xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision c81e3f21d1ea05e5123278b15d9d5e1257b6ba99)
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 (id >= (REG_C_7 - start_reg))
446 				return rte_flow_error_set(error, EINVAL,
447 						       RTE_FLOW_ERROR_TYPE_ITEM,
448 							NULL, "invalid tag id");
449 			if (config->flow_mreg_c
450 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
451 				return config->flow_mreg_c
452 					       [id + 1 + start_reg - REG_C_0];
453 			return rte_flow_error_set(error, ENOTSUP,
454 						  RTE_FLOW_ERROR_TYPE_ITEM,
455 						  NULL, "unsupported tag id");
456 		}
457 		return config->flow_mreg_c[id + start_reg - REG_C_0];
458 	}
459 	MLX5_ASSERT(false);
460 	return rte_flow_error_set(error, EINVAL,
461 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
462 				  NULL, "invalid feature name");
463 }
464 
465 /**
466  * Check extensive flow metadata register support.
467  *
468  * @param dev
469  *   Pointer to rte_eth_dev structure.
470  *
471  * @return
472  *   True if device supports extensive flow metadata register, otherwise false.
473  */
474 bool
475 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
476 {
477 	struct mlx5_priv *priv = dev->data->dev_private;
478 	struct mlx5_dev_config *config = &priv->config;
479 
480 	/*
481 	 * Having available reg_c can be regarded inclusively as supporting
482 	 * extensive flow metadata register, which could mean,
483 	 * - metadata register copy action by modify header.
484 	 * - 16 modify header actions is supported.
485 	 * - reg_c's are preserved across different domain (FDB and NIC) on
486 	 *   packet loopback by flow lookup miss.
487 	 */
488 	return config->flow_mreg_c[2] != REG_NONE;
489 }
490 
491 /**
492  * Discover the maximum number of priority available.
493  *
494  * @param[in] dev
495  *   Pointer to the Ethernet device structure.
496  *
497  * @return
498  *   number of supported flow priority on success, a negative errno
499  *   value otherwise and rte_errno is set.
500  */
501 int
502 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
503 {
504 	struct mlx5_priv *priv = dev->data->dev_private;
505 	struct {
506 		struct ibv_flow_attr attr;
507 		struct ibv_flow_spec_eth eth;
508 		struct ibv_flow_spec_action_drop drop;
509 	} flow_attr = {
510 		.attr = {
511 			.num_of_specs = 2,
512 			.port = (uint8_t)priv->ibv_port,
513 		},
514 		.eth = {
515 			.type = IBV_FLOW_SPEC_ETH,
516 			.size = sizeof(struct ibv_flow_spec_eth),
517 		},
518 		.drop = {
519 			.size = sizeof(struct ibv_flow_spec_action_drop),
520 			.type = IBV_FLOW_SPEC_ACTION_DROP,
521 		},
522 	};
523 	struct ibv_flow *flow;
524 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
525 	uint16_t vprio[] = { 8, 16 };
526 	int i;
527 	int priority = 0;
528 
529 	if (!drop) {
530 		rte_errno = ENOTSUP;
531 		return -rte_errno;
532 	}
533 	for (i = 0; i != RTE_DIM(vprio); i++) {
534 		flow_attr.attr.priority = vprio[i] - 1;
535 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
536 		if (!flow)
537 			break;
538 		claim_zero(mlx5_glue->destroy_flow(flow));
539 		priority = vprio[i];
540 	}
541 	mlx5_hrxq_drop_release(dev);
542 	switch (priority) {
543 	case 8:
544 		priority = RTE_DIM(priority_map_3);
545 		break;
546 	case 16:
547 		priority = RTE_DIM(priority_map_5);
548 		break;
549 	default:
550 		rte_errno = ENOTSUP;
551 		DRV_LOG(ERR,
552 			"port %u verbs maximum priority: %d expected 8/16",
553 			dev->data->port_id, priority);
554 		return -rte_errno;
555 	}
556 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
557 		dev->data->port_id, priority);
558 	return priority;
559 }
560 
561 /**
562  * Adjust flow priority based on the highest layer and the request priority.
563  *
564  * @param[in] dev
565  *   Pointer to the Ethernet device structure.
566  * @param[in] priority
567  *   The rule base priority.
568  * @param[in] subpriority
569  *   The priority based on the items.
570  *
571  * @return
572  *   The new priority.
573  */
574 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
575 				   uint32_t subpriority)
576 {
577 	uint32_t res = 0;
578 	struct mlx5_priv *priv = dev->data->dev_private;
579 
580 	switch (priv->config.flow_prio) {
581 	case RTE_DIM(priority_map_3):
582 		res = priority_map_3[priority][subpriority];
583 		break;
584 	case RTE_DIM(priority_map_5):
585 		res = priority_map_5[priority][subpriority];
586 		break;
587 	}
588 	return  res;
589 }
590 
591 /**
592  * Verify the @p item specifications (spec, last, mask) are compatible with the
593  * NIC capabilities.
594  *
595  * @param[in] item
596  *   Item specification.
597  * @param[in] mask
598  *   @p item->mask or flow default bit-masks.
599  * @param[in] nic_mask
600  *   Bit-masks covering supported fields by the NIC to compare with user mask.
601  * @param[in] size
602  *   Bit-masks size in bytes.
603  * @param[out] error
604  *   Pointer to error structure.
605  *
606  * @return
607  *   0 on success, a negative errno value otherwise and rte_errno is set.
608  */
609 int
610 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
611 			  const uint8_t *mask,
612 			  const uint8_t *nic_mask,
613 			  unsigned int size,
614 			  struct rte_flow_error *error)
615 {
616 	unsigned int i;
617 
618 	MLX5_ASSERT(nic_mask);
619 	for (i = 0; i < size; ++i)
620 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
621 			return rte_flow_error_set(error, ENOTSUP,
622 						  RTE_FLOW_ERROR_TYPE_ITEM,
623 						  item,
624 						  "mask enables non supported"
625 						  " bits");
626 	if (!item->spec && (item->mask || item->last))
627 		return rte_flow_error_set(error, EINVAL,
628 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
629 					  "mask/last without a spec is not"
630 					  " supported");
631 	if (item->spec && item->last) {
632 		uint8_t spec[size];
633 		uint8_t last[size];
634 		unsigned int i;
635 		int ret;
636 
637 		for (i = 0; i < size; ++i) {
638 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
639 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
640 		}
641 		ret = memcmp(spec, last, size);
642 		if (ret != 0)
643 			return rte_flow_error_set(error, EINVAL,
644 						  RTE_FLOW_ERROR_TYPE_ITEM,
645 						  item,
646 						  "range is not valid");
647 	}
648 	return 0;
649 }
650 
651 /**
652  * Adjust the hash fields according to the @p flow information.
653  *
654  * @param[in] dev_flow.
655  *   Pointer to the mlx5_flow.
656  * @param[in] tunnel
657  *   1 when the hash field is for a tunnel item.
658  * @param[in] layer_types
659  *   ETH_RSS_* types.
660  * @param[in] hash_fields
661  *   Item hash fields.
662  *
663  * @return
664  *   The hash fields that should be used.
665  */
666 uint64_t
667 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
668 			    int tunnel __rte_unused, uint64_t layer_types,
669 			    uint64_t hash_fields)
670 {
671 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
672 	int rss_request_inner = rss_desc->level >= 2;
673 
674 	/* Check RSS hash level for tunnel. */
675 	if (tunnel && rss_request_inner)
676 		hash_fields |= IBV_RX_HASH_INNER;
677 	else if (tunnel || rss_request_inner)
678 		return 0;
679 #endif
680 	/* Check if requested layer matches RSS hash fields. */
681 	if (!(rss_desc->types & layer_types))
682 		return 0;
683 	return hash_fields;
684 }
685 
686 /**
687  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
688  * if several tunnel rules are used on this queue, the tunnel ptype will be
689  * cleared.
690  *
691  * @param rxq_ctrl
692  *   Rx queue to update.
693  */
694 static void
695 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
696 {
697 	unsigned int i;
698 	uint32_t tunnel_ptype = 0;
699 
700 	/* Look up for the ptype to use. */
701 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
702 		if (!rxq_ctrl->flow_tunnels_n[i])
703 			continue;
704 		if (!tunnel_ptype) {
705 			tunnel_ptype = tunnels_info[i].ptype;
706 		} else {
707 			tunnel_ptype = 0;
708 			break;
709 		}
710 	}
711 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
712 }
713 
714 /**
715  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
716  * flow.
717  *
718  * @param[in] dev
719  *   Pointer to the Ethernet device structure.
720  * @param[in] dev_handle
721  *   Pointer to device flow handle structure.
722  */
723 static void
724 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
725 		       struct mlx5_flow_handle *dev_handle)
726 {
727 	struct mlx5_priv *priv = dev->data->dev_private;
728 	const int mark = dev_handle->mark;
729 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
730 	struct mlx5_hrxq *hrxq;
731 	unsigned int i;
732 
733 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
734 		return;
735 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
736 			      dev_handle->rix_hrxq);
737 	if (!hrxq)
738 		return;
739 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
740 		int idx = hrxq->ind_table->queues[i];
741 		struct mlx5_rxq_ctrl *rxq_ctrl =
742 			container_of((*priv->rxqs)[idx],
743 				     struct mlx5_rxq_ctrl, rxq);
744 
745 		/*
746 		 * To support metadata register copy on Tx loopback,
747 		 * this must be always enabled (metadata may arive
748 		 * from other port - not from local flows only.
749 		 */
750 		if (priv->config.dv_flow_en &&
751 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
752 		    mlx5_flow_ext_mreg_supported(dev)) {
753 			rxq_ctrl->rxq.mark = 1;
754 			rxq_ctrl->flow_mark_n = 1;
755 		} else if (mark) {
756 			rxq_ctrl->rxq.mark = 1;
757 			rxq_ctrl->flow_mark_n++;
758 		}
759 		if (tunnel) {
760 			unsigned int j;
761 
762 			/* Increase the counter matching the flow. */
763 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
764 				if ((tunnels_info[j].tunnel &
765 				     dev_handle->layers) ==
766 				    tunnels_info[j].tunnel) {
767 					rxq_ctrl->flow_tunnels_n[j]++;
768 					break;
769 				}
770 			}
771 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
772 		}
773 	}
774 }
775 
776 /**
777  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
778  *
779  * @param[in] dev
780  *   Pointer to the Ethernet device structure.
781  * @param[in] flow
782  *   Pointer to flow structure.
783  */
784 static void
785 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
786 {
787 	struct mlx5_priv *priv = dev->data->dev_private;
788 	uint32_t handle_idx;
789 	struct mlx5_flow_handle *dev_handle;
790 
791 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
792 		       handle_idx, dev_handle, next)
793 		flow_drv_rxq_flags_set(dev, dev_handle);
794 }
795 
796 /**
797  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
798  * device flow if no other flow uses it with the same kind of request.
799  *
800  * @param dev
801  *   Pointer to Ethernet device.
802  * @param[in] dev_handle
803  *   Pointer to the device flow handle structure.
804  */
805 static void
806 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
807 			struct mlx5_flow_handle *dev_handle)
808 {
809 	struct mlx5_priv *priv = dev->data->dev_private;
810 	const int mark = dev_handle->mark;
811 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
812 	struct mlx5_hrxq *hrxq;
813 	unsigned int i;
814 
815 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
816 		return;
817 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
818 			      dev_handle->rix_hrxq);
819 	if (!hrxq)
820 		return;
821 	MLX5_ASSERT(dev->data->dev_started);
822 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
823 		int idx = hrxq->ind_table->queues[i];
824 		struct mlx5_rxq_ctrl *rxq_ctrl =
825 			container_of((*priv->rxqs)[idx],
826 				     struct mlx5_rxq_ctrl, rxq);
827 
828 		if (priv->config.dv_flow_en &&
829 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
830 		    mlx5_flow_ext_mreg_supported(dev)) {
831 			rxq_ctrl->rxq.mark = 1;
832 			rxq_ctrl->flow_mark_n = 1;
833 		} else if (mark) {
834 			rxq_ctrl->flow_mark_n--;
835 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
836 		}
837 		if (tunnel) {
838 			unsigned int j;
839 
840 			/* Decrease the counter matching the flow. */
841 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
842 				if ((tunnels_info[j].tunnel &
843 				     dev_handle->layers) ==
844 				    tunnels_info[j].tunnel) {
845 					rxq_ctrl->flow_tunnels_n[j]--;
846 					break;
847 				}
848 			}
849 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
850 		}
851 	}
852 }
853 
854 /**
855  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
856  * @p flow if no other flow uses it with the same kind of request.
857  *
858  * @param dev
859  *   Pointer to Ethernet device.
860  * @param[in] flow
861  *   Pointer to the flow.
862  */
863 static void
864 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
865 {
866 	struct mlx5_priv *priv = dev->data->dev_private;
867 	uint32_t handle_idx;
868 	struct mlx5_flow_handle *dev_handle;
869 
870 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
871 		       handle_idx, dev_handle, next)
872 		flow_drv_rxq_flags_trim(dev, dev_handle);
873 }
874 
875 /**
876  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
877  *
878  * @param dev
879  *   Pointer to Ethernet device.
880  */
881 static void
882 flow_rxq_flags_clear(struct rte_eth_dev *dev)
883 {
884 	struct mlx5_priv *priv = dev->data->dev_private;
885 	unsigned int i;
886 
887 	for (i = 0; i != priv->rxqs_n; ++i) {
888 		struct mlx5_rxq_ctrl *rxq_ctrl;
889 		unsigned int j;
890 
891 		if (!(*priv->rxqs)[i])
892 			continue;
893 		rxq_ctrl = container_of((*priv->rxqs)[i],
894 					struct mlx5_rxq_ctrl, rxq);
895 		rxq_ctrl->flow_mark_n = 0;
896 		rxq_ctrl->rxq.mark = 0;
897 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
898 			rxq_ctrl->flow_tunnels_n[j] = 0;
899 		rxq_ctrl->rxq.tunnel = 0;
900 	}
901 }
902 
903 /**
904  * Set the Rx queue dynamic metadata (mask and offset) for a flow
905  *
906  * @param[in] dev
907  *   Pointer to the Ethernet device structure.
908  */
909 void
910 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
911 {
912 	struct mlx5_priv *priv = dev->data->dev_private;
913 	struct mlx5_rxq_data *data;
914 	unsigned int i;
915 
916 	for (i = 0; i != priv->rxqs_n; ++i) {
917 		if (!(*priv->rxqs)[i])
918 			continue;
919 		data = (*priv->rxqs)[i];
920 		if (!rte_flow_dynf_metadata_avail()) {
921 			data->dynf_meta = 0;
922 			data->flow_meta_mask = 0;
923 			data->flow_meta_offset = -1;
924 		} else {
925 			data->dynf_meta = 1;
926 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
927 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
928 		}
929 	}
930 }
931 
932 /*
933  * return a pointer to the desired action in the list of actions.
934  *
935  * @param[in] actions
936  *   The list of actions to search the action in.
937  * @param[in] action
938  *   The action to find.
939  *
940  * @return
941  *   Pointer to the action in the list, if found. NULL otherwise.
942  */
943 const struct rte_flow_action *
944 mlx5_flow_find_action(const struct rte_flow_action *actions,
945 		      enum rte_flow_action_type action)
946 {
947 	if (actions == NULL)
948 		return NULL;
949 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
950 		if (actions->type == action)
951 			return actions;
952 	return NULL;
953 }
954 
955 /*
956  * Validate the flag action.
957  *
958  * @param[in] action_flags
959  *   Bit-fields that holds the actions detected until now.
960  * @param[in] attr
961  *   Attributes of flow that includes this action.
962  * @param[out] error
963  *   Pointer to error structure.
964  *
965  * @return
966  *   0 on success, a negative errno value otherwise and rte_errno is set.
967  */
968 int
969 mlx5_flow_validate_action_flag(uint64_t action_flags,
970 			       const struct rte_flow_attr *attr,
971 			       struct rte_flow_error *error)
972 {
973 	if (action_flags & MLX5_FLOW_ACTION_MARK)
974 		return rte_flow_error_set(error, EINVAL,
975 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 					  "can't mark and flag in same flow");
977 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
978 		return rte_flow_error_set(error, EINVAL,
979 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
980 					  "can't have 2 flag"
981 					  " actions in same flow");
982 	if (attr->egress)
983 		return rte_flow_error_set(error, ENOTSUP,
984 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
985 					  "flag action not supported for "
986 					  "egress");
987 	return 0;
988 }
989 
990 /*
991  * Validate the mark action.
992  *
993  * @param[in] action
994  *   Pointer to the queue action.
995  * @param[in] action_flags
996  *   Bit-fields that holds the actions detected until now.
997  * @param[in] attr
998  *   Attributes of flow that includes this action.
999  * @param[out] error
1000  *   Pointer to error structure.
1001  *
1002  * @return
1003  *   0 on success, a negative errno value otherwise and rte_errno is set.
1004  */
1005 int
1006 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1007 			       uint64_t action_flags,
1008 			       const struct rte_flow_attr *attr,
1009 			       struct rte_flow_error *error)
1010 {
1011 	const struct rte_flow_action_mark *mark = action->conf;
1012 
1013 	if (!mark)
1014 		return rte_flow_error_set(error, EINVAL,
1015 					  RTE_FLOW_ERROR_TYPE_ACTION,
1016 					  action,
1017 					  "configuration cannot be null");
1018 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1019 		return rte_flow_error_set(error, EINVAL,
1020 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1021 					  &mark->id,
1022 					  "mark id must in 0 <= id < "
1023 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1024 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1025 		return rte_flow_error_set(error, EINVAL,
1026 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1027 					  "can't flag and mark in same flow");
1028 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1029 		return rte_flow_error_set(error, EINVAL,
1030 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1031 					  "can't have 2 mark actions in same"
1032 					  " flow");
1033 	if (attr->egress)
1034 		return rte_flow_error_set(error, ENOTSUP,
1035 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1036 					  "mark action not supported for "
1037 					  "egress");
1038 	return 0;
1039 }
1040 
1041 /*
1042  * Validate the drop action.
1043  *
1044  * @param[in] action_flags
1045  *   Bit-fields that holds the actions detected until now.
1046  * @param[in] attr
1047  *   Attributes of flow that includes this action.
1048  * @param[out] error
1049  *   Pointer to error structure.
1050  *
1051  * @return
1052  *   0 on success, a negative errno value otherwise and rte_errno is set.
1053  */
1054 int
1055 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1056 			       const struct rte_flow_attr *attr,
1057 			       struct rte_flow_error *error)
1058 {
1059 	if (attr->egress)
1060 		return rte_flow_error_set(error, ENOTSUP,
1061 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1062 					  "drop action not supported for "
1063 					  "egress");
1064 	return 0;
1065 }
1066 
1067 /*
1068  * Validate the queue action.
1069  *
1070  * @param[in] action
1071  *   Pointer to the queue action.
1072  * @param[in] action_flags
1073  *   Bit-fields that holds the actions detected until now.
1074  * @param[in] dev
1075  *   Pointer to the Ethernet device structure.
1076  * @param[in] attr
1077  *   Attributes of flow that includes this action.
1078  * @param[out] error
1079  *   Pointer to error structure.
1080  *
1081  * @return
1082  *   0 on success, a negative errno value otherwise and rte_errno is set.
1083  */
1084 int
1085 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1086 				uint64_t action_flags,
1087 				struct rte_eth_dev *dev,
1088 				const struct rte_flow_attr *attr,
1089 				struct rte_flow_error *error)
1090 {
1091 	struct mlx5_priv *priv = dev->data->dev_private;
1092 	const struct rte_flow_action_queue *queue = action->conf;
1093 
1094 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1095 		return rte_flow_error_set(error, EINVAL,
1096 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1097 					  "can't have 2 fate actions in"
1098 					  " same flow");
1099 	if (!priv->rxqs_n)
1100 		return rte_flow_error_set(error, EINVAL,
1101 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1102 					  NULL, "No Rx queues configured");
1103 	if (queue->index >= priv->rxqs_n)
1104 		return rte_flow_error_set(error, EINVAL,
1105 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1106 					  &queue->index,
1107 					  "queue index out of range");
1108 	if (!(*priv->rxqs)[queue->index])
1109 		return rte_flow_error_set(error, EINVAL,
1110 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1111 					  &queue->index,
1112 					  "queue is not configured");
1113 	if (attr->egress)
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1116 					  "queue action not supported for "
1117 					  "egress");
1118 	return 0;
1119 }
1120 
1121 /*
1122  * Validate the rss action.
1123  *
1124  * @param[in] action
1125  *   Pointer to the queue action.
1126  * @param[in] action_flags
1127  *   Bit-fields that holds the actions detected until now.
1128  * @param[in] dev
1129  *   Pointer to the Ethernet device structure.
1130  * @param[in] attr
1131  *   Attributes of flow that includes this action.
1132  * @param[in] item_flags
1133  *   Items that were detected.
1134  * @param[out] error
1135  *   Pointer to error structure.
1136  *
1137  * @return
1138  *   0 on success, a negative errno value otherwise and rte_errno is set.
1139  */
1140 int
1141 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1142 			      uint64_t action_flags,
1143 			      struct rte_eth_dev *dev,
1144 			      const struct rte_flow_attr *attr,
1145 			      uint64_t item_flags,
1146 			      struct rte_flow_error *error)
1147 {
1148 	struct mlx5_priv *priv = dev->data->dev_private;
1149 	const struct rte_flow_action_rss *rss = action->conf;
1150 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1151 	unsigned int i;
1152 
1153 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1154 		return rte_flow_error_set(error, EINVAL,
1155 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1156 					  "can't have 2 fate actions"
1157 					  " in same flow");
1158 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1159 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1160 		return rte_flow_error_set(error, ENOTSUP,
1161 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 					  &rss->func,
1163 					  "RSS hash function not supported");
1164 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1165 	if (rss->level > 2)
1166 #else
1167 	if (rss->level > 1)
1168 #endif
1169 		return rte_flow_error_set(error, ENOTSUP,
1170 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1171 					  &rss->level,
1172 					  "tunnel RSS is not supported");
1173 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1174 	if (rss->key_len == 0 && rss->key != NULL)
1175 		return rte_flow_error_set(error, ENOTSUP,
1176 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177 					  &rss->key_len,
1178 					  "RSS hash key length 0");
1179 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1180 		return rte_flow_error_set(error, ENOTSUP,
1181 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1182 					  &rss->key_len,
1183 					  "RSS hash key too small");
1184 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1185 		return rte_flow_error_set(error, ENOTSUP,
1186 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1187 					  &rss->key_len,
1188 					  "RSS hash key too large");
1189 	if (rss->queue_num > priv->config.ind_table_max_size)
1190 		return rte_flow_error_set(error, ENOTSUP,
1191 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1192 					  &rss->queue_num,
1193 					  "number of queues too large");
1194 	if (rss->types & MLX5_RSS_HF_MASK)
1195 		return rte_flow_error_set(error, ENOTSUP,
1196 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1197 					  &rss->types,
1198 					  "some RSS protocols are not"
1199 					  " supported");
1200 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1201 	    !(rss->types & ETH_RSS_IP))
1202 		return rte_flow_error_set(error, EINVAL,
1203 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204 					  "L3 partial RSS requested but L3 RSS"
1205 					  " type not specified");
1206 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1207 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1208 		return rte_flow_error_set(error, EINVAL,
1209 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1210 					  "L4 partial RSS requested but L4 RSS"
1211 					  " type not specified");
1212 	if (!priv->rxqs_n)
1213 		return rte_flow_error_set(error, EINVAL,
1214 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1215 					  NULL, "No Rx queues configured");
1216 	if (!rss->queue_num)
1217 		return rte_flow_error_set(error, EINVAL,
1218 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1219 					  NULL, "No queues configured");
1220 	for (i = 0; i != rss->queue_num; ++i) {
1221 		if (rss->queue[i] >= priv->rxqs_n)
1222 			return rte_flow_error_set
1223 				(error, EINVAL,
1224 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1225 				 &rss->queue[i], "queue index out of range");
1226 		if (!(*priv->rxqs)[rss->queue[i]])
1227 			return rte_flow_error_set
1228 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1229 				 &rss->queue[i], "queue is not configured");
1230 	}
1231 	if (attr->egress)
1232 		return rte_flow_error_set(error, ENOTSUP,
1233 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1234 					  "rss action not supported for "
1235 					  "egress");
1236 	if (rss->level > 1 &&  !tunnel)
1237 		return rte_flow_error_set(error, EINVAL,
1238 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1239 					  "inner RSS is not supported for "
1240 					  "non-tunnel flows");
1241 	return 0;
1242 }
1243 
1244 /*
1245  * Validate the count action.
1246  *
1247  * @param[in] dev
1248  *   Pointer to the Ethernet device structure.
1249  * @param[in] attr
1250  *   Attributes of flow that includes this action.
1251  * @param[out] error
1252  *   Pointer to error structure.
1253  *
1254  * @return
1255  *   0 on success, a negative errno value otherwise and rte_errno is set.
1256  */
1257 int
1258 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1259 				const struct rte_flow_attr *attr,
1260 				struct rte_flow_error *error)
1261 {
1262 	if (attr->egress)
1263 		return rte_flow_error_set(error, ENOTSUP,
1264 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1265 					  "count action not supported for "
1266 					  "egress");
1267 	return 0;
1268 }
1269 
1270 /**
1271  * Verify the @p attributes will be correctly understood by the NIC and store
1272  * them in the @p flow if everything is correct.
1273  *
1274  * @param[in] dev
1275  *   Pointer to the Ethernet device structure.
1276  * @param[in] attributes
1277  *   Pointer to flow attributes
1278  * @param[out] error
1279  *   Pointer to error structure.
1280  *
1281  * @return
1282  *   0 on success, a negative errno value otherwise and rte_errno is set.
1283  */
1284 int
1285 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1286 			      const struct rte_flow_attr *attributes,
1287 			      struct rte_flow_error *error)
1288 {
1289 	struct mlx5_priv *priv = dev->data->dev_private;
1290 	uint32_t priority_max = priv->config.flow_prio - 1;
1291 
1292 	if (attributes->group)
1293 		return rte_flow_error_set(error, ENOTSUP,
1294 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1295 					  NULL, "groups is not supported");
1296 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1297 	    attributes->priority >= priority_max)
1298 		return rte_flow_error_set(error, ENOTSUP,
1299 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1300 					  NULL, "priority out of range");
1301 	if (attributes->egress)
1302 		return rte_flow_error_set(error, ENOTSUP,
1303 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1304 					  "egress is not supported");
1305 	if (attributes->transfer && !priv->config.dv_esw_en)
1306 		return rte_flow_error_set(error, ENOTSUP,
1307 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1308 					  NULL, "transfer is not supported");
1309 	if (!attributes->ingress)
1310 		return rte_flow_error_set(error, EINVAL,
1311 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1312 					  NULL,
1313 					  "ingress attribute is mandatory");
1314 	return 0;
1315 }
1316 
1317 /**
1318  * Validate ICMP6 item.
1319  *
1320  * @param[in] item
1321  *   Item specification.
1322  * @param[in] item_flags
1323  *   Bit-fields that holds the items detected until now.
1324  * @param[out] error
1325  *   Pointer to error structure.
1326  *
1327  * @return
1328  *   0 on success, a negative errno value otherwise and rte_errno is set.
1329  */
1330 int
1331 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1332 			       uint64_t item_flags,
1333 			       uint8_t target_protocol,
1334 			       struct rte_flow_error *error)
1335 {
1336 	const struct rte_flow_item_icmp6 *mask = item->mask;
1337 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1338 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1339 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1340 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1341 				      MLX5_FLOW_LAYER_OUTER_L4;
1342 	int ret;
1343 
1344 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1345 		return rte_flow_error_set(error, EINVAL,
1346 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1347 					  "protocol filtering not compatible"
1348 					  " with ICMP6 layer");
1349 	if (!(item_flags & l3m))
1350 		return rte_flow_error_set(error, EINVAL,
1351 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1352 					  "IPv6 is mandatory to filter on"
1353 					  " ICMP6");
1354 	if (item_flags & l4m)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 					  "multiple L4 layers not supported");
1358 	if (!mask)
1359 		mask = &rte_flow_item_icmp6_mask;
1360 	ret = mlx5_flow_item_acceptable
1361 		(item, (const uint8_t *)mask,
1362 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1363 		 sizeof(struct rte_flow_item_icmp6), error);
1364 	if (ret < 0)
1365 		return ret;
1366 	return 0;
1367 }
1368 
1369 /**
1370  * Validate ICMP item.
1371  *
1372  * @param[in] item
1373  *   Item specification.
1374  * @param[in] item_flags
1375  *   Bit-fields that holds the items detected until now.
1376  * @param[out] error
1377  *   Pointer to error structure.
1378  *
1379  * @return
1380  *   0 on success, a negative errno value otherwise and rte_errno is set.
1381  */
1382 int
1383 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1384 			     uint64_t item_flags,
1385 			     uint8_t target_protocol,
1386 			     struct rte_flow_error *error)
1387 {
1388 	const struct rte_flow_item_icmp *mask = item->mask;
1389 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1390 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1391 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1392 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1393 				      MLX5_FLOW_LAYER_OUTER_L4;
1394 	int ret;
1395 
1396 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1397 		return rte_flow_error_set(error, EINVAL,
1398 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1399 					  "protocol filtering not compatible"
1400 					  " with ICMP layer");
1401 	if (!(item_flags & l3m))
1402 		return rte_flow_error_set(error, EINVAL,
1403 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1404 					  "IPv4 is mandatory to filter"
1405 					  " on ICMP");
1406 	if (item_flags & l4m)
1407 		return rte_flow_error_set(error, EINVAL,
1408 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1409 					  "multiple L4 layers not supported");
1410 	if (!mask)
1411 		mask = &rte_flow_item_icmp_mask;
1412 	ret = mlx5_flow_item_acceptable
1413 		(item, (const uint8_t *)mask,
1414 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1415 		 sizeof(struct rte_flow_item_icmp), error);
1416 	if (ret < 0)
1417 		return ret;
1418 	return 0;
1419 }
1420 
1421 /**
1422  * Validate Ethernet item.
1423  *
1424  * @param[in] item
1425  *   Item specification.
1426  * @param[in] item_flags
1427  *   Bit-fields that holds the items detected until now.
1428  * @param[out] error
1429  *   Pointer to error structure.
1430  *
1431  * @return
1432  *   0 on success, a negative errno value otherwise and rte_errno is set.
1433  */
1434 int
1435 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1436 			    uint64_t item_flags,
1437 			    struct rte_flow_error *error)
1438 {
1439 	const struct rte_flow_item_eth *mask = item->mask;
1440 	const struct rte_flow_item_eth nic_mask = {
1441 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1442 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1443 		.type = RTE_BE16(0xffff),
1444 	};
1445 	int ret;
1446 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1447 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1448 				       MLX5_FLOW_LAYER_OUTER_L2;
1449 
1450 	if (item_flags & ethm)
1451 		return rte_flow_error_set(error, ENOTSUP,
1452 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1453 					  "multiple L2 layers not supported");
1454 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1455 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1456 		return rte_flow_error_set(error, EINVAL,
1457 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1458 					  "L2 layer should not follow "
1459 					  "L3 layers");
1460 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1461 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1462 		return rte_flow_error_set(error, EINVAL,
1463 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1464 					  "L2 layer should not follow VLAN");
1465 	if (!mask)
1466 		mask = &rte_flow_item_eth_mask;
1467 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1468 					(const uint8_t *)&nic_mask,
1469 					sizeof(struct rte_flow_item_eth),
1470 					error);
1471 	return ret;
1472 }
1473 
1474 /**
1475  * Validate VLAN item.
1476  *
1477  * @param[in] item
1478  *   Item specification.
1479  * @param[in] item_flags
1480  *   Bit-fields that holds the items detected until now.
1481  * @param[in] dev
1482  *   Ethernet device flow is being created on.
1483  * @param[out] error
1484  *   Pointer to error structure.
1485  *
1486  * @return
1487  *   0 on success, a negative errno value otherwise and rte_errno is set.
1488  */
1489 int
1490 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1491 			     uint64_t item_flags,
1492 			     struct rte_eth_dev *dev,
1493 			     struct rte_flow_error *error)
1494 {
1495 	const struct rte_flow_item_vlan *spec = item->spec;
1496 	const struct rte_flow_item_vlan *mask = item->mask;
1497 	const struct rte_flow_item_vlan nic_mask = {
1498 		.tci = RTE_BE16(UINT16_MAX),
1499 		.inner_type = RTE_BE16(UINT16_MAX),
1500 	};
1501 	uint16_t vlan_tag = 0;
1502 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1503 	int ret;
1504 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1505 					MLX5_FLOW_LAYER_INNER_L4) :
1506 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1507 					MLX5_FLOW_LAYER_OUTER_L4);
1508 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1509 					MLX5_FLOW_LAYER_OUTER_VLAN;
1510 
1511 	if (item_flags & vlanm)
1512 		return rte_flow_error_set(error, EINVAL,
1513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1514 					  "multiple VLAN layers not supported");
1515 	else if ((item_flags & l34m) != 0)
1516 		return rte_flow_error_set(error, EINVAL,
1517 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1518 					  "VLAN cannot follow L3/L4 layer");
1519 	if (!mask)
1520 		mask = &rte_flow_item_vlan_mask;
1521 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1522 					(const uint8_t *)&nic_mask,
1523 					sizeof(struct rte_flow_item_vlan),
1524 					error);
1525 	if (ret)
1526 		return ret;
1527 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1528 		struct mlx5_priv *priv = dev->data->dev_private;
1529 
1530 		if (priv->vmwa_context) {
1531 			/*
1532 			 * Non-NULL context means we have a virtual machine
1533 			 * and SR-IOV enabled, we have to create VLAN interface
1534 			 * to make hypervisor to setup E-Switch vport
1535 			 * context correctly. We avoid creating the multiple
1536 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1537 			 */
1538 			return rte_flow_error_set(error, EINVAL,
1539 						  RTE_FLOW_ERROR_TYPE_ITEM,
1540 						  item,
1541 						  "VLAN tag mask is not"
1542 						  " supported in virtual"
1543 						  " environment");
1544 		}
1545 	}
1546 	if (spec) {
1547 		vlan_tag = spec->tci;
1548 		vlan_tag &= mask->tci;
1549 	}
1550 	/*
1551 	 * From verbs perspective an empty VLAN is equivalent
1552 	 * to a packet without VLAN layer.
1553 	 */
1554 	if (!vlan_tag)
1555 		return rte_flow_error_set(error, EINVAL,
1556 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1557 					  item->spec,
1558 					  "VLAN cannot be empty");
1559 	return 0;
1560 }
1561 
1562 /**
1563  * Validate IPV4 item.
1564  *
1565  * @param[in] item
1566  *   Item specification.
1567  * @param[in] item_flags
1568  *   Bit-fields that holds the items detected until now.
1569  * @param[in] acc_mask
1570  *   Acceptable mask, if NULL default internal default mask
1571  *   will be used to check whether item fields are supported.
1572  * @param[out] error
1573  *   Pointer to error structure.
1574  *
1575  * @return
1576  *   0 on success, a negative errno value otherwise and rte_errno is set.
1577  */
1578 int
1579 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1580 			     uint64_t item_flags,
1581 			     uint64_t last_item,
1582 			     uint16_t ether_type,
1583 			     const struct rte_flow_item_ipv4 *acc_mask,
1584 			     struct rte_flow_error *error)
1585 {
1586 	const struct rte_flow_item_ipv4 *mask = item->mask;
1587 	const struct rte_flow_item_ipv4 *spec = item->spec;
1588 	const struct rte_flow_item_ipv4 nic_mask = {
1589 		.hdr = {
1590 			.src_addr = RTE_BE32(0xffffffff),
1591 			.dst_addr = RTE_BE32(0xffffffff),
1592 			.type_of_service = 0xff,
1593 			.next_proto_id = 0xff,
1594 		},
1595 	};
1596 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1597 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1598 				      MLX5_FLOW_LAYER_OUTER_L3;
1599 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1600 				      MLX5_FLOW_LAYER_OUTER_L4;
1601 	int ret;
1602 	uint8_t next_proto = 0xFF;
1603 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1604 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1605 				  MLX5_FLOW_LAYER_INNER_VLAN);
1606 
1607 	if ((last_item & l2_vlan) && ether_type &&
1608 	    ether_type != RTE_ETHER_TYPE_IPV4)
1609 		return rte_flow_error_set(error, EINVAL,
1610 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1611 					  "IPv4 cannot follow L2/VLAN layer "
1612 					  "which ether type is not IPv4");
1613 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1614 		if (mask && spec)
1615 			next_proto = mask->hdr.next_proto_id &
1616 				     spec->hdr.next_proto_id;
1617 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1618 			return rte_flow_error_set(error, EINVAL,
1619 						  RTE_FLOW_ERROR_TYPE_ITEM,
1620 						  item,
1621 						  "multiple tunnel "
1622 						  "not supported");
1623 	}
1624 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1625 		return rte_flow_error_set(error, EINVAL,
1626 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1627 					  "wrong tunnel type - IPv6 specified "
1628 					  "but IPv4 item provided");
1629 	if (item_flags & l3m)
1630 		return rte_flow_error_set(error, ENOTSUP,
1631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1632 					  "multiple L3 layers not supported");
1633 	else if (item_flags & l4m)
1634 		return rte_flow_error_set(error, EINVAL,
1635 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1636 					  "L3 cannot follow an L4 layer.");
1637 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1638 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1639 		return rte_flow_error_set(error, EINVAL,
1640 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1641 					  "L3 cannot follow an NVGRE layer.");
1642 	if (!mask)
1643 		mask = &rte_flow_item_ipv4_mask;
1644 	else if (mask->hdr.next_proto_id != 0 &&
1645 		 mask->hdr.next_proto_id != 0xff)
1646 		return rte_flow_error_set(error, EINVAL,
1647 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1648 					  "partial mask is not supported"
1649 					  " for protocol");
1650 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1651 					acc_mask ? (const uint8_t *)acc_mask
1652 						 : (const uint8_t *)&nic_mask,
1653 					sizeof(struct rte_flow_item_ipv4),
1654 					error);
1655 	if (ret < 0)
1656 		return ret;
1657 	return 0;
1658 }
1659 
1660 /**
1661  * Validate IPV6 item.
1662  *
1663  * @param[in] item
1664  *   Item specification.
1665  * @param[in] item_flags
1666  *   Bit-fields that holds the items detected until now.
1667  * @param[in] acc_mask
1668  *   Acceptable mask, if NULL default internal default mask
1669  *   will be used to check whether item fields are supported.
1670  * @param[out] error
1671  *   Pointer to error structure.
1672  *
1673  * @return
1674  *   0 on success, a negative errno value otherwise and rte_errno is set.
1675  */
1676 int
1677 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1678 			     uint64_t item_flags,
1679 			     uint64_t last_item,
1680 			     uint16_t ether_type,
1681 			     const struct rte_flow_item_ipv6 *acc_mask,
1682 			     struct rte_flow_error *error)
1683 {
1684 	const struct rte_flow_item_ipv6 *mask = item->mask;
1685 	const struct rte_flow_item_ipv6 *spec = item->spec;
1686 	const struct rte_flow_item_ipv6 nic_mask = {
1687 		.hdr = {
1688 			.src_addr =
1689 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1690 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1691 			.dst_addr =
1692 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1693 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1694 			.vtc_flow = RTE_BE32(0xffffffff),
1695 			.proto = 0xff,
1696 		},
1697 	};
1698 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1699 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1700 				      MLX5_FLOW_LAYER_OUTER_L3;
1701 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1702 				      MLX5_FLOW_LAYER_OUTER_L4;
1703 	int ret;
1704 	uint8_t next_proto = 0xFF;
1705 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1706 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1707 				  MLX5_FLOW_LAYER_INNER_VLAN);
1708 
1709 	if ((last_item & l2_vlan) && ether_type &&
1710 	    ether_type != RTE_ETHER_TYPE_IPV6)
1711 		return rte_flow_error_set(error, EINVAL,
1712 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1713 					  "IPv6 cannot follow L2/VLAN layer "
1714 					  "which ether type is not IPv6");
1715 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1716 		if (mask && spec)
1717 			next_proto = mask->hdr.proto & spec->hdr.proto;
1718 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1719 			return rte_flow_error_set(error, EINVAL,
1720 						  RTE_FLOW_ERROR_TYPE_ITEM,
1721 						  item,
1722 						  "multiple tunnel "
1723 						  "not supported");
1724 	}
1725 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1726 		return rte_flow_error_set(error, EINVAL,
1727 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1728 					  "wrong tunnel type - IPv4 specified "
1729 					  "but IPv6 item provided");
1730 	if (item_flags & l3m)
1731 		return rte_flow_error_set(error, ENOTSUP,
1732 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1733 					  "multiple L3 layers not supported");
1734 	else if (item_flags & l4m)
1735 		return rte_flow_error_set(error, EINVAL,
1736 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1737 					  "L3 cannot follow an L4 layer.");
1738 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1739 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1740 		return rte_flow_error_set(error, EINVAL,
1741 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1742 					  "L3 cannot follow an NVGRE layer.");
1743 	if (!mask)
1744 		mask = &rte_flow_item_ipv6_mask;
1745 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1746 					acc_mask ? (const uint8_t *)acc_mask
1747 						 : (const uint8_t *)&nic_mask,
1748 					sizeof(struct rte_flow_item_ipv6),
1749 					error);
1750 	if (ret < 0)
1751 		return ret;
1752 	return 0;
1753 }
1754 
1755 /**
1756  * Validate UDP item.
1757  *
1758  * @param[in] item
1759  *   Item specification.
1760  * @param[in] item_flags
1761  *   Bit-fields that holds the items detected until now.
1762  * @param[in] target_protocol
1763  *   The next protocol in the previous item.
1764  * @param[in] flow_mask
1765  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1766  * @param[out] error
1767  *   Pointer to error structure.
1768  *
1769  * @return
1770  *   0 on success, a negative errno value otherwise and rte_errno is set.
1771  */
1772 int
1773 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1774 			    uint64_t item_flags,
1775 			    uint8_t target_protocol,
1776 			    struct rte_flow_error *error)
1777 {
1778 	const struct rte_flow_item_udp *mask = item->mask;
1779 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1780 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1781 				      MLX5_FLOW_LAYER_OUTER_L3;
1782 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1783 				      MLX5_FLOW_LAYER_OUTER_L4;
1784 	int ret;
1785 
1786 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1787 		return rte_flow_error_set(error, EINVAL,
1788 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1789 					  "protocol filtering not compatible"
1790 					  " with UDP layer");
1791 	if (!(item_flags & l3m))
1792 		return rte_flow_error_set(error, EINVAL,
1793 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1794 					  "L3 is mandatory to filter on L4");
1795 	if (item_flags & l4m)
1796 		return rte_flow_error_set(error, EINVAL,
1797 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1798 					  "multiple L4 layers not supported");
1799 	if (!mask)
1800 		mask = &rte_flow_item_udp_mask;
1801 	ret = mlx5_flow_item_acceptable
1802 		(item, (const uint8_t *)mask,
1803 		 (const uint8_t *)&rte_flow_item_udp_mask,
1804 		 sizeof(struct rte_flow_item_udp), error);
1805 	if (ret < 0)
1806 		return ret;
1807 	return 0;
1808 }
1809 
1810 /**
1811  * Validate TCP item.
1812  *
1813  * @param[in] item
1814  *   Item specification.
1815  * @param[in] item_flags
1816  *   Bit-fields that holds the items detected until now.
1817  * @param[in] target_protocol
1818  *   The next protocol in the previous item.
1819  * @param[out] error
1820  *   Pointer to error structure.
1821  *
1822  * @return
1823  *   0 on success, a negative errno value otherwise and rte_errno is set.
1824  */
1825 int
1826 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1827 			    uint64_t item_flags,
1828 			    uint8_t target_protocol,
1829 			    const struct rte_flow_item_tcp *flow_mask,
1830 			    struct rte_flow_error *error)
1831 {
1832 	const struct rte_flow_item_tcp *mask = item->mask;
1833 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1834 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1835 				      MLX5_FLOW_LAYER_OUTER_L3;
1836 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1837 				      MLX5_FLOW_LAYER_OUTER_L4;
1838 	int ret;
1839 
1840 	MLX5_ASSERT(flow_mask);
1841 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1842 		return rte_flow_error_set(error, EINVAL,
1843 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1844 					  "protocol filtering not compatible"
1845 					  " with TCP layer");
1846 	if (!(item_flags & l3m))
1847 		return rte_flow_error_set(error, EINVAL,
1848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1849 					  "L3 is mandatory to filter on L4");
1850 	if (item_flags & l4m)
1851 		return rte_flow_error_set(error, EINVAL,
1852 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 					  "multiple L4 layers not supported");
1854 	if (!mask)
1855 		mask = &rte_flow_item_tcp_mask;
1856 	ret = mlx5_flow_item_acceptable
1857 		(item, (const uint8_t *)mask,
1858 		 (const uint8_t *)flow_mask,
1859 		 sizeof(struct rte_flow_item_tcp), error);
1860 	if (ret < 0)
1861 		return ret;
1862 	return 0;
1863 }
1864 
1865 /**
1866  * Validate VXLAN item.
1867  *
1868  * @param[in] item
1869  *   Item specification.
1870  * @param[in] item_flags
1871  *   Bit-fields that holds the items detected until now.
1872  * @param[in] target_protocol
1873  *   The next protocol in the previous item.
1874  * @param[out] error
1875  *   Pointer to error structure.
1876  *
1877  * @return
1878  *   0 on success, a negative errno value otherwise and rte_errno is set.
1879  */
1880 int
1881 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1882 			      uint64_t item_flags,
1883 			      struct rte_flow_error *error)
1884 {
1885 	const struct rte_flow_item_vxlan *spec = item->spec;
1886 	const struct rte_flow_item_vxlan *mask = item->mask;
1887 	int ret;
1888 	union vni {
1889 		uint32_t vlan_id;
1890 		uint8_t vni[4];
1891 	} id = { .vlan_id = 0, };
1892 
1893 
1894 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1895 		return rte_flow_error_set(error, ENOTSUP,
1896 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1897 					  "multiple tunnel layers not"
1898 					  " supported");
1899 	/*
1900 	 * Verify only UDPv4 is present as defined in
1901 	 * https://tools.ietf.org/html/rfc7348
1902 	 */
1903 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1904 		return rte_flow_error_set(error, EINVAL,
1905 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1906 					  "no outer UDP layer found");
1907 	if (!mask)
1908 		mask = &rte_flow_item_vxlan_mask;
1909 	ret = mlx5_flow_item_acceptable
1910 		(item, (const uint8_t *)mask,
1911 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1912 		 sizeof(struct rte_flow_item_vxlan),
1913 		 error);
1914 	if (ret < 0)
1915 		return ret;
1916 	if (spec) {
1917 		memcpy(&id.vni[1], spec->vni, 3);
1918 		memcpy(&id.vni[1], mask->vni, 3);
1919 	}
1920 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1921 		return rte_flow_error_set(error, ENOTSUP,
1922 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1923 					  "VXLAN tunnel must be fully defined");
1924 	return 0;
1925 }
1926 
1927 /**
1928  * Validate VXLAN_GPE item.
1929  *
1930  * @param[in] item
1931  *   Item specification.
1932  * @param[in] item_flags
1933  *   Bit-fields that holds the items detected until now.
1934  * @param[in] priv
1935  *   Pointer to the private data structure.
1936  * @param[in] target_protocol
1937  *   The next protocol in the previous item.
1938  * @param[out] error
1939  *   Pointer to error structure.
1940  *
1941  * @return
1942  *   0 on success, a negative errno value otherwise and rte_errno is set.
1943  */
1944 int
1945 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1946 				  uint64_t item_flags,
1947 				  struct rte_eth_dev *dev,
1948 				  struct rte_flow_error *error)
1949 {
1950 	struct mlx5_priv *priv = dev->data->dev_private;
1951 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1952 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1953 	int ret;
1954 	union vni {
1955 		uint32_t vlan_id;
1956 		uint8_t vni[4];
1957 	} id = { .vlan_id = 0, };
1958 
1959 	if (!priv->config.l3_vxlan_en)
1960 		return rte_flow_error_set(error, ENOTSUP,
1961 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1962 					  "L3 VXLAN is not enabled by device"
1963 					  " parameter and/or not configured in"
1964 					  " firmware");
1965 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1966 		return rte_flow_error_set(error, ENOTSUP,
1967 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1968 					  "multiple tunnel layers not"
1969 					  " supported");
1970 	/*
1971 	 * Verify only UDPv4 is present as defined in
1972 	 * https://tools.ietf.org/html/rfc7348
1973 	 */
1974 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1975 		return rte_flow_error_set(error, EINVAL,
1976 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 					  "no outer UDP layer found");
1978 	if (!mask)
1979 		mask = &rte_flow_item_vxlan_gpe_mask;
1980 	ret = mlx5_flow_item_acceptable
1981 		(item, (const uint8_t *)mask,
1982 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1983 		 sizeof(struct rte_flow_item_vxlan_gpe),
1984 		 error);
1985 	if (ret < 0)
1986 		return ret;
1987 	if (spec) {
1988 		if (spec->protocol)
1989 			return rte_flow_error_set(error, ENOTSUP,
1990 						  RTE_FLOW_ERROR_TYPE_ITEM,
1991 						  item,
1992 						  "VxLAN-GPE protocol"
1993 						  " not supported");
1994 		memcpy(&id.vni[1], spec->vni, 3);
1995 		memcpy(&id.vni[1], mask->vni, 3);
1996 	}
1997 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1998 		return rte_flow_error_set(error, ENOTSUP,
1999 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2000 					  "VXLAN-GPE tunnel must be fully"
2001 					  " defined");
2002 	return 0;
2003 }
2004 /**
2005  * Validate GRE Key item.
2006  *
2007  * @param[in] item
2008  *   Item specification.
2009  * @param[in] item_flags
2010  *   Bit flags to mark detected items.
2011  * @param[in] gre_item
2012  *   Pointer to gre_item
2013  * @param[out] error
2014  *   Pointer to error structure.
2015  *
2016  * @return
2017  *   0 on success, a negative errno value otherwise and rte_errno is set.
2018  */
2019 int
2020 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2021 				uint64_t item_flags,
2022 				const struct rte_flow_item *gre_item,
2023 				struct rte_flow_error *error)
2024 {
2025 	const rte_be32_t *mask = item->mask;
2026 	int ret = 0;
2027 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2028 	const struct rte_flow_item_gre *gre_spec;
2029 	const struct rte_flow_item_gre *gre_mask;
2030 
2031 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2032 		return rte_flow_error_set(error, ENOTSUP,
2033 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2034 					  "Multiple GRE key not support");
2035 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2036 		return rte_flow_error_set(error, ENOTSUP,
2037 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2038 					  "No preceding GRE header");
2039 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2040 		return rte_flow_error_set(error, ENOTSUP,
2041 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2042 					  "GRE key following a wrong item");
2043 	gre_mask = gre_item->mask;
2044 	if (!gre_mask)
2045 		gre_mask = &rte_flow_item_gre_mask;
2046 	gre_spec = gre_item->spec;
2047 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2048 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2049 		return rte_flow_error_set(error, EINVAL,
2050 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2051 					  "Key bit must be on");
2052 
2053 	if (!mask)
2054 		mask = &gre_key_default_mask;
2055 	ret = mlx5_flow_item_acceptable
2056 		(item, (const uint8_t *)mask,
2057 		 (const uint8_t *)&gre_key_default_mask,
2058 		 sizeof(rte_be32_t), error);
2059 	return ret;
2060 }
2061 
2062 /**
2063  * Validate GRE item.
2064  *
2065  * @param[in] item
2066  *   Item specification.
2067  * @param[in] item_flags
2068  *   Bit flags to mark detected items.
2069  * @param[in] target_protocol
2070  *   The next protocol in the previous item.
2071  * @param[out] error
2072  *   Pointer to error structure.
2073  *
2074  * @return
2075  *   0 on success, a negative errno value otherwise and rte_errno is set.
2076  */
2077 int
2078 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2079 			    uint64_t item_flags,
2080 			    uint8_t target_protocol,
2081 			    struct rte_flow_error *error)
2082 {
2083 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2084 	const struct rte_flow_item_gre *mask = item->mask;
2085 	int ret;
2086 	const struct rte_flow_item_gre nic_mask = {
2087 		.c_rsvd0_ver = RTE_BE16(0xB000),
2088 		.protocol = RTE_BE16(UINT16_MAX),
2089 	};
2090 
2091 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2092 		return rte_flow_error_set(error, EINVAL,
2093 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2094 					  "protocol filtering not compatible"
2095 					  " with this GRE layer");
2096 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2097 		return rte_flow_error_set(error, ENOTSUP,
2098 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2099 					  "multiple tunnel layers not"
2100 					  " supported");
2101 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2102 		return rte_flow_error_set(error, ENOTSUP,
2103 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2104 					  "L3 Layer is missing");
2105 	if (!mask)
2106 		mask = &rte_flow_item_gre_mask;
2107 	ret = mlx5_flow_item_acceptable
2108 		(item, (const uint8_t *)mask,
2109 		 (const uint8_t *)&nic_mask,
2110 		 sizeof(struct rte_flow_item_gre), error);
2111 	if (ret < 0)
2112 		return ret;
2113 #ifndef HAVE_MLX5DV_DR
2114 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2115 	if (spec && (spec->protocol & mask->protocol))
2116 		return rte_flow_error_set(error, ENOTSUP,
2117 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2118 					  "without MPLS support the"
2119 					  " specification cannot be used for"
2120 					  " filtering");
2121 #endif
2122 #endif
2123 	return 0;
2124 }
2125 
2126 /**
2127  * Validate Geneve item.
2128  *
2129  * @param[in] item
2130  *   Item specification.
2131  * @param[in] itemFlags
2132  *   Bit-fields that holds the items detected until now.
2133  * @param[in] enPriv
2134  *   Pointer to the private data structure.
2135  * @param[out] error
2136  *   Pointer to error structure.
2137  *
2138  * @return
2139  *   0 on success, a negative errno value otherwise and rte_errno is set.
2140  */
2141 
2142 int
2143 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2144 			       uint64_t item_flags,
2145 			       struct rte_eth_dev *dev,
2146 			       struct rte_flow_error *error)
2147 {
2148 	struct mlx5_priv *priv = dev->data->dev_private;
2149 	const struct rte_flow_item_geneve *spec = item->spec;
2150 	const struct rte_flow_item_geneve *mask = item->mask;
2151 	int ret;
2152 	uint16_t gbhdr;
2153 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2154 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2155 	const struct rte_flow_item_geneve nic_mask = {
2156 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2157 		.vni = "\xff\xff\xff",
2158 		.protocol = RTE_BE16(UINT16_MAX),
2159 	};
2160 
2161 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2162 		return rte_flow_error_set(error, ENOTSUP,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "L3 Geneve is not enabled by device"
2165 					  " parameter and/or not configured in"
2166 					  " firmware");
2167 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2168 		return rte_flow_error_set(error, ENOTSUP,
2169 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2170 					  "multiple tunnel layers not"
2171 					  " supported");
2172 	/*
2173 	 * Verify only UDPv4 is present as defined in
2174 	 * https://tools.ietf.org/html/rfc7348
2175 	 */
2176 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2177 		return rte_flow_error_set(error, EINVAL,
2178 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2179 					  "no outer UDP layer found");
2180 	if (!mask)
2181 		mask = &rte_flow_item_geneve_mask;
2182 	ret = mlx5_flow_item_acceptable
2183 				  (item, (const uint8_t *)mask,
2184 				   (const uint8_t *)&nic_mask,
2185 				   sizeof(struct rte_flow_item_geneve), error);
2186 	if (ret)
2187 		return ret;
2188 	if (spec) {
2189 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2190 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2191 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2192 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2193 			return rte_flow_error_set(error, ENOTSUP,
2194 						  RTE_FLOW_ERROR_TYPE_ITEM,
2195 						  item,
2196 						  "Geneve protocol unsupported"
2197 						  " fields are being used");
2198 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2199 			return rte_flow_error_set
2200 					(error, ENOTSUP,
2201 					 RTE_FLOW_ERROR_TYPE_ITEM,
2202 					 item,
2203 					 "Unsupported Geneve options length");
2204 	}
2205 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2206 		return rte_flow_error_set
2207 				    (error, ENOTSUP,
2208 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2209 				     "Geneve tunnel must be fully defined");
2210 	return 0;
2211 }
2212 
2213 /**
2214  * Validate MPLS item.
2215  *
2216  * @param[in] dev
2217  *   Pointer to the rte_eth_dev structure.
2218  * @param[in] item
2219  *   Item specification.
2220  * @param[in] item_flags
2221  *   Bit-fields that holds the items detected until now.
2222  * @param[in] prev_layer
2223  *   The protocol layer indicated in previous item.
2224  * @param[out] error
2225  *   Pointer to error structure.
2226  *
2227  * @return
2228  *   0 on success, a negative errno value otherwise and rte_errno is set.
2229  */
2230 int
2231 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2232 			     const struct rte_flow_item *item __rte_unused,
2233 			     uint64_t item_flags __rte_unused,
2234 			     uint64_t prev_layer __rte_unused,
2235 			     struct rte_flow_error *error)
2236 {
2237 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2238 	const struct rte_flow_item_mpls *mask = item->mask;
2239 	struct mlx5_priv *priv = dev->data->dev_private;
2240 	int ret;
2241 
2242 	if (!priv->config.mpls_en)
2243 		return rte_flow_error_set(error, ENOTSUP,
2244 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2245 					  "MPLS not supported or"
2246 					  " disabled in firmware"
2247 					  " configuration.");
2248 	/* MPLS over IP, UDP, GRE is allowed */
2249 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2250 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2251 			    MLX5_FLOW_LAYER_GRE)))
2252 		return rte_flow_error_set(error, EINVAL,
2253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 					  "protocol filtering not compatible"
2255 					  " with MPLS layer");
2256 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2257 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2258 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2259 		return rte_flow_error_set(error, ENOTSUP,
2260 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2261 					  "multiple tunnel layers not"
2262 					  " supported");
2263 	if (!mask)
2264 		mask = &rte_flow_item_mpls_mask;
2265 	ret = mlx5_flow_item_acceptable
2266 		(item, (const uint8_t *)mask,
2267 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2268 		 sizeof(struct rte_flow_item_mpls), error);
2269 	if (ret < 0)
2270 		return ret;
2271 	return 0;
2272 #endif
2273 	return rte_flow_error_set(error, ENOTSUP,
2274 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2275 				  "MPLS is not supported by Verbs, please"
2276 				  " update.");
2277 }
2278 
2279 /**
2280  * Validate NVGRE item.
2281  *
2282  * @param[in] item
2283  *   Item specification.
2284  * @param[in] item_flags
2285  *   Bit flags to mark detected items.
2286  * @param[in] target_protocol
2287  *   The next protocol in the previous item.
2288  * @param[out] error
2289  *   Pointer to error structure.
2290  *
2291  * @return
2292  *   0 on success, a negative errno value otherwise and rte_errno is set.
2293  */
2294 int
2295 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2296 			      uint64_t item_flags,
2297 			      uint8_t target_protocol,
2298 			      struct rte_flow_error *error)
2299 {
2300 	const struct rte_flow_item_nvgre *mask = item->mask;
2301 	int ret;
2302 
2303 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2304 		return rte_flow_error_set(error, EINVAL,
2305 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2306 					  "protocol filtering not compatible"
2307 					  " with this GRE layer");
2308 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2309 		return rte_flow_error_set(error, ENOTSUP,
2310 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2311 					  "multiple tunnel layers not"
2312 					  " supported");
2313 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2314 		return rte_flow_error_set(error, ENOTSUP,
2315 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2316 					  "L3 Layer is missing");
2317 	if (!mask)
2318 		mask = &rte_flow_item_nvgre_mask;
2319 	ret = mlx5_flow_item_acceptable
2320 		(item, (const uint8_t *)mask,
2321 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2322 		 sizeof(struct rte_flow_item_nvgre), error);
2323 	if (ret < 0)
2324 		return ret;
2325 	return 0;
2326 }
2327 
2328 /* Allocate unique ID for the split Q/RSS subflows. */
2329 static uint32_t
2330 flow_qrss_get_id(struct rte_eth_dev *dev)
2331 {
2332 	struct mlx5_priv *priv = dev->data->dev_private;
2333 	uint32_t qrss_id, ret;
2334 
2335 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2336 	if (ret)
2337 		return 0;
2338 	MLX5_ASSERT(qrss_id);
2339 	return qrss_id;
2340 }
2341 
2342 /* Free unique ID for the split Q/RSS subflows. */
2343 static void
2344 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2345 {
2346 	struct mlx5_priv *priv = dev->data->dev_private;
2347 
2348 	if (qrss_id)
2349 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2350 }
2351 
2352 /**
2353  * Release resource related QUEUE/RSS action split.
2354  *
2355  * @param dev
2356  *   Pointer to Ethernet device.
2357  * @param flow
2358  *   Flow to release id's from.
2359  */
2360 static void
2361 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2362 			     struct rte_flow *flow)
2363 {
2364 	struct mlx5_priv *priv = dev->data->dev_private;
2365 	uint32_t handle_idx;
2366 	struct mlx5_flow_handle *dev_handle;
2367 
2368 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2369 		       handle_idx, dev_handle, next)
2370 		if (dev_handle->split_flow_id)
2371 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2372 }
2373 
2374 static int
2375 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2376 		   const struct rte_flow_attr *attr __rte_unused,
2377 		   const struct rte_flow_item items[] __rte_unused,
2378 		   const struct rte_flow_action actions[] __rte_unused,
2379 		   bool external __rte_unused,
2380 		   int hairpin __rte_unused,
2381 		   struct rte_flow_error *error)
2382 {
2383 	return rte_flow_error_set(error, ENOTSUP,
2384 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2385 }
2386 
2387 static struct mlx5_flow *
2388 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2389 		  const struct rte_flow_attr *attr __rte_unused,
2390 		  const struct rte_flow_item items[] __rte_unused,
2391 		  const struct rte_flow_action actions[] __rte_unused,
2392 		  struct rte_flow_error *error)
2393 {
2394 	rte_flow_error_set(error, ENOTSUP,
2395 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2396 	return NULL;
2397 }
2398 
2399 static int
2400 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2401 		    struct mlx5_flow *dev_flow __rte_unused,
2402 		    const struct rte_flow_attr *attr __rte_unused,
2403 		    const struct rte_flow_item items[] __rte_unused,
2404 		    const struct rte_flow_action actions[] __rte_unused,
2405 		    struct rte_flow_error *error)
2406 {
2407 	return rte_flow_error_set(error, ENOTSUP,
2408 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2409 }
2410 
2411 static int
2412 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2413 		struct rte_flow *flow __rte_unused,
2414 		struct rte_flow_error *error)
2415 {
2416 	return rte_flow_error_set(error, ENOTSUP,
2417 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2418 }
2419 
2420 static void
2421 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2422 		 struct rte_flow *flow __rte_unused)
2423 {
2424 }
2425 
2426 static void
2427 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2428 		  struct rte_flow *flow __rte_unused)
2429 {
2430 }
2431 
2432 static int
2433 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2434 		struct rte_flow *flow __rte_unused,
2435 		const struct rte_flow_action *actions __rte_unused,
2436 		void *data __rte_unused,
2437 		struct rte_flow_error *error)
2438 {
2439 	return rte_flow_error_set(error, ENOTSUP,
2440 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2441 }
2442 
2443 /* Void driver to protect from null pointer reference. */
2444 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2445 	.validate = flow_null_validate,
2446 	.prepare = flow_null_prepare,
2447 	.translate = flow_null_translate,
2448 	.apply = flow_null_apply,
2449 	.remove = flow_null_remove,
2450 	.destroy = flow_null_destroy,
2451 	.query = flow_null_query,
2452 };
2453 
2454 /**
2455  * Select flow driver type according to flow attributes and device
2456  * configuration.
2457  *
2458  * @param[in] dev
2459  *   Pointer to the dev structure.
2460  * @param[in] attr
2461  *   Pointer to the flow attributes.
2462  *
2463  * @return
2464  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2465  */
2466 static enum mlx5_flow_drv_type
2467 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2468 {
2469 	struct mlx5_priv *priv = dev->data->dev_private;
2470 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2471 
2472 	if (attr->transfer && priv->config.dv_esw_en)
2473 		type = MLX5_FLOW_TYPE_DV;
2474 	if (!attr->transfer)
2475 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2476 						 MLX5_FLOW_TYPE_VERBS;
2477 	return type;
2478 }
2479 
2480 #define flow_get_drv_ops(type) flow_drv_ops[type]
2481 
2482 /**
2483  * Flow driver validation API. This abstracts calling driver specific functions.
2484  * The type of flow driver is determined according to flow attributes.
2485  *
2486  * @param[in] dev
2487  *   Pointer to the dev structure.
2488  * @param[in] attr
2489  *   Pointer to the flow attributes.
2490  * @param[in] items
2491  *   Pointer to the list of items.
2492  * @param[in] actions
2493  *   Pointer to the list of actions.
2494  * @param[in] external
2495  *   This flow rule is created by request external to PMD.
2496  * @param[in] hairpin
2497  *   Number of hairpin TX actions, 0 means classic flow.
2498  * @param[out] error
2499  *   Pointer to the error structure.
2500  *
2501  * @return
2502  *   0 on success, a negative errno value otherwise and rte_errno is set.
2503  */
2504 static inline int
2505 flow_drv_validate(struct rte_eth_dev *dev,
2506 		  const struct rte_flow_attr *attr,
2507 		  const struct rte_flow_item items[],
2508 		  const struct rte_flow_action actions[],
2509 		  bool external, int hairpin, struct rte_flow_error *error)
2510 {
2511 	const struct mlx5_flow_driver_ops *fops;
2512 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2513 
2514 	fops = flow_get_drv_ops(type);
2515 	return fops->validate(dev, attr, items, actions, external,
2516 			      hairpin, error);
2517 }
2518 
2519 /**
2520  * Flow driver preparation API. This abstracts calling driver specific
2521  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2522  * calculates the size of memory required for device flow, allocates the memory,
2523  * initializes the device flow and returns the pointer.
2524  *
2525  * @note
2526  *   This function initializes device flow structure such as dv or verbs in
2527  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2528  *   rest. For example, adding returning device flow to flow->dev_flow list and
2529  *   setting backward reference to the flow should be done out of this function.
2530  *   layers field is not filled either.
2531  *
2532  * @param[in] dev
2533  *   Pointer to the dev structure.
2534  * @param[in] attr
2535  *   Pointer to the flow attributes.
2536  * @param[in] items
2537  *   Pointer to the list of items.
2538  * @param[in] actions
2539  *   Pointer to the list of actions.
2540  * @param[in] flow_idx
2541  *   This memory pool index to the flow.
2542  * @param[out] error
2543  *   Pointer to the error structure.
2544  *
2545  * @return
2546  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2547  */
2548 static inline struct mlx5_flow *
2549 flow_drv_prepare(struct rte_eth_dev *dev,
2550 		 const struct rte_flow *flow,
2551 		 const struct rte_flow_attr *attr,
2552 		 const struct rte_flow_item items[],
2553 		 const struct rte_flow_action actions[],
2554 		 uint32_t flow_idx,
2555 		 struct rte_flow_error *error)
2556 {
2557 	const struct mlx5_flow_driver_ops *fops;
2558 	enum mlx5_flow_drv_type type = flow->drv_type;
2559 	struct mlx5_flow *mlx5_flow = NULL;
2560 
2561 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2562 	fops = flow_get_drv_ops(type);
2563 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2564 	if (mlx5_flow)
2565 		mlx5_flow->flow_idx = flow_idx;
2566 	return mlx5_flow;
2567 }
2568 
2569 /**
2570  * Flow driver translation API. This abstracts calling driver specific
2571  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2572  * translates a generic flow into a driver flow. flow_drv_prepare() must
2573  * precede.
2574  *
2575  * @note
2576  *   dev_flow->layers could be filled as a result of parsing during translation
2577  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2578  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2579  *   flow->actions could be overwritten even though all the expanded dev_flows
2580  *   have the same actions.
2581  *
2582  * @param[in] dev
2583  *   Pointer to the rte dev structure.
2584  * @param[in, out] dev_flow
2585  *   Pointer to the mlx5 flow.
2586  * @param[in] attr
2587  *   Pointer to the flow attributes.
2588  * @param[in] items
2589  *   Pointer to the list of items.
2590  * @param[in] actions
2591  *   Pointer to the list of actions.
2592  * @param[out] error
2593  *   Pointer to the error structure.
2594  *
2595  * @return
2596  *   0 on success, a negative errno value otherwise and rte_errno is set.
2597  */
2598 static inline int
2599 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2600 		   const struct rte_flow_attr *attr,
2601 		   const struct rte_flow_item items[],
2602 		   const struct rte_flow_action actions[],
2603 		   struct rte_flow_error *error)
2604 {
2605 	const struct mlx5_flow_driver_ops *fops;
2606 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2607 
2608 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2609 	fops = flow_get_drv_ops(type);
2610 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2611 }
2612 
2613 /**
2614  * Flow driver apply API. This abstracts calling driver specific functions.
2615  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2616  * translated driver flows on to device. flow_drv_translate() must precede.
2617  *
2618  * @param[in] dev
2619  *   Pointer to Ethernet device structure.
2620  * @param[in, out] flow
2621  *   Pointer to flow structure.
2622  * @param[out] error
2623  *   Pointer to error structure.
2624  *
2625  * @return
2626  *   0 on success, a negative errno value otherwise and rte_errno is set.
2627  */
2628 static inline int
2629 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2630 	       struct rte_flow_error *error)
2631 {
2632 	const struct mlx5_flow_driver_ops *fops;
2633 	enum mlx5_flow_drv_type type = flow->drv_type;
2634 
2635 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2636 	fops = flow_get_drv_ops(type);
2637 	return fops->apply(dev, flow, error);
2638 }
2639 
2640 /**
2641  * Flow driver remove API. This abstracts calling driver specific functions.
2642  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2643  * on device. All the resources of the flow should be freed by calling
2644  * flow_drv_destroy().
2645  *
2646  * @param[in] dev
2647  *   Pointer to Ethernet device.
2648  * @param[in, out] flow
2649  *   Pointer to flow structure.
2650  */
2651 static inline void
2652 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2653 {
2654 	const struct mlx5_flow_driver_ops *fops;
2655 	enum mlx5_flow_drv_type type = flow->drv_type;
2656 
2657 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2658 	fops = flow_get_drv_ops(type);
2659 	fops->remove(dev, flow);
2660 }
2661 
2662 /**
2663  * Flow driver destroy API. This abstracts calling driver specific functions.
2664  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2665  * on device and releases resources of the flow.
2666  *
2667  * @param[in] dev
2668  *   Pointer to Ethernet device.
2669  * @param[in, out] flow
2670  *   Pointer to flow structure.
2671  */
2672 static inline void
2673 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2674 {
2675 	const struct mlx5_flow_driver_ops *fops;
2676 	enum mlx5_flow_drv_type type = flow->drv_type;
2677 
2678 	flow_mreg_split_qrss_release(dev, flow);
2679 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2680 	fops = flow_get_drv_ops(type);
2681 	fops->destroy(dev, flow);
2682 }
2683 
2684 /**
2685  * Get RSS action from the action list.
2686  *
2687  * @param[in] actions
2688  *   Pointer to the list of actions.
2689  *
2690  * @return
2691  *   Pointer to the RSS action if exist, else return NULL.
2692  */
2693 static const struct rte_flow_action_rss*
2694 flow_get_rss_action(const struct rte_flow_action actions[])
2695 {
2696 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2697 		switch (actions->type) {
2698 		case RTE_FLOW_ACTION_TYPE_RSS:
2699 			return (const struct rte_flow_action_rss *)
2700 			       actions->conf;
2701 		default:
2702 			break;
2703 		}
2704 	}
2705 	return NULL;
2706 }
2707 
2708 static unsigned int
2709 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2710 {
2711 	const struct rte_flow_item *item;
2712 	unsigned int has_vlan = 0;
2713 
2714 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2715 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2716 			has_vlan = 1;
2717 			break;
2718 		}
2719 	}
2720 	if (has_vlan)
2721 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2722 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2723 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2724 			       MLX5_EXPANSION_ROOT_OUTER;
2725 }
2726 
2727 /**
2728  *  Get layer flags from the prefix flow.
2729  *
2730  *  Some flows may be split to several subflows, the prefix subflow gets the
2731  *  match items and the suffix sub flow gets the actions.
2732  *  Some actions need the user defined match item flags to get the detail for
2733  *  the action.
2734  *  This function helps the suffix flow to get the item layer flags from prefix
2735  *  subflow.
2736  *
2737  * @param[in] dev_flow
2738  *   Pointer the created preifx subflow.
2739  *
2740  * @return
2741  *   The layers get from prefix subflow.
2742  */
2743 static inline uint64_t
2744 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2745 {
2746 	uint64_t layers = 0;
2747 
2748 	/*
2749 	 * Layers bits could be localization, but usually the compiler will
2750 	 * help to do the optimization work for source code.
2751 	 * If no decap actions, use the layers directly.
2752 	 */
2753 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2754 		return dev_flow->handle->layers;
2755 	/* Convert L3 layers with decap action. */
2756 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2757 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2758 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2759 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2760 	/* Convert L4 layers with decap action.  */
2761 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2762 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2763 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2764 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2765 	return layers;
2766 }
2767 
2768 /**
2769  * Get metadata split action information.
2770  *
2771  * @param[in] actions
2772  *   Pointer to the list of actions.
2773  * @param[out] qrss
2774  *   Pointer to the return pointer.
2775  * @param[out] qrss_type
2776  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2777  *   if no QUEUE/RSS is found.
2778  * @param[out] encap_idx
2779  *   Pointer to the index of the encap action if exists, otherwise the last
2780  *   action index.
2781  *
2782  * @return
2783  *   Total number of actions.
2784  */
2785 static int
2786 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2787 				       const struct rte_flow_action **qrss,
2788 				       int *encap_idx)
2789 {
2790 	const struct rte_flow_action_raw_encap *raw_encap;
2791 	int actions_n = 0;
2792 	int raw_decap_idx = -1;
2793 
2794 	*encap_idx = -1;
2795 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2796 		switch (actions->type) {
2797 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2798 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2799 			*encap_idx = actions_n;
2800 			break;
2801 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2802 			raw_decap_idx = actions_n;
2803 			break;
2804 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2805 			raw_encap = actions->conf;
2806 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2807 				*encap_idx = raw_decap_idx != -1 ?
2808 						      raw_decap_idx : actions_n;
2809 			break;
2810 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2811 		case RTE_FLOW_ACTION_TYPE_RSS:
2812 			*qrss = actions;
2813 			break;
2814 		default:
2815 			break;
2816 		}
2817 		actions_n++;
2818 	}
2819 	if (*encap_idx == -1)
2820 		*encap_idx = actions_n;
2821 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2822 	return actions_n + 1;
2823 }
2824 
2825 /**
2826  * Check meter action from the action list.
2827  *
2828  * @param[in] actions
2829  *   Pointer to the list of actions.
2830  * @param[out] mtr
2831  *   Pointer to the meter exist flag.
2832  *
2833  * @return
2834  *   Total number of actions.
2835  */
2836 static int
2837 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2838 {
2839 	int actions_n = 0;
2840 
2841 	MLX5_ASSERT(mtr);
2842 	*mtr = 0;
2843 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2844 		switch (actions->type) {
2845 		case RTE_FLOW_ACTION_TYPE_METER:
2846 			*mtr = 1;
2847 			break;
2848 		default:
2849 			break;
2850 		}
2851 		actions_n++;
2852 	}
2853 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2854 	return actions_n + 1;
2855 }
2856 
2857 /**
2858  * Check if the flow should be splited due to hairpin.
2859  * The reason for the split is that in current HW we can't
2860  * support encap on Rx, so if a flow have encap we move it
2861  * to Tx.
2862  *
2863  * @param dev
2864  *   Pointer to Ethernet device.
2865  * @param[in] attr
2866  *   Flow rule attributes.
2867  * @param[in] actions
2868  *   Associated actions (list terminated by the END action).
2869  *
2870  * @return
2871  *   > 0 the number of actions and the flow should be split,
2872  *   0 when no split required.
2873  */
2874 static int
2875 flow_check_hairpin_split(struct rte_eth_dev *dev,
2876 			 const struct rte_flow_attr *attr,
2877 			 const struct rte_flow_action actions[])
2878 {
2879 	int queue_action = 0;
2880 	int action_n = 0;
2881 	int encap = 0;
2882 	const struct rte_flow_action_queue *queue;
2883 	const struct rte_flow_action_rss *rss;
2884 	const struct rte_flow_action_raw_encap *raw_encap;
2885 
2886 	if (!attr->ingress)
2887 		return 0;
2888 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2889 		switch (actions->type) {
2890 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2891 			queue = actions->conf;
2892 			if (queue == NULL)
2893 				return 0;
2894 			if (mlx5_rxq_get_type(dev, queue->index) !=
2895 			    MLX5_RXQ_TYPE_HAIRPIN)
2896 				return 0;
2897 			queue_action = 1;
2898 			action_n++;
2899 			break;
2900 		case RTE_FLOW_ACTION_TYPE_RSS:
2901 			rss = actions->conf;
2902 			if (rss == NULL || rss->queue_num == 0)
2903 				return 0;
2904 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2905 			    MLX5_RXQ_TYPE_HAIRPIN)
2906 				return 0;
2907 			queue_action = 1;
2908 			action_n++;
2909 			break;
2910 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2911 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2912 			encap = 1;
2913 			action_n++;
2914 			break;
2915 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2916 			raw_encap = actions->conf;
2917 			if (raw_encap->size >
2918 			    (sizeof(struct rte_flow_item_eth) +
2919 			     sizeof(struct rte_flow_item_ipv4)))
2920 				encap = 1;
2921 			action_n++;
2922 			break;
2923 		default:
2924 			action_n++;
2925 			break;
2926 		}
2927 	}
2928 	if (encap == 1 && queue_action)
2929 		return action_n;
2930 	return 0;
2931 }
2932 
2933 /* Declare flow create/destroy prototype in advance. */
2934 static uint32_t
2935 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2936 		 const struct rte_flow_attr *attr,
2937 		 const struct rte_flow_item items[],
2938 		 const struct rte_flow_action actions[],
2939 		 bool external, struct rte_flow_error *error);
2940 
2941 static void
2942 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2943 		  uint32_t flow_idx);
2944 
2945 /**
2946  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2947  *
2948  * As mark_id is unique, if there's already a registered flow for the mark_id,
2949  * return by increasing the reference counter of the resource. Otherwise, create
2950  * the resource (mcp_res) and flow.
2951  *
2952  * Flow looks like,
2953  *   - If ingress port is ANY and reg_c[1] is mark_id,
2954  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2955  *
2956  * For default flow (zero mark_id), flow is like,
2957  *   - If ingress port is ANY,
2958  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2959  *
2960  * @param dev
2961  *   Pointer to Ethernet device.
2962  * @param mark_id
2963  *   ID of MARK action, zero means default flow for META.
2964  * @param[out] error
2965  *   Perform verbose error reporting if not NULL.
2966  *
2967  * @return
2968  *   Associated resource on success, NULL otherwise and rte_errno is set.
2969  */
2970 static struct mlx5_flow_mreg_copy_resource *
2971 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2972 			  struct rte_flow_error *error)
2973 {
2974 	struct mlx5_priv *priv = dev->data->dev_private;
2975 	struct rte_flow_attr attr = {
2976 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2977 		.ingress = 1,
2978 	};
2979 	struct mlx5_rte_flow_item_tag tag_spec = {
2980 		.data = mark_id,
2981 	};
2982 	struct rte_flow_item items[] = {
2983 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2984 	};
2985 	struct rte_flow_action_mark ftag = {
2986 		.id = mark_id,
2987 	};
2988 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2989 		.dst = REG_B,
2990 		.src = 0,
2991 	};
2992 	struct rte_flow_action_jump jump = {
2993 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2994 	};
2995 	struct rte_flow_action actions[] = {
2996 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2997 	};
2998 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2999 	uint32_t idx = 0;
3000 	int ret;
3001 
3002 	/* Fill the register fileds in the flow. */
3003 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3004 	if (ret < 0)
3005 		return NULL;
3006 	tag_spec.id = ret;
3007 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3008 	if (ret < 0)
3009 		return NULL;
3010 	cp_mreg.src = ret;
3011 	/* Check if already registered. */
3012 	MLX5_ASSERT(priv->mreg_cp_tbl);
3013 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3014 	if (mcp_res) {
3015 		/* For non-default rule. */
3016 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3017 			mcp_res->refcnt++;
3018 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3019 			    mcp_res->refcnt == 1);
3020 		return mcp_res;
3021 	}
3022 	/* Provide the full width of FLAG specific value. */
3023 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3024 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3025 	/* Build a new flow. */
3026 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3027 		items[0] = (struct rte_flow_item){
3028 			.type = (enum rte_flow_item_type)
3029 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3030 			.spec = &tag_spec,
3031 		};
3032 		items[1] = (struct rte_flow_item){
3033 			.type = RTE_FLOW_ITEM_TYPE_END,
3034 		};
3035 		actions[0] = (struct rte_flow_action){
3036 			.type = (enum rte_flow_action_type)
3037 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3038 			.conf = &ftag,
3039 		};
3040 		actions[1] = (struct rte_flow_action){
3041 			.type = (enum rte_flow_action_type)
3042 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3043 			.conf = &cp_mreg,
3044 		};
3045 		actions[2] = (struct rte_flow_action){
3046 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3047 			.conf = &jump,
3048 		};
3049 		actions[3] = (struct rte_flow_action){
3050 			.type = RTE_FLOW_ACTION_TYPE_END,
3051 		};
3052 	} else {
3053 		/* Default rule, wildcard match. */
3054 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3055 		items[0] = (struct rte_flow_item){
3056 			.type = RTE_FLOW_ITEM_TYPE_END,
3057 		};
3058 		actions[0] = (struct rte_flow_action){
3059 			.type = (enum rte_flow_action_type)
3060 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3061 			.conf = &cp_mreg,
3062 		};
3063 		actions[1] = (struct rte_flow_action){
3064 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3065 			.conf = &jump,
3066 		};
3067 		actions[2] = (struct rte_flow_action){
3068 			.type = RTE_FLOW_ACTION_TYPE_END,
3069 		};
3070 	}
3071 	/* Build a new entry. */
3072 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3073 	if (!mcp_res) {
3074 		rte_errno = ENOMEM;
3075 		return NULL;
3076 	}
3077 	mcp_res->idx = idx;
3078 	/*
3079 	 * The copy Flows are not included in any list. There
3080 	 * ones are referenced from other Flows and can not
3081 	 * be applied, removed, deleted in ardbitrary order
3082 	 * by list traversing.
3083 	 */
3084 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3085 					 actions, false, error);
3086 	if (!mcp_res->rix_flow)
3087 		goto error;
3088 	mcp_res->refcnt++;
3089 	mcp_res->hlist_ent.key = mark_id;
3090 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3091 				&mcp_res->hlist_ent);
3092 	MLX5_ASSERT(!ret);
3093 	if (ret)
3094 		goto error;
3095 	return mcp_res;
3096 error:
3097 	if (mcp_res->rix_flow)
3098 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3099 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3100 	return NULL;
3101 }
3102 
3103 /**
3104  * Release flow in RX_CP_TBL.
3105  *
3106  * @param dev
3107  *   Pointer to Ethernet device.
3108  * @flow
3109  *   Parent flow for wich copying is provided.
3110  */
3111 static void
3112 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3113 			  struct rte_flow *flow)
3114 {
3115 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3116 	struct mlx5_priv *priv = dev->data->dev_private;
3117 
3118 	if (!flow->rix_mreg_copy)
3119 		return;
3120 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3121 				 flow->rix_mreg_copy);
3122 	if (!mcp_res || !priv->mreg_cp_tbl)
3123 		return;
3124 	if (flow->copy_applied) {
3125 		MLX5_ASSERT(mcp_res->appcnt);
3126 		flow->copy_applied = 0;
3127 		--mcp_res->appcnt;
3128 		if (!mcp_res->appcnt) {
3129 			struct rte_flow *mcp_flow = mlx5_ipool_get
3130 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3131 					mcp_res->rix_flow);
3132 
3133 			if (mcp_flow)
3134 				flow_drv_remove(dev, mcp_flow);
3135 		}
3136 	}
3137 	/*
3138 	 * We do not check availability of metadata registers here,
3139 	 * because copy resources are not allocated in this case.
3140 	 */
3141 	if (--mcp_res->refcnt)
3142 		return;
3143 	MLX5_ASSERT(mcp_res->rix_flow);
3144 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3145 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3146 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3147 	flow->rix_mreg_copy = 0;
3148 }
3149 
3150 /**
3151  * Start flow in RX_CP_TBL.
3152  *
3153  * @param dev
3154  *   Pointer to Ethernet device.
3155  * @flow
3156  *   Parent flow for wich copying is provided.
3157  *
3158  * @return
3159  *   0 on success, a negative errno value otherwise and rte_errno is set.
3160  */
3161 static int
3162 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3163 			    struct rte_flow *flow)
3164 {
3165 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3166 	struct mlx5_priv *priv = dev->data->dev_private;
3167 	int ret;
3168 
3169 	if (!flow->rix_mreg_copy || flow->copy_applied)
3170 		return 0;
3171 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3172 				 flow->rix_mreg_copy);
3173 	if (!mcp_res)
3174 		return 0;
3175 	if (!mcp_res->appcnt) {
3176 		struct rte_flow *mcp_flow = mlx5_ipool_get
3177 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3178 				mcp_res->rix_flow);
3179 
3180 		if (mcp_flow) {
3181 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3182 			if (ret)
3183 				return ret;
3184 		}
3185 	}
3186 	++mcp_res->appcnt;
3187 	flow->copy_applied = 1;
3188 	return 0;
3189 }
3190 
3191 /**
3192  * Stop flow in RX_CP_TBL.
3193  *
3194  * @param dev
3195  *   Pointer to Ethernet device.
3196  * @flow
3197  *   Parent flow for wich copying is provided.
3198  */
3199 static void
3200 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3201 			   struct rte_flow *flow)
3202 {
3203 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3204 	struct mlx5_priv *priv = dev->data->dev_private;
3205 
3206 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3207 		return;
3208 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3209 				 flow->rix_mreg_copy);
3210 	if (!mcp_res)
3211 		return;
3212 	MLX5_ASSERT(mcp_res->appcnt);
3213 	--mcp_res->appcnt;
3214 	flow->copy_applied = 0;
3215 	if (!mcp_res->appcnt) {
3216 		struct rte_flow *mcp_flow = mlx5_ipool_get
3217 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3218 				mcp_res->rix_flow);
3219 
3220 		if (mcp_flow)
3221 			flow_drv_remove(dev, mcp_flow);
3222 	}
3223 }
3224 
3225 /**
3226  * Remove the default copy action from RX_CP_TBL.
3227  *
3228  * @param dev
3229  *   Pointer to Ethernet device.
3230  */
3231 static void
3232 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3233 {
3234 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3235 	struct mlx5_priv *priv = dev->data->dev_private;
3236 
3237 	/* Check if default flow is registered. */
3238 	if (!priv->mreg_cp_tbl)
3239 		return;
3240 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3241 					    MLX5_DEFAULT_COPY_ID);
3242 	if (!mcp_res)
3243 		return;
3244 	MLX5_ASSERT(mcp_res->rix_flow);
3245 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3246 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3247 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3248 }
3249 
3250 /**
3251  * Add the default copy action in in RX_CP_TBL.
3252  *
3253  * @param dev
3254  *   Pointer to Ethernet device.
3255  * @param[out] error
3256  *   Perform verbose error reporting if not NULL.
3257  *
3258  * @return
3259  *   0 for success, negative value otherwise and rte_errno is set.
3260  */
3261 static int
3262 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3263 				  struct rte_flow_error *error)
3264 {
3265 	struct mlx5_priv *priv = dev->data->dev_private;
3266 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3267 
3268 	/* Check whether extensive metadata feature is engaged. */
3269 	if (!priv->config.dv_flow_en ||
3270 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3271 	    !mlx5_flow_ext_mreg_supported(dev) ||
3272 	    !priv->sh->dv_regc0_mask)
3273 		return 0;
3274 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3275 	if (!mcp_res)
3276 		return -rte_errno;
3277 	return 0;
3278 }
3279 
3280 /**
3281  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3282  *
3283  * All the flow having Q/RSS action should be split by
3284  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3285  * performs the following,
3286  *   - CQE->flow_tag := reg_c[1] (MARK)
3287  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3288  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3289  * but there should be a flow per each MARK ID set by MARK action.
3290  *
3291  * For the aforementioned reason, if there's a MARK action in flow's action
3292  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3293  * the MARK ID to CQE's flow_tag like,
3294  *   - If reg_c[1] is mark_id,
3295  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3296  *
3297  * For SET_META action which stores value in reg_c[0], as the destination is
3298  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3299  * MARK ID means the default flow. The default flow looks like,
3300  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3301  *
3302  * @param dev
3303  *   Pointer to Ethernet device.
3304  * @param flow
3305  *   Pointer to flow structure.
3306  * @param[in] actions
3307  *   Pointer to the list of actions.
3308  * @param[out] error
3309  *   Perform verbose error reporting if not NULL.
3310  *
3311  * @return
3312  *   0 on success, negative value otherwise and rte_errno is set.
3313  */
3314 static int
3315 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3316 			    struct rte_flow *flow,
3317 			    const struct rte_flow_action *actions,
3318 			    struct rte_flow_error *error)
3319 {
3320 	struct mlx5_priv *priv = dev->data->dev_private;
3321 	struct mlx5_dev_config *config = &priv->config;
3322 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3323 	const struct rte_flow_action_mark *mark;
3324 
3325 	/* Check whether extensive metadata feature is engaged. */
3326 	if (!config->dv_flow_en ||
3327 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3328 	    !mlx5_flow_ext_mreg_supported(dev) ||
3329 	    !priv->sh->dv_regc0_mask)
3330 		return 0;
3331 	/* Find MARK action. */
3332 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3333 		switch (actions->type) {
3334 		case RTE_FLOW_ACTION_TYPE_FLAG:
3335 			mcp_res = flow_mreg_add_copy_action
3336 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3337 			if (!mcp_res)
3338 				return -rte_errno;
3339 			flow->rix_mreg_copy = mcp_res->idx;
3340 			if (dev->data->dev_started) {
3341 				mcp_res->appcnt++;
3342 				flow->copy_applied = 1;
3343 			}
3344 			return 0;
3345 		case RTE_FLOW_ACTION_TYPE_MARK:
3346 			mark = (const struct rte_flow_action_mark *)
3347 				actions->conf;
3348 			mcp_res =
3349 				flow_mreg_add_copy_action(dev, mark->id, error);
3350 			if (!mcp_res)
3351 				return -rte_errno;
3352 			flow->rix_mreg_copy = mcp_res->idx;
3353 			if (dev->data->dev_started) {
3354 				mcp_res->appcnt++;
3355 				flow->copy_applied = 1;
3356 			}
3357 			return 0;
3358 		default:
3359 			break;
3360 		}
3361 	}
3362 	return 0;
3363 }
3364 
3365 #define MLX5_MAX_SPLIT_ACTIONS 24
3366 #define MLX5_MAX_SPLIT_ITEMS 24
3367 
3368 /**
3369  * Split the hairpin flow.
3370  * Since HW can't support encap on Rx we move the encap to Tx.
3371  * If the count action is after the encap then we also
3372  * move the count action. in this case the count will also measure
3373  * the outer bytes.
3374  *
3375  * @param dev
3376  *   Pointer to Ethernet device.
3377  * @param[in] actions
3378  *   Associated actions (list terminated by the END action).
3379  * @param[out] actions_rx
3380  *   Rx flow actions.
3381  * @param[out] actions_tx
3382  *   Tx flow actions..
3383  * @param[out] pattern_tx
3384  *   The pattern items for the Tx flow.
3385  * @param[out] flow_id
3386  *   The flow ID connected to this flow.
3387  *
3388  * @return
3389  *   0 on success.
3390  */
3391 static int
3392 flow_hairpin_split(struct rte_eth_dev *dev,
3393 		   const struct rte_flow_action actions[],
3394 		   struct rte_flow_action actions_rx[],
3395 		   struct rte_flow_action actions_tx[],
3396 		   struct rte_flow_item pattern_tx[],
3397 		   uint32_t *flow_id)
3398 {
3399 	struct mlx5_priv *priv = dev->data->dev_private;
3400 	const struct rte_flow_action_raw_encap *raw_encap;
3401 	const struct rte_flow_action_raw_decap *raw_decap;
3402 	struct mlx5_rte_flow_action_set_tag *set_tag;
3403 	struct rte_flow_action *tag_action;
3404 	struct mlx5_rte_flow_item_tag *tag_item;
3405 	struct rte_flow_item *item;
3406 	char *addr;
3407 	int encap = 0;
3408 
3409 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3410 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3411 		switch (actions->type) {
3412 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3413 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3414 			rte_memcpy(actions_tx, actions,
3415 			       sizeof(struct rte_flow_action));
3416 			actions_tx++;
3417 			break;
3418 		case RTE_FLOW_ACTION_TYPE_COUNT:
3419 			if (encap) {
3420 				rte_memcpy(actions_tx, actions,
3421 					   sizeof(struct rte_flow_action));
3422 				actions_tx++;
3423 			} else {
3424 				rte_memcpy(actions_rx, actions,
3425 					   sizeof(struct rte_flow_action));
3426 				actions_rx++;
3427 			}
3428 			break;
3429 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3430 			raw_encap = actions->conf;
3431 			if (raw_encap->size >
3432 			    (sizeof(struct rte_flow_item_eth) +
3433 			     sizeof(struct rte_flow_item_ipv4))) {
3434 				memcpy(actions_tx, actions,
3435 				       sizeof(struct rte_flow_action));
3436 				actions_tx++;
3437 				encap = 1;
3438 			} else {
3439 				rte_memcpy(actions_rx, actions,
3440 					   sizeof(struct rte_flow_action));
3441 				actions_rx++;
3442 			}
3443 			break;
3444 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3445 			raw_decap = actions->conf;
3446 			if (raw_decap->size <
3447 			    (sizeof(struct rte_flow_item_eth) +
3448 			     sizeof(struct rte_flow_item_ipv4))) {
3449 				memcpy(actions_tx, actions,
3450 				       sizeof(struct rte_flow_action));
3451 				actions_tx++;
3452 			} else {
3453 				rte_memcpy(actions_rx, actions,
3454 					   sizeof(struct rte_flow_action));
3455 				actions_rx++;
3456 			}
3457 			break;
3458 		default:
3459 			rte_memcpy(actions_rx, actions,
3460 				   sizeof(struct rte_flow_action));
3461 			actions_rx++;
3462 			break;
3463 		}
3464 	}
3465 	/* Add set meta action and end action for the Rx flow. */
3466 	tag_action = actions_rx;
3467 	tag_action->type = (enum rte_flow_action_type)
3468 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3469 	actions_rx++;
3470 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3471 	actions_rx++;
3472 	set_tag = (void *)actions_rx;
3473 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3474 	MLX5_ASSERT(set_tag->id > REG_NONE);
3475 	set_tag->data = *flow_id;
3476 	tag_action->conf = set_tag;
3477 	/* Create Tx item list. */
3478 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3479 	addr = (void *)&pattern_tx[2];
3480 	item = pattern_tx;
3481 	item->type = (enum rte_flow_item_type)
3482 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3483 	tag_item = (void *)addr;
3484 	tag_item->data = *flow_id;
3485 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3486 	MLX5_ASSERT(set_tag->id > REG_NONE);
3487 	item->spec = tag_item;
3488 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3489 	tag_item = (void *)addr;
3490 	tag_item->data = UINT32_MAX;
3491 	tag_item->id = UINT16_MAX;
3492 	item->mask = tag_item;
3493 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3494 	item->last = NULL;
3495 	item++;
3496 	item->type = RTE_FLOW_ITEM_TYPE_END;
3497 	return 0;
3498 }
3499 
3500 /**
3501  * The last stage of splitting chain, just creates the subflow
3502  * without any modification.
3503  *
3504  * @param[in] dev
3505  *   Pointer to Ethernet device.
3506  * @param[in] flow
3507  *   Parent flow structure pointer.
3508  * @param[in, out] sub_flow
3509  *   Pointer to return the created subflow, may be NULL.
3510  * @param[in] prefix_layers
3511  *   Prefix subflow layers, may be 0.
3512  * @param[in] attr
3513  *   Flow rule attributes.
3514  * @param[in] items
3515  *   Pattern specification (list terminated by the END pattern item).
3516  * @param[in] actions
3517  *   Associated actions (list terminated by the END action).
3518  * @param[in] external
3519  *   This flow rule is created by request external to PMD.
3520  * @param[in] flow_idx
3521  *   This memory pool index to the flow.
3522  * @param[out] error
3523  *   Perform verbose error reporting if not NULL.
3524  * @return
3525  *   0 on success, negative value otherwise
3526  */
3527 static int
3528 flow_create_split_inner(struct rte_eth_dev *dev,
3529 			struct rte_flow *flow,
3530 			struct mlx5_flow **sub_flow,
3531 			uint64_t prefix_layers,
3532 			const struct rte_flow_attr *attr,
3533 			const struct rte_flow_item items[],
3534 			const struct rte_flow_action actions[],
3535 			bool external, uint32_t flow_idx,
3536 			struct rte_flow_error *error)
3537 {
3538 	struct mlx5_flow *dev_flow;
3539 
3540 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3541 		flow_idx, error);
3542 	if (!dev_flow)
3543 		return -rte_errno;
3544 	dev_flow->flow = flow;
3545 	dev_flow->external = external;
3546 	/* Subflow object was created, we must include one in the list. */
3547 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3548 		      dev_flow->handle, next);
3549 	/*
3550 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3551 	 * flow may need some user defined item layer flags.
3552 	 */
3553 	if (prefix_layers)
3554 		dev_flow->handle->layers = prefix_layers;
3555 	if (sub_flow)
3556 		*sub_flow = dev_flow;
3557 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3558 }
3559 
3560 /**
3561  * Split the meter flow.
3562  *
3563  * As meter flow will split to three sub flow, other than meter
3564  * action, the other actions make sense to only meter accepts
3565  * the packet. If it need to be dropped, no other additional
3566  * actions should be take.
3567  *
3568  * One kind of special action which decapsulates the L3 tunnel
3569  * header will be in the prefix sub flow, as not to take the
3570  * L3 tunnel header into account.
3571  *
3572  * @param dev
3573  *   Pointer to Ethernet device.
3574  * @param[in] items
3575  *   Pattern specification (list terminated by the END pattern item).
3576  * @param[out] sfx_items
3577  *   Suffix flow match items (list terminated by the END pattern item).
3578  * @param[in] actions
3579  *   Associated actions (list terminated by the END action).
3580  * @param[out] actions_sfx
3581  *   Suffix flow actions.
3582  * @param[out] actions_pre
3583  *   Prefix flow actions.
3584  * @param[out] pattern_sfx
3585  *   The pattern items for the suffix flow.
3586  * @param[out] tag_sfx
3587  *   Pointer to suffix flow tag.
3588  *
3589  * @return
3590  *   0 on success.
3591  */
3592 static int
3593 flow_meter_split_prep(struct rte_eth_dev *dev,
3594 		 const struct rte_flow_item items[],
3595 		 struct rte_flow_item sfx_items[],
3596 		 const struct rte_flow_action actions[],
3597 		 struct rte_flow_action actions_sfx[],
3598 		 struct rte_flow_action actions_pre[])
3599 {
3600 	struct rte_flow_action *tag_action = NULL;
3601 	struct rte_flow_item *tag_item;
3602 	struct mlx5_rte_flow_action_set_tag *set_tag;
3603 	struct rte_flow_error error;
3604 	const struct rte_flow_action_raw_encap *raw_encap;
3605 	const struct rte_flow_action_raw_decap *raw_decap;
3606 	struct mlx5_rte_flow_item_tag *tag_spec;
3607 	struct mlx5_rte_flow_item_tag *tag_mask;
3608 	uint32_t tag_id;
3609 	bool copy_vlan = false;
3610 
3611 	/* Prepare the actions for prefix and suffix flow. */
3612 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3613 		struct rte_flow_action **action_cur = NULL;
3614 
3615 		switch (actions->type) {
3616 		case RTE_FLOW_ACTION_TYPE_METER:
3617 			/* Add the extra tag action first. */
3618 			tag_action = actions_pre;
3619 			tag_action->type = (enum rte_flow_action_type)
3620 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3621 			actions_pre++;
3622 			action_cur = &actions_pre;
3623 			break;
3624 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3625 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3626 			action_cur = &actions_pre;
3627 			break;
3628 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3629 			raw_encap = actions->conf;
3630 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3631 				action_cur = &actions_pre;
3632 			break;
3633 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3634 			raw_decap = actions->conf;
3635 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3636 				action_cur = &actions_pre;
3637 			break;
3638 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3639 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3640 			copy_vlan = true;
3641 			break;
3642 		default:
3643 			break;
3644 		}
3645 		if (!action_cur)
3646 			action_cur = &actions_sfx;
3647 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3648 		(*action_cur)++;
3649 	}
3650 	/* Add end action to the actions. */
3651 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3652 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3653 	actions_pre++;
3654 	/* Set the tag. */
3655 	set_tag = (void *)actions_pre;
3656 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3657 	/*
3658 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3659 	 */
3660 	tag_id = flow_qrss_get_id(dev);
3661 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3662 	assert(tag_action);
3663 	tag_action->conf = set_tag;
3664 	/* Prepare the suffix subflow items. */
3665 	tag_item = sfx_items++;
3666 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3667 		int item_type = items->type;
3668 
3669 		switch (item_type) {
3670 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3671 			memcpy(sfx_items, items, sizeof(*sfx_items));
3672 			sfx_items++;
3673 			break;
3674 		case RTE_FLOW_ITEM_TYPE_VLAN:
3675 			if (copy_vlan) {
3676 				memcpy(sfx_items, items, sizeof(*sfx_items));
3677 				/*
3678 				 * Convert to internal match item, it is used
3679 				 * for vlan push and set vid.
3680 				 */
3681 				sfx_items->type = (enum rte_flow_item_type)
3682 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3683 				sfx_items++;
3684 			}
3685 			break;
3686 		default:
3687 			break;
3688 		}
3689 	}
3690 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3691 	sfx_items++;
3692 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3693 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3694 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3695 	tag_mask = tag_spec + 1;
3696 	tag_mask->data = 0xffffff00;
3697 	tag_item->type = (enum rte_flow_item_type)
3698 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3699 	tag_item->spec = tag_spec;
3700 	tag_item->last = NULL;
3701 	tag_item->mask = tag_mask;
3702 	return tag_id;
3703 }
3704 
3705 /**
3706  * Split action list having QUEUE/RSS for metadata register copy.
3707  *
3708  * Once Q/RSS action is detected in user's action list, the flow action
3709  * should be split in order to copy metadata registers, which will happen in
3710  * RX_CP_TBL like,
3711  *   - CQE->flow_tag := reg_c[1] (MARK)
3712  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3713  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3714  * This is because the last action of each flow must be a terminal action
3715  * (QUEUE, RSS or DROP).
3716  *
3717  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3718  * stored and kept in the mlx5_flow structure per each sub_flow.
3719  *
3720  * The Q/RSS action is replaced with,
3721  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3722  * And the following JUMP action is added at the end,
3723  *   - JUMP, to RX_CP_TBL.
3724  *
3725  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3726  * flow_create_split_metadata() routine. The flow will look like,
3727  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3728  *
3729  * @param dev
3730  *   Pointer to Ethernet device.
3731  * @param[out] split_actions
3732  *   Pointer to store split actions to jump to CP_TBL.
3733  * @param[in] actions
3734  *   Pointer to the list of original flow actions.
3735  * @param[in] qrss
3736  *   Pointer to the Q/RSS action.
3737  * @param[in] actions_n
3738  *   Number of original actions.
3739  * @param[out] error
3740  *   Perform verbose error reporting if not NULL.
3741  *
3742  * @return
3743  *   non-zero unique flow_id on success, otherwise 0 and
3744  *   error/rte_error are set.
3745  */
3746 static uint32_t
3747 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3748 			  struct rte_flow_action *split_actions,
3749 			  const struct rte_flow_action *actions,
3750 			  const struct rte_flow_action *qrss,
3751 			  int actions_n, struct rte_flow_error *error)
3752 {
3753 	struct mlx5_rte_flow_action_set_tag *set_tag;
3754 	struct rte_flow_action_jump *jump;
3755 	const int qrss_idx = qrss - actions;
3756 	uint32_t flow_id = 0;
3757 	int ret = 0;
3758 
3759 	/*
3760 	 * Given actions will be split
3761 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3762 	 * - Add jump to mreg CP_TBL.
3763 	 * As a result, there will be one more action.
3764 	 */
3765 	++actions_n;
3766 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3767 	set_tag = (void *)(split_actions + actions_n);
3768 	/*
3769 	 * If tag action is not set to void(it means we are not the meter
3770 	 * suffix flow), add the tag action. Since meter suffix flow already
3771 	 * has the tag added.
3772 	 */
3773 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3774 		/*
3775 		 * Allocate the new subflow ID. This one is unique within
3776 		 * device and not shared with representors. Otherwise,
3777 		 * we would have to resolve multi-thread access synch
3778 		 * issue. Each flow on the shared device is appended
3779 		 * with source vport identifier, so the resulting
3780 		 * flows will be unique in the shared (by master and
3781 		 * representors) domain even if they have coinciding
3782 		 * IDs.
3783 		 */
3784 		flow_id = flow_qrss_get_id(dev);
3785 		if (!flow_id)
3786 			return rte_flow_error_set(error, ENOMEM,
3787 						  RTE_FLOW_ERROR_TYPE_ACTION,
3788 						  NULL, "can't allocate id "
3789 						  "for split Q/RSS subflow");
3790 		/* Internal SET_TAG action to set flow ID. */
3791 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3792 			.data = flow_id,
3793 		};
3794 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3795 		if (ret < 0)
3796 			return ret;
3797 		set_tag->id = ret;
3798 		/* Construct new actions array. */
3799 		/* Replace QUEUE/RSS action. */
3800 		split_actions[qrss_idx] = (struct rte_flow_action){
3801 			.type = (enum rte_flow_action_type)
3802 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3803 			.conf = set_tag,
3804 		};
3805 	}
3806 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3807 	jump = (void *)(set_tag + 1);
3808 	*jump = (struct rte_flow_action_jump){
3809 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3810 	};
3811 	split_actions[actions_n - 2] = (struct rte_flow_action){
3812 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3813 		.conf = jump,
3814 	};
3815 	split_actions[actions_n - 1] = (struct rte_flow_action){
3816 		.type = RTE_FLOW_ACTION_TYPE_END,
3817 	};
3818 	return flow_id;
3819 }
3820 
3821 /**
3822  * Extend the given action list for Tx metadata copy.
3823  *
3824  * Copy the given action list to the ext_actions and add flow metadata register
3825  * copy action in order to copy reg_a set by WQE to reg_c[0].
3826  *
3827  * @param[out] ext_actions
3828  *   Pointer to the extended action list.
3829  * @param[in] actions
3830  *   Pointer to the list of actions.
3831  * @param[in] actions_n
3832  *   Number of actions in the list.
3833  * @param[out] error
3834  *   Perform verbose error reporting if not NULL.
3835  * @param[in] encap_idx
3836  *   The encap action inndex.
3837  *
3838  * @return
3839  *   0 on success, negative value otherwise
3840  */
3841 static int
3842 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3843 		       struct rte_flow_action *ext_actions,
3844 		       const struct rte_flow_action *actions,
3845 		       int actions_n, struct rte_flow_error *error,
3846 		       int encap_idx)
3847 {
3848 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3849 		(struct mlx5_flow_action_copy_mreg *)
3850 			(ext_actions + actions_n + 1);
3851 	int ret;
3852 
3853 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3854 	if (ret < 0)
3855 		return ret;
3856 	cp_mreg->dst = ret;
3857 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3858 	if (ret < 0)
3859 		return ret;
3860 	cp_mreg->src = ret;
3861 	if (encap_idx != 0)
3862 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3863 	if (encap_idx == actions_n - 1) {
3864 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3865 			.type = (enum rte_flow_action_type)
3866 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3867 			.conf = cp_mreg,
3868 		};
3869 		ext_actions[actions_n] = (struct rte_flow_action){
3870 			.type = RTE_FLOW_ACTION_TYPE_END,
3871 		};
3872 	} else {
3873 		ext_actions[encap_idx] = (struct rte_flow_action){
3874 			.type = (enum rte_flow_action_type)
3875 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3876 			.conf = cp_mreg,
3877 		};
3878 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3879 				sizeof(*ext_actions) * (actions_n - encap_idx));
3880 	}
3881 	return 0;
3882 }
3883 
3884 /**
3885  * The splitting for metadata feature.
3886  *
3887  * - Q/RSS action on NIC Rx should be split in order to pass by
3888  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3889  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3890  *
3891  * - All the actions on NIC Tx should have a mreg copy action to
3892  *   copy reg_a from WQE to reg_c[0].
3893  *
3894  * @param dev
3895  *   Pointer to Ethernet device.
3896  * @param[in] flow
3897  *   Parent flow structure pointer.
3898  * @param[in] prefix_layers
3899  *   Prefix flow layer flags.
3900  * @param[in] attr
3901  *   Flow rule attributes.
3902  * @param[in] items
3903  *   Pattern specification (list terminated by the END pattern item).
3904  * @param[in] actions
3905  *   Associated actions (list terminated by the END action).
3906  * @param[in] external
3907  *   This flow rule is created by request external to PMD.
3908  * @param[in] flow_idx
3909  *   This memory pool index to the flow.
3910  * @param[out] error
3911  *   Perform verbose error reporting if not NULL.
3912  * @return
3913  *   0 on success, negative value otherwise
3914  */
3915 static int
3916 flow_create_split_metadata(struct rte_eth_dev *dev,
3917 			   struct rte_flow *flow,
3918 			   uint64_t prefix_layers,
3919 			   const struct rte_flow_attr *attr,
3920 			   const struct rte_flow_item items[],
3921 			   const struct rte_flow_action actions[],
3922 			   bool external, uint32_t flow_idx,
3923 			   struct rte_flow_error *error)
3924 {
3925 	struct mlx5_priv *priv = dev->data->dev_private;
3926 	struct mlx5_dev_config *config = &priv->config;
3927 	const struct rte_flow_action *qrss = NULL;
3928 	struct rte_flow_action *ext_actions = NULL;
3929 	struct mlx5_flow *dev_flow = NULL;
3930 	uint32_t qrss_id = 0;
3931 	int mtr_sfx = 0;
3932 	size_t act_size;
3933 	int actions_n;
3934 	int encap_idx;
3935 	int ret;
3936 
3937 	/* Check whether extensive metadata feature is engaged. */
3938 	if (!config->dv_flow_en ||
3939 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3940 	    !mlx5_flow_ext_mreg_supported(dev))
3941 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3942 					       attr, items, actions, external,
3943 					       flow_idx, error);
3944 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3945 							   &encap_idx);
3946 	if (qrss) {
3947 		/* Exclude hairpin flows from splitting. */
3948 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3949 			const struct rte_flow_action_queue *queue;
3950 
3951 			queue = qrss->conf;
3952 			if (mlx5_rxq_get_type(dev, queue->index) ==
3953 			    MLX5_RXQ_TYPE_HAIRPIN)
3954 				qrss = NULL;
3955 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3956 			const struct rte_flow_action_rss *rss;
3957 
3958 			rss = qrss->conf;
3959 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3960 			    MLX5_RXQ_TYPE_HAIRPIN)
3961 				qrss = NULL;
3962 		}
3963 	}
3964 	if (qrss) {
3965 		/* Check if it is in meter suffix table. */
3966 		mtr_sfx = attr->group == (attr->transfer ?
3967 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3968 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3969 		/*
3970 		 * Q/RSS action on NIC Rx should be split in order to pass by
3971 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3972 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3973 		 */
3974 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3975 			   sizeof(struct rte_flow_action_set_tag) +
3976 			   sizeof(struct rte_flow_action_jump);
3977 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3978 		if (!ext_actions)
3979 			return rte_flow_error_set(error, ENOMEM,
3980 						  RTE_FLOW_ERROR_TYPE_ACTION,
3981 						  NULL, "no memory to split "
3982 						  "metadata flow");
3983 		/*
3984 		 * If we are the suffix flow of meter, tag already exist.
3985 		 * Set the tag action to void.
3986 		 */
3987 		if (mtr_sfx)
3988 			ext_actions[qrss - actions].type =
3989 						RTE_FLOW_ACTION_TYPE_VOID;
3990 		else
3991 			ext_actions[qrss - actions].type =
3992 						(enum rte_flow_action_type)
3993 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3994 		/*
3995 		 * Create the new actions list with removed Q/RSS action
3996 		 * and appended set tag and jump to register copy table
3997 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3998 		 * in advance, because it is needed for set tag action.
3999 		 */
4000 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4001 						    qrss, actions_n, error);
4002 		if (!mtr_sfx && !qrss_id) {
4003 			ret = -rte_errno;
4004 			goto exit;
4005 		}
4006 	} else if (attr->egress && !attr->transfer) {
4007 		/*
4008 		 * All the actions on NIC Tx should have a metadata register
4009 		 * copy action to copy reg_a from WQE to reg_c[meta]
4010 		 */
4011 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4012 			   sizeof(struct mlx5_flow_action_copy_mreg);
4013 		ext_actions = rte_zmalloc(__func__, act_size, 0);
4014 		if (!ext_actions)
4015 			return rte_flow_error_set(error, ENOMEM,
4016 						  RTE_FLOW_ERROR_TYPE_ACTION,
4017 						  NULL, "no memory to split "
4018 						  "metadata flow");
4019 		/* Create the action list appended with copy register. */
4020 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4021 					     actions_n, error, encap_idx);
4022 		if (ret < 0)
4023 			goto exit;
4024 	}
4025 	/* Add the unmodified original or prefix subflow. */
4026 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4027 				      items, ext_actions ? ext_actions :
4028 				      actions, external, flow_idx, error);
4029 	if (ret < 0)
4030 		goto exit;
4031 	MLX5_ASSERT(dev_flow);
4032 	if (qrss) {
4033 		const struct rte_flow_attr q_attr = {
4034 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4035 			.ingress = 1,
4036 		};
4037 		/* Internal PMD action to set register. */
4038 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4039 			.data = qrss_id,
4040 			.id = 0,
4041 		};
4042 		struct rte_flow_item q_items[] = {
4043 			{
4044 				.type = (enum rte_flow_item_type)
4045 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4046 				.spec = &q_tag_spec,
4047 				.last = NULL,
4048 				.mask = NULL,
4049 			},
4050 			{
4051 				.type = RTE_FLOW_ITEM_TYPE_END,
4052 			},
4053 		};
4054 		struct rte_flow_action q_actions[] = {
4055 			{
4056 				.type = qrss->type,
4057 				.conf = qrss->conf,
4058 			},
4059 			{
4060 				.type = RTE_FLOW_ACTION_TYPE_END,
4061 			},
4062 		};
4063 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4064 
4065 		/*
4066 		 * Configure the tag item only if there is no meter subflow.
4067 		 * Since tag is already marked in the meter suffix subflow
4068 		 * we can just use the meter suffix items as is.
4069 		 */
4070 		if (qrss_id) {
4071 			/* Not meter subflow. */
4072 			MLX5_ASSERT(!mtr_sfx);
4073 			/*
4074 			 * Put unique id in prefix flow due to it is destroyed
4075 			 * after suffix flow and id will be freed after there
4076 			 * is no actual flows with this id and identifier
4077 			 * reallocation becomes possible (for example, for
4078 			 * other flows in other threads).
4079 			 */
4080 			dev_flow->handle->split_flow_id = qrss_id;
4081 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4082 						   error);
4083 			if (ret < 0)
4084 				goto exit;
4085 			q_tag_spec.id = ret;
4086 		}
4087 		dev_flow = NULL;
4088 		/* Add suffix subflow to execute Q/RSS. */
4089 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4090 					      &q_attr, mtr_sfx ? items :
4091 					      q_items, q_actions,
4092 					      external, flow_idx, error);
4093 		if (ret < 0)
4094 			goto exit;
4095 		/* qrss ID should be freed if failed. */
4096 		qrss_id = 0;
4097 		MLX5_ASSERT(dev_flow);
4098 	}
4099 
4100 exit:
4101 	/*
4102 	 * We do not destroy the partially created sub_flows in case of error.
4103 	 * These ones are included into parent flow list and will be destroyed
4104 	 * by flow_drv_destroy.
4105 	 */
4106 	flow_qrss_free_id(dev, qrss_id);
4107 	rte_free(ext_actions);
4108 	return ret;
4109 }
4110 
4111 /**
4112  * The splitting for meter feature.
4113  *
4114  * - The meter flow will be split to two flows as prefix and
4115  *   suffix flow. The packets make sense only it pass the prefix
4116  *   meter action.
4117  *
4118  * - Reg_C_5 is used for the packet to match betweend prefix and
4119  *   suffix flow.
4120  *
4121  * @param dev
4122  *   Pointer to Ethernet device.
4123  * @param[in] flow
4124  *   Parent flow structure pointer.
4125  * @param[in] attr
4126  *   Flow rule attributes.
4127  * @param[in] items
4128  *   Pattern specification (list terminated by the END pattern item).
4129  * @param[in] actions
4130  *   Associated actions (list terminated by the END action).
4131  * @param[in] external
4132  *   This flow rule is created by request external to PMD.
4133  * @param[in] flow_idx
4134  *   This memory pool index to the flow.
4135  * @param[out] error
4136  *   Perform verbose error reporting if not NULL.
4137  * @return
4138  *   0 on success, negative value otherwise
4139  */
4140 static int
4141 flow_create_split_meter(struct rte_eth_dev *dev,
4142 			   struct rte_flow *flow,
4143 			   const struct rte_flow_attr *attr,
4144 			   const struct rte_flow_item items[],
4145 			   const struct rte_flow_action actions[],
4146 			   bool external, uint32_t flow_idx,
4147 			   struct rte_flow_error *error)
4148 {
4149 	struct mlx5_priv *priv = dev->data->dev_private;
4150 	struct rte_flow_action *sfx_actions = NULL;
4151 	struct rte_flow_action *pre_actions = NULL;
4152 	struct rte_flow_item *sfx_items = NULL;
4153 	struct mlx5_flow *dev_flow = NULL;
4154 	struct rte_flow_attr sfx_attr = *attr;
4155 	uint32_t mtr = 0;
4156 	uint32_t mtr_tag_id = 0;
4157 	size_t act_size;
4158 	size_t item_size;
4159 	int actions_n = 0;
4160 	int ret;
4161 
4162 	if (priv->mtr_en)
4163 		actions_n = flow_check_meter_action(actions, &mtr);
4164 	if (mtr) {
4165 		/* The five prefix actions: meter, decap, encap, tag, end. */
4166 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4167 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4168 		/* tag, vlan, port id, end. */
4169 #define METER_SUFFIX_ITEM 4
4170 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4171 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4172 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4173 		if (!sfx_actions)
4174 			return rte_flow_error_set(error, ENOMEM,
4175 						  RTE_FLOW_ERROR_TYPE_ACTION,
4176 						  NULL, "no memory to split "
4177 						  "meter flow");
4178 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4179 			     act_size);
4180 		pre_actions = sfx_actions + actions_n;
4181 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4182 						   actions, sfx_actions,
4183 						   pre_actions);
4184 		if (!mtr_tag_id) {
4185 			ret = -rte_errno;
4186 			goto exit;
4187 		}
4188 		/* Add the prefix subflow. */
4189 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4190 					      items, pre_actions, external,
4191 					      flow_idx, error);
4192 		if (ret) {
4193 			ret = -rte_errno;
4194 			goto exit;
4195 		}
4196 		dev_flow->handle->split_flow_id = mtr_tag_id;
4197 		/* Setting the sfx group atrr. */
4198 		sfx_attr.group = sfx_attr.transfer ?
4199 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4200 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4201 	}
4202 	/* Add the prefix subflow. */
4203 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4204 					 flow_get_prefix_layer_flags(dev_flow) :
4205 					 0, &sfx_attr,
4206 					 sfx_items ? sfx_items : items,
4207 					 sfx_actions ? sfx_actions : actions,
4208 					 external, flow_idx, error);
4209 exit:
4210 	if (sfx_actions)
4211 		rte_free(sfx_actions);
4212 	return ret;
4213 }
4214 
4215 /**
4216  * Split the flow to subflow set. The splitters might be linked
4217  * in the chain, like this:
4218  * flow_create_split_outer() calls:
4219  *   flow_create_split_meter() calls:
4220  *     flow_create_split_metadata(meter_subflow_0) calls:
4221  *       flow_create_split_inner(metadata_subflow_0)
4222  *       flow_create_split_inner(metadata_subflow_1)
4223  *       flow_create_split_inner(metadata_subflow_2)
4224  *     flow_create_split_metadata(meter_subflow_1) calls:
4225  *       flow_create_split_inner(metadata_subflow_0)
4226  *       flow_create_split_inner(metadata_subflow_1)
4227  *       flow_create_split_inner(metadata_subflow_2)
4228  *
4229  * This provide flexible way to add new levels of flow splitting.
4230  * The all of successfully created subflows are included to the
4231  * parent flow dev_flow list.
4232  *
4233  * @param dev
4234  *   Pointer to Ethernet device.
4235  * @param[in] flow
4236  *   Parent flow structure pointer.
4237  * @param[in] attr
4238  *   Flow rule attributes.
4239  * @param[in] items
4240  *   Pattern specification (list terminated by the END pattern item).
4241  * @param[in] actions
4242  *   Associated actions (list terminated by the END action).
4243  * @param[in] external
4244  *   This flow rule is created by request external to PMD.
4245  * @param[in] flow_idx
4246  *   This memory pool index to the flow.
4247  * @param[out] error
4248  *   Perform verbose error reporting if not NULL.
4249  * @return
4250  *   0 on success, negative value otherwise
4251  */
4252 static int
4253 flow_create_split_outer(struct rte_eth_dev *dev,
4254 			struct rte_flow *flow,
4255 			const struct rte_flow_attr *attr,
4256 			const struct rte_flow_item items[],
4257 			const struct rte_flow_action actions[],
4258 			bool external, uint32_t flow_idx,
4259 			struct rte_flow_error *error)
4260 {
4261 	int ret;
4262 
4263 	ret = flow_create_split_meter(dev, flow, attr, items,
4264 					 actions, external, flow_idx, error);
4265 	MLX5_ASSERT(ret <= 0);
4266 	return ret;
4267 }
4268 
4269 /**
4270  * Create a flow and add it to @p list.
4271  *
4272  * @param dev
4273  *   Pointer to Ethernet device.
4274  * @param list
4275  *   Pointer to a TAILQ flow list. If this parameter NULL,
4276  *   no list insertion occurred, flow is just created,
4277  *   this is caller's responsibility to track the
4278  *   created flow.
4279  * @param[in] attr
4280  *   Flow rule attributes.
4281  * @param[in] items
4282  *   Pattern specification (list terminated by the END pattern item).
4283  * @param[in] actions
4284  *   Associated actions (list terminated by the END action).
4285  * @param[in] external
4286  *   This flow rule is created by request external to PMD.
4287  * @param[out] error
4288  *   Perform verbose error reporting if not NULL.
4289  *
4290  * @return
4291  *   A flow index on success, 0 otherwise and rte_errno is set.
4292  */
4293 static uint32_t
4294 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4295 		 const struct rte_flow_attr *attr,
4296 		 const struct rte_flow_item items[],
4297 		 const struct rte_flow_action actions[],
4298 		 bool external, struct rte_flow_error *error)
4299 {
4300 	struct mlx5_priv *priv = dev->data->dev_private;
4301 	struct rte_flow *flow = NULL;
4302 	struct mlx5_flow *dev_flow;
4303 	const struct rte_flow_action_rss *rss;
4304 	union {
4305 		struct rte_flow_expand_rss buf;
4306 		uint8_t buffer[2048];
4307 	} expand_buffer;
4308 	union {
4309 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4310 		uint8_t buffer[2048];
4311 	} actions_rx;
4312 	union {
4313 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4314 		uint8_t buffer[2048];
4315 	} actions_hairpin_tx;
4316 	union {
4317 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4318 		uint8_t buffer[2048];
4319 	} items_tx;
4320 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4321 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4322 					      priv->rss_desc)[!!priv->flow_idx];
4323 	const struct rte_flow_action *p_actions_rx = actions;
4324 	uint32_t i;
4325 	uint32_t idx = 0;
4326 	int hairpin_flow;
4327 	uint32_t hairpin_id = 0;
4328 	struct rte_flow_attr attr_tx = { .priority = 0 };
4329 	int ret;
4330 
4331 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4332 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4333 				external, hairpin_flow, error);
4334 	if (ret < 0)
4335 		return 0;
4336 	if (hairpin_flow > 0) {
4337 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4338 			rte_errno = EINVAL;
4339 			return 0;
4340 		}
4341 		flow_hairpin_split(dev, actions, actions_rx.actions,
4342 				   actions_hairpin_tx.actions, items_tx.items,
4343 				   &hairpin_id);
4344 		p_actions_rx = actions_rx.actions;
4345 	}
4346 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4347 	if (!flow) {
4348 		rte_errno = ENOMEM;
4349 		goto error_before_flow;
4350 	}
4351 	flow->drv_type = flow_get_drv_type(dev, attr);
4352 	if (hairpin_id != 0)
4353 		flow->hairpin_flow_id = hairpin_id;
4354 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4355 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4356 	memset(rss_desc, 0, sizeof(*rss_desc));
4357 	rss = flow_get_rss_action(p_actions_rx);
4358 	if (rss) {
4359 		/*
4360 		 * The following information is required by
4361 		 * mlx5_flow_hashfields_adjust() in advance.
4362 		 */
4363 		rss_desc->level = rss->level;
4364 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4365 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4366 	}
4367 	flow->dev_handles = 0;
4368 	if (rss && rss->types) {
4369 		unsigned int graph_root;
4370 
4371 		graph_root = find_graph_root(items, rss->level);
4372 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4373 					  items, rss->types,
4374 					  mlx5_support_expansion,
4375 					  graph_root);
4376 		MLX5_ASSERT(ret > 0 &&
4377 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4378 	} else {
4379 		buf->entries = 1;
4380 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4381 	}
4382 	/*
4383 	 * Record the start index when there is a nested call. All sub-flows
4384 	 * need to be translated before another calling.
4385 	 * No need to use ping-pong buffer to save memory here.
4386 	 */
4387 	if (priv->flow_idx) {
4388 		MLX5_ASSERT(!priv->flow_nested_idx);
4389 		priv->flow_nested_idx = priv->flow_idx;
4390 	}
4391 	for (i = 0; i < buf->entries; ++i) {
4392 		/*
4393 		 * The splitter may create multiple dev_flows,
4394 		 * depending on configuration. In the simplest
4395 		 * case it just creates unmodified original flow.
4396 		 */
4397 		ret = flow_create_split_outer(dev, flow, attr,
4398 					      buf->entry[i].pattern,
4399 					      p_actions_rx, external, idx,
4400 					      error);
4401 		if (ret < 0)
4402 			goto error;
4403 	}
4404 	/* Create the tx flow. */
4405 	if (hairpin_flow) {
4406 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4407 		attr_tx.ingress = 0;
4408 		attr_tx.egress = 1;
4409 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4410 					 actions_hairpin_tx.actions,
4411 					 idx, error);
4412 		if (!dev_flow)
4413 			goto error;
4414 		dev_flow->flow = flow;
4415 		dev_flow->external = 0;
4416 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4417 			      dev_flow->handle, next);
4418 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4419 					 items_tx.items,
4420 					 actions_hairpin_tx.actions, error);
4421 		if (ret < 0)
4422 			goto error;
4423 	}
4424 	/*
4425 	 * Update the metadata register copy table. If extensive
4426 	 * metadata feature is enabled and registers are supported
4427 	 * we might create the extra rte_flow for each unique
4428 	 * MARK/FLAG action ID.
4429 	 *
4430 	 * The table is updated for ingress Flows only, because
4431 	 * the egress Flows belong to the different device and
4432 	 * copy table should be updated in peer NIC Rx domain.
4433 	 */
4434 	if (attr->ingress &&
4435 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4436 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4437 		if (ret)
4438 			goto error;
4439 	}
4440 	/*
4441 	 * If the flow is external (from application) OR device is started, then
4442 	 * the flow will be applied immediately.
4443 	 */
4444 	if (external || dev->data->dev_started) {
4445 		ret = flow_drv_apply(dev, flow, error);
4446 		if (ret < 0)
4447 			goto error;
4448 	}
4449 	if (list)
4450 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4451 			     flow, next);
4452 	flow_rxq_flags_set(dev, flow);
4453 	/* Nested flow creation index recovery. */
4454 	priv->flow_idx = priv->flow_nested_idx;
4455 	if (priv->flow_nested_idx)
4456 		priv->flow_nested_idx = 0;
4457 	return idx;
4458 error:
4459 	MLX5_ASSERT(flow);
4460 	ret = rte_errno; /* Save rte_errno before cleanup. */
4461 	flow_mreg_del_copy_action(dev, flow);
4462 	flow_drv_destroy(dev, flow);
4463 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4464 	rte_errno = ret; /* Restore rte_errno. */
4465 error_before_flow:
4466 	ret = rte_errno;
4467 	if (hairpin_id)
4468 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4469 				     hairpin_id);
4470 	rte_errno = ret;
4471 	priv->flow_idx = priv->flow_nested_idx;
4472 	if (priv->flow_nested_idx)
4473 		priv->flow_nested_idx = 0;
4474 	return 0;
4475 }
4476 
4477 /**
4478  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4479  * incoming packets to table 1.
4480  *
4481  * Other flow rules, requested for group n, will be created in
4482  * e-switch table n+1.
4483  * Jump action to e-switch group n will be created to group n+1.
4484  *
4485  * Used when working in switchdev mode, to utilise advantages of table 1
4486  * and above.
4487  *
4488  * @param dev
4489  *   Pointer to Ethernet device.
4490  *
4491  * @return
4492  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4493  */
4494 struct rte_flow *
4495 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4496 {
4497 	const struct rte_flow_attr attr = {
4498 		.group = 0,
4499 		.priority = 0,
4500 		.ingress = 1,
4501 		.egress = 0,
4502 		.transfer = 1,
4503 	};
4504 	const struct rte_flow_item pattern = {
4505 		.type = RTE_FLOW_ITEM_TYPE_END,
4506 	};
4507 	struct rte_flow_action_jump jump = {
4508 		.group = 1,
4509 	};
4510 	const struct rte_flow_action actions[] = {
4511 		{
4512 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4513 			.conf = &jump,
4514 		},
4515 		{
4516 			.type = RTE_FLOW_ACTION_TYPE_END,
4517 		},
4518 	};
4519 	struct mlx5_priv *priv = dev->data->dev_private;
4520 	struct rte_flow_error error;
4521 
4522 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4523 						   &attr, &pattern,
4524 						   actions, false, &error);
4525 }
4526 
4527 /**
4528  * Validate a flow supported by the NIC.
4529  *
4530  * @see rte_flow_validate()
4531  * @see rte_flow_ops
4532  */
4533 int
4534 mlx5_flow_validate(struct rte_eth_dev *dev,
4535 		   const struct rte_flow_attr *attr,
4536 		   const struct rte_flow_item items[],
4537 		   const struct rte_flow_action actions[],
4538 		   struct rte_flow_error *error)
4539 {
4540 	int hairpin_flow;
4541 
4542 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4543 	return flow_drv_validate(dev, attr, items, actions,
4544 				true, hairpin_flow, error);
4545 }
4546 
4547 /**
4548  * Create a flow.
4549  *
4550  * @see rte_flow_create()
4551  * @see rte_flow_ops
4552  */
4553 struct rte_flow *
4554 mlx5_flow_create(struct rte_eth_dev *dev,
4555 		 const struct rte_flow_attr *attr,
4556 		 const struct rte_flow_item items[],
4557 		 const struct rte_flow_action actions[],
4558 		 struct rte_flow_error *error)
4559 {
4560 	struct mlx5_priv *priv = dev->data->dev_private;
4561 
4562 	/*
4563 	 * If the device is not started yet, it is not allowed to created a
4564 	 * flow from application. PMD default flows and traffic control flows
4565 	 * are not affected.
4566 	 */
4567 	if (unlikely(!dev->data->dev_started)) {
4568 		DRV_LOG(DEBUG, "port %u is not started when "
4569 			"inserting a flow", dev->data->port_id);
4570 		rte_flow_error_set(error, ENODEV,
4571 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4572 				   NULL,
4573 				   "port not started");
4574 		return NULL;
4575 	}
4576 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4577 				  attr, items, actions, true, error);
4578 }
4579 
4580 /**
4581  * Destroy a flow in a list.
4582  *
4583  * @param dev
4584  *   Pointer to Ethernet device.
4585  * @param list
4586  *   Pointer to the Indexed flow list. If this parameter NULL,
4587  *   there is no flow removal from the list. Be noted that as
4588  *   flow is add to the indexed list, memory of the indexed
4589  *   list points to maybe changed as flow destroyed.
4590  * @param[in] flow_idx
4591  *   Index of flow to destroy.
4592  */
4593 static void
4594 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4595 		  uint32_t flow_idx)
4596 {
4597 	struct mlx5_priv *priv = dev->data->dev_private;
4598 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4599 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4600 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4601 
4602 	if (!flow)
4603 		return;
4604 	/*
4605 	 * Update RX queue flags only if port is started, otherwise it is
4606 	 * already clean.
4607 	 */
4608 	if (dev->data->dev_started)
4609 		flow_rxq_flags_trim(dev, flow);
4610 	if (flow->hairpin_flow_id)
4611 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4612 				     flow->hairpin_flow_id);
4613 	flow_drv_destroy(dev, flow);
4614 	if (list)
4615 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4616 			     flow_idx, flow, next);
4617 	flow_mreg_del_copy_action(dev, flow);
4618 	if (flow->fdir) {
4619 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4620 			if (priv_fdir_flow->rix_flow == flow_idx)
4621 				break;
4622 		}
4623 		if (priv_fdir_flow) {
4624 			LIST_REMOVE(priv_fdir_flow, next);
4625 			rte_free(priv_fdir_flow->fdir);
4626 			rte_free(priv_fdir_flow);
4627 		}
4628 	}
4629 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4630 }
4631 
4632 /**
4633  * Destroy all flows.
4634  *
4635  * @param dev
4636  *   Pointer to Ethernet device.
4637  * @param list
4638  *   Pointer to the Indexed flow list.
4639  * @param active
4640  *   If flushing is called avtively.
4641  */
4642 void
4643 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4644 {
4645 	uint32_t num_flushed = 0;
4646 
4647 	while (*list) {
4648 		flow_list_destroy(dev, list, *list);
4649 		num_flushed++;
4650 	}
4651 	if (active) {
4652 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4653 			dev->data->port_id, num_flushed);
4654 	}
4655 }
4656 
4657 /**
4658  * Remove all flows.
4659  *
4660  * @param dev
4661  *   Pointer to Ethernet device.
4662  * @param list
4663  *   Pointer to the Indexed flow list.
4664  */
4665 void
4666 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4667 {
4668 	struct mlx5_priv *priv = dev->data->dev_private;
4669 	struct rte_flow *flow = NULL;
4670 	uint32_t idx;
4671 
4672 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4673 		      flow, next) {
4674 		flow_drv_remove(dev, flow);
4675 		flow_mreg_stop_copy_action(dev, flow);
4676 	}
4677 	flow_mreg_del_default_copy_action(dev);
4678 	flow_rxq_flags_clear(dev);
4679 }
4680 
4681 /**
4682  * Add all flows.
4683  *
4684  * @param dev
4685  *   Pointer to Ethernet device.
4686  * @param list
4687  *   Pointer to the Indexed flow list.
4688  *
4689  * @return
4690  *   0 on success, a negative errno value otherwise and rte_errno is set.
4691  */
4692 int
4693 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4694 {
4695 	struct mlx5_priv *priv = dev->data->dev_private;
4696 	struct rte_flow *flow = NULL;
4697 	struct rte_flow_error error;
4698 	uint32_t idx;
4699 	int ret = 0;
4700 
4701 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4702 	ret = flow_mreg_add_default_copy_action(dev, &error);
4703 	if (ret < 0)
4704 		return -rte_errno;
4705 	/* Apply Flows created by application. */
4706 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4707 		      flow, next) {
4708 		ret = flow_mreg_start_copy_action(dev, flow);
4709 		if (ret < 0)
4710 			goto error;
4711 		ret = flow_drv_apply(dev, flow, &error);
4712 		if (ret < 0)
4713 			goto error;
4714 		flow_rxq_flags_set(dev, flow);
4715 	}
4716 	return 0;
4717 error:
4718 	ret = rte_errno; /* Save rte_errno before cleanup. */
4719 	mlx5_flow_stop(dev, list);
4720 	rte_errno = ret; /* Restore rte_errno. */
4721 	return -rte_errno;
4722 }
4723 
4724 /**
4725  * Stop all default actions for flows.
4726  *
4727  * @param dev
4728  *   Pointer to Ethernet device.
4729  */
4730 void
4731 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4732 {
4733 	flow_mreg_del_default_copy_action(dev);
4734 	flow_rxq_flags_clear(dev);
4735 }
4736 
4737 /**
4738  * Start all default actions for flows.
4739  *
4740  * @param dev
4741  *   Pointer to Ethernet device.
4742  * @return
4743  *   0 on success, a negative errno value otherwise and rte_errno is set.
4744  */
4745 int
4746 mlx5_flow_start_default(struct rte_eth_dev *dev)
4747 {
4748 	struct rte_flow_error error;
4749 
4750 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4751 	return flow_mreg_add_default_copy_action(dev, &error);
4752 }
4753 
4754 /**
4755  * Allocate intermediate resources for flow creation.
4756  *
4757  * @param dev
4758  *   Pointer to Ethernet device.
4759  */
4760 void
4761 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4762 {
4763 	struct mlx5_priv *priv = dev->data->dev_private;
4764 
4765 	if (!priv->inter_flows) {
4766 		priv->inter_flows = rte_calloc(__func__, 1,
4767 				    MLX5_NUM_MAX_DEV_FLOWS *
4768 				    sizeof(struct mlx5_flow) +
4769 				    (sizeof(struct mlx5_flow_rss_desc) +
4770 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4771 		if (!priv->inter_flows) {
4772 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4773 			return;
4774 		}
4775 	}
4776 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4777 			 [MLX5_NUM_MAX_DEV_FLOWS];
4778 	/* Reset the index. */
4779 	priv->flow_idx = 0;
4780 	priv->flow_nested_idx = 0;
4781 }
4782 
4783 /**
4784  * Free intermediate resources for flows.
4785  *
4786  * @param dev
4787  *   Pointer to Ethernet device.
4788  */
4789 void
4790 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4791 {
4792 	struct mlx5_priv *priv = dev->data->dev_private;
4793 
4794 	rte_free(priv->inter_flows);
4795 	priv->inter_flows = NULL;
4796 }
4797 
4798 /**
4799  * Verify the flow list is empty
4800  *
4801  * @param dev
4802  *  Pointer to Ethernet device.
4803  *
4804  * @return the number of flows not released.
4805  */
4806 int
4807 mlx5_flow_verify(struct rte_eth_dev *dev)
4808 {
4809 	struct mlx5_priv *priv = dev->data->dev_private;
4810 	struct rte_flow *flow;
4811 	uint32_t idx;
4812 	int ret = 0;
4813 
4814 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4815 		      flow, next) {
4816 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4817 			dev->data->port_id, (void *)flow);
4818 		++ret;
4819 	}
4820 	return ret;
4821 }
4822 
4823 /**
4824  * Enable default hairpin egress flow.
4825  *
4826  * @param dev
4827  *   Pointer to Ethernet device.
4828  * @param queue
4829  *   The queue index.
4830  *
4831  * @return
4832  *   0 on success, a negative errno value otherwise and rte_errno is set.
4833  */
4834 int
4835 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4836 			    uint32_t queue)
4837 {
4838 	struct mlx5_priv *priv = dev->data->dev_private;
4839 	const struct rte_flow_attr attr = {
4840 		.egress = 1,
4841 		.priority = 0,
4842 	};
4843 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4844 		.queue = queue,
4845 	};
4846 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4847 		.queue = UINT32_MAX,
4848 	};
4849 	struct rte_flow_item items[] = {
4850 		{
4851 			.type = (enum rte_flow_item_type)
4852 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4853 			.spec = &queue_spec,
4854 			.last = NULL,
4855 			.mask = &queue_mask,
4856 		},
4857 		{
4858 			.type = RTE_FLOW_ITEM_TYPE_END,
4859 		},
4860 	};
4861 	struct rte_flow_action_jump jump = {
4862 		.group = MLX5_HAIRPIN_TX_TABLE,
4863 	};
4864 	struct rte_flow_action actions[2];
4865 	uint32_t flow_idx;
4866 	struct rte_flow_error error;
4867 
4868 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4869 	actions[0].conf = &jump;
4870 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4871 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4872 				&attr, items, actions, false, &error);
4873 	if (!flow_idx) {
4874 		DRV_LOG(DEBUG,
4875 			"Failed to create ctrl flow: rte_errno(%d),"
4876 			" type(%d), message(%s)",
4877 			rte_errno, error.type,
4878 			error.message ? error.message : " (no stated reason)");
4879 		return -rte_errno;
4880 	}
4881 	return 0;
4882 }
4883 
4884 /**
4885  * Enable a control flow configured from the control plane.
4886  *
4887  * @param dev
4888  *   Pointer to Ethernet device.
4889  * @param eth_spec
4890  *   An Ethernet flow spec to apply.
4891  * @param eth_mask
4892  *   An Ethernet flow mask to apply.
4893  * @param vlan_spec
4894  *   A VLAN flow spec to apply.
4895  * @param vlan_mask
4896  *   A VLAN flow mask to apply.
4897  *
4898  * @return
4899  *   0 on success, a negative errno value otherwise and rte_errno is set.
4900  */
4901 int
4902 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4903 		    struct rte_flow_item_eth *eth_spec,
4904 		    struct rte_flow_item_eth *eth_mask,
4905 		    struct rte_flow_item_vlan *vlan_spec,
4906 		    struct rte_flow_item_vlan *vlan_mask)
4907 {
4908 	struct mlx5_priv *priv = dev->data->dev_private;
4909 	const struct rte_flow_attr attr = {
4910 		.ingress = 1,
4911 		.priority = MLX5_FLOW_PRIO_RSVD,
4912 	};
4913 	struct rte_flow_item items[] = {
4914 		{
4915 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4916 			.spec = eth_spec,
4917 			.last = NULL,
4918 			.mask = eth_mask,
4919 		},
4920 		{
4921 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4922 					      RTE_FLOW_ITEM_TYPE_END,
4923 			.spec = vlan_spec,
4924 			.last = NULL,
4925 			.mask = vlan_mask,
4926 		},
4927 		{
4928 			.type = RTE_FLOW_ITEM_TYPE_END,
4929 		},
4930 	};
4931 	uint16_t queue[priv->reta_idx_n];
4932 	struct rte_flow_action_rss action_rss = {
4933 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4934 		.level = 0,
4935 		.types = priv->rss_conf.rss_hf,
4936 		.key_len = priv->rss_conf.rss_key_len,
4937 		.queue_num = priv->reta_idx_n,
4938 		.key = priv->rss_conf.rss_key,
4939 		.queue = queue,
4940 	};
4941 	struct rte_flow_action actions[] = {
4942 		{
4943 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4944 			.conf = &action_rss,
4945 		},
4946 		{
4947 			.type = RTE_FLOW_ACTION_TYPE_END,
4948 		},
4949 	};
4950 	uint32_t flow_idx;
4951 	struct rte_flow_error error;
4952 	unsigned int i;
4953 
4954 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4955 		return 0;
4956 	}
4957 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4958 		action_rss.types = 0;
4959 	for (i = 0; i != priv->reta_idx_n; ++i)
4960 		queue[i] = (*priv->reta_idx)[i];
4961 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4962 				&attr, items, actions, false, &error);
4963 	if (!flow_idx)
4964 		return -rte_errno;
4965 	return 0;
4966 }
4967 
4968 /**
4969  * Enable a flow control configured from the control plane.
4970  *
4971  * @param dev
4972  *   Pointer to Ethernet device.
4973  * @param eth_spec
4974  *   An Ethernet flow spec to apply.
4975  * @param eth_mask
4976  *   An Ethernet flow mask to apply.
4977  *
4978  * @return
4979  *   0 on success, a negative errno value otherwise and rte_errno is set.
4980  */
4981 int
4982 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4983 	       struct rte_flow_item_eth *eth_spec,
4984 	       struct rte_flow_item_eth *eth_mask)
4985 {
4986 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4987 }
4988 
4989 /**
4990  * Destroy a flow.
4991  *
4992  * @see rte_flow_destroy()
4993  * @see rte_flow_ops
4994  */
4995 int
4996 mlx5_flow_destroy(struct rte_eth_dev *dev,
4997 		  struct rte_flow *flow,
4998 		  struct rte_flow_error *error __rte_unused)
4999 {
5000 	struct mlx5_priv *priv = dev->data->dev_private;
5001 
5002 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5003 	return 0;
5004 }
5005 
5006 /**
5007  * Destroy all flows.
5008  *
5009  * @see rte_flow_flush()
5010  * @see rte_flow_ops
5011  */
5012 int
5013 mlx5_flow_flush(struct rte_eth_dev *dev,
5014 		struct rte_flow_error *error __rte_unused)
5015 {
5016 	struct mlx5_priv *priv = dev->data->dev_private;
5017 
5018 	mlx5_flow_list_flush(dev, &priv->flows, false);
5019 	return 0;
5020 }
5021 
5022 /**
5023  * Isolated mode.
5024  *
5025  * @see rte_flow_isolate()
5026  * @see rte_flow_ops
5027  */
5028 int
5029 mlx5_flow_isolate(struct rte_eth_dev *dev,
5030 		  int enable,
5031 		  struct rte_flow_error *error)
5032 {
5033 	struct mlx5_priv *priv = dev->data->dev_private;
5034 
5035 	if (dev->data->dev_started) {
5036 		rte_flow_error_set(error, EBUSY,
5037 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5038 				   NULL,
5039 				   "port must be stopped first");
5040 		return -rte_errno;
5041 	}
5042 	priv->isolated = !!enable;
5043 	if (enable)
5044 		dev->dev_ops = &mlx5_dev_ops_isolate;
5045 	else
5046 		dev->dev_ops = &mlx5_dev_ops;
5047 	return 0;
5048 }
5049 
5050 /**
5051  * Query a flow.
5052  *
5053  * @see rte_flow_query()
5054  * @see rte_flow_ops
5055  */
5056 static int
5057 flow_drv_query(struct rte_eth_dev *dev,
5058 	       uint32_t flow_idx,
5059 	       const struct rte_flow_action *actions,
5060 	       void *data,
5061 	       struct rte_flow_error *error)
5062 {
5063 	struct mlx5_priv *priv = dev->data->dev_private;
5064 	const struct mlx5_flow_driver_ops *fops;
5065 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5066 					       [MLX5_IPOOL_RTE_FLOW],
5067 					       flow_idx);
5068 	enum mlx5_flow_drv_type ftype;
5069 
5070 	if (!flow) {
5071 		return rte_flow_error_set(error, ENOENT,
5072 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5073 			  NULL,
5074 			  "invalid flow handle");
5075 	}
5076 	ftype = flow->drv_type;
5077 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5078 	fops = flow_get_drv_ops(ftype);
5079 
5080 	return fops->query(dev, flow, actions, data, error);
5081 }
5082 
5083 /**
5084  * Query a flow.
5085  *
5086  * @see rte_flow_query()
5087  * @see rte_flow_ops
5088  */
5089 int
5090 mlx5_flow_query(struct rte_eth_dev *dev,
5091 		struct rte_flow *flow,
5092 		const struct rte_flow_action *actions,
5093 		void *data,
5094 		struct rte_flow_error *error)
5095 {
5096 	int ret;
5097 
5098 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5099 			     error);
5100 	if (ret < 0)
5101 		return ret;
5102 	return 0;
5103 }
5104 
5105 /**
5106  * Convert a flow director filter to a generic flow.
5107  *
5108  * @param dev
5109  *   Pointer to Ethernet device.
5110  * @param fdir_filter
5111  *   Flow director filter to add.
5112  * @param attributes
5113  *   Generic flow parameters structure.
5114  *
5115  * @return
5116  *   0 on success, a negative errno value otherwise and rte_errno is set.
5117  */
5118 static int
5119 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5120 			 const struct rte_eth_fdir_filter *fdir_filter,
5121 			 struct mlx5_fdir *attributes)
5122 {
5123 	struct mlx5_priv *priv = dev->data->dev_private;
5124 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5125 	const struct rte_eth_fdir_masks *mask =
5126 		&dev->data->dev_conf.fdir_conf.mask;
5127 
5128 	/* Validate queue number. */
5129 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5130 		DRV_LOG(ERR, "port %u invalid queue number %d",
5131 			dev->data->port_id, fdir_filter->action.rx_queue);
5132 		rte_errno = EINVAL;
5133 		return -rte_errno;
5134 	}
5135 	attributes->attr.ingress = 1;
5136 	attributes->items[0] = (struct rte_flow_item) {
5137 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5138 		.spec = &attributes->l2,
5139 		.mask = &attributes->l2_mask,
5140 	};
5141 	switch (fdir_filter->action.behavior) {
5142 	case RTE_ETH_FDIR_ACCEPT:
5143 		attributes->actions[0] = (struct rte_flow_action){
5144 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5145 			.conf = &attributes->queue,
5146 		};
5147 		break;
5148 	case RTE_ETH_FDIR_REJECT:
5149 		attributes->actions[0] = (struct rte_flow_action){
5150 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5151 		};
5152 		break;
5153 	default:
5154 		DRV_LOG(ERR, "port %u invalid behavior %d",
5155 			dev->data->port_id,
5156 			fdir_filter->action.behavior);
5157 		rte_errno = ENOTSUP;
5158 		return -rte_errno;
5159 	}
5160 	attributes->queue.index = fdir_filter->action.rx_queue;
5161 	/* Handle L3. */
5162 	switch (fdir_filter->input.flow_type) {
5163 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5164 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5165 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5166 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5167 			.src_addr = input->flow.ip4_flow.src_ip,
5168 			.dst_addr = input->flow.ip4_flow.dst_ip,
5169 			.time_to_live = input->flow.ip4_flow.ttl,
5170 			.type_of_service = input->flow.ip4_flow.tos,
5171 		};
5172 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5173 			.src_addr = mask->ipv4_mask.src_ip,
5174 			.dst_addr = mask->ipv4_mask.dst_ip,
5175 			.time_to_live = mask->ipv4_mask.ttl,
5176 			.type_of_service = mask->ipv4_mask.tos,
5177 			.next_proto_id = mask->ipv4_mask.proto,
5178 		};
5179 		attributes->items[1] = (struct rte_flow_item){
5180 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5181 			.spec = &attributes->l3,
5182 			.mask = &attributes->l3_mask,
5183 		};
5184 		break;
5185 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5186 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5187 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5188 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5189 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5190 			.proto = input->flow.ipv6_flow.proto,
5191 		};
5192 
5193 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5194 		       input->flow.ipv6_flow.src_ip,
5195 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5196 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5197 		       input->flow.ipv6_flow.dst_ip,
5198 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5199 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5200 		       mask->ipv6_mask.src_ip,
5201 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5202 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5203 		       mask->ipv6_mask.dst_ip,
5204 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5205 		attributes->items[1] = (struct rte_flow_item){
5206 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5207 			.spec = &attributes->l3,
5208 			.mask = &attributes->l3_mask,
5209 		};
5210 		break;
5211 	default:
5212 		DRV_LOG(ERR, "port %u invalid flow type%d",
5213 			dev->data->port_id, fdir_filter->input.flow_type);
5214 		rte_errno = ENOTSUP;
5215 		return -rte_errno;
5216 	}
5217 	/* Handle L4. */
5218 	switch (fdir_filter->input.flow_type) {
5219 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5220 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5221 			.src_port = input->flow.udp4_flow.src_port,
5222 			.dst_port = input->flow.udp4_flow.dst_port,
5223 		};
5224 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5225 			.src_port = mask->src_port_mask,
5226 			.dst_port = mask->dst_port_mask,
5227 		};
5228 		attributes->items[2] = (struct rte_flow_item){
5229 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5230 			.spec = &attributes->l4,
5231 			.mask = &attributes->l4_mask,
5232 		};
5233 		break;
5234 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5235 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5236 			.src_port = input->flow.tcp4_flow.src_port,
5237 			.dst_port = input->flow.tcp4_flow.dst_port,
5238 		};
5239 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5240 			.src_port = mask->src_port_mask,
5241 			.dst_port = mask->dst_port_mask,
5242 		};
5243 		attributes->items[2] = (struct rte_flow_item){
5244 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5245 			.spec = &attributes->l4,
5246 			.mask = &attributes->l4_mask,
5247 		};
5248 		break;
5249 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5250 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5251 			.src_port = input->flow.udp6_flow.src_port,
5252 			.dst_port = input->flow.udp6_flow.dst_port,
5253 		};
5254 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5255 			.src_port = mask->src_port_mask,
5256 			.dst_port = mask->dst_port_mask,
5257 		};
5258 		attributes->items[2] = (struct rte_flow_item){
5259 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5260 			.spec = &attributes->l4,
5261 			.mask = &attributes->l4_mask,
5262 		};
5263 		break;
5264 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5265 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5266 			.src_port = input->flow.tcp6_flow.src_port,
5267 			.dst_port = input->flow.tcp6_flow.dst_port,
5268 		};
5269 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5270 			.src_port = mask->src_port_mask,
5271 			.dst_port = mask->dst_port_mask,
5272 		};
5273 		attributes->items[2] = (struct rte_flow_item){
5274 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5275 			.spec = &attributes->l4,
5276 			.mask = &attributes->l4_mask,
5277 		};
5278 		break;
5279 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5280 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5281 		break;
5282 	default:
5283 		DRV_LOG(ERR, "port %u invalid flow type%d",
5284 			dev->data->port_id, fdir_filter->input.flow_type);
5285 		rte_errno = ENOTSUP;
5286 		return -rte_errno;
5287 	}
5288 	return 0;
5289 }
5290 
5291 #define FLOW_FDIR_CMP(f1, f2, fld) \
5292 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5293 
5294 /**
5295  * Compare two FDIR flows. If items and actions are identical, the two flows are
5296  * regarded as same.
5297  *
5298  * @param dev
5299  *   Pointer to Ethernet device.
5300  * @param f1
5301  *   FDIR flow to compare.
5302  * @param f2
5303  *   FDIR flow to compare.
5304  *
5305  * @return
5306  *   Zero on match, 1 otherwise.
5307  */
5308 static int
5309 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5310 {
5311 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5312 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5313 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5314 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5315 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5316 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5317 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5318 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5319 		return 1;
5320 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5321 	    FLOW_FDIR_CMP(f1, f2, queue))
5322 		return 1;
5323 	return 0;
5324 }
5325 
5326 /**
5327  * Search device flow list to find out a matched FDIR flow.
5328  *
5329  * @param dev
5330  *   Pointer to Ethernet device.
5331  * @param fdir_flow
5332  *   FDIR flow to lookup.
5333  *
5334  * @return
5335  *   Index of flow if found, 0 otherwise.
5336  */
5337 static uint32_t
5338 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5339 {
5340 	struct mlx5_priv *priv = dev->data->dev_private;
5341 	uint32_t flow_idx = 0;
5342 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5343 
5344 	MLX5_ASSERT(fdir_flow);
5345 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5346 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5347 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5348 				dev->data->port_id, flow_idx);
5349 			flow_idx = priv_fdir_flow->rix_flow;
5350 			break;
5351 		}
5352 	}
5353 	return flow_idx;
5354 }
5355 
5356 /**
5357  * Add new flow director filter and store it in list.
5358  *
5359  * @param dev
5360  *   Pointer to Ethernet device.
5361  * @param fdir_filter
5362  *   Flow director filter to add.
5363  *
5364  * @return
5365  *   0 on success, a negative errno value otherwise and rte_errno is set.
5366  */
5367 static int
5368 flow_fdir_filter_add(struct rte_eth_dev *dev,
5369 		     const struct rte_eth_fdir_filter *fdir_filter)
5370 {
5371 	struct mlx5_priv *priv = dev->data->dev_private;
5372 	struct mlx5_fdir *fdir_flow;
5373 	struct rte_flow *flow;
5374 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5375 	uint32_t flow_idx;
5376 	int ret;
5377 
5378 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5379 	if (!fdir_flow) {
5380 		rte_errno = ENOMEM;
5381 		return -rte_errno;
5382 	}
5383 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5384 	if (ret)
5385 		goto error;
5386 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5387 	if (flow_idx) {
5388 		rte_errno = EEXIST;
5389 		goto error;
5390 	}
5391 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5392 				     0);
5393 	if (!priv_fdir_flow) {
5394 		rte_errno = ENOMEM;
5395 		goto error;
5396 	}
5397 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5398 				    fdir_flow->items, fdir_flow->actions, true,
5399 				    NULL);
5400 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5401 	if (!flow)
5402 		goto error;
5403 	flow->fdir = 1;
5404 	priv_fdir_flow->fdir = fdir_flow;
5405 	priv_fdir_flow->rix_flow = flow_idx;
5406 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5407 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5408 		dev->data->port_id, (void *)flow);
5409 	return 0;
5410 error:
5411 	rte_free(priv_fdir_flow);
5412 	rte_free(fdir_flow);
5413 	return -rte_errno;
5414 }
5415 
5416 /**
5417  * Delete specific filter.
5418  *
5419  * @param dev
5420  *   Pointer to Ethernet device.
5421  * @param fdir_filter
5422  *   Filter to be deleted.
5423  *
5424  * @return
5425  *   0 on success, a negative errno value otherwise and rte_errno is set.
5426  */
5427 static int
5428 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5429 			const struct rte_eth_fdir_filter *fdir_filter)
5430 {
5431 	struct mlx5_priv *priv = dev->data->dev_private;
5432 	uint32_t flow_idx;
5433 	struct mlx5_fdir fdir_flow = {
5434 		.attr.group = 0,
5435 	};
5436 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5437 	int ret;
5438 
5439 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5440 	if (ret)
5441 		return -rte_errno;
5442 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5443 		/* Find the fdir in priv list */
5444 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5445 			break;
5446 	}
5447 	if (!priv_fdir_flow)
5448 		return 0;
5449 	LIST_REMOVE(priv_fdir_flow, next);
5450 	flow_idx = priv_fdir_flow->rix_flow;
5451 	flow_list_destroy(dev, &priv->flows, flow_idx);
5452 	rte_free(priv_fdir_flow->fdir);
5453 	rte_free(priv_fdir_flow);
5454 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5455 		dev->data->port_id, flow_idx);
5456 	return 0;
5457 }
5458 
5459 /**
5460  * Update queue for specific filter.
5461  *
5462  * @param dev
5463  *   Pointer to Ethernet device.
5464  * @param fdir_filter
5465  *   Filter to be updated.
5466  *
5467  * @return
5468  *   0 on success, a negative errno value otherwise and rte_errno is set.
5469  */
5470 static int
5471 flow_fdir_filter_update(struct rte_eth_dev *dev,
5472 			const struct rte_eth_fdir_filter *fdir_filter)
5473 {
5474 	int ret;
5475 
5476 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5477 	if (ret)
5478 		return ret;
5479 	return flow_fdir_filter_add(dev, fdir_filter);
5480 }
5481 
5482 /**
5483  * Flush all filters.
5484  *
5485  * @param dev
5486  *   Pointer to Ethernet device.
5487  */
5488 static void
5489 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5490 {
5491 	struct mlx5_priv *priv = dev->data->dev_private;
5492 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5493 
5494 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5495 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5496 		LIST_REMOVE(priv_fdir_flow, next);
5497 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5498 		rte_free(priv_fdir_flow->fdir);
5499 		rte_free(priv_fdir_flow);
5500 	}
5501 }
5502 
5503 /**
5504  * Get flow director information.
5505  *
5506  * @param dev
5507  *   Pointer to Ethernet device.
5508  * @param[out] fdir_info
5509  *   Resulting flow director information.
5510  */
5511 static void
5512 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5513 {
5514 	struct rte_eth_fdir_masks *mask =
5515 		&dev->data->dev_conf.fdir_conf.mask;
5516 
5517 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5518 	fdir_info->guarant_spc = 0;
5519 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5520 	fdir_info->max_flexpayload = 0;
5521 	fdir_info->flow_types_mask[0] = 0;
5522 	fdir_info->flex_payload_unit = 0;
5523 	fdir_info->max_flex_payload_segment_num = 0;
5524 	fdir_info->flex_payload_limit = 0;
5525 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5526 }
5527 
5528 /**
5529  * Deal with flow director operations.
5530  *
5531  * @param dev
5532  *   Pointer to Ethernet device.
5533  * @param filter_op
5534  *   Operation to perform.
5535  * @param arg
5536  *   Pointer to operation-specific structure.
5537  *
5538  * @return
5539  *   0 on success, a negative errno value otherwise and rte_errno is set.
5540  */
5541 static int
5542 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5543 		    void *arg)
5544 {
5545 	enum rte_fdir_mode fdir_mode =
5546 		dev->data->dev_conf.fdir_conf.mode;
5547 
5548 	if (filter_op == RTE_ETH_FILTER_NOP)
5549 		return 0;
5550 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5551 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5552 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5553 			dev->data->port_id, fdir_mode);
5554 		rte_errno = EINVAL;
5555 		return -rte_errno;
5556 	}
5557 	switch (filter_op) {
5558 	case RTE_ETH_FILTER_ADD:
5559 		return flow_fdir_filter_add(dev, arg);
5560 	case RTE_ETH_FILTER_UPDATE:
5561 		return flow_fdir_filter_update(dev, arg);
5562 	case RTE_ETH_FILTER_DELETE:
5563 		return flow_fdir_filter_delete(dev, arg);
5564 	case RTE_ETH_FILTER_FLUSH:
5565 		flow_fdir_filter_flush(dev);
5566 		break;
5567 	case RTE_ETH_FILTER_INFO:
5568 		flow_fdir_info_get(dev, arg);
5569 		break;
5570 	default:
5571 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5572 			dev->data->port_id, filter_op);
5573 		rte_errno = EINVAL;
5574 		return -rte_errno;
5575 	}
5576 	return 0;
5577 }
5578 
5579 /**
5580  * Manage filter operations.
5581  *
5582  * @param dev
5583  *   Pointer to Ethernet device structure.
5584  * @param filter_type
5585  *   Filter type.
5586  * @param filter_op
5587  *   Operation to perform.
5588  * @param arg
5589  *   Pointer to operation-specific structure.
5590  *
5591  * @return
5592  *   0 on success, a negative errno value otherwise and rte_errno is set.
5593  */
5594 int
5595 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5596 		     enum rte_filter_type filter_type,
5597 		     enum rte_filter_op filter_op,
5598 		     void *arg)
5599 {
5600 	switch (filter_type) {
5601 	case RTE_ETH_FILTER_GENERIC:
5602 		if (filter_op != RTE_ETH_FILTER_GET) {
5603 			rte_errno = EINVAL;
5604 			return -rte_errno;
5605 		}
5606 		*(const void **)arg = &mlx5_flow_ops;
5607 		return 0;
5608 	case RTE_ETH_FILTER_FDIR:
5609 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5610 	default:
5611 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5612 			dev->data->port_id, filter_type);
5613 		rte_errno = ENOTSUP;
5614 		return -rte_errno;
5615 	}
5616 	return 0;
5617 }
5618 
5619 /**
5620  * Create the needed meter and suffix tables.
5621  *
5622  * @param[in] dev
5623  *   Pointer to Ethernet device.
5624  * @param[in] fm
5625  *   Pointer to the flow meter.
5626  *
5627  * @return
5628  *   Pointer to table set on success, NULL otherwise.
5629  */
5630 struct mlx5_meter_domains_infos *
5631 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5632 			  const struct mlx5_flow_meter *fm)
5633 {
5634 	const struct mlx5_flow_driver_ops *fops;
5635 
5636 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5637 	return fops->create_mtr_tbls(dev, fm);
5638 }
5639 
5640 /**
5641  * Destroy the meter table set.
5642  *
5643  * @param[in] dev
5644  *   Pointer to Ethernet device.
5645  * @param[in] tbl
5646  *   Pointer to the meter table set.
5647  *
5648  * @return
5649  *   0 on success.
5650  */
5651 int
5652 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5653 			   struct mlx5_meter_domains_infos *tbls)
5654 {
5655 	const struct mlx5_flow_driver_ops *fops;
5656 
5657 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5658 	return fops->destroy_mtr_tbls(dev, tbls);
5659 }
5660 
5661 /**
5662  * Create policer rules.
5663  *
5664  * @param[in] dev
5665  *   Pointer to Ethernet device.
5666  * @param[in] fm
5667  *   Pointer to flow meter structure.
5668  * @param[in] attr
5669  *   Pointer to flow attributes.
5670  *
5671  * @return
5672  *   0 on success, -1 otherwise.
5673  */
5674 int
5675 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5676 			       struct mlx5_flow_meter *fm,
5677 			       const struct rte_flow_attr *attr)
5678 {
5679 	const struct mlx5_flow_driver_ops *fops;
5680 
5681 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5682 	return fops->create_policer_rules(dev, fm, attr);
5683 }
5684 
5685 /**
5686  * Destroy policer rules.
5687  *
5688  * @param[in] fm
5689  *   Pointer to flow meter structure.
5690  * @param[in] attr
5691  *   Pointer to flow attributes.
5692  *
5693  * @return
5694  *   0 on success, -1 otherwise.
5695  */
5696 int
5697 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5698 				struct mlx5_flow_meter *fm,
5699 				const struct rte_flow_attr *attr)
5700 {
5701 	const struct mlx5_flow_driver_ops *fops;
5702 
5703 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5704 	return fops->destroy_policer_rules(dev, fm, attr);
5705 }
5706 
5707 /**
5708  * Allocate a counter.
5709  *
5710  * @param[in] dev
5711  *   Pointer to Ethernet device structure.
5712  *
5713  * @return
5714  *   Index to allocated counter  on success, 0 otherwise.
5715  */
5716 uint32_t
5717 mlx5_counter_alloc(struct rte_eth_dev *dev)
5718 {
5719 	const struct mlx5_flow_driver_ops *fops;
5720 	struct rte_flow_attr attr = { .transfer = 0 };
5721 
5722 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5723 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5724 		return fops->counter_alloc(dev);
5725 	}
5726 	DRV_LOG(ERR,
5727 		"port %u counter allocate is not supported.",
5728 		 dev->data->port_id);
5729 	return 0;
5730 }
5731 
5732 /**
5733  * Free a counter.
5734  *
5735  * @param[in] dev
5736  *   Pointer to Ethernet device structure.
5737  * @param[in] cnt
5738  *   Index to counter to be free.
5739  */
5740 void
5741 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5742 {
5743 	const struct mlx5_flow_driver_ops *fops;
5744 	struct rte_flow_attr attr = { .transfer = 0 };
5745 
5746 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5747 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5748 		fops->counter_free(dev, cnt);
5749 		return;
5750 	}
5751 	DRV_LOG(ERR,
5752 		"port %u counter free is not supported.",
5753 		 dev->data->port_id);
5754 }
5755 
5756 /**
5757  * Query counter statistics.
5758  *
5759  * @param[in] dev
5760  *   Pointer to Ethernet device structure.
5761  * @param[in] cnt
5762  *   Index to counter to query.
5763  * @param[in] clear
5764  *   Set to clear counter statistics.
5765  * @param[out] pkts
5766  *   The counter hits packets number to save.
5767  * @param[out] bytes
5768  *   The counter hits bytes number to save.
5769  *
5770  * @return
5771  *   0 on success, a negative errno value otherwise.
5772  */
5773 int
5774 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5775 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5776 {
5777 	const struct mlx5_flow_driver_ops *fops;
5778 	struct rte_flow_attr attr = { .transfer = 0 };
5779 
5780 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5781 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5782 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5783 	}
5784 	DRV_LOG(ERR,
5785 		"port %u counter query is not supported.",
5786 		 dev->data->port_id);
5787 	return -ENOTSUP;
5788 }
5789 
5790 #define MLX5_POOL_QUERY_FREQ_US 1000000
5791 
5792 /**
5793  * Get number of all validate pools.
5794  *
5795  * @param[in] sh
5796  *   Pointer to mlx5_ibv_shared object.
5797  *
5798  * @return
5799  *   The number of all validate pools.
5800  */
5801 static uint32_t
5802 mlx5_get_all_valid_pool_count(struct mlx5_ibv_shared *sh)
5803 {
5804 	uint8_t age, i;
5805 	uint32_t pools_n = 0;
5806 	struct mlx5_pools_container *cont;
5807 
5808 	for (age = 0; age < RTE_DIM(sh->cmng.ccont[0]); ++age) {
5809 		for (i = 0; i < 2 ; ++i) {
5810 			cont = MLX5_CNT_CONTAINER(sh, i, 0, age);
5811 			pools_n += rte_atomic16_read(&cont->n_valid);
5812 		}
5813 	}
5814 	return pools_n;
5815 }
5816 
5817 /**
5818  * Set the periodic procedure for triggering asynchronous batch queries for all
5819  * the counter pools.
5820  *
5821  * @param[in] sh
5822  *   Pointer to mlx5_ibv_shared object.
5823  */
5824 void
5825 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5826 {
5827 	uint32_t pools_n, us;
5828 
5829 	pools_n = mlx5_get_all_valid_pool_count(sh);
5830 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5831 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5832 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5833 		sh->cmng.query_thread_on = 0;
5834 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5835 	} else {
5836 		sh->cmng.query_thread_on = 1;
5837 	}
5838 }
5839 
5840 /**
5841  * The periodic procedure for triggering asynchronous batch queries for all the
5842  * counter pools. This function is probably called by the host thread.
5843  *
5844  * @param[in] arg
5845  *   The parameter for the alarm process.
5846  */
5847 void
5848 mlx5_flow_query_alarm(void *arg)
5849 {
5850 	struct mlx5_ibv_shared *sh = arg;
5851 	struct mlx5_devx_obj *dcs;
5852 	uint16_t offset;
5853 	int ret;
5854 	uint8_t batch = sh->cmng.batch;
5855 	uint8_t age = sh->cmng.age;
5856 	uint16_t pool_index = sh->cmng.pool_index;
5857 	struct mlx5_pools_container *cont;
5858 	struct mlx5_pools_container *mcont;
5859 	struct mlx5_flow_counter_pool *pool;
5860 
5861 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5862 		goto set_alarm;
5863 next_container:
5864 	cont = MLX5_CNT_CONTAINER(sh, batch, 1, age);
5865 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0, age);
5866 	/* Check if resize was done and need to flip a container. */
5867 	if (cont != mcont) {
5868 		if (cont->pools) {
5869 			/* Clean the old container. */
5870 			rte_free(cont->pools);
5871 			memset(cont, 0, sizeof(*cont));
5872 		}
5873 		rte_cio_wmb();
5874 		 /* Flip the host container. */
5875 		sh->cmng.mhi[batch][age] ^= (uint8_t)2;
5876 		cont = mcont;
5877 	}
5878 	if (!cont->pools) {
5879 		/* 2 empty containers case is unexpected. */
5880 		if (unlikely(batch != sh->cmng.batch) &&
5881 			unlikely(age != sh->cmng.age)) {
5882 			goto set_alarm;
5883 		}
5884 		batch ^= 0x1;
5885 		pool_index = 0;
5886 		if (batch == 0 && pool_index == 0) {
5887 			age ^= 0x1;
5888 			sh->cmng.batch = batch;
5889 			sh->cmng.age = age;
5890 		}
5891 		goto next_container;
5892 	}
5893 	pool = cont->pools[pool_index];
5894 	if (pool->raw_hw)
5895 		/* There is a pool query in progress. */
5896 		goto set_alarm;
5897 	pool->raw_hw =
5898 		LIST_FIRST(&sh->cmng.free_stat_raws);
5899 	if (!pool->raw_hw)
5900 		/* No free counter statistics raw memory. */
5901 		goto set_alarm;
5902 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5903 							      (&pool->a64_dcs);
5904 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5905 	/*
5906 	 * Identify the counters released between query trigger and query
5907 	 * handle more effiecntly. The counter released in this gap period
5908 	 * should wait for a new round of query as the new arrived packets
5909 	 * will not be taken into account.
5910 	 */
5911 	rte_atomic64_add(&pool->start_query_gen, 1);
5912 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5913 					       offset, NULL, NULL,
5914 					       pool->raw_hw->mem_mng->dm->id,
5915 					       (void *)(uintptr_t)
5916 					       (pool->raw_hw->data + offset),
5917 					       sh->devx_comp,
5918 					       (uint64_t)(uintptr_t)pool);
5919 	if (ret) {
5920 		rte_atomic64_sub(&pool->start_query_gen, 1);
5921 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5922 			" %d", pool->min_dcs->id);
5923 		pool->raw_hw = NULL;
5924 		goto set_alarm;
5925 	}
5926 	pool->raw_hw->min_dcs_id = dcs->id;
5927 	LIST_REMOVE(pool->raw_hw, next);
5928 	sh->cmng.pending_queries++;
5929 	pool_index++;
5930 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5931 		batch ^= 0x1;
5932 		pool_index = 0;
5933 		if (batch == 0 && pool_index == 0)
5934 			age ^= 0x1;
5935 	}
5936 set_alarm:
5937 	sh->cmng.batch = batch;
5938 	sh->cmng.pool_index = pool_index;
5939 	sh->cmng.age = age;
5940 	mlx5_set_query_alarm(sh);
5941 }
5942 
5943 /**
5944  * Check and callback event for new aged flow in the counter pool
5945  *
5946  * @param[in] sh
5947  *   Pointer to mlx5_ibv_shared object.
5948  * @param[in] pool
5949  *   Pointer to Current counter pool.
5950  */
5951 static void
5952 mlx5_flow_aging_check(struct mlx5_ibv_shared *sh,
5953 		   struct mlx5_flow_counter_pool *pool)
5954 {
5955 	struct mlx5_priv *priv;
5956 	struct mlx5_flow_counter *cnt;
5957 	struct mlx5_age_info *age_info;
5958 	struct mlx5_age_param *age_param;
5959 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
5960 	struct mlx5_counter_stats_raw *prev = pool->raw;
5961 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
5962 	uint32_t i;
5963 
5964 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
5965 		cnt = MLX5_POOL_GET_CNT(pool, i);
5966 		age_param = MLX5_CNT_TO_AGE(cnt);
5967 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
5968 			continue;
5969 		if (cur->data[i].hits != prev->data[i].hits) {
5970 			age_param->expire = curr + age_param->timeout;
5971 			continue;
5972 		}
5973 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
5974 			continue;
5975 		/**
5976 		 * Hold the lock first, or if between the
5977 		 * state AGE_TMOUT and tailq operation the
5978 		 * release happened, the release procedure
5979 		 * may delete a non-existent tailq node.
5980 		 */
5981 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
5982 		age_info = GET_PORT_AGE_INFO(priv);
5983 		rte_spinlock_lock(&age_info->aged_sl);
5984 		/* If the cpmset fails, release happens. */
5985 		if (rte_atomic16_cmpset((volatile uint16_t *)
5986 					&age_param->state,
5987 					AGE_CANDIDATE,
5988 					AGE_TMOUT) ==
5989 					AGE_CANDIDATE) {
5990 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
5991 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
5992 		}
5993 		rte_spinlock_unlock(&age_info->aged_sl);
5994 	}
5995 	for (i = 0; i < sh->max_port; i++) {
5996 		age_info = &sh->port[i].age_info;
5997 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
5998 			continue;
5999 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6000 			_rte_eth_dev_callback_process
6001 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
6002 				RTE_ETH_EVENT_FLOW_AGED, NULL);
6003 		age_info->flags = 0;
6004 	}
6005 }
6006 
6007 /**
6008  * Handler for the HW respond about ready values from an asynchronous batch
6009  * query. This function is probably called by the host thread.
6010  *
6011  * @param[in] sh
6012  *   The pointer to the shared IB device context.
6013  * @param[in] async_id
6014  *   The Devx async ID.
6015  * @param[in] status
6016  *   The status of the completion.
6017  */
6018 void
6019 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
6020 				  uint64_t async_id, int status)
6021 {
6022 	struct mlx5_flow_counter_pool *pool =
6023 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6024 	struct mlx5_counter_stats_raw *raw_to_free;
6025 
6026 	if (unlikely(status)) {
6027 		rte_atomic64_sub(&pool->start_query_gen, 1);
6028 		raw_to_free = pool->raw_hw;
6029 	} else {
6030 		raw_to_free = pool->raw;
6031 		if (IS_AGE_POOL(pool))
6032 			mlx5_flow_aging_check(sh, pool);
6033 		rte_spinlock_lock(&pool->sl);
6034 		pool->raw = pool->raw_hw;
6035 		rte_spinlock_unlock(&pool->sl);
6036 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
6037 			    rte_atomic64_read(&pool->start_query_gen));
6038 		rte_atomic64_set(&pool->end_query_gen,
6039 				 rte_atomic64_read(&pool->start_query_gen));
6040 		/* Be sure the new raw counters data is updated in memory. */
6041 		rte_cio_wmb();
6042 	}
6043 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6044 	pool->raw_hw = NULL;
6045 	sh->cmng.pending_queries--;
6046 }
6047 
6048 /**
6049  * Translate the rte_flow group index to HW table value.
6050  *
6051  * @param[in] attributes
6052  *   Pointer to flow attributes
6053  * @param[in] external
6054  *   Value is part of flow rule created by request external to PMD.
6055  * @param[in] group
6056  *   rte_flow group index value.
6057  * @param[out] fdb_def_rule
6058  *   Whether fdb jump to table 1 is configured.
6059  * @param[out] table
6060  *   HW table value.
6061  * @param[out] error
6062  *   Pointer to error structure.
6063  *
6064  * @return
6065  *   0 on success, a negative errno value otherwise and rte_errno is set.
6066  */
6067 int
6068 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6069 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6070 			 struct rte_flow_error *error)
6071 {
6072 	if (attributes->transfer && external && fdb_def_rule) {
6073 		if (group == UINT32_MAX)
6074 			return rte_flow_error_set
6075 						(error, EINVAL,
6076 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6077 						 NULL,
6078 						 "group index not supported");
6079 		*table = group + 1;
6080 	} else {
6081 		*table = group;
6082 	}
6083 	return 0;
6084 }
6085 
6086 /**
6087  * Discover availability of metadata reg_c's.
6088  *
6089  * Iteratively use test flows to check availability.
6090  *
6091  * @param[in] dev
6092  *   Pointer to the Ethernet device structure.
6093  *
6094  * @return
6095  *   0 on success, a negative errno value otherwise and rte_errno is set.
6096  */
6097 int
6098 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6099 {
6100 	struct mlx5_priv *priv = dev->data->dev_private;
6101 	struct mlx5_dev_config *config = &priv->config;
6102 	enum modify_reg idx;
6103 	int n = 0;
6104 
6105 	/* reg_c[0] and reg_c[1] are reserved. */
6106 	config->flow_mreg_c[n++] = REG_C_0;
6107 	config->flow_mreg_c[n++] = REG_C_1;
6108 	/* Discover availability of other reg_c's. */
6109 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6110 		struct rte_flow_attr attr = {
6111 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6112 			.priority = MLX5_FLOW_PRIO_RSVD,
6113 			.ingress = 1,
6114 		};
6115 		struct rte_flow_item items[] = {
6116 			[0] = {
6117 				.type = RTE_FLOW_ITEM_TYPE_END,
6118 			},
6119 		};
6120 		struct rte_flow_action actions[] = {
6121 			[0] = {
6122 				.type = (enum rte_flow_action_type)
6123 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6124 				.conf = &(struct mlx5_flow_action_copy_mreg){
6125 					.src = REG_C_1,
6126 					.dst = idx,
6127 				},
6128 			},
6129 			[1] = {
6130 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6131 				.conf = &(struct rte_flow_action_jump){
6132 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6133 				},
6134 			},
6135 			[2] = {
6136 				.type = RTE_FLOW_ACTION_TYPE_END,
6137 			},
6138 		};
6139 		uint32_t flow_idx;
6140 		struct rte_flow *flow;
6141 		struct rte_flow_error error;
6142 
6143 		if (!config->dv_flow_en)
6144 			break;
6145 		/* Create internal flow, validation skips copy action. */
6146 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6147 					    actions, false, &error);
6148 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6149 				      flow_idx);
6150 		if (!flow)
6151 			continue;
6152 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6153 			config->flow_mreg_c[n++] = idx;
6154 		flow_list_destroy(dev, NULL, flow_idx);
6155 	}
6156 	for (; n < MLX5_MREG_C_NUM; ++n)
6157 		config->flow_mreg_c[n] = REG_NONE;
6158 	return 0;
6159 }
6160 
6161 /**
6162  * Dump flow raw hw data to file
6163  *
6164  * @param[in] dev
6165  *    The pointer to Ethernet device.
6166  * @param[in] file
6167  *   A pointer to a file for output.
6168  * @param[out] error
6169  *   Perform verbose error reporting if not NULL. PMDs initialize this
6170  *   structure in case of error only.
6171  * @return
6172  *   0 on success, a nagative value otherwise.
6173  */
6174 int
6175 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6176 		   FILE *file,
6177 		   struct rte_flow_error *error __rte_unused)
6178 {
6179 	struct mlx5_priv *priv = dev->data->dev_private;
6180 	struct mlx5_ibv_shared *sh = priv->sh;
6181 
6182 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6183 				       sh->tx_domain, file);
6184 }
6185 
6186 /**
6187  * Get aged-out flows.
6188  *
6189  * @param[in] dev
6190  *   Pointer to the Ethernet device structure.
6191  * @param[in] context
6192  *   The address of an array of pointers to the aged-out flows contexts.
6193  * @param[in] nb_countexts
6194  *   The length of context array pointers.
6195  * @param[out] error
6196  *   Perform verbose error reporting if not NULL. Initialized in case of
6197  *   error only.
6198  *
6199  * @return
6200  *   how many contexts get in success, otherwise negative errno value.
6201  *   if nb_contexts is 0, return the amount of all aged contexts.
6202  *   if nb_contexts is not 0 , return the amount of aged flows reported
6203  *   in the context array.
6204  */
6205 int
6206 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6207 			uint32_t nb_contexts, struct rte_flow_error *error)
6208 {
6209 	const struct mlx5_flow_driver_ops *fops;
6210 	struct rte_flow_attr attr = { .transfer = 0 };
6211 
6212 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6213 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6214 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6215 						    error);
6216 	}
6217 	DRV_LOG(ERR,
6218 		"port %u get aged flows is not supported.",
6219 		 dev->data->port_id);
6220 	return -ENOTSUP;
6221 }
6222