(error, ENOTSUP,
2240 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2241 					  "MPLS not supported or"
2242 					  " disabled in firmware"
2243 					  " configuration.");
2244 	/* MPLS over IP, UDP, GRE is allowed */
2245 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2246 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2247 			    MLX5_FLOW_LAYER_GRE)))
2248 		return rte_flow_error_set(error, EINVAL,
2249 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2250 					  "protocol filtering not compatible"
2251 					  " with MPLS layer");
2252 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2253 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2254 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2255 		return rte_flow_error_set(error, ENOTSUP,
2256 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2257 					  "multiple tunnel layers not"
2258 					  " supported");
2259 	if (!mask)
2260 		mask = &rte_flow_item_mpls_mask;
2261 	ret = mlx5_flow_item_acceptable
2262 		(item, (const uint8_t *)mask,
2263 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2264 		 sizeof(struct rte_flow_item_mpls), error);
2265 	if (ret < 0)
2266 		return ret;
2267 	return 0;
2268 #endif
2269 	return rte_flow_error_set(error, ENOTSUP,
2270 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 				  "MPLS is not supported by Verbs, please"
2272 				  " update.");
2273 }
2274 
2275 /**
2276  * Validate NVGRE item.
2277  *
2278  * @param[in] item
2279  *   Item specification.
2280  * @param[in] item_flags
2281  *   Bit flags to mark detected items.
2282  * @param[in] target_protocol
2283  *   The next protocol in the previous item.
2284  * @param[out] error
2285  *   Pointer to error structure.
2286  *
2287  * @return
2288  *   0 on success, a negative errno value otherwise and rte_errno is set.
2289  */
2290 int
2291 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2292 			      uint64_t item_flags,
2293 			      uint8_t target_protocol,
2294 			      struct rte_flow_error *error)
2295 {
2296 	const struct rte_flow_item_nvgre *mask = item->mask;
2297 	int ret;
2298 
2299 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2300 		return rte_flow_error_set(error, EINVAL,
2301 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2302 					  "protocol filtering not compatible"
2303 					  " with this GRE layer");
2304 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2305 		return rte_flow_error_set(error, ENOTSUP,
2306 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2307 					  "multiple tunnel layers not"
2308 					  " supported");
2309 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2310 		return rte_flow_error_set(error, ENOTSUP,
2311 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2312 					  "L3 Layer is missing");
2313 	if (!mask)
2314 		mask = &rte_flow_item_nvgre_mask;
2315 	ret = mlx5_flow_item_acceptable
2316 		(item, (const uint8_t *)mask,
2317 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2318 		 sizeof(struct rte_flow_item_nvgre), error);
2319 	if (ret < 0)
2320 		return ret;
2321 	return 0;
2322 }
2323 
2324 /* Allocate unique ID for the split Q/RSS subflows. */
2325 static uint32_t
2326 flow_qrss_get_id(struct rte_eth_dev *dev)
2327 {
2328 	struct mlx5_priv *priv = dev->data->dev_private;
2329 	uint32_t qrss_id, ret;
2330 
2331 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2332 	if (ret)
2333 		return 0;
2334 	MLX5_ASSERT(qrss_id);
2335 	return qrss_id;
2336 }
2337 
2338 /* Free unique ID for the split Q/RSS subflows. */
2339 static void
2340 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2341 {
2342 	struct mlx5_priv *priv = dev->data->dev_private;
2343 
2344 	if (qrss_id)
2345 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2346 }
2347 
2348 /**
2349  * Release resource related QUEUE/RSS action split.
2350  *
2351  * @param dev
2352  *   Pointer to Ethernet device.
2353  * @param flow
2354  *   Flow to release id's from.
2355  */
2356 static void
2357 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2358 			     struct rte_flow *flow)
2359 {
2360 	struct mlx5_priv *priv = dev->data->dev_private;
2361 	uint32_t handle_idx;
2362 	struct mlx5_flow_handle *dev_handle;
2363 
2364 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2365 		       handle_idx, dev_handle, next)
2366 		if (dev_handle->split_flow_id)
2367 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2368 }
2369 
2370 static int
2371 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2372 		   const struct rte_flow_attr *attr __rte_unused,
2373 		   const struct rte_flow_item items[] __rte_unused,
2374 		   const struct rte_flow_action actions[] __rte_unused,
2375 		   bool external __rte_unused,
2376 		   int hairpin __rte_unused,
2377 		   struct rte_flow_error *error)
2378 {
2379 	return rte_flow_error_set(error, ENOTSUP,
2380 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2381 }
2382 
2383 static struct mlx5_flow *
2384 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2385 		  const struct rte_flow_attr *attr __rte_unused,
2386 		  const struct rte_flow_item items[] __rte_unused,
2387 		  const struct rte_flow_action actions[] __rte_unused,
2388 		  struct rte_flow_error *error)
2389 {
2390 	rte_flow_error_set(error, ENOTSUP,
2391 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2392 	return NULL;
2393 }
2394 
2395 static int
2396 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2397 		    struct mlx5_flow *dev_flow __rte_unused,
2398 		    const struct rte_flow_attr *attr __rte_unused,
2399 		    const struct rte_flow_item items[] __rte_unused,
2400 		    const struct rte_flow_action actions[] __rte_unused,
2401 		    struct rte_flow_error *error)
2402 {
2403 	return rte_flow_error_set(error, ENOTSUP,
2404 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2405 }
2406 
2407 static int
2408 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2409 		struct rte_flow *flow __rte_unused,
2410 		struct rte_flow_error *error)
2411 {
2412 	return rte_flow_error_set(error, ENOTSUP,
2413 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2414 }
2415 
2416 static void
2417 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2418 		 struct rte_flow *flow __rte_unused)
2419 {
2420 }
2421 
2422 static void
2423 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2424 		  struct rte_flow *flow __rte_unused)
2425 {
2426 }
2427 
2428 static int
2429 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2430 		struct rte_flow *flow __rte_unused,
2431 		const struct rte_flow_action *actions __rte_unused,
2432 		void *data __rte_unused,
2433 		struct rte_flow_error *error)
2434 {
2435 	return rte_flow_error_set(error, ENOTSUP,
2436 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2437 }
2438 
2439 /* Void driver to protect from null pointer reference. */
2440 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2441 	.validate = flow_null_validate,
2442 	.prepare = flow_null_prepare,
2443 	.translate = flow_null_translate,
2444 	.apply = flow_null_apply,
2445 	.remove = flow_null_remove,
2446 	.destroy = flow_null_destroy,
2447 	.query = flow_null_query,
2448 };
2449 
2450 /**
2451  * Select flow driver type according to flow attributes and device
2452  * configuration.
2453  *
2454  * @param[in] dev
2455  *   Pointer to the dev structure.
2456  * @param[in] attr
2457  *   Pointer to the flow attributes.
2458  *
2459  * @return
2460  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2461  */
2462 static enum mlx5_flow_drv_type
2463 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2464 {
2465 	struct mlx5_priv *priv = dev->data->dev_private;
2466 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2467 
2468 	if (attr->transfer && priv->config.dv_esw_en)
2469 		type = MLX5_FLOW_TYPE_DV;
2470 	if (!attr->transfer)
2471 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2472 						 MLX5_FLOW_TYPE_VERBS;
2473 	return type;
2474 }
2475 
2476 #define flow_get_drv_ops(type) flow_drv_ops[type]
2477 
2478 /**
2479  * Flow driver validation API. This abstracts calling driver specific functions.
2480  * The type of flow driver is determined according to flow attributes.
2481  *
2482  * @param[in] dev
2483  *   Pointer to the dev structure.
2484  * @param[in] attr
2485  *   Pointer to the flow attributes.
2486  * @param[in] items
2487  *   Pointer to the list of items.
2488  * @param[in] actions
2489  *   Pointer to the list of actions.
2490  * @param[in] external
2491  *   This flow rule is created by request external to PMD.
2492  * @param[in] hairpin
2493  *   Number of hairpin TX actions, 0 means classic flow.
2494  * @param[out] error
2495  *   Pointer to the error structure.
2496  *
2497  * @return
2498  *   0 on success, a negative errno value otherwise and rte_errno is set.
2499  */
2500 static inline int
2501 flow_drv_validate(struct rte_eth_dev *dev,
2502 		  const struct rte_flow_attr *attr,
2503 		  const struct rte_flow_item items[],
2504 		  const struct rte_flow_action actions[],
2505 		  bool external, int hairpin, struct rte_flow_error *error)
2506 {
2507 	const struct mlx5_flow_driver_ops *fops;
2508 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2509 
2510 	fops = flow_get_drv_ops(type);
2511 	return fops->validate(dev, attr, items, actions, external,
2512 			      hairpin, error);
2513 }
2514 
2515 /**
2516  * Flow driver preparation API. This abstracts calling driver specific
2517  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2518  * calculates the size of memory required for device flow, allocates the memory,
2519  * initializes the device flow and returns the pointer.
2520  *
2521  * @note
2522  *   This function initializes device flow structure such as dv or verbs in
2523  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2524  *   rest. For example, adding returning device flow to flow->dev_flow list and
2525  *   setting backward reference to the flow should be done out of this function.
2526  *   layers field is not filled either.
2527  *
2528  * @param[in] dev
2529  *   Pointer to the dev structure.
2530  * @param[in] attr
2531  *   Pointer to the flow attributes.
2532  * @param[in] items
2533  *   Pointer to the list of items.
2534  * @param[in] actions
2535  *   Pointer to the list of actions.
2536  * @param[in] flow_idx
2537  *   This memory pool index to the flow.
2538  * @param[out] error
2539  *   Pointer to the error structure.
2540  *
2541  * @return
2542  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2543  */
2544 static inline struct mlx5_flow *
2545 flow_drv_prepare(struct rte_eth_dev *dev,
2546 		 const struct rte_flow *flow,
2547 		 const struct rte_flow_attr *attr,
2548 		 const struct rte_flow_item items[],
2549 		 const struct rte_flow_action actions[],
2550 		 uint32_t flow_idx,
2551 		 struct rte_flow_error *error)
2552 {
2553 	const struct mlx5_flow_driver_ops *fops;
2554 	enum mlx5_flow_drv_type type = flow->drv_type;
2555 	struct mlx5_flow *mlx5_flow = NULL;
2556 
2557 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2558 	fops = flow_get_drv_ops(type);
2559 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2560 	if (mlx5_flow)
2561 		mlx5_flow->flow_idx = flow_idx;
2562 	return mlx5_flow;
2563 }
2564 
2565 /**
2566  * Flow driver translation API. This abstracts calling driver specific
2567  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2568  * translates a generic flow into a driver flow. flow_drv_prepare() must
2569  * precede.
2570  *
2571  * @note
2572  *   dev_flow->layers could be filled as a result of parsing during translation
2573  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2574  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2575  *   flow->actions could be overwritten even though all the expanded dev_flows
2576  *   have the same actions.
2577  *
2578  * @param[in] dev
2579  *   Pointer to the rte dev structure.
2580  * @param[in, out] dev_flow
2581  *   Pointer to the mlx5 flow.
2582  * @param[in] attr
2583  *   Pointer to the flow attributes.
2584  * @param[in] items
2585  *   Pointer to the list of items.
2586  * @param[in] actions
2587  *   Pointer to the list of actions.
2588  * @param[out] error
2589  *   Pointer to the error structure.
2590  *
2591  * @return
2592  *   0 on success, a negative errno value otherwise and rte_errno is set.
2593  */
2594 static inline int
2595 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2596 		   const struct rte_flow_attr *attr,
2597 		   const struct rte_flow_item items[],
2598 		   const struct rte_flow_action actions[],
2599 		   struct rte_flow_error *error)
2600 {
2601 	const struct mlx5_flow_driver_ops *fops;
2602 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2603 
2604 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2605 	fops = flow_get_drv_ops(type);
2606 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2607 }
2608 
2609 /**
2610  * Flow driver apply API. This abstracts calling driver specific functions.
2611  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2612  * translated driver flows on to device. flow_drv_translate() must precede.
2613  *
2614  * @param[in] dev
2615  *   Pointer to Ethernet device structure.
2616  * @param[in, out] flow
2617  *   Pointer to flow structure.
2618  * @param[out] error
2619  *   Pointer to error structure.
2620  *
2621  * @return
2622  *   0 on success, a negative errno value otherwise and rte_errno is set.
2623  */
2624 static inline int
2625 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2626 	       struct rte_flow_error *error)
2627 {
2628 	const struct mlx5_flow_driver_ops *fops;
2629 	enum mlx5_flow_drv_type type = flow->drv_type;
2630 
2631 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2632 	fops = flow_get_drv_ops(type);
2633 	return fops->apply(dev, flow, error);
2634 }
2635 
2636 /**
2637  * Flow driver remove API. This abstracts calling driver specific functions.
2638  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2639  * on device. All the resources of the flow should be freed by calling
2640  * flow_drv_destroy().
2641  *
2642  * @param[in] dev
2643  *   Pointer to Ethernet device.
2644  * @param[in, out] flow
2645  *   Pointer to flow structure.
2646  */
2647 static inline void
2648 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2649 {
2650 	const struct mlx5_flow_driver_ops *fops;
2651 	enum mlx5_flow_drv_type type = flow->drv_type;
2652 
2653 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2654 	fops = flow_get_drv_ops(type);
2655 	fops->remove(dev, flow);
2656 }
2657 
2658 /**
2659  * Flow driver destroy API. This abstracts calling driver specific functions.
2660  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2661  * on device and releases resources of the flow.
2662  *
2663  * @param[in] dev
2664  *   Pointer to Ethernet device.
2665  * @param[in, out] flow
2666  *   Pointer to flow structure.
2667  */
2668 static inline void
2669 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2670 {
2671 	const struct mlx5_flow_driver_ops *fops;
2672 	enum mlx5_flow_drv_type type = flow->drv_type;
2673 
2674 	flow_mreg_split_qrss_release(dev, flow);
2675 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2676 	fops = flow_get_drv_ops(type);
2677 	fops->destroy(dev, flow);
2678 }
2679 
2680 /**
2681  * Get RSS action from the action list.
2682  *
2683  * @param[in] actions
2684  *   Pointer to the list of actions.
2685  *
2686  * @return
2687  *   Pointer to the RSS action if exist, else return NULL.
2688  */
2689 static const struct rte_flow_action_rss*
2690 flow_get_rss_action(const struct rte_flow_action actions[])
2691 {
2692 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2693 		switch (actions->type) {
2694 		case RTE_FLOW_ACTION_TYPE_RSS:
2695 			return (const struct rte_flow_action_rss *)
2696 			       actions->conf;
2697 		default:
2698 			break;
2699 		}
2700 	}
2701 	return NULL;
2702 }
2703 
2704 static unsigned int
2705 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2706 {
2707 	const struct rte_flow_item *item;
2708 	unsigned int has_vlan = 0;
2709 
2710 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2711 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2712 			has_vlan = 1;
2713 			break;
2714 		}
2715 	}
2716 	if (has_vlan)
2717 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2718 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2719 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2720 			       MLX5_EXPANSION_ROOT_OUTER;
2721 }
2722 
2723 /**
2724  *  Get layer flags from the prefix flow.
2725  *
2726  *  Some flows may be split to several subflows, the prefix subflow gets the
2727  *  match items and the suffix sub flow gets the actions.
2728  *  Some actions need the user defined match item flags to get the detail for
2729  *  the action.
2730  *  This function helps the suffix flow to get the item layer flags from prefix
2731  *  subflow.
2732  *
2733  * @param[in] dev_flow
2734  *   Pointer the created preifx subflow.
2735  *
2736  * @return
2737  *   The layers get from prefix subflow.
2738  */
2739 static inline uint64_t
2740 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2741 {
2742 	uint64_t layers = 0;
2743 
2744 	/*
2745 	 * Layers bits could be localization, but usually the compiler will
2746 	 * help to do the optimization work for source code.
2747 	 * If no decap actions, use the layers directly.
2748 	 */
2749 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2750 		return dev_flow->handle->layers;
2751 	/* Convert L3 layers with decap action. */
2752 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2753 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2754 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2755 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2756 	/* Convert L4 layers with decap action.  */
2757 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2758 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2759 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2760 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2761 	return layers;
2762 }
2763 
2764 /**
2765  * Get metadata split action information.
2766  *
2767  * @param[in] actions
2768  *   Pointer to the list of actions.
2769  * @param[out] qrss
2770  *   Pointer to the return pointer.
2771  * @param[out] qrss_type
2772  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2773  *   if no QUEUE/RSS is found.
2774  * @param[out] encap_idx
2775  *   Pointer to the index of the encap action if exists, otherwise the last
2776  *   action index.
2777  *
2778  * @return
2779  *   Total number of actions.
2780  */
2781 static int
2782 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2783 				       const struct rte_flow_action **qrss,
2784 				       int *encap_idx)
2785 {
2786 	const struct rte_flow_action_raw_encap *raw_encap;
2787 	int actions_n = 0;
2788 	int raw_decap_idx = -1;
2789 
2790 	*encap_idx = -1;
2791 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2792 		switch (actions->type) {
2793 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2794 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2795 			*encap_idx = actions_n;
2796 			break;
2797 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2798 			raw_decap_idx = actions_n;
2799 			break;
2800 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2801 			raw_encap = actions->conf;
2802 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2803 				*encap_idx = raw_decap_idx != -1 ?
2804 						      raw_decap_idx : actions_n;
2805 			break;
2806 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2807 		case RTE_FLOW_ACTION_TYPE_RSS:
2808 			*qrss = actions;
2809 			break;
2810 		default:
2811 			break;
2812 		}
2813 		actions_n++;
2814 	}
2815 	if (*encap_idx == -1)
2816 		*encap_idx = actions_n;
2817 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2818 	return actions_n + 1;
2819 }
2820 
2821 /**
2822  * Check meter action from the action list.
2823  *
2824  * @param[in] actions
2825  *   Pointer to the list of actions.
2826  * @param[out] mtr
2827  *   Pointer to the meter exist flag.
2828  *
2829  * @return
2830  *   Total number of actions.
2831  */
2832 static int
2833 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2834 {
2835 	int actions_n = 0;
2836 
2837 	MLX5_ASSERT(mtr);
2838 	*mtr = 0;
2839 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2840 		switch (actions->type) {
2841 		case RTE_FLOW_ACTION_TYPE_METER:
2842 			*mtr = 1;
2843 			break;
2844 		default:
2845 			break;
2846 		}
2847 		actions_n++;
2848 	}
2849 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2850 	return actions_n + 1;
2851 }
2852 
2853 /**
2854  * Check if the flow should be splited due to hairpin.
2855  * The reason for the split is that in current HW we can't
2856  * support encap on Rx, so if a flow have encap we move it
2857  * to Tx.
2858  *
2859  * @param dev
2860  *   Pointer to Ethernet device.
2861  * @param[in] attr
2862  *   Flow rule attributes.
2863  * @param[in] actions
2864  *   Associated actions (list terminated by the END action).
2865  *
2866  * @return
2867  *   > 0 the number of actions and the flow should be split,
2868  *   0 when no split required.
2869  */
2870 static int
2871 flow_check_hairpin_split(struct rte_eth_dev *dev,
2872 			 const struct rte_flow_attr *attr,
2873 			 const struct rte_flow_action actions[])
2874 {
2875 	int queue_action = 0;
2876 	int action_n = 0;
2877 	int encap = 0;
2878 	const struct rte_flow_action_queue *queue;
2879 	const struct rte_flow_action_rss *rss;
2880 	const struct rte_flow_action_raw_encap *raw_encap;
2881 
2882 	if (!attr->ingress)
2883 		return 0;
2884 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2885 		switch (actions->type) {
2886 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2887 			queue = actions->conf;
2888 			if (queue == NULL)
2889 				return 0;
2890 			if (mlx5_rxq_get_type(dev, queue->index) !=
2891 			    MLX5_RXQ_TYPE_HAIRPIN)
2892 				return 0;
2893 			queue_action = 1;
2894 			action_n++;
2895 			break;
2896 		case RTE_FLOW_ACTION_TYPE_RSS:
2897 			rss = actions->conf;
2898 			if (rss == NULL || rss->queue_num == 0)
2899 				return 0;
2900 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2901 			    MLX5_RXQ_TYPE_HAIRPIN)
2902 				return 0;
2903 			queue_action = 1;
2904 			action_n++;
2905 			break;
2906 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2907 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2908 			encap = 1;
2909 			action_n++;
2910 			break;
2911 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2912 			raw_encap = actions->conf;
2913 			if (raw_encap->size >
2914 			    (sizeof(struct rte_flow_item_eth) +
2915 			     sizeof(struct rte_flow_item_ipv4)))
2916 				encap = 1;
2917 			action_n++;
2918 			break;
2919 		default:
2920 			action_n++;
2921 			break;
2922 		}
2923 	}
2924 	if (encap == 1 && queue_action)
2925 		return action_n;
2926 	return 0;
2927 }
2928 
2929 /* Declare flow create/destroy prototype in advance. */
2930 static uint32_t
2931 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2932 		 const struct rte_flow_attr *attr,
2933 		 const struct rte_flow_item items[],
2934 		 const struct rte_flow_action actions[],
2935 		 bool external, struct rte_flow_error *error);
2936 
2937 static void
2938 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2939 		  uint32_t flow_idx);
2940 
2941 /**
2942  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2943  *
2944  * As mark_id is unique, if there's already a registered flow for the mark_id,
2945  * return by increasing the reference counter of the resource. Otherwise, create
2946  * the resource (mcp_res) and flow.
2947  *
2948  * Flow looks like,
2949  *   - If ingress port is ANY and reg_c[1] is mark_id,
2950  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2951  *
2952  * For default flow (zero mark_id), flow is like,
2953  *   - If ingress port is ANY,
2954  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2955  *
2956  * @param dev
2957  *   Pointer to Ethernet device.
2958  * @param mark_id
2959  *   ID of MARK action, zero means default flow for META.
2960  * @param[out] error
2961  *   Perform verbose error reporting if not NULL.
2962  *
2963  * @return
2964  *   Associated resource on success, NULL otherwise and rte_errno is set.
2965  */
2966 static struct mlx5_flow_mreg_copy_resource *
2967 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2968 			  struct rte_flow_error *error)
2969 {
2970 	struct mlx5_priv *priv = dev->data->dev_private;
2971 	struct rte_flow_attr attr = {
2972 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2973 		.ingress = 1,
2974 	};
2975 	struct mlx5_rte_flow_item_tag tag_spec = {
2976 		.data = mark_id,
2977 	};
2978 	struct rte_flow_item items[] = {
2979 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2980 	};
2981 	struct rte_flow_action_mark ftag = {
2982 		.id = mark_id,
2983 	};
2984 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2985 		.dst = REG_B,
2986 		.src = 0,
2987 	};
2988 	struct rte_flow_action_jump jump = {
2989 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2990 	};
2991 	struct rte_flow_action actions[] = {
2992 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2993 	};
2994 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2995 	uint32_t idx = 0;
2996 	int ret;
2997 
2998 	/* Fill the register fileds in the flow. */
2999 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3000 	if (ret < 0)
3001 		return NULL;
3002 	tag_spec.id = ret;
3003 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3004 	if (ret < 0)
3005 		return NULL;
3006 	cp_mreg.src = ret;
3007 	/* Check if already registered. */
3008 	MLX5_ASSERT(priv->mreg_cp_tbl);
3009 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3010 	if (mcp_res) {
3011 		/* For non-default rule. */
3012 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3013 			mcp_res->refcnt++;
3014 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3015 			    mcp_res->refcnt == 1);
3016 		return mcp_res;
3017 	}
3018 	/* Provide the full width of FLAG specific value. */
3019 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3020 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3021 	/* Build a new flow. */
3022 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3023 		items[0] = (struct rte_flow_item){
3024 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3025 			.spec = &tag_spec,
3026 		};
3027 		items[1] = (struct rte_flow_item){
3028 			.type = RTE_FLOW_ITEM_TYPE_END,
3029 		};
3030 		actions[0] = (struct rte_flow_action){
3031 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3032 			.conf = &ftag,
3033 		};
3034 		actions[1] = (struct rte_flow_action){
3035 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3036 			.conf = &cp_mreg,
3037 		};
3038 		actions[2] = (struct rte_flow_action){
3039 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3040 			.conf = &jump,
3041 		};
3042 		actions[3] = (struct rte_flow_action){
3043 			.type = RTE_FLOW_ACTION_TYPE_END,
3044 		};
3045 	} else {
3046 		/* Default rule, wildcard match. */
3047 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3048 		items[0] = (struct rte_flow_item){
3049 			.type = RTE_FLOW_ITEM_TYPE_END,
3050 		};
3051 		actions[0] = (struct rte_flow_action){
3052 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3053 			.conf = &cp_mreg,
3054 		};
3055 		actions[1] = (struct rte_flow_action){
3056 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3057 			.conf = &jump,
3058 		};
3059 		actions[2] = (struct rte_flow_action){
3060 			.type = RTE_FLOW_ACTION_TYPE_END,
3061 		};
3062 	}
3063 	/* Build a new entry. */
3064 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3065 	if (!mcp_res) {
3066 		rte_errno = ENOMEM;
3067 		return NULL;
3068 	}
3069 	mcp_res->idx = idx;
3070 	/*
3071 	 * The copy Flows are not included in any list. There
3072 	 * ones are referenced from other Flows and can not
3073 	 * be applied, removed, deleted in ardbitrary order
3074 	 * by list traversing.
3075 	 */
3076 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3077 					 actions, false, error);
3078 	if (!mcp_res->rix_flow)
3079 		goto error;
3080 	mcp_res->refcnt++;
3081 	mcp_res->hlist_ent.key = mark_id;
3082 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3083 				&mcp_res->hlist_ent);
3084 	MLX5_ASSERT(!ret);
3085 	if (ret)
3086 		goto error;
3087 	return mcp_res;
3088 error:
3089 	if (mcp_res->rix_flow)
3090 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3091 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3092 	return NULL;
3093 }
3094 
3095 /**
3096  * Release flow in RX_CP_TBL.
3097  *
3098  * @param dev
3099  *   Pointer to Ethernet device.
3100  * @flow
3101  *   Parent flow for wich copying is provided.
3102  */
3103 static void
3104 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3105 			  struct rte_flow *flow)
3106 {
3107 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3108 	struct mlx5_priv *priv = dev->data->dev_private;
3109 
3110 	if (!flow->rix_mreg_copy)
3111 		return;
3112 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3113 				 flow->rix_mreg_copy);
3114 	if (!mcp_res || !priv->mreg_cp_tbl)
3115 		return;
3116 	if (flow->copy_applied) {
3117 		MLX5_ASSERT(mcp_res->appcnt);
3118 		flow->copy_applied = 0;
3119 		--mcp_res->appcnt;
3120 		if (!mcp_res->appcnt) {
3121 			struct rte_flow *mcp_flow = mlx5_ipool_get
3122 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3123 					mcp_res->rix_flow);
3124 
3125 			if (mcp_flow)
3126 				flow_drv_remove(dev, mcp_flow);
3127 		}
3128 	}
3129 	/*
3130 	 * We do not check availability of metadata registers here,
3131 	 * because copy resources are not allocated in this case.
3132 	 */
3133 	if (--mcp_res->refcnt)
3134 		return;
3135 	MLX5_ASSERT(mcp_res->rix_flow);
3136 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3137 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3138 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3139 	flow->rix_mreg_copy = 0;
3140 }
3141 
3142 /**
3143  * Start flow in RX_CP_TBL.
3144  *
3145  * @param dev
3146  *   Pointer to Ethernet device.
3147  * @flow
3148  *   Parent flow for wich copying is provided.
3149  *
3150  * @return
3151  *   0 on success, a negative errno value otherwise and rte_errno is set.
3152  */
3153 static int
3154 flow_mreg_start_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 	int ret;
3160 
3161 	if (!flow->rix_mreg_copy || flow->copy_applied)
3162 		return 0;
3163 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3164 				 flow->rix_mreg_copy);
3165 	if (!mcp_res)
3166 		return 0;
3167 	if (!mcp_res->appcnt) {
3168 		struct rte_flow *mcp_flow = mlx5_ipool_get
3169 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3170 				mcp_res->rix_flow);
3171 
3172 		if (mcp_flow) {
3173 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3174 			if (ret)
3175 				return ret;
3176 		}
3177 	}
3178 	++mcp_res->appcnt;
3179 	flow->copy_applied = 1;
3180 	return 0;
3181 }
3182 
3183 /**
3184  * Stop flow in RX_CP_TBL.
3185  *
3186  * @param dev
3187  *   Pointer to Ethernet device.
3188  * @flow
3189  *   Parent flow for wich copying is provided.
3190  */
3191 static void
3192 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3193 			   struct rte_flow *flow)
3194 {
3195 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3196 	struct mlx5_priv *priv = dev->data->dev_private;
3197 
3198 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3199 		return;
3200 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3201 				 flow->rix_mreg_copy);
3202 	if (!mcp_res)
3203 		return;
3204 	MLX5_ASSERT(mcp_res->appcnt);
3205 	--mcp_res->appcnt;
3206 	flow->copy_applied = 0;
3207 	if (!mcp_res->appcnt) {
3208 		struct rte_flow *mcp_flow = mlx5_ipool_get
3209 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3210 				mcp_res->rix_flow);
3211 
3212 		if (mcp_flow)
3213 			flow_drv_remove(dev, mcp_flow);
3214 	}
3215 }
3216 
3217 /**
3218  * Remove the default copy action from RX_CP_TBL.
3219  *
3220  * @param dev
3221  *   Pointer to Ethernet device.
3222  */
3223 static void
3224 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3225 {
3226 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3227 	struct mlx5_priv *priv = dev->data->dev_private;
3228 
3229 	/* Check if default flow is registered. */
3230 	if (!priv->mreg_cp_tbl)
3231 		return;
3232 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3233 					    MLX5_DEFAULT_COPY_ID);
3234 	if (!mcp_res)
3235 		return;
3236 	MLX5_ASSERT(mcp_res->rix_flow);
3237 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3238 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3239 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3240 }
3241 
3242 /**
3243  * Add the default copy action in in RX_CP_TBL.
3244  *
3245  * @param dev
3246  *   Pointer to Ethernet device.
3247  * @param[out] error
3248  *   Perform verbose error reporting if not NULL.
3249  *
3250  * @return
3251  *   0 for success, negative value otherwise and rte_errno is set.
3252  */
3253 static int
3254 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3255 				  struct rte_flow_error *error)
3256 {
3257 	struct mlx5_priv *priv = dev->data->dev_private;
3258 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3259 
3260 	/* Check whether extensive metadata feature is engaged. */
3261 	if (!priv->config.dv_flow_en ||
3262 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3263 	    !mlx5_flow_ext_mreg_supported(dev) ||
3264 	    !priv->sh->dv_regc0_mask)
3265 		return 0;
3266 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3267 	if (!mcp_res)
3268 		return -rte_errno;
3269 	return 0;
3270 }
3271 
3272 /**
3273  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3274  *
3275  * All the flow having Q/RSS action should be split by
3276  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3277  * performs the following,
3278  *   - CQE->flow_tag := reg_c[1] (MARK)
3279  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3280  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3281  * but there should be a flow per each MARK ID set by MARK action.
3282  *
3283  * For the aforementioned reason, if there's a MARK action in flow's action
3284  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3285  * the MARK ID to CQE's flow_tag like,
3286  *   - If reg_c[1] is mark_id,
3287  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3288  *
3289  * For SET_META action which stores value in reg_c[0], as the destination is
3290  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3291  * MARK ID means the default flow. The default flow looks like,
3292  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3293  *
3294  * @param dev
3295  *   Pointer to Ethernet device.
3296  * @param flow
3297  *   Pointer to flow structure.
3298  * @param[in] actions
3299  *   Pointer to the list of actions.
3300  * @param[out] error
3301  *   Perform verbose error reporting if not NULL.
3302  *
3303  * @return
3304  *   0 on success, negative value otherwise and rte_errno is set.
3305  */
3306 static int
3307 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3308 			    struct rte_flow *flow,
3309 			    const struct rte_flow_action *actions,
3310 			    struct rte_flow_error *error)
3311 {
3312 	struct mlx5_priv *priv = dev->data->dev_private;
3313 	struct mlx5_dev_config *config = &priv->config;
3314 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3315 	const struct rte_flow_action_mark *mark;
3316 
3317 	/* Check whether extensive metadata feature is engaged. */
3318 	if (!config->dv_flow_en ||
3319 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3320 	    !mlx5_flow_ext_mreg_supported(dev) ||
3321 	    !priv->sh->dv_regc0_mask)
3322 		return 0;
3323 	/* Find MARK action. */
3324 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3325 		switch (actions->type) {
3326 		case RTE_FLOW_ACTION_TYPE_FLAG:
3327 			mcp_res = flow_mreg_add_copy_action
3328 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3329 			if (!mcp_res)
3330 				return -rte_errno;
3331 			flow->rix_mreg_copy = mcp_res->idx;
3332 			if (dev->data->dev_started) {
3333 				mcp_res->appcnt++;
3334 				flow->copy_applied = 1;
3335 			}
3336 			return 0;
3337 		case RTE_FLOW_ACTION_TYPE_MARK:
3338 			mark = (const struct rte_flow_action_mark *)
3339 				actions->conf;
3340 			mcp_res =
3341 				flow_mreg_add_copy_action(dev, mark->id, error);
3342 			if (!mcp_res)
3343 				return -rte_errno;
3344 			flow->rix_mreg_copy = mcp_res->idx;
3345 			if (dev->data->dev_started) {
3346 				mcp_res->appcnt++;
3347 				flow->copy_applied = 1;
3348 			}
3349 			return 0;
3350 		default:
3351 			break;
3352 		}
3353 	}
3354 	return 0;
3355 }
3356 
3357 #define MLX5_MAX_SPLIT_ACTIONS 24
3358 #define MLX5_MAX_SPLIT_ITEMS 24
3359 
3360 /**
3361  * Split the hairpin flow.
3362  * Since HW can't support encap on Rx we move the encap to Tx.
3363  * If the count action is after the encap then we also
3364  * move the count action. in this case the count will also measure
3365  * the outer bytes.
3366  *
3367  * @param dev
3368  *   Pointer to Ethernet device.
3369  * @param[in] actions
3370  *   Associated actions (list terminated by the END action).
3371  * @param[out] actions_rx
3372  *   Rx flow actions.
3373  * @param[out] actions_tx
3374  *   Tx flow actions..
3375  * @param[out] pattern_tx
3376  *   The pattern items for the Tx flow.
3377  * @param[out] flow_id
3378  *   The flow ID connected to this flow.
3379  *
3380  * @return
3381  *   0 on success.
3382  */
3383 static int
3384 flow_hairpin_split(struct rte_eth_dev *dev,
3385 		   const struct rte_flow_action actions[],
3386 		   struct rte_flow_action actions_rx[],
3387 		   struct rte_flow_action actions_tx[],
3388 		   struct rte_flow_item pattern_tx[],
3389 		   uint32_t *flow_id)
3390 {
3391 	struct mlx5_priv *priv = dev->data->dev_private;
3392 	const struct rte_flow_action_raw_encap *raw_encap;
3393 	const struct rte_flow_action_raw_decap *raw_decap;
3394 	struct mlx5_rte_flow_action_set_tag *set_tag;
3395 	struct rte_flow_action *tag_action;
3396 	struct mlx5_rte_flow_item_tag *tag_item;
3397 	struct rte_flow_item *item;
3398 	char *addr;
3399 	int encap = 0;
3400 
3401 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3402 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3403 		switch (actions->type) {
3404 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3405 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3406 			rte_memcpy(actions_tx, actions,
3407 			       sizeof(struct rte_flow_action));
3408 			actions_tx++;
3409 			break;
3410 		case RTE_FLOW_ACTION_TYPE_COUNT:
3411 			if (encap) {
3412 				rte_memcpy(actions_tx, actions,
3413 					   sizeof(struct rte_flow_action));
3414 				actions_tx++;
3415 			} else {
3416 				rte_memcpy(actions_rx, actions,
3417 					   sizeof(struct rte_flow_action));
3418 				actions_rx++;
3419 			}
3420 			break;
3421 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3422 			raw_encap = actions->conf;
3423 			if (raw_encap->size >
3424 			    (sizeof(struct rte_flow_item_eth) +
3425 			     sizeof(struct rte_flow_item_ipv4))) {
3426 				memcpy(actions_tx, actions,
3427 				       sizeof(struct rte_flow_action));
3428 				actions_tx++;
3429 				encap = 1;
3430 			} else {
3431 				rte_memcpy(actions_rx, actions,
3432 					   sizeof(struct rte_flow_action));
3433 				actions_rx++;
3434 			}
3435 			break;
3436 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3437 			raw_decap = actions->conf;
3438 			if (raw_decap->size <
3439 			    (sizeof(struct rte_flow_item_eth) +
3440 			     sizeof(struct rte_flow_item_ipv4))) {
3441 				memcpy(actions_tx, actions,
3442 				       sizeof(struct rte_flow_action));
3443 				actions_tx++;
3444 			} else {
3445 				rte_memcpy(actions_rx, actions,
3446 					   sizeof(struct rte_flow_action));
3447 				actions_rx++;
3448 			}
3449 			break;
3450 		default:
3451 			rte_memcpy(actions_rx, actions,
3452 				   sizeof(struct rte_flow_action));
3453 			actions_rx++;
3454 			break;
3455 		}
3456 	}
3457 	/* Add set meta action and end action for the Rx flow. */
3458 	tag_action = actions_rx;
3459 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3460 	actions_rx++;
3461 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3462 	actions_rx++;
3463 	set_tag = (void *)actions_rx;
3464 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3465 	MLX5_ASSERT(set_tag->id > REG_NONE);
3466 	set_tag->data = *flow_id;
3467 	tag_action->conf = set_tag;
3468 	/* Create Tx item list. */
3469 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3470 	addr = (void *)&pattern_tx[2];
3471 	item = pattern_tx;
3472 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3473 	tag_item = (void *)addr;
3474 	tag_item->data = *flow_id;
3475 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3476 	MLX5_ASSERT(set_tag->id > REG_NONE);
3477 	item->spec = tag_item;
3478 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3479 	tag_item = (void *)addr;
3480 	tag_item->data = UINT32_MAX;
3481 	tag_item->id = UINT16_MAX;
3482 	item->mask = tag_item;
3483 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3484 	item->last = NULL;
3485 	item++;
3486 	item->type = RTE_FLOW_ITEM_TYPE_END;
3487 	return 0;
3488 }
3489 
3490 /**
3491  * The last stage of splitting chain, just creates the subflow
3492  * without any modification.
3493  *
3494  * @param[in] dev
3495  *   Pointer to Ethernet device.
3496  * @param[in] flow
3497  *   Parent flow structure pointer.
3498  * @param[in, out] sub_flow
3499  *   Pointer to return the created subflow, may be NULL.
3500  * @param[in] prefix_layers
3501  *   Prefix subflow layers, may be 0.
3502  * @param[in] attr
3503  *   Flow rule attributes.
3504  * @param[in] items
3505  *   Pattern specification (list terminated by the END pattern item).
3506  * @param[in] actions
3507  *   Associated actions (list terminated by the END action).
3508  * @param[in] external
3509  *   This flow rule is created by request external to PMD.
3510  * @param[in] flow_idx
3511  *   This memory pool index to the flow.
3512  * @param[out] error
3513  *   Perform verbose error reporting if not NULL.
3514  * @return
3515  *   0 on success, negative value otherwise
3516  */
3517 static int
3518 flow_create_split_inner(struct rte_eth_dev *dev,
3519 			struct rte_flow *flow,
3520 			struct mlx5_flow **sub_flow,
3521 			uint64_t prefix_layers,
3522 			const struct rte_flow_attr *attr,
3523 			const struct rte_flow_item items[],
3524 			const struct rte_flow_action actions[],
3525 			bool external, uint32_t flow_idx,
3526 			struct rte_flow_error *error)
3527 {
3528 	struct mlx5_flow *dev_flow;
3529 
3530 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3531 		flow_idx, error);
3532 	if (!dev_flow)
3533 		return -rte_errno;
3534 	dev_flow->flow = flow;
3535 	dev_flow->external = external;
3536 	/* Subflow object was created, we must include one in the list. */
3537 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3538 		      dev_flow->handle, next);
3539 	/*
3540 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3541 	 * flow may need some user defined item layer flags.
3542 	 */
3543 	if (prefix_layers)
3544 		dev_flow->handle->layers = prefix_layers;
3545 	if (sub_flow)
3546 		*sub_flow = dev_flow;
3547 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3548 }
3549 
3550 /**
3551  * Split the meter flow.
3552  *
3553  * As meter flow will split to three sub flow, other than meter
3554  * action, the other actions make sense to only meter accepts
3555  * the packet. If it need to be dropped, no other additional
3556  * actions should be take.
3557  *
3558  * One kind of special action which decapsulates the L3 tunnel
3559  * header will be in the prefix sub flow, as not to take the
3560  * L3 tunnel header into account.
3561  *
3562  * @param dev
3563  *   Pointer to Ethernet device.
3564  * @param[in] items
3565  *   Pattern specification (list terminated by the END pattern item).
3566  * @param[out] sfx_items
3567  *   Suffix flow match items (list terminated by the END pattern item).
3568  * @param[in] actions
3569  *   Associated actions (list terminated by the END action).
3570  * @param[out] actions_sfx
3571  *   Suffix flow actions.
3572  * @param[out] actions_pre
3573  *   Prefix flow actions.
3574  * @param[out] pattern_sfx
3575  *   The pattern items for the suffix flow.
3576  * @param[out] tag_sfx
3577  *   Pointer to suffix flow tag.
3578  *
3579  * @return
3580  *   0 on success.
3581  */
3582 static int
3583 flow_meter_split_prep(struct rte_eth_dev *dev,
3584 		 const struct rte_flow_item items[],
3585 		 struct rte_flow_item sfx_items[],
3586 		 const struct rte_flow_action actions[],
3587 		 struct rte_flow_action actions_sfx[],
3588 		 struct rte_flow_action actions_pre[])
3589 {
3590 	struct rte_flow_action *tag_action = NULL;
3591 	struct rte_flow_item *tag_item;
3592 	struct mlx5_rte_flow_action_set_tag *set_tag;
3593 	struct rte_flow_error error;
3594 	const struct rte_flow_action_raw_encap *raw_encap;
3595 	const struct rte_flow_action_raw_decap *raw_decap;
3596 	struct mlx5_rte_flow_item_tag *tag_spec;
3597 	struct mlx5_rte_flow_item_tag *tag_mask;
3598 	uint32_t tag_id;
3599 	bool copy_vlan = false;
3600 
3601 	/* Prepare the actions for prefix and suffix flow. */
3602 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3603 		struct rte_flow_action **action_cur = NULL;
3604 
3605 		switch (actions->type) {
3606 		case RTE_FLOW_ACTION_TYPE_METER:
3607 			/* Add the extra tag action first. */
3608 			tag_action = actions_pre;
3609 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3610 			actions_pre++;
3611 			action_cur = &actions_pre;
3612 			break;
3613 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3614 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3615 			action_cur = &actions_pre;
3616 			break;
3617 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3618 			raw_encap = actions->conf;
3619 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3620 				action_cur = &actions_pre;
3621 			break;
3622 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3623 			raw_decap = actions->conf;
3624 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3625 				action_cur = &actions_pre;
3626 			break;
3627 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3628 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3629 			copy_vlan = true;
3630 			break;
3631 		default:
3632 			break;
3633 		}
3634 		if (!action_cur)
3635 			action_cur = &actions_sfx;
3636 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3637 		(*action_cur)++;
3638 	}
3639 	/* Add end action to the actions. */
3640 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3641 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3642 	actions_pre++;
3643 	/* Set the tag. */
3644 	set_tag = (void *)actions_pre;
3645 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3646 	/*
3647 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3648 	 */
3649 	tag_id = flow_qrss_get_id(dev);
3650 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3651 	assert(tag_action);
3652 	tag_action->conf = set_tag;
3653 	/* Prepare the suffix subflow items. */
3654 	tag_item = sfx_items++;
3655 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3656 		int item_type = items->type;
3657 
3658 		switch (item_type) {
3659 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3660 			memcpy(sfx_items, items, sizeof(*sfx_items));
3661 			sfx_items++;
3662 			break;
3663 		case RTE_FLOW_ITEM_TYPE_VLAN:
3664 			if (copy_vlan) {
3665 				memcpy(sfx_items, items, sizeof(*sfx_items));
3666 				/*
3667 				 * Convert to internal match item, it is used
3668 				 * for vlan push and set vid.
3669 				 */
3670 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3671 				sfx_items++;
3672 			}
3673 			break;
3674 		default:
3675 			break;
3676 		}
3677 	}
3678 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3679 	sfx_items++;
3680 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3681 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3682 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3683 	tag_mask = tag_spec + 1;
3684 	tag_mask->data = 0xffffff00;
3685 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3686 	tag_item->spec = tag_spec;
3687 	tag_item->last = NULL;
3688 	tag_item->mask = tag_mask;
3689 	return tag_id;
3690 }
3691 
3692 /**
3693  * Split action list having QUEUE/RSS for metadata register copy.
3694  *
3695  * Once Q/RSS action is detected in user's action list, the flow action
3696  * should be split in order to copy metadata registers, which will happen in
3697  * RX_CP_TBL like,
3698  *   - CQE->flow_tag := reg_c[1] (MARK)
3699  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3700  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3701  * This is because the last action of each flow must be a terminal action
3702  * (QUEUE, RSS or DROP).
3703  *
3704  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3705  * stored and kept in the mlx5_flow structure per each sub_flow.
3706  *
3707  * The Q/RSS action is replaced with,
3708  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3709  * And the following JUMP action is added at the end,
3710  *   - JUMP, to RX_CP_TBL.
3711  *
3712  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3713  * flow_create_split_metadata() routine. The flow will look like,
3714  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3715  *
3716  * @param dev
3717  *   Pointer to Ethernet device.
3718  * @param[out] split_actions
3719  *   Pointer to store split actions to jump to CP_TBL.
3720  * @param[in] actions
3721  *   Pointer to the list of original flow actions.
3722  * @param[in] qrss
3723  *   Pointer to the Q/RSS action.
3724  * @param[in] actions_n
3725  *   Number of original actions.
3726  * @param[out] error
3727  *   Perform verbose error reporting if not NULL.
3728  *
3729  * @return
3730  *   non-zero unique flow_id on success, otherwise 0 and
3731  *   error/rte_error are set.
3732  */
3733 static uint32_t
3734 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3735 			  struct rte_flow_action *split_actions,
3736 			  const struct rte_flow_action *actions,
3737 			  const struct rte_flow_action *qrss,
3738 			  int actions_n, struct rte_flow_error *error)
3739 {
3740 	struct mlx5_rte_flow_action_set_tag *set_tag;
3741 	struct rte_flow_action_jump *jump;
3742 	const int qrss_idx = qrss - actions;
3743 	uint32_t flow_id = 0;
3744 	int ret = 0;
3745 
3746 	/*
3747 	 * Given actions will be split
3748 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3749 	 * - Add jump to mreg CP_TBL.
3750 	 * As a result, there will be one more action.
3751 	 */
3752 	++actions_n;
3753 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3754 	set_tag = (void *)(split_actions + actions_n);
3755 	/*
3756 	 * If tag action is not set to void(it means we are not the meter
3757 	 * suffix flow), add the tag action. Since meter suffix flow already
3758 	 * has the tag added.
3759 	 */
3760 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3761 		/*
3762 		 * Allocate the new subflow ID. This one is unique within
3763 		 * device and not shared with representors. Otherwise,
3764 		 * we would have to resolve multi-thread access synch
3765 		 * issue. Each flow on the shared device is appended
3766 		 * with source vport identifier, so the resulting
3767 		 * flows will be unique in the shared (by master and
3768 		 * representors) domain even if they have coinciding
3769 		 * IDs.
3770 		 */
3771 		flow_id = flow_qrss_get_id(dev);
3772 		if (!flow_id)
3773 			return rte_flow_error_set(error, ENOMEM,
3774 						  RTE_FLOW_ERROR_TYPE_ACTION,
3775 						  NULL, "can't allocate id "
3776 						  "for split Q/RSS subflow");
3777 		/* Internal SET_TAG action to set flow ID. */
3778 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3779 			.data = flow_id,
3780 		};
3781 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3782 		if (ret < 0)
3783 			return ret;
3784 		set_tag->id = ret;
3785 		/* Construct new actions array. */
3786 		/* Replace QUEUE/RSS action. */
3787 		split_actions[qrss_idx] = (struct rte_flow_action){
3788 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3789 			.conf = set_tag,
3790 		};
3791 	}
3792 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3793 	jump = (void *)(set_tag + 1);
3794 	*jump = (struct rte_flow_action_jump){
3795 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3796 	};
3797 	split_actions[actions_n - 2] = (struct rte_flow_action){
3798 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3799 		.conf = jump,
3800 	};
3801 	split_actions[actions_n - 1] = (struct rte_flow_action){
3802 		.type = RTE_FLOW_ACTION_TYPE_END,
3803 	};
3804 	return flow_id;
3805 }
3806 
3807 /**
3808  * Extend the given action list for Tx metadata copy.
3809  *
3810  * Copy the given action list to the ext_actions and add flow metadata register
3811  * copy action in order to copy reg_a set by WQE to reg_c[0].
3812  *
3813  * @param[out] ext_actions
3814  *   Pointer to the extended action list.
3815  * @param[in] actions
3816  *   Pointer to the list of actions.
3817  * @param[in] actions_n
3818  *   Number of actions in the list.
3819  * @param[out] error
3820  *   Perform verbose error reporting if not NULL.
3821  * @param[in] encap_idx
3822  *   The encap action inndex.
3823  *
3824  * @return
3825  *   0 on success, negative value otherwise
3826  */
3827 static int
3828 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3829 		       struct rte_flow_action *ext_actions,
3830 		       const struct rte_flow_action *actions,
3831 		       int actions_n, struct rte_flow_error *error,
3832 		       int encap_idx)
3833 {
3834 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3835 		(struct mlx5_flow_action_copy_mreg *)
3836 			(ext_actions + actions_n + 1);
3837 	int ret;
3838 
3839 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3840 	if (ret < 0)
3841 		return ret;
3842 	cp_mreg->dst = ret;
3843 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3844 	if (ret < 0)
3845 		return ret;
3846 	cp_mreg->src = ret;
3847 	if (encap_idx != 0)
3848 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3849 	if (encap_idx == actions_n - 1) {
3850 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3851 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3852 			.conf = cp_mreg,
3853 		};
3854 		ext_actions[actions_n] = (struct rte_flow_action){
3855 			.type = RTE_FLOW_ACTION_TYPE_END,
3856 		};
3857 	} else {
3858 		ext_actions[encap_idx] = (struct rte_flow_action){
3859 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3860 			.conf = cp_mreg,
3861 		};
3862 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3863 				sizeof(*ext_actions) * (actions_n - encap_idx));
3864 	}
3865 	return 0;
3866 }
3867 
3868 /**
3869  * The splitting for metadata feature.
3870  *
3871  * - Q/RSS action on NIC Rx should be split in order to pass by
3872  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3873  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3874  *
3875  * - All the actions on NIC Tx should have a mreg copy action to
3876  *   copy reg_a from WQE to reg_c[0].
3877  *
3878  * @param dev
3879  *   Pointer to Ethernet device.
3880  * @param[in] flow
3881  *   Parent flow structure pointer.
3882  * @param[in] prefix_layers
3883  *   Prefix flow layer flags.
3884  * @param[in] attr
3885  *   Flow rule attributes.
3886  * @param[in] items
3887  *   Pattern specification (list terminated by the END pattern item).
3888  * @param[in] actions
3889  *   Associated actions (list terminated by the END action).
3890  * @param[in] external
3891  *   This flow rule is created by request external to PMD.
3892  * @param[in] flow_idx
3893  *   This memory pool index to the flow.
3894  * @param[out] error
3895  *   Perform verbose error reporting if not NULL.
3896  * @return
3897  *   0 on success, negative value otherwise
3898  */
3899 static int
3900 flow_create_split_metadata(struct rte_eth_dev *dev,
3901 			   struct rte_flow *flow,
3902 			   uint64_t prefix_layers,
3903 			   const struct rte_flow_attr *attr,
3904 			   const struct rte_flow_item items[],
3905 			   const struct rte_flow_action actions[],
3906 			   bool external, uint32_t flow_idx,
3907 			   struct rte_flow_error *error)
3908 {
3909 	struct mlx5_priv *priv = dev->data->dev_private;
3910 	struct mlx5_dev_config *config = &priv->config;
3911 	const struct rte_flow_action *qrss = NULL;
3912 	struct rte_flow_action *ext_actions = NULL;
3913 	struct mlx5_flow *dev_flow = NULL;
3914 	uint32_t qrss_id = 0;
3915 	int mtr_sfx = 0;
3916 	size_t act_size;
3917 	int actions_n;
3918 	int encap_idx;
3919 	int ret;
3920 
3921 	/* Check whether extensive metadata feature is engaged. */
3922 	if (!config->dv_flow_en ||
3923 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3924 	    !mlx5_flow_ext_mreg_supported(dev))
3925 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3926 					       attr, items, actions, external,
3927 					       flow_idx, error);
3928 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3929 							   &encap_idx);
3930 	if (qrss) {
3931 		/* Exclude hairpin flows from splitting. */
3932 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3933 			const struct rte_flow_action_queue *queue;
3934 
3935 			queue = qrss->conf;
3936 			if (mlx5_rxq_get_type(dev, queue->index) ==
3937 			    MLX5_RXQ_TYPE_HAIRPIN)
3938 				qrss = NULL;
3939 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3940 			const struct rte_flow_action_rss *rss;
3941 
3942 			rss = qrss->conf;
3943 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3944 			    MLX5_RXQ_TYPE_HAIRPIN)
3945 				qrss = NULL;
3946 		}
3947 	}
3948 	if (qrss) {
3949 		/* Check if it is in meter suffix table. */
3950 		mtr_sfx = attr->group == (attr->transfer ?
3951 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3952 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3953 		/*
3954 		 * Q/RSS action on NIC Rx should be split in order to pass by
3955 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3956 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3957 		 */
3958 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3959 			   sizeof(struct rte_flow_action_set_tag) +
3960 			   sizeof(struct rte_flow_action_jump);
3961 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3962 		if (!ext_actions)
3963 			return rte_flow_error_set(error, ENOMEM,
3964 						  RTE_FLOW_ERROR_TYPE_ACTION,
3965 						  NULL, "no memory to split "
3966 						  "metadata flow");
3967 		/*
3968 		 * If we are the suffix flow of meter, tag already exist.
3969 		 * Set the tag action to void.
3970 		 */
3971 		if (mtr_sfx)
3972 			ext_actions[qrss - actions].type =
3973 						RTE_FLOW_ACTION_TYPE_VOID;
3974 		else
3975 			ext_actions[qrss - actions].type =
3976 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3977 		/*
3978 		 * Create the new actions list with removed Q/RSS action
3979 		 * and appended set tag and jump to register copy table
3980 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3981 		 * in advance, because it is needed for set tag action.
3982 		 */
3983 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3984 						    qrss, actions_n, error);
3985 		if (!mtr_sfx && !qrss_id) {
3986 			ret = -rte_errno;
3987 			goto exit;
3988 		}
3989 	} else if (attr->egress && !attr->transfer) {
3990 		/*
3991 		 * All the actions on NIC Tx should have a metadata register
3992 		 * copy action to copy reg_a from WQE to reg_c[meta]
3993 		 */
3994 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3995 			   sizeof(struct mlx5_flow_action_copy_mreg);
3996 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3997 		if (!ext_actions)
3998 			return rte_flow_error_set(error, ENOMEM,
3999 						  RTE_FLOW_ERROR_TYPE_ACTION,
4000 						  NULL, "no memory to split "
4001 						  "metadata flow");
4002 		/* Create the action list appended with copy register. */
4003 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4004 					     actions_n, error, encap_idx);
4005 		if (ret < 0)
4006 			goto exit;
4007 	}
4008 	/* Add the unmodified original or prefix subflow. */
4009 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4010 				      items, ext_actions ? ext_actions :
4011 				      actions, external, flow_idx, error);
4012 	if (ret < 0)
4013 		goto exit;
4014 	MLX5_ASSERT(dev_flow);
4015 	if (qrss) {
4016 		const struct rte_flow_attr q_attr = {
4017 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4018 			.ingress = 1,
4019 		};
4020 		/* Internal PMD action to set register. */
4021 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4022 			.data = qrss_id,
4023 			.id = 0,
4024 		};
4025 		struct rte_flow_item q_items[] = {
4026 			{
4027 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4028 				.spec = &q_tag_spec,
4029 				.last = NULL,
4030 				.mask = NULL,
4031 			},
4032 			{
4033 				.type = RTE_FLOW_ITEM_TYPE_END,
4034 			},
4035 		};
4036 		struct rte_flow_action q_actions[] = {
4037 			{
4038 				.type = qrss->type,
4039 				.conf = qrss->conf,
4040 			},
4041 			{
4042 				.type = RTE_FLOW_ACTION_TYPE_END,
4043 			},
4044 		};
4045 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4046 
4047 		/*
4048 		 * Configure the tag item only if there is no meter subflow.
4049 		 * Since tag is already marked in the meter suffix subflow
4050 		 * we can just use the meter suffix items as is.
4051 		 */
4052 		if (qrss_id) {
4053 			/* Not meter subflow. */
4054 			MLX5_ASSERT(!mtr_sfx);
4055 			/*
4056 			 * Put unique id in prefix flow due to it is destroyed
4057 			 * after suffix flow and id will be freed after there
4058 			 * is no actual flows with this id and identifier
4059 			 * reallocation becomes possible (for example, for
4060 			 * other flows in other threads).
4061 			 */
4062 			dev_flow->handle->split_flow_id = qrss_id;
4063 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4064 						   error);
4065 			if (ret < 0)
4066 				goto exit;
4067 			q_tag_spec.id = ret;
4068 		}
4069 		dev_flow = NULL;
4070 		/* Add suffix subflow to execute Q/RSS. */
4071 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4072 					      &q_attr, mtr_sfx ? items :
4073 					      q_items, q_actions,
4074 					      external, flow_idx, error);
4075 		if (ret < 0)
4076 			goto exit;
4077 		/* qrss ID should be freed if failed. */
4078 		qrss_id = 0;
4079 		MLX5_ASSERT(dev_flow);
4080 	}
4081 
4082 exit:
4083 	/*
4084 	 * We do not destroy the partially created sub_flows in case of error.
4085 	 * These ones are included into parent flow list and will be destroyed
4086 	 * by flow_drv_destroy.
4087 	 */
4088 	flow_qrss_free_id(dev, qrss_id);
4089 	rte_free(ext_actions);
4090 	return ret;
4091 }
4092 
4093 /**
4094  * The splitting for meter feature.
4095  *
4096  * - The meter flow will be split to two flows as prefix and
4097  *   suffix flow. The packets make sense only it pass the prefix
4098  *   meter action.
4099  *
4100  * - Reg_C_5 is used for the packet to match betweend prefix and
4101  *   suffix flow.
4102  *
4103  * @param dev
4104  *   Pointer to Ethernet device.
4105  * @param[in] flow
4106  *   Parent flow structure pointer.
4107  * @param[in] attr
4108  *   Flow rule attributes.
4109  * @param[in] items
4110  *   Pattern specification (list terminated by the END pattern item).
4111  * @param[in] actions
4112  *   Associated actions (list terminated by the END action).
4113  * @param[in] external
4114  *   This flow rule is created by request external to PMD.
4115  * @param[in] flow_idx
4116  *   This memory pool index to the flow.
4117  * @param[out] error
4118  *   Perform verbose error reporting if not NULL.
4119  * @return
4120  *   0 on success, negative value otherwise
4121  */
4122 static int
4123 flow_create_split_meter(struct rte_eth_dev *dev,
4124 			   struct rte_flow *flow,
4125 			   const struct rte_flow_attr *attr,
4126 			   const struct rte_flow_item items[],
4127 			   const struct rte_flow_action actions[],
4128 			   bool external, uint32_t flow_idx,
4129 			   struct rte_flow_error *error)
4130 {
4131 	struct mlx5_priv *priv = dev->data->dev_private;
4132 	struct rte_flow_action *sfx_actions = NULL;
4133 	struct rte_flow_action *pre_actions = NULL;
4134 	struct rte_flow_item *sfx_items = NULL;
4135 	struct mlx5_flow *dev_flow = NULL;
4136 	struct rte_flow_attr sfx_attr = *attr;
4137 	uint32_t mtr = 0;
4138 	uint32_t mtr_tag_id = 0;
4139 	size_t act_size;
4140 	size_t item_size;
4141 	int actions_n = 0;
4142 	int ret;
4143 
4144 	if (priv->mtr_en)
4145 		actions_n = flow_check_meter_action(actions, &mtr);
4146 	if (mtr) {
4147 		/* The five prefix actions: meter, decap, encap, tag, end. */
4148 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4149 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4150 		/* tag, vlan, port id, end. */
4151 #define METER_SUFFIX_ITEM 4
4152 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4153 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4154 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4155 		if (!sfx_actions)
4156 			return rte_flow_error_set(error, ENOMEM,
4157 						  RTE_FLOW_ERROR_TYPE_ACTION,
4158 						  NULL, "no memory to split "
4159 						  "meter flow");
4160 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4161 			     act_size);
4162 		pre_actions = sfx_actions + actions_n;
4163 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4164 						   actions, sfx_actions,
4165 						   pre_actions);
4166 		if (!mtr_tag_id) {
4167 			ret = -rte_errno;
4168 			goto exit;
4169 		}
4170 		/* Add the prefix subflow. */
4171 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4172 					      items, pre_actions, external,
4173 					      flow_idx, error);
4174 		if (ret) {
4175 			ret = -rte_errno;
4176 			goto exit;
4177 		}
4178 		dev_flow->handle->split_flow_id = mtr_tag_id;
4179 		/* Setting the sfx group atrr. */
4180 		sfx_attr.group = sfx_attr.transfer ?
4181 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4182 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4183 	}
4184 	/* Add the prefix subflow. */
4185 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4186 					 flow_get_prefix_layer_flags(dev_flow) :
4187 					 0, &sfx_attr,
4188 					 sfx_items ? sfx_items : items,
4189 					 sfx_actions ? sfx_actions : actions,
4190 					 external, flow_idx, error);
4191 exit:
4192 	if (sfx_actions)
4193 		rte_free(sfx_actions);
4194 	return ret;
4195 }
4196 
4197 /**
4198  * Split the flow to subflow set. The splitters might be linked
4199  * in the chain, like this:
4200  * flow_create_split_outer() calls:
4201  *   flow_create_split_meter() calls:
4202  *     flow_create_split_metadata(meter_subflow_0) calls:
4203  *       flow_create_split_inner(metadata_subflow_0)
4204  *       flow_create_split_inner(metadata_subflow_1)
4205  *       flow_create_split_inner(metadata_subflow_2)
4206  *     flow_create_split_metadata(meter_subflow_1) calls:
4207  *       flow_create_split_inner(metadata_subflow_0)
4208  *       flow_create_split_inner(metadata_subflow_1)
4209  *       flow_create_split_inner(metadata_subflow_2)
4210  *
4211  * This provide flexible way to add new levels of flow splitting.
4212  * The all of successfully created subflows are included to the
4213  * parent flow dev_flow list.
4214  *
4215  * @param dev
4216  *   Pointer to Ethernet device.
4217  * @param[in] flow
4218  *   Parent flow structure pointer.
4219  * @param[in] attr
4220  *   Flow rule attributes.
4221  * @param[in] items
4222  *   Pattern specification (list terminated by the END pattern item).
4223  * @param[in] actions
4224  *   Associated actions (list terminated by the END action).
4225  * @param[in] external
4226  *   This flow rule is created by request external to PMD.
4227  * @param[in] flow_idx
4228  *   This memory pool index to the flow.
4229  * @param[out] error
4230  *   Perform verbose error reporting if not NULL.
4231  * @return
4232  *   0 on success, negative value otherwise
4233  */
4234 static int
4235 flow_create_split_outer(struct rte_eth_dev *dev,
4236 			struct rte_flow *flow,
4237 			const struct rte_flow_attr *attr,
4238 			const struct rte_flow_item items[],
4239 			const struct rte_flow_action actions[],
4240 			bool external, uint32_t flow_idx,
4241 			struct rte_flow_error *error)
4242 {
4243 	int ret;
4244 
4245 	ret = flow_create_split_meter(dev, flow, attr, items,
4246 					 actions, external, flow_idx, error);
4247 	MLX5_ASSERT(ret <= 0);
4248 	return ret;
4249 }
4250 
4251 /**
4252  * Create a flow and add it to @p list.
4253  *
4254  * @param dev
4255  *   Pointer to Ethernet device.
4256  * @param list
4257  *   Pointer to a TAILQ flow list. If this parameter NULL,
4258  *   no list insertion occurred, flow is just created,
4259  *   this is caller's responsibility to track the
4260  *   created flow.
4261  * @param[in] attr
4262  *   Flow rule attributes.
4263  * @param[in] items
4264  *   Pattern specification (list terminated by the END pattern item).
4265  * @param[in] actions
4266  *   Associated actions (list terminated by the END action).
4267  * @param[in] external
4268  *   This flow rule is created by request external to PMD.
4269  * @param[out] error
4270  *   Perform verbose error reporting if not NULL.
4271  *
4272  * @return
4273  *   A flow index on success, 0 otherwise and rte_errno is set.
4274  */
4275 static uint32_t
4276 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4277 		 const struct rte_flow_attr *attr,
4278 		 const struct rte_flow_item items[],
4279 		 const struct rte_flow_action actions[],
4280 		 bool external, struct rte_flow_error *error)
4281 {
4282 	struct mlx5_priv *priv = dev->data->dev_private;
4283 	struct rte_flow *flow = NULL;
4284 	struct mlx5_flow *dev_flow;
4285 	const struct rte_flow_action_rss *rss;
4286 	union {
4287 		struct rte_flow_expand_rss buf;
4288 		uint8_t buffer[2048];
4289 	} expand_buffer;
4290 	union {
4291 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4292 		uint8_t buffer[2048];
4293 	} actions_rx;
4294 	union {
4295 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4296 		uint8_t buffer[2048];
4297 	} actions_hairpin_tx;
4298 	union {
4299 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4300 		uint8_t buffer[2048];
4301 	} items_tx;
4302 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4303 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4304 					      priv->rss_desc)[!!priv->flow_idx];
4305 	const struct rte_flow_action *p_actions_rx = actions;
4306 	uint32_t i;
4307 	uint32_t idx = 0;
4308 	int hairpin_flow;
4309 	uint32_t hairpin_id = 0;
4310 	struct rte_flow_attr attr_tx = { .priority = 0 };
4311 	int ret;
4312 
4313 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4314 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4315 				external, hairpin_flow, error);
4316 	if (ret < 0)
4317 		return 0;
4318 	if (hairpin_flow > 0) {
4319 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4320 			rte_errno = EINVAL;
4321 			return 0;
4322 		}
4323 		flow_hairpin_split(dev, actions, actions_rx.actions,
4324 				   actions_hairpin_tx.actions, items_tx.items,
4325 				   &hairpin_id);
4326 		p_actions_rx = actions_rx.actions;
4327 	}
4328 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4329 	if (!flow) {
4330 		rte_errno = ENOMEM;
4331 		goto error_before_flow;
4332 	}
4333 	flow->drv_type = flow_get_drv_type(dev, attr);
4334 	if (hairpin_id != 0)
4335 		flow->hairpin_flow_id = hairpin_id;
4336 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4337 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4338 	memset(rss_desc, 0, sizeof(*rss_desc));
4339 	rss = flow_get_rss_action(p_actions_rx);
4340 	if (rss) {
4341 		/*
4342 		 * The following information is required by
4343 		 * mlx5_flow_hashfields_adjust() in advance.
4344 		 */
4345 		rss_desc->level = rss->level;
4346 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4347 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4348 	}
4349 	flow->dev_handles = 0;
4350 	if (rss && rss->types) {
4351 		unsigned int graph_root;
4352 
4353 		graph_root = find_graph_root(items, rss->level);
4354 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4355 					  items, rss->types,
4356 					  mlx5_support_expansion,
4357 					  graph_root);
4358 		MLX5_ASSERT(ret > 0 &&
4359 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4360 	} else {
4361 		buf->entries = 1;
4362 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4363 	}
4364 	/*
4365 	 * Record the start index when there is a nested call. All sub-flows
4366 	 * need to be translated before another calling.
4367 	 * No need to use ping-pong buffer to save memory here.
4368 	 */
4369 	if (priv->flow_idx) {
4370 		MLX5_ASSERT(!priv->flow_nested_idx);
4371 		priv->flow_nested_idx = priv->flow_idx;
4372 	}
4373 	for (i = 0; i < buf->entries; ++i) {
4374 		/*
4375 		 * The splitter may create multiple dev_flows,
4376 		 * depending on configuration. In the simplest
4377 		 * case it just creates unmodified original flow.
4378 		 */
4379 		ret = flow_create_split_outer(dev, flow, attr,
4380 					      buf->entry[i].pattern,
4381 					      p_actions_rx, external, idx,
4382 					      error);
4383 		if (ret < 0)
4384 			goto error;
4385 	}
4386 	/* Create the tx flow. */
4387 	if (hairpin_flow) {
4388 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4389 		attr_tx.ingress = 0;
4390 		attr_tx.egress = 1;
4391 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4392 					 actions_hairpin_tx.actions,
4393 					 idx, error);
4394 		if (!dev_flow)
4395 			goto error;
4396 		dev_flow->flow = flow;
4397 		dev_flow->external = 0;
4398 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4399 			      dev_flow->handle, next);
4400 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4401 					 items_tx.items,
4402 					 actions_hairpin_tx.actions, error);
4403 		if (ret < 0)
4404 			goto error;
4405 	}
4406 	/*
4407 	 * Update the metadata register copy table. If extensive
4408 	 * metadata feature is enabled and registers are supported
4409 	 * we might create the extra rte_flow for each unique
4410 	 * MARK/FLAG action ID.
4411 	 *
4412 	 * The table is updated for ingress Flows only, because
4413 	 * the egress Flows belong to the different device and
4414 	 * copy table should be updated in peer NIC Rx domain.
4415 	 */
4416 	if (attr->ingress &&
4417 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4418 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4419 		if (ret)
4420 			goto error;
4421 	}
4422 	/*
4423 	 * If the flow is external (from application) OR device is started, then
4424 	 * the flow will be applied immediately.
4425 	 */
4426 	if (external || dev->data->dev_started) {
4427 		ret = flow_drv_apply(dev, flow, error);
4428 		if (ret < 0)
4429 			goto error;
4430 	}
4431 	if (list)
4432 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4433 			     flow, next);
4434 	flow_rxq_flags_set(dev, flow);
4435 	/* Nested flow creation index recovery. */
4436 	priv->flow_idx = priv->flow_nested_idx;
4437 	if (priv->flow_nested_idx)
4438 		priv->flow_nested_idx = 0;
4439 	return idx;
4440 error:
4441 	MLX5_ASSERT(flow);
4442 	ret = rte_errno; /* Save rte_errno before cleanup. */
4443 	flow_mreg_del_copy_action(dev, flow);
4444 	flow_drv_destroy(dev, flow);
4445 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4446 	rte_errno = ret; /* Restore rte_errno. */
4447 error_before_flow:
4448 	ret = rte_errno;
4449 	if (hairpin_id)
4450 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4451 				     hairpin_id);
4452 	rte_errno = ret;
4453 	priv->flow_idx = priv->flow_nested_idx;
4454 	if (priv->flow_nested_idx)
4455 		priv->flow_nested_idx = 0;
4456 	return 0;
4457 }
4458 
4459 /**
4460  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4461  * incoming packets to table 1.
4462  *
4463  * Other flow rules, requested for group n, will be created in
4464  * e-switch table n+1.
4465  * Jump action to e-switch group n will be created to group n+1.
4466  *
4467  * Used when working in switchdev mode, to utilise advantages of table 1
4468  * and above.
4469  *
4470  * @param dev
4471  *   Pointer to Ethernet device.
4472  *
4473  * @return
4474  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4475  */
4476 struct rte_flow *
4477 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4478 {
4479 	const struct rte_flow_attr attr = {
4480 		.group = 0,
4481 		.priority = 0,
4482 		.ingress = 1,
4483 		.egress = 0,
4484 		.transfer = 1,
4485 	};
4486 	const struct rte_flow_item pattern = {
4487 		.type = RTE_FLOW_ITEM_TYPE_END,
4488 	};
4489 	struct rte_flow_action_jump jump = {
4490 		.group = 1,
4491 	};
4492 	const struct rte_flow_action actions[] = {
4493 		{
4494 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4495 			.conf = &jump,
4496 		},
4497 		{
4498 			.type = RTE_FLOW_ACTION_TYPE_END,
4499 		},
4500 	};
4501 	struct mlx5_priv *priv = dev->data->dev_private;
4502 	struct rte_flow_error error;
4503 
4504 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4505 						   &attr, &pattern,
4506 						   actions, false, &error);
4507 }
4508 
4509 /**
4510  * Validate a flow supported by the NIC.
4511  *
4512  * @see rte_flow_validate()
4513  * @see rte_flow_ops
4514  */
4515 int
4516 mlx5_flow_validate(struct rte_eth_dev *dev,
4517 		   const struct rte_flow_attr *attr,
4518 		   const struct rte_flow_item items[],
4519 		   const struct rte_flow_action actions[],
4520 		   struct rte_flow_error *error)
4521 {
4522 	int hairpin_flow;
4523 
4524 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4525 	return flow_drv_validate(dev, attr, items, actions,
4526 				true, hairpin_flow, error);
4527 }
4528 
4529 /**
4530  * Create a flow.
4531  *
4532  * @see rte_flow_create()
4533  * @see rte_flow_ops
4534  */
4535 struct rte_flow *
4536 mlx5_flow_create(struct rte_eth_dev *dev,
4537 		 const struct rte_flow_attr *attr,
4538 		 const struct rte_flow_item items[],
4539 		 const struct rte_flow_action actions[],
4540 		 struct rte_flow_error *error)
4541 {
4542 	struct mlx5_priv *priv = dev->data->dev_private;
4543 
4544 	/*
4545 	 * If the device is not started yet, it is not allowed to created a
4546 	 * flow from application. PMD default flows and traffic control flows
4547 	 * are not affected.
4548 	 */
4549 	if (unlikely(!dev->data->dev_started)) {
4550 		DRV_LOG(DEBUG, "port %u is not started when "
4551 			"inserting a flow", dev->data->port_id);
4552 		rte_flow_error_set(error, ENODEV,
4553 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4554 				   NULL,
4555 				   "port not started");
4556 		return NULL;
4557 	}
4558 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4559 				  attr, items, actions, true, error);
4560 }
4561 
4562 /**
4563  * Destroy a flow in a list.
4564  *
4565  * @param dev
4566  *   Pointer to Ethernet device.
4567  * @param list
4568  *   Pointer to the Indexed flow list. If this parameter NULL,
4569  *   there is no flow removal from the list. Be noted that as
4570  *   flow is add to the indexed list, memory of the indexed
4571  *   list points to maybe changed as flow destroyed.
4572  * @param[in] flow_idx
4573  *   Index of flow to destroy.
4574  */
4575 static void
4576 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4577 		  uint32_t flow_idx)
4578 {
4579 	struct mlx5_priv *priv = dev->data->dev_private;
4580 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4581 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4582 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4583 
4584 	if (!flow)
4585 		return;
4586 	/*
4587 	 * Update RX queue flags only if port is started, otherwise it is
4588 	 * already clean.
4589 	 */
4590 	if (dev->data->dev_started)
4591 		flow_rxq_flags_trim(dev, flow);
4592 	if (flow->hairpin_flow_id)
4593 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4594 				     flow->hairpin_flow_id);
4595 	flow_drv_destroy(dev, flow);
4596 	if (list)
4597 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4598 			     flow_idx, flow, next);
4599 	flow_mreg_del_copy_action(dev, flow);
4600 	if (flow->fdir) {
4601 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4602 			if (priv_fdir_flow->rix_flow == flow_idx)
4603 				break;
4604 		}
4605 		if (priv_fdir_flow) {
4606 			LIST_REMOVE(priv_fdir_flow, next);
4607 			rte_free(priv_fdir_flow->fdir);
4608 			rte_free(priv_fdir_flow);
4609 		}
4610 	}
4611 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4612 }
4613 
4614 /**
4615  * Destroy all flows.
4616  *
4617  * @param dev
4618  *   Pointer to Ethernet device.
4619  * @param list
4620  *   Pointer to the Indexed flow list.
4621  * @param active
4622  *   If flushing is called avtively.
4623  */
4624 void
4625 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4626 {
4627 	uint32_t num_flushed = 0;
4628 
4629 	while (*list) {
4630 		flow_list_destroy(dev, list, *list);
4631 		num_flushed++;
4632 	}
4633 	if (active) {
4634 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4635 			dev->data->port_id, num_flushed);
4636 	}
4637 }
4638 
4639 /**
4640  * Remove all flows.
4641  *
4642  * @param dev
4643  *   Pointer to Ethernet device.
4644  * @param list
4645  *   Pointer to the Indexed flow list.
4646  */
4647 void
4648 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4649 {
4650 	struct mlx5_priv *priv = dev->data->dev_private;
4651 	struct rte_flow *flow = NULL;
4652 	uint32_t idx;
4653 
4654 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4655 		      flow, next) {
4656 		flow_drv_remove(dev, flow);
4657 		flow_mreg_stop_copy_action(dev, flow);
4658 	}
4659 	flow_mreg_del_default_copy_action(dev);
4660 	flow_rxq_flags_clear(dev);
4661 }
4662 
4663 /**
4664  * Add all flows.
4665  *
4666  * @param dev
4667  *   Pointer to Ethernet device.
4668  * @param list
4669  *   Pointer to the Indexed flow list.
4670  *
4671  * @return
4672  *   0 on success, a negative errno value otherwise and rte_errno is set.
4673  */
4674 int
4675 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4676 {
4677 	struct mlx5_priv *priv = dev->data->dev_private;
4678 	struct rte_flow *flow = NULL;
4679 	struct rte_flow_error error;
4680 	uint32_t idx;
4681 	int ret = 0;
4682 
4683 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4684 	ret = flow_mreg_add_default_copy_action(dev, &error);
4685 	if (ret < 0)
4686 		return -rte_errno;
4687 	/* Apply Flows created by application. */
4688 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4689 		      flow, next) {
4690 		ret = flow_mreg_start_copy_action(dev, flow);
4691 		if (ret < 0)
4692 			goto error;
4693 		ret = flow_drv_apply(dev, flow, &error);
4694 		if (ret < 0)
4695 			goto error;
4696 		flow_rxq_flags_set(dev, flow);
4697 	}
4698 	return 0;
4699 error:
4700 	ret = rte_errno; /* Save rte_errno before cleanup. */
4701 	mlx5_flow_stop(dev, list);
4702 	rte_errno = ret; /* Restore rte_errno. */
4703 	return -rte_errno;
4704 }
4705 
4706 /**
4707  * Stop all default actions for flows.
4708  *
4709  * @param dev
4710  *   Pointer to Ethernet device.
4711  */
4712 void
4713 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4714 {
4715 	flow_mreg_del_default_copy_action(dev);
4716 	flow_rxq_flags_clear(dev);
4717 }
4718 
4719 /**
4720  * Start all default actions for flows.
4721  *
4722  * @param dev
4723  *   Pointer to Ethernet device.
4724  * @return
4725  *   0 on success, a negative errno value otherwise and rte_errno is set.
4726  */
4727 int
4728 mlx5_flow_start_default(struct rte_eth_dev *dev)
4729 {
4730 	struct rte_flow_error error;
4731 
4732 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4733 	return flow_mreg_add_default_copy_action(dev, &error);
4734 }
4735 
4736 /**
4737  * Allocate intermediate resources for flow creation.
4738  *
4739  * @param dev
4740  *   Pointer to Ethernet device.
4741  */
4742 void
4743 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4744 {
4745 	struct mlx5_priv *priv = dev->data->dev_private;
4746 
4747 	if (!priv->inter_flows) {
4748 		priv->inter_flows = rte_calloc(__func__, 1,
4749 				    MLX5_NUM_MAX_DEV_FLOWS *
4750 				    sizeof(struct mlx5_flow) +
4751 				    (sizeof(struct mlx5_flow_rss_desc) +
4752 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4753 		if (!priv->inter_flows) {
4754 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4755 			return;
4756 		}
4757 	}
4758 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4759 			 [MLX5_NUM_MAX_DEV_FLOWS];
4760 	/* Reset the index. */
4761 	priv->flow_idx = 0;
4762 	priv->flow_nested_idx = 0;
4763 }
4764 
4765 /**
4766  * Free intermediate resources for flows.
4767  *
4768  * @param dev
4769  *   Pointer to Ethernet device.
4770  */
4771 void
4772 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4773 {
4774 	struct mlx5_priv *priv = dev->data->dev_private;
4775 
4776 	rte_free(priv->inter_flows);
4777 	priv->inter_flows = NULL;
4778 }
4779 
4780 /**
4781  * Verify the flow list is empty
4782  *
4783  * @param dev
4784  *  Pointer to Ethernet device.
4785  *
4786  * @return the number of flows not released.
4787  */
4788 int
4789 mlx5_flow_verify(struct rte_eth_dev *dev)
4790 {
4791 	struct mlx5_priv *priv = dev->data->dev_private;
4792 	struct rte_flow *flow;
4793 	uint32_t idx;
4794 	int ret = 0;
4795 
4796 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4797 		      flow, next) {
4798 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4799 			dev->data->port_id, (void *)flow);
4800 		++ret;
4801 	}
4802 	return ret;
4803 }
4804 
4805 /**
4806  * Enable default hairpin egress flow.
4807  *
4808  * @param dev
4809  *   Pointer to Ethernet device.
4810  * @param queue
4811  *   The queue index.
4812  *
4813  * @return
4814  *   0 on success, a negative errno value otherwise and rte_errno is set.
4815  */
4816 int
4817 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4818 			    uint32_t queue)
4819 {
4820 	struct mlx5_priv *priv = dev->data->dev_private;
4821 	const struct rte_flow_attr attr = {
4822 		.egress = 1,
4823 		.priority = 0,
4824 	};
4825 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4826 		.queue = queue,
4827 	};
4828 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4829 		.queue = UINT32_MAX,
4830 	};
4831 	struct rte_flow_item items[] = {
4832 		{
4833 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4834 			.spec = &queue_spec,
4835 			.last = NULL,
4836 			.mask = &queue_mask,
4837 		},
4838 		{
4839 			.type = RTE_FLOW_ITEM_TYPE_END,
4840 		},
4841 	};
4842 	struct rte_flow_action_jump jump = {
4843 		.group = MLX5_HAIRPIN_TX_TABLE,
4844 	};
4845 	struct rte_flow_action actions[2];
4846 	uint32_t flow_idx;
4847 	struct rte_flow_error error;
4848 
4849 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4850 	actions[0].conf = &jump;
4851 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4852 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4853 				&attr, items, actions, false, &error);
4854 	if (!flow_idx) {
4855 		DRV_LOG(DEBUG,
4856 			"Failed to create ctrl flow: rte_errno(%d),"
4857 			" type(%d), message(%s)",
4858 			rte_errno, error.type,
4859 			error.message ? error.message : " (no stated reason)");
4860 		return -rte_errno;
4861 	}
4862 	return 0;
4863 }
4864 
4865 /**
4866  * Enable a control flow configured from the control plane.
4867  *
4868  * @param dev
4869  *   Pointer to Ethernet device.
4870  * @param eth_spec
4871  *   An Ethernet flow spec to apply.
4872  * @param eth_mask
4873  *   An Ethernet flow mask to apply.
4874  * @param vlan_spec
4875  *   A VLAN flow spec to apply.
4876  * @param vlan_mask
4877  *   A VLAN flow mask to apply.
4878  *
4879  * @return
4880  *   0 on success, a negative errno value otherwise and rte_errno is set.
4881  */
4882 int
4883 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4884 		    struct rte_flow_item_eth *eth_spec,
4885 		    struct rte_flow_item_eth *eth_mask,
4886 		    struct rte_flow_item_vlan *vlan_spec,
4887 		    struct rte_flow_item_vlan *vlan_mask)
4888 {
4889 	struct mlx5_priv *priv = dev->data->dev_private;
4890 	const struct rte_flow_attr attr = {
4891 		.ingress = 1,
4892 		.priority = MLX5_FLOW_PRIO_RSVD,
4893 	};
4894 	struct rte_flow_item items[] = {
4895 		{
4896 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4897 			.spec = eth_spec,
4898 			.last = NULL,
4899 			.mask = eth_mask,
4900 		},
4901 		{
4902 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4903 					      RTE_FLOW_ITEM_TYPE_END,
4904 			.spec = vlan_spec,
4905 			.last = NULL,
4906 			.mask = vlan_mask,
4907 		},
4908 		{
4909 			.type = RTE_FLOW_ITEM_TYPE_END,
4910 		},
4911 	};
4912 	uint16_t queue[priv->reta_idx_n];
4913 	struct rte_flow_action_rss action_rss = {
4914 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4915 		.level = 0,
4916 		.types = priv->rss_conf.rss_hf,
4917 		.key_len = priv->rss_conf.rss_key_len,
4918 		.queue_num = priv->reta_idx_n,
4919 		.key = priv->rss_conf.rss_key,
4920 		.queue = queue,
4921 	};
4922 	struct rte_flow_action actions[] = {
4923 		{
4924 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4925 			.conf = &action_rss,
4926 		},
4927 		{
4928 			.type = RTE_FLOW_ACTION_TYPE_END,
4929 		},
4930 	};
4931 	uint32_t flow_idx;
4932 	struct rte_flow_error error;
4933 	unsigned int i;
4934 
4935 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4936 		return 0;
4937 	}
4938 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4939 		action_rss.types = 0;
4940 	for (i = 0; i != priv->reta_idx_n; ++i)
4941 		queue[i] = (*priv->reta_idx)[i];
4942 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4943 				&attr, items, actions, false, &error);
4944 	if (!flow_idx)
4945 		return -rte_errno;
4946 	return 0;
4947 }
4948 
4949 /**
4950  * Enable a flow control configured from the control plane.
4951  *
4952  * @param dev
4953  *   Pointer to Ethernet device.
4954  * @param eth_spec
4955  *   An Ethernet flow spec to apply.
4956  * @param eth_mask
4957  *   An Ethernet flow mask to apply.
4958  *
4959  * @return
4960  *   0 on success, a negative errno value otherwise and rte_errno is set.
4961  */
4962 int
4963 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4964 	       struct rte_flow_item_eth *eth_spec,
4965 	       struct rte_flow_item_eth *eth_mask)
4966 {
4967 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4968 }
4969 
4970 /**
4971  * Destroy a flow.
4972  *
4973  * @see rte_flow_destroy()
4974  * @see rte_flow_ops
4975  */
4976 int
4977 mlx5_flow_destroy(struct rte_eth_dev *dev,
4978 		  struct rte_flow *flow,
4979 		  struct rte_flow_error *error __rte_unused)
4980 {
4981 	struct mlx5_priv *priv = dev->data->dev_private;
4982 
4983 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
4984 	return 0;
4985 }
4986 
4987 /**
4988  * Destroy all flows.
4989  *
4990  * @see rte_flow_flush()
4991  * @see rte_flow_ops
4992  */
4993 int
4994 mlx5_flow_flush(struct rte_eth_dev *dev,
4995 		struct rte_flow_error *error __rte_unused)
4996 {
4997 	struct mlx5_priv *priv = dev->data->dev_private;
4998 
4999 	mlx5_flow_list_flush(dev, &priv->flows, false);
5000 	return 0;
5001 }
5002 
5003 /**
5004  * Isolated mode.
5005  *
5006  * @see rte_flow_isolate()
5007  * @see rte_flow_ops
5008  */
5009 int
5010 mlx5_flow_isolate(struct rte_eth_dev *dev,
5011 		  int enable,
5012 		  struct rte_flow_error *error)
5013 {
5014 	struct mlx5_priv *priv = dev->data->dev_private;
5015 
5016 	if (dev->data->dev_started) {
5017 		rte_flow_error_set(error, EBUSY,
5018 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5019 				   NULL,
5020 				   "port must be stopped first");
5021 		return -rte_errno;
5022 	}
5023 	priv->isolated = !!enable;
5024 	if (enable)
5025 		dev->dev_ops = &mlx5_dev_ops_isolate;
5026 	else
5027 		dev->dev_ops = &mlx5_dev_ops;
5028 	return 0;
5029 }
5030 
5031 /**
5032  * Query a flow.
5033  *
5034  * @see rte_flow_query()
5035  * @see rte_flow_ops
5036  */
5037 static int
5038 flow_drv_query(struct rte_eth_dev *dev,
5039 	       uint32_t flow_idx,
5040 	       const struct rte_flow_action *actions,
5041 	       void *data,
5042 	       struct rte_flow_error *error)
5043 {
5044 	struct mlx5_priv *priv = dev->data->dev_private;
5045 	const struct mlx5_flow_driver_ops *fops;
5046 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5047 					       [MLX5_IPOOL_RTE_FLOW],
5048 					       flow_idx);
5049 	enum mlx5_flow_drv_type ftype;
5050 
5051 	if (!flow) {
5052 		return rte_flow_error_set(error, ENOENT,
5053 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5054 			  NULL,
5055 			  "invalid flow handle");
5056 	}
5057 	ftype = flow->drv_type;
5058 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5059 	fops = flow_get_drv_ops(ftype);
5060 
5061 	return fops->query(dev, flow, actions, data, error);
5062 }
5063 
5064 /**
5065  * Query a flow.
5066  *
5067  * @see rte_flow_query()
5068  * @see rte_flow_ops
5069  */
5070 int
5071 mlx5_flow_query(struct rte_eth_dev *dev,
5072 		struct rte_flow *flow,
5073 		const struct rte_flow_action *actions,
5074 		void *data,
5075 		struct rte_flow_error *error)
5076 {
5077 	int ret;
5078 
5079 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5080 			     error);
5081 	if (ret < 0)
5082 		return ret;
5083 	return 0;
5084 }
5085 
5086 /**
5087  * Convert a flow director filter to a generic flow.
5088  *
5089  * @param dev
5090  *   Pointer to Ethernet device.
5091  * @param fdir_filter
5092  *   Flow director filter to add.
5093  * @param attributes
5094  *   Generic flow parameters structure.
5095  *
5096  * @return
5097  *   0 on success, a negative errno value otherwise and rte_errno is set.
5098  */
5099 static int
5100 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5101 			 const struct rte_eth_fdir_filter *fdir_filter,
5102 			 struct mlx5_fdir *attributes)
5103 {
5104 	struct mlx5_priv *priv = dev->data->dev_private;
5105 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5106 	const struct rte_eth_fdir_masks *mask =
5107 		&dev->data->dev_conf.fdir_conf.mask;
5108 
5109 	/* Validate queue number. */
5110 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5111 		DRV_LOG(ERR, "port %u invalid queue number %d",
5112 			dev->data->port_id, fdir_filter->action.rx_queue);
5113 		rte_errno = EINVAL;
5114 		return -rte_errno;
5115 	}
5116 	attributes->attr.ingress = 1;
5117 	attributes->items[0] = (struct rte_flow_item) {
5118 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5119 		.spec = &attributes->l2,
5120 		.mask = &attributes->l2_mask,
5121 	};
5122 	switch (fdir_filter->action.behavior) {
5123 	case RTE_ETH_FDIR_ACCEPT:
5124 		attributes->actions[0] = (struct rte_flow_action){
5125 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5126 			.conf = &attributes->queue,
5127 		};
5128 		break;
5129 	case RTE_ETH_FDIR_REJECT:
5130 		attributes->actions[0] = (struct rte_flow_action){
5131 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5132 		};
5133 		break;
5134 	default:
5135 		DRV_LOG(ERR, "port %u invalid behavior %d",
5136 			dev->data->port_id,
5137 			fdir_filter->action.behavior);
5138 		rte_errno = ENOTSUP;
5139 		return -rte_errno;
5140 	}
5141 	attributes->queue.index = fdir_filter->action.rx_queue;
5142 	/* Handle L3. */
5143 	switch (fdir_filter->input.flow_type) {
5144 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5145 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5146 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5147 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5148 			.src_addr = input->flow.ip4_flow.src_ip,
5149 			.dst_addr = input->flow.ip4_flow.dst_ip,
5150 			.time_to_live = input->flow.ip4_flow.ttl,
5151 			.type_of_service = input->flow.ip4_flow.tos,
5152 		};
5153 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5154 			.src_addr = mask->ipv4_mask.src_ip,
5155 			.dst_addr = mask->ipv4_mask.dst_ip,
5156 			.time_to_live = mask->ipv4_mask.ttl,
5157 			.type_of_service = mask->ipv4_mask.tos,
5158 			.next_proto_id = mask->ipv4_mask.proto,
5159 		};
5160 		attributes->items[1] = (struct rte_flow_item){
5161 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5162 			.spec = &attributes->l3,
5163 			.mask = &attributes->l3_mask,
5164 		};
5165 		break;
5166 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5167 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5168 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5169 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5170 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5171 			.proto = input->flow.ipv6_flow.proto,
5172 		};
5173 
5174 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5175 		       input->flow.ipv6_flow.src_ip,
5176 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5177 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5178 		       input->flow.ipv6_flow.dst_ip,
5179 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5180 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5181 		       mask->ipv6_mask.src_ip,
5182 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5183 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5184 		       mask->ipv6_mask.dst_ip,
5185 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5186 		attributes->items[1] = (struct rte_flow_item){
5187 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5188 			.spec = &attributes->l3,
5189 			.mask = &attributes->l3_mask,
5190 		};
5191 		break;
5192 	default:
5193 		DRV_LOG(ERR, "port %u invalid flow type%d",
5194 			dev->data->port_id, fdir_filter->input.flow_type);
5195 		rte_errno = ENOTSUP;
5196 		return -rte_errno;
5197 	}
5198 	/* Handle L4. */
5199 	switch (fdir_filter->input.flow_type) {
5200 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5201 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5202 			.src_port = input->flow.udp4_flow.src_port,
5203 			.dst_port = input->flow.udp4_flow.dst_port,
5204 		};
5205 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5206 			.src_port = mask->src_port_mask,
5207 			.dst_port = mask->dst_port_mask,
5208 		};
5209 		attributes->items[2] = (struct rte_flow_item){
5210 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5211 			.spec = &attributes->l4,
5212 			.mask = &attributes->l4_mask,
5213 		};
5214 		break;
5215 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5216 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5217 			.src_port = input->flow.tcp4_flow.src_port,
5218 			.dst_port = input->flow.tcp4_flow.dst_port,
5219 		};
5220 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5221 			.src_port = mask->src_port_mask,
5222 			.dst_port = mask->dst_port_mask,
5223 		};
5224 		attributes->items[2] = (struct rte_flow_item){
5225 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5226 			.spec = &attributes->l4,
5227 			.mask = &attributes->l4_mask,
5228 		};
5229 		break;
5230 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5231 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5232 			.src_port = input->flow.udp6_flow.src_port,
5233 			.dst_port = input->flow.udp6_flow.dst_port,
5234 		};
5235 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5236 			.src_port = mask->src_port_mask,
5237 			.dst_port = mask->dst_port_mask,
5238 		};
5239 		attributes->items[2] = (struct rte_flow_item){
5240 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5241 			.spec = &attributes->l4,
5242 			.mask = &attributes->l4_mask,
5243 		};
5244 		break;
5245 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5246 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5247 			.src_port = input->flow.tcp6_flow.src_port,
5248 			.dst_port = input->flow.tcp6_flow.dst_port,
5249 		};
5250 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5251 			.src_port = mask->src_port_mask,
5252 			.dst_port = mask->dst_port_mask,
5253 		};
5254 		attributes->items[2] = (struct rte_flow_item){
5255 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5256 			.spec = &attributes->l4,
5257 			.mask = &attributes->l4_mask,
5258 		};
5259 		break;
5260 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5261 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5262 		break;
5263 	default:
5264 		DRV_LOG(ERR, "port %u invalid flow type%d",
5265 			dev->data->port_id, fdir_filter->input.flow_type);
5266 		rte_errno = ENOTSUP;
5267 		return -rte_errno;
5268 	}
5269 	return 0;
5270 }
5271 
5272 #define FLOW_FDIR_CMP(f1, f2, fld) \
5273 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5274 
5275 /**
5276  * Compare two FDIR flows. If items and actions are identical, the two flows are
5277  * regarded as same.
5278  *
5279  * @param dev
5280  *   Pointer to Ethernet device.
5281  * @param f1
5282  *   FDIR flow to compare.
5283  * @param f2
5284  *   FDIR flow to compare.
5285  *
5286  * @return
5287  *   Zero on match, 1 otherwise.
5288  */
5289 static int
5290 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5291 {
5292 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5293 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5294 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5295 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5296 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5297 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5298 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5299 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5300 		return 1;
5301 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5302 	    FLOW_FDIR_CMP(f1, f2, queue))
5303 		return 1;
5304 	return 0;
5305 }
5306 
5307 /**
5308  * Search device flow list to find out a matched FDIR flow.
5309  *
5310  * @param dev
5311  *   Pointer to Ethernet device.
5312  * @param fdir_flow
5313  *   FDIR flow to lookup.
5314  *
5315  * @return
5316  *   Index of flow if found, 0 otherwise.
5317  */
5318 static uint32_t
5319 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5320 {
5321 	struct mlx5_priv *priv = dev->data->dev_private;
5322 	uint32_t flow_idx = 0;
5323 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5324 
5325 	MLX5_ASSERT(fdir_flow);
5326 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5327 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5328 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5329 				dev->data->port_id, flow_idx);
5330 			flow_idx = priv_fdir_flow->rix_flow;
5331 			break;
5332 		}
5333 	}
5334 	return flow_idx;
5335 }
5336 
5337 /**
5338  * Add new flow director filter and store it in list.
5339  *
5340  * @param dev
5341  *   Pointer to Ethernet device.
5342  * @param fdir_filter
5343  *   Flow director filter to add.
5344  *
5345  * @return
5346  *   0 on success, a negative errno value otherwise and rte_errno is set.
5347  */
5348 static int
5349 flow_fdir_filter_add(struct rte_eth_dev *dev,
5350 		     const struct rte_eth_fdir_filter *fdir_filter)
5351 {
5352 	struct mlx5_priv *priv = dev->data->dev_private;
5353 	struct mlx5_fdir *fdir_flow;
5354 	struct rte_flow *flow;
5355 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5356 	uint32_t flow_idx;
5357 	int ret;
5358 
5359 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5360 	if (!fdir_flow) {
5361 		rte_errno = ENOMEM;
5362 		return -rte_errno;
5363 	}
5364 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5365 	if (ret)
5366 		goto error;
5367 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5368 	if (flow_idx) {
5369 		rte_errno = EEXIST;
5370 		goto error;
5371 	}
5372 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5373 				     0);
5374 	if (!priv_fdir_flow) {
5375 		rte_errno = ENOMEM;
5376 		goto error;
5377 	}
5378 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5379 				    fdir_flow->items, fdir_flow->actions, true,
5380 				    NULL);
5381 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5382 	if (!flow)
5383 		goto error;
5384 	flow->fdir = 1;
5385 	priv_fdir_flow->fdir = fdir_flow;
5386 	priv_fdir_flow->rix_flow = flow_idx;
5387 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5388 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5389 		dev->data->port_id, (void *)flow);
5390 	return 0;
5391 error:
5392 	rte_free(priv_fdir_flow);
5393 	rte_free(fdir_flow);
5394 	return -rte_errno;
5395 }
5396 
5397 /**
5398  * Delete specific filter.
5399  *
5400  * @param dev
5401  *   Pointer to Ethernet device.
5402  * @param fdir_filter
5403  *   Filter to be deleted.
5404  *
5405  * @return
5406  *   0 on success, a negative errno value otherwise and rte_errno is set.
5407  */
5408 static int
5409 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5410 			const struct rte_eth_fdir_filter *fdir_filter)
5411 {
5412 	struct mlx5_priv *priv = dev->data->dev_private;
5413 	uint32_t flow_idx;
5414 	struct mlx5_fdir fdir_flow = {
5415 		.attr.group = 0,
5416 	};
5417 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5418 	int ret;
5419 
5420 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5421 	if (ret)
5422 		return -rte_errno;
5423 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5424 		/* Find the fdir in priv list */
5425 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5426 			break;
5427 	}
5428 	if (!priv_fdir_flow)
5429 		return 0;
5430 	LIST_REMOVE(priv_fdir_flow, next);
5431 	flow_idx = priv_fdir_flow->rix_flow;
5432 	flow_list_destroy(dev, &priv->flows, flow_idx);
5433 	rte_free(priv_fdir_flow->fdir);
5434 	rte_free(priv_fdir_flow);
5435 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5436 		dev->data->port_id, flow_idx);
5437 	return 0;
5438 }
5439 
5440 /**
5441  * Update queue for specific filter.
5442  *
5443  * @param dev
5444  *   Pointer to Ethernet device.
5445  * @param fdir_filter
5446  *   Filter to be updated.
5447  *
5448  * @return
5449  *   0 on success, a negative errno value otherwise and rte_errno is set.
5450  */
5451 static int
5452 flow_fdir_filter_update(struct rte_eth_dev *dev,
5453 			const struct rte_eth_fdir_filter *fdir_filter)
5454 {
5455 	int ret;
5456 
5457 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5458 	if (ret)
5459 		return ret;
5460 	return flow_fdir_filter_add(dev, fdir_filter);
5461 }
5462 
5463 /**
5464  * Flush all filters.
5465  *
5466  * @param dev
5467  *   Pointer to Ethernet device.
5468  */
5469 static void
5470 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5471 {
5472 	struct mlx5_priv *priv = dev->data->dev_private;
5473 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5474 
5475 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5476 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5477 		LIST_REMOVE(priv_fdir_flow, next);
5478 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5479 		rte_free(priv_fdir_flow->fdir);
5480 		rte_free(priv_fdir_flow);
5481 	}
5482 }
5483 
5484 /**
5485  * Get flow director information.
5486  *
5487  * @param dev
5488  *   Pointer to Ethernet device.
5489  * @param[out] fdir_info
5490  *   Resulting flow director information.
5491  */
5492 static void
5493 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5494 {
5495 	struct rte_eth_fdir_masks *mask =
5496 		&dev->data->dev_conf.fdir_conf.mask;
5497 
5498 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5499 	fdir_info->guarant_spc = 0;
5500 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5501 	fdir_info->max_flexpayload = 0;
5502 	fdir_info->flow_types_mask[0] = 0;
5503 	fdir_info->flex_payload_unit = 0;
5504 	fdir_info->max_flex_payload_segment_num = 0;
5505 	fdir_info->flex_payload_limit = 0;
5506 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5507 }
5508 
5509 /**
5510  * Deal with flow director operations.
5511  *
5512  * @param dev
5513  *   Pointer to Ethernet device.
5514  * @param filter_op
5515  *   Operation to perform.
5516  * @param arg
5517  *   Pointer to operation-specific structure.
5518  *
5519  * @return
5520  *   0 on success, a negative errno value otherwise and rte_errno is set.
5521  */
5522 static int
5523 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5524 		    void *arg)
5525 {
5526 	enum rte_fdir_mode fdir_mode =
5527 		dev->data->dev_conf.fdir_conf.mode;
5528 
5529 	if (filter_op == RTE_ETH_FILTER_NOP)
5530 		return 0;
5531 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5532 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5533 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5534 			dev->data->port_id, fdir_mode);
5535 		rte_errno = EINVAL;
5536 		return -rte_errno;
5537 	}
5538 	switch (filter_op) {
5539 	case RTE_ETH_FILTER_ADD:
5540 		return flow_fdir_filter_add(dev, arg);
5541 	case RTE_ETH_FILTER_UPDATE:
5542 		return flow_fdir_filter_update(dev, arg);
5543 	case RTE_ETH_FILTER_DELETE:
5544 		return flow_fdir_filter_delete(dev, arg);
5545 	case RTE_ETH_FILTER_FLUSH:
5546 		flow_fdir_filter_flush(dev);
5547 		break;
5548 	case RTE_ETH_FILTER_INFO:
5549 		flow_fdir_info_get(dev, arg);
5550 		break;
5551 	default:
5552 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5553 			dev->data->port_id, filter_op);
5554 		rte_errno = EINVAL;
5555 		return -rte_errno;
5556 	}
5557 	return 0;
5558 }
5559 
5560 /**
5561  * Manage filter operations.
5562  *
5563  * @param dev
5564  *   Pointer to Ethernet device structure.
5565  * @param filter_type
5566  *   Filter type.
5567  * @param filter_op
5568  *   Operation to perform.
5569  * @param arg
5570  *   Pointer to operation-specific structure.
5571  *
5572  * @return
5573  *   0 on success, a negative errno value otherwise and rte_errno is set.
5574  */
5575 int
5576 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5577 		     enum rte_filter_type filter_type,
5578 		     enum rte_filter_op filter_op,
5579 		     void *arg)
5580 {
5581 	switch (filter_type) {
5582 	case RTE_ETH_FILTER_GENERIC:
5583 		if (filter_op != RTE_ETH_FILTER_GET) {
5584 			rte_errno = EINVAL;
5585 			return -rte_errno;
5586 		}
5587 		*(const void **)arg = &mlx5_flow_ops;
5588 		return 0;
5589 	case RTE_ETH_FILTER_FDIR:
5590 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5591 	default:
5592 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5593 			dev->data->port_id, filter_type);
5594 		rte_errno = ENOTSUP;
5595 		return -rte_errno;
5596 	}
5597 	return 0;
5598 }
5599 
5600 /**
5601  * Create the needed meter and suffix tables.
5602  *
5603  * @param[in] dev
5604  *   Pointer to Ethernet device.
5605  * @param[in] fm
5606  *   Pointer to the flow meter.
5607  *
5608  * @return
5609  *   Pointer to table set on success, NULL otherwise.
5610  */
5611 struct mlx5_meter_domains_infos *
5612 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5613 			  const struct mlx5_flow_meter *fm)
5614 {
5615 	const struct mlx5_flow_driver_ops *fops;
5616 
5617 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5618 	return fops->create_mtr_tbls(dev, fm);
5619 }
5620 
5621 /**
5622  * Destroy the meter table set.
5623  *
5624  * @param[in] dev
5625  *   Pointer to Ethernet device.
5626  * @param[in] tbl
5627  *   Pointer to the meter table set.
5628  *
5629  * @return
5630  *   0 on success.
5631  */
5632 int
5633 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5634 			   struct mlx5_meter_domains_infos *tbls)
5635 {
5636 	const struct mlx5_flow_driver_ops *fops;
5637 
5638 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5639 	return fops->destroy_mtr_tbls(dev, tbls);
5640 }
5641 
5642 /**
5643  * Create policer rules.
5644  *
5645  * @param[in] dev
5646  *   Pointer to Ethernet device.
5647  * @param[in] fm
5648  *   Pointer to flow meter structure.
5649  * @param[in] attr
5650  *   Pointer to flow attributes.
5651  *
5652  * @return
5653  *   0 on success, -1 otherwise.
5654  */
5655 int
5656 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5657 			       struct mlx5_flow_meter *fm,
5658 			       const struct rte_flow_attr *attr)
5659 {
5660 	const struct mlx5_flow_driver_ops *fops;
5661 
5662 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5663 	return fops->create_policer_rules(dev, fm, attr);
5664 }
5665 
5666 /**
5667  * Destroy policer rules.
5668  *
5669  * @param[in] fm
5670  *   Pointer to flow meter structure.
5671  * @param[in] attr
5672  *   Pointer to flow attributes.
5673  *
5674  * @return
5675  *   0 on success, -1 otherwise.
5676  */
5677 int
5678 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5679 				struct mlx5_flow_meter *fm,
5680 				const struct rte_flow_attr *attr)
5681 {
5682 	const struct mlx5_flow_driver_ops *fops;
5683 
5684 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5685 	return fops->destroy_policer_rules(dev, fm, attr);
5686 }
5687 
5688 /**
5689  * Allocate a counter.
5690  *
5691  * @param[in] dev
5692  *   Pointer to Ethernet device structure.
5693  *
5694  * @return
5695  *   Index to allocated counter  on success, 0 otherwise.
5696  */
5697 uint32_t
5698 mlx5_counter_alloc(struct rte_eth_dev *dev)
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_alloc(dev);
5706 	}
5707 	DRV_LOG(ERR,
5708 		"port %u counter allocate is not supported.",
5709 		 dev->data->port_id);
5710 	return 0;
5711 }
5712 
5713 /**
5714  * Free a counter.
5715  *
5716  * @param[in] dev
5717  *   Pointer to Ethernet device structure.
5718  * @param[in] cnt
5719  *   Index to counter to be free.
5720  */
5721 void
5722 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5723 {
5724 	const struct mlx5_flow_driver_ops *fops;
5725 	struct rte_flow_attr attr = { .transfer = 0 };
5726 
5727 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5728 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5729 		fops->counter_free(dev, cnt);
5730 		return;
5731 	}
5732 	DRV_LOG(ERR,
5733 		"port %u counter free is not supported.",
5734 		 dev->data->port_id);
5735 }
5736 
5737 /**
5738  * Query counter statistics.
5739  *
5740  * @param[in] dev
5741  *   Pointer to Ethernet device structure.
5742  * @param[in] cnt
5743  *   Index to counter to query.
5744  * @param[in] clear
5745  *   Set to clear counter statistics.
5746  * @param[out] pkts
5747  *   The counter hits packets number to save.
5748  * @param[out] bytes
5749  *   The counter hits bytes number to save.
5750  *
5751  * @return
5752  *   0 on success, a negative errno value otherwise.
5753  */
5754 int
5755 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5756 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5757 {
5758 	const struct mlx5_flow_driver_ops *fops;
5759 	struct rte_flow_attr attr = { .transfer = 0 };
5760 
5761 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5762 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5763 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5764 	}
5765 	DRV_LOG(ERR,
5766 		"port %u counter query is not supported.",
5767 		 dev->data->port_id);
5768 	return -ENOTSUP;
5769 }
5770 
5771 #define MLX5_POOL_QUERY_FREQ_US 1000000
5772 
5773 /**
5774  * Get number of all validate pools.
5775  *
5776  * @param[in] sh
5777  *   Pointer to mlx5_ibv_shared object.
5778  *
5779  * @return
5780  *   The number of all validate pools.
5781  */
5782 static uint32_t
5783 mlx5_get_all_valid_pool_count(struct mlx5_ibv_shared *sh)
5784 {
5785 	uint8_t age, i;
5786 	uint32_t pools_n = 0;
5787 	struct mlx5_pools_container *cont;
5788 
5789 	for (age = 0; age < RTE_DIM(sh->cmng.ccont[0]); ++age) {
5790 		for (i = 0; i < 2 ; ++i) {
5791 			cont = MLX5_CNT_CONTAINER(sh, i, 0, age);
5792 			pools_n += rte_atomic16_read(&cont->n_valid);
5793 		}
5794 	}
5795 	return pools_n;
5796 }
5797 
5798 /**
5799  * Set the periodic procedure for triggering asynchronous batch queries for all
5800  * the counter pools.
5801  *
5802  * @param[in] sh
5803  *   Pointer to mlx5_ibv_shared object.
5804  */
5805 void
5806 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5807 {
5808 	uint32_t pools_n, us;
5809 
5810 	pools_n = mlx5_get_all_valid_pool_count(sh);
5811 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5812 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5813 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5814 		sh->cmng.query_thread_on = 0;
5815 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5816 	} else {
5817 		sh->cmng.query_thread_on = 1;
5818 	}
5819 }
5820 
5821 /**
5822  * The periodic procedure for triggering asynchronous batch queries for all the
5823  * counter pools. This function is probably called by the host thread.
5824  *
5825  * @param[in] arg
5826  *   The parameter for the alarm process.
5827  */
5828 void
5829 mlx5_flow_query_alarm(void *arg)
5830 {
5831 	struct mlx5_ibv_shared *sh = arg;
5832 	struct mlx5_devx_obj *dcs;
5833 	uint16_t offset;
5834 	int ret;
5835 	uint8_t batch = sh->cmng.batch;
5836 	uint8_t age = sh->cmng.age;
5837 	uint16_t pool_index = sh->cmng.pool_index;
5838 	struct mlx5_pools_container *cont;
5839 	struct mlx5_pools_container *mcont;
5840 	struct mlx5_flow_counter_pool *pool;
5841 
5842 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5843 		goto set_alarm;
5844 next_container:
5845 	cont = MLX5_CNT_CONTAINER(sh, batch, 1, age);
5846 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0, age);
5847 	/* Check if resize was done and need to flip a container. */
5848 	if (cont != mcont) {
5849 		if (cont->pools) {
5850 			/* Clean the old container. */
5851 			rte_free(cont->pools);
5852 			memset(cont, 0, sizeof(*cont));
5853 		}
5854 		rte_cio_wmb();
5855 		 /* Flip the host container. */
5856 		sh->cmng.mhi[batch][age] ^= (uint8_t)2;
5857 		cont = mcont;
5858 	}
5859 	if (!cont->pools) {
5860 		/* 2 empty containers case is unexpected. */
5861 		if (unlikely(batch != sh->cmng.batch) &&
5862 			unlikely(age != sh->cmng.age)) {
5863 			goto set_alarm;
5864 		}
5865 		batch ^= 0x1;
5866 		pool_index = 0;
5867 		if (batch == 0 && pool_index == 0) {
5868 			age ^= 0x1;
5869 			sh->cmng.batch = batch;
5870 			sh->cmng.age = age;
5871 		}
5872 		goto next_container;
5873 	}
5874 	pool = cont->pools[pool_index];
5875 	if (pool->raw_hw)
5876 		/* There is a pool query in progress. */
5877 		goto set_alarm;
5878 	pool->raw_hw =
5879 		LIST_FIRST(&sh->cmng.free_stat_raws);
5880 	if (!pool->raw_hw)
5881 		/* No free counter statistics raw memory. */
5882 		goto set_alarm;
5883 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5884 							      (&pool->a64_dcs);
5885 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5886 	/*
5887 	 * Identify the counters released between query trigger and query
5888 	 * handle more effiecntly. The counter released in this gap period
5889 	 * should wait for a new round of query as the new arrived packets
5890 	 * will not be taken into account.
5891 	 */
5892 	rte_atomic64_add(&pool->start_query_gen, 1);
5893 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5894 					       offset, NULL, NULL,
5895 					       pool->raw_hw->mem_mng->dm->id,
5896 					       (void *)(uintptr_t)
5897 					       (pool->raw_hw->data + offset),
5898 					       sh->devx_comp,
5899 					       (uint64_t)(uintptr_t)pool);
5900 	if (ret) {
5901 		rte_atomic64_sub(&pool->start_query_gen, 1);
5902 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5903 			" %d", pool->min_dcs->id);
5904 		pool->raw_hw = NULL;
5905 		goto set_alarm;
5906 	}
5907 	pool->raw_hw->min_dcs_id = dcs->id;
5908 	LIST_REMOVE(pool->raw_hw, next);
5909 	sh->cmng.pending_queries++;
5910 	pool_index++;
5911 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5912 		batch ^= 0x1;
5913 		pool_index = 0;
5914 		if (batch == 0 && pool_index == 0)
5915 			age ^= 0x1;
5916 	}
5917 set_alarm:
5918 	sh->cmng.batch = batch;
5919 	sh->cmng.pool_index = pool_index;
5920 	sh->cmng.age = age;
5921 	mlx5_set_query_alarm(sh);
5922 }
5923 
5924 /**
5925  * Check and callback event for new aged flow in the counter pool
5926  *
5927  * @param[in] sh
5928  *   Pointer to mlx5_ibv_shared object.
5929  * @param[in] pool
5930  *   Pointer to Current counter pool.
5931  */
5932 static void
5933 mlx5_flow_aging_check(struct mlx5_ibv_shared *sh,
5934 		   struct mlx5_flow_counter_pool *pool)
5935 {
5936 	struct mlx5_priv *priv;
5937 	struct mlx5_flow_counter *cnt;
5938 	struct mlx5_age_info *age_info;
5939 	struct mlx5_age_param *age_param;
5940 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
5941 	struct mlx5_counter_stats_raw *prev = pool->raw;
5942 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
5943 	uint32_t i;
5944 
5945 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
5946 		cnt = MLX5_POOL_GET_CNT(pool, i);
5947 		age_param = MLX5_CNT_TO_AGE(cnt);
5948 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
5949 			continue;
5950 		if (cur->data[i].hits != prev->data[i].hits) {
5951 			age_param->expire = curr + age_param->timeout;
5952 			continue;
5953 		}
5954 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
5955 			continue;
5956 		/**
5957 		 * Hold the lock first, or if between the
5958 		 * state AGE_TMOUT and tailq operation the
5959 		 * release happened, the release procedure
5960 		 * may delete a non-existent tailq node.
5961 		 */
5962 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
5963 		age_info = GET_PORT_AGE_INFO(priv);
5964 		rte_spinlock_lock(&age_info->aged_sl);
5965 		/* If the cpmset fails, release happens. */
5966 		if (rte_atomic16_cmpset((volatile uint16_t *)
5967 					&age_param->state,
5968 					AGE_CANDIDATE,
5969 					AGE_TMOUT) ==
5970 					AGE_CANDIDATE) {
5971 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
5972 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
5973 		}
5974 		rte_spinlock_unlock(&age_info->aged_sl);
5975 	}
5976 	for (i = 0; i < sh->max_port; i++) {
5977 		age_info = &sh->port[i].age_info;
5978 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
5979 			continue;
5980 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
5981 			_rte_eth_dev_callback_process
5982 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
5983 				RTE_ETH_EVENT_FLOW_AGED, NULL);
5984 		age_info->flags = 0;
5985 	}
5986 }
5987 
5988 /**
5989  * Handler for the HW respond about ready values from an asynchronous batch
5990  * query. This function is probably called by the host thread.
5991  *
5992  * @param[in] sh
5993  *   The pointer to the shared IB device context.
5994  * @param[in] async_id
5995  *   The Devx async ID.
5996  * @param[in] status
5997  *   The status of the completion.
5998  */
5999 void
6000 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
6001 				  uint64_t async_id, int status)
6002 {
6003 	struct mlx5_flow_counter_pool *pool =
6004 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6005 	struct mlx5_counter_stats_raw *raw_to_free;
6006 
6007 	if (unlikely(status)) {
6008 		rte_atomic64_sub(&pool->start_query_gen, 1);
6009 		raw_to_free = pool->raw_hw;
6010 	} else {
6011 		raw_to_free = pool->raw;
6012 		if (IS_AGE_POOL(pool))
6013 			mlx5_flow_aging_check(sh, pool);
6014 		rte_spinlock_lock(&pool->sl);
6015 		pool->raw = pool->raw_hw;
6016 		rte_spinlock_unlock(&pool->sl);
6017 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
6018 			    rte_atomic64_read(&pool->start_query_gen));
6019 		rte_atomic64_set(&pool->end_query_gen,
6020 				 rte_atomic64_read(&pool->start_query_gen));
6021 		/* Be sure the new raw counters data is updated in memory. */
6022 		rte_cio_wmb();
6023 	}
6024 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6025 	pool->raw_hw = NULL;
6026 	sh->cmng.pending_queries--;
6027 }
6028 
6029 /**
6030  * Translate the rte_flow group index to HW table value.
6031  *
6032  * @param[in] attributes
6033  *   Pointer to flow attributes
6034  * @param[in] external
6035  *   Value is part of flow rule created by request external to PMD.
6036  * @param[in] group
6037  *   rte_flow group index value.
6038  * @param[out] fdb_def_rule
6039  *   Whether fdb jump to table 1 is configured.
6040  * @param[out] table
6041  *   HW table value.
6042  * @param[out] error
6043  *   Pointer to error structure.
6044  *
6045  * @return
6046  *   0 on success, a negative errno value otherwise and rte_errno is set.
6047  */
6048 int
6049 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6050 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6051 			 struct rte_flow_error *error)
6052 {
6053 	if (attributes->transfer && external && fdb_def_rule) {
6054 		if (group == UINT32_MAX)
6055 			return rte_flow_error_set
6056 						(error, EINVAL,
6057 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6058 						 NULL,
6059 						 "group index not supported");
6060 		*table = group + 1;
6061 	} else {
6062 		*table = group;
6063 	}
6064 	return 0;
6065 }
6066 
6067 /**
6068  * Discover availability of metadata reg_c's.
6069  *
6070  * Iteratively use test flows to check availability.
6071  *
6072  * @param[in] dev
6073  *   Pointer to the Ethernet device structure.
6074  *
6075  * @return
6076  *   0 on success, a negative errno value otherwise and rte_errno is set.
6077  */
6078 int
6079 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6080 {
6081 	struct mlx5_priv *priv = dev->data->dev_private;
6082 	struct mlx5_dev_config *config = &priv->config;
6083 	enum modify_reg idx;
6084 	int n = 0;
6085 
6086 	/* reg_c[0] and reg_c[1] are reserved. */
6087 	config->flow_mreg_c[n++] = REG_C_0;
6088 	config->flow_mreg_c[n++] = REG_C_1;
6089 	/* Discover availability of other reg_c's. */
6090 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6091 		struct rte_flow_attr attr = {
6092 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6093 			.priority = MLX5_FLOW_PRIO_RSVD,
6094 			.ingress = 1,
6095 		};
6096 		struct rte_flow_item items[] = {
6097 			[0] = {
6098 				.type = RTE_FLOW_ITEM_TYPE_END,
6099 			},
6100 		};
6101 		struct rte_flow_action actions[] = {
6102 			[0] = {
6103 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6104 				.conf = &(struct mlx5_flow_action_copy_mreg){
6105 					.src = REG_C_1,
6106 					.dst = idx,
6107 				},
6108 			},
6109 			[1] = {
6110 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6111 				.conf = &(struct rte_flow_action_jump){
6112 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6113 				},
6114 			},
6115 			[2] = {
6116 				.type = RTE_FLOW_ACTION_TYPE_END,
6117 			},
6118 		};
6119 		uint32_t flow_idx;
6120 		struct rte_flow *flow;
6121 		struct rte_flow_error error;
6122 
6123 		if (!config->dv_flow_en)
6124 			break;
6125 		/* Create internal flow, validation skips copy action. */
6126 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6127 					    actions, false, &error);
6128 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6129 				      flow_idx);
6130 		if (!flow)
6131 			continue;
6132 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6133 			config->flow_mreg_c[n++] = idx;
6134 		flow_list_destroy(dev, NULL, flow_idx);
6135 	}
6136 	for (; n < MLX5_MREG_C_NUM; ++n)
6137 		config->flow_mreg_c[n] = REG_NONE;
6138 	return 0;
6139 }
6140 
6141 /**
6142  * Dump flow raw hw data to file
6143  *
6144  * @param[in] dev
6145  *    The pointer to Ethernet device.
6146  * @param[in] file
6147  *   A pointer to a file for output.
6148  * @param[out] error
6149  *   Perform verbose error reporting if not NULL. PMDs initialize this
6150  *   structure in case of error only.
6151  * @return
6152  *   0 on success, a nagative value otherwise.
6153  */
6154 int
6155 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6156 		   FILE *file,
6157 		   struct rte_flow_error *error __rte_unused)
6158 {
6159 	struct mlx5_priv *priv = dev->data->dev_private;
6160 	struct mlx5_ibv_shared *sh = priv->sh;
6161 
6162 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6163 				       sh->tx_domain, file);
6164 }
6165 
6166 /**
6167  * Get aged-out flows.
6168  *
6169  * @param[in] dev
6170  *   Pointer to the Ethernet device structure.
6171  * @param[in] context
6172  *   The address of an array of pointers to the aged-out flows contexts.
6173  * @param[in] nb_countexts
6174  *   The length of context array pointers.
6175  * @param[out] error
6176  *   Perform verbose error reporting if not NULL. Initialized in case of
6177  *   error only.
6178  *
6179  * @return
6180  *   how many contexts get in success, otherwise negative errno value.
6181  *   if nb_contexts is 0, return the amount of all aged contexts.
6182  *   if nb_contexts is not 0 , return the amount of aged flows reported
6183  *   in the context array.
6184  */
6185 int
6186 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6187 			uint32_t nb_contexts, struct rte_flow_error *error)
6188 {
6189 	const struct mlx5_flow_driver_ops *fops;
6190 	struct rte_flow_attr attr = { .transfer = 0 };
6191 
6192 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6193 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6194 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6195 						    error);
6196 	}
6197 	DRV_LOG(ERR,
6198 		"port %u get aged flows is not supported.",
6199 		 dev->data->port_id);
6200 	return -ENOTSUP;
6201 }
6202