xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 543e218fa58dcfb9b342142b9aeb3bd7f547ab86)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 
12 /* Verbs header. */
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #ifdef PEDANTIC
15 #pragma GCC diagnostic ignored "-Wpedantic"
16 #endif
17 #include <infiniband/verbs.h>
18 #ifdef PEDANTIC
19 #pragma GCC diagnostic error "-Wpedantic"
20 #endif
21 
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
25 #include <rte_flow.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
28 #include <rte_ip.h>
29 
30 #include "mlx5.h"
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
34 #include "mlx5_devx_cmds.h"
35 #include "mlx5_prm.h"
36 #include "mlx5_rxtx.h"
37 
38 /* Dev ops structure defined in mlx5.c */
39 extern const struct eth_dev_ops mlx5_dev_ops;
40 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
41 
42 /** Device flow drivers. */
43 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
44 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
45 #endif
46 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
47 
48 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
49 
50 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
51 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
52 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
53 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
54 #endif
55 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
56 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
57 };
58 
59 enum mlx5_expansion {
60 	MLX5_EXPANSION_ROOT,
61 	MLX5_EXPANSION_ROOT_OUTER,
62 	MLX5_EXPANSION_ROOT_ETH_VLAN,
63 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
64 	MLX5_EXPANSION_OUTER_ETH,
65 	MLX5_EXPANSION_OUTER_ETH_VLAN,
66 	MLX5_EXPANSION_OUTER_VLAN,
67 	MLX5_EXPANSION_OUTER_IPV4,
68 	MLX5_EXPANSION_OUTER_IPV4_UDP,
69 	MLX5_EXPANSION_OUTER_IPV4_TCP,
70 	MLX5_EXPANSION_OUTER_IPV6,
71 	MLX5_EXPANSION_OUTER_IPV6_UDP,
72 	MLX5_EXPANSION_OUTER_IPV6_TCP,
73 	MLX5_EXPANSION_VXLAN,
74 	MLX5_EXPANSION_VXLAN_GPE,
75 	MLX5_EXPANSION_GRE,
76 	MLX5_EXPANSION_MPLS,
77 	MLX5_EXPANSION_ETH,
78 	MLX5_EXPANSION_ETH_VLAN,
79 	MLX5_EXPANSION_VLAN,
80 	MLX5_EXPANSION_IPV4,
81 	MLX5_EXPANSION_IPV4_UDP,
82 	MLX5_EXPANSION_IPV4_TCP,
83 	MLX5_EXPANSION_IPV6,
84 	MLX5_EXPANSION_IPV6_UDP,
85 	MLX5_EXPANSION_IPV6_TCP,
86 };
87 
88 /** Supported expansion of items. */
89 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
90 	[MLX5_EXPANSION_ROOT] = {
91 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
92 						 MLX5_EXPANSION_IPV4,
93 						 MLX5_EXPANSION_IPV6),
94 		.type = RTE_FLOW_ITEM_TYPE_END,
95 	},
96 	[MLX5_EXPANSION_ROOT_OUTER] = {
97 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
98 						 MLX5_EXPANSION_OUTER_IPV4,
99 						 MLX5_EXPANSION_OUTER_IPV6),
100 		.type = RTE_FLOW_ITEM_TYPE_END,
101 	},
102 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
103 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
104 		.type = RTE_FLOW_ITEM_TYPE_END,
105 	},
106 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
107 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
108 		.type = RTE_FLOW_ITEM_TYPE_END,
109 	},
110 	[MLX5_EXPANSION_OUTER_ETH] = {
111 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
112 						 MLX5_EXPANSION_OUTER_IPV6,
113 						 MLX5_EXPANSION_MPLS),
114 		.type = RTE_FLOW_ITEM_TYPE_ETH,
115 		.rss_types = 0,
116 	},
117 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
118 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
119 		.type = RTE_FLOW_ITEM_TYPE_ETH,
120 		.rss_types = 0,
121 	},
122 	[MLX5_EXPANSION_OUTER_VLAN] = {
123 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
124 						 MLX5_EXPANSION_OUTER_IPV6),
125 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
126 	},
127 	[MLX5_EXPANSION_OUTER_IPV4] = {
128 		.next = RTE_FLOW_EXPAND_RSS_NEXT
129 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
130 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
131 			 MLX5_EXPANSION_GRE,
132 			 MLX5_EXPANSION_IPV4,
133 			 MLX5_EXPANSION_IPV6),
134 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
135 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
136 			ETH_RSS_NONFRAG_IPV4_OTHER,
137 	},
138 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
139 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
140 						 MLX5_EXPANSION_VXLAN_GPE),
141 		.type = RTE_FLOW_ITEM_TYPE_UDP,
142 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
143 	},
144 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
145 		.type = RTE_FLOW_ITEM_TYPE_TCP,
146 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
147 	},
148 	[MLX5_EXPANSION_OUTER_IPV6] = {
149 		.next = RTE_FLOW_EXPAND_RSS_NEXT
150 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
151 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
152 			 MLX5_EXPANSION_IPV4,
153 			 MLX5_EXPANSION_IPV6),
154 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
155 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
156 			ETH_RSS_NONFRAG_IPV6_OTHER,
157 	},
158 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
159 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
160 						 MLX5_EXPANSION_VXLAN_GPE),
161 		.type = RTE_FLOW_ITEM_TYPE_UDP,
162 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
163 	},
164 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
165 		.type = RTE_FLOW_ITEM_TYPE_TCP,
166 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
167 	},
168 	[MLX5_EXPANSION_VXLAN] = {
169 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
170 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
171 	},
172 	[MLX5_EXPANSION_VXLAN_GPE] = {
173 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
174 						 MLX5_EXPANSION_IPV4,
175 						 MLX5_EXPANSION_IPV6),
176 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
177 	},
178 	[MLX5_EXPANSION_GRE] = {
179 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
180 		.type = RTE_FLOW_ITEM_TYPE_GRE,
181 	},
182 	[MLX5_EXPANSION_MPLS] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
184 						 MLX5_EXPANSION_IPV6),
185 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
186 	},
187 	[MLX5_EXPANSION_ETH] = {
188 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
189 						 MLX5_EXPANSION_IPV6),
190 		.type = RTE_FLOW_ITEM_TYPE_ETH,
191 	},
192 	[MLX5_EXPANSION_ETH_VLAN] = {
193 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
194 		.type = RTE_FLOW_ITEM_TYPE_ETH,
195 	},
196 	[MLX5_EXPANSION_VLAN] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
198 						 MLX5_EXPANSION_IPV6),
199 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
200 	},
201 	[MLX5_EXPANSION_IPV4] = {
202 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
203 						 MLX5_EXPANSION_IPV4_TCP),
204 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
205 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
206 			ETH_RSS_NONFRAG_IPV4_OTHER,
207 	},
208 	[MLX5_EXPANSION_IPV4_UDP] = {
209 		.type = RTE_FLOW_ITEM_TYPE_UDP,
210 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
211 	},
212 	[MLX5_EXPANSION_IPV4_TCP] = {
213 		.type = RTE_FLOW_ITEM_TYPE_TCP,
214 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
215 	},
216 	[MLX5_EXPANSION_IPV6] = {
217 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
218 						 MLX5_EXPANSION_IPV6_TCP),
219 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
220 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
221 			ETH_RSS_NONFRAG_IPV6_OTHER,
222 	},
223 	[MLX5_EXPANSION_IPV6_UDP] = {
224 		.type = RTE_FLOW_ITEM_TYPE_UDP,
225 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
226 	},
227 	[MLX5_EXPANSION_IPV6_TCP] = {
228 		.type = RTE_FLOW_ITEM_TYPE_TCP,
229 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
230 	},
231 };
232 
233 static const struct rte_flow_ops mlx5_flow_ops = {
234 	.validate = mlx5_flow_validate,
235 	.create = mlx5_flow_create,
236 	.destroy = mlx5_flow_destroy,
237 	.flush = mlx5_flow_flush,
238 	.isolate = mlx5_flow_isolate,
239 	.query = mlx5_flow_query,
240 	.dev_dump = mlx5_flow_dev_dump,
241 };
242 
243 /* Convert FDIR request to Generic flow. */
244 struct mlx5_fdir {
245 	struct rte_flow_attr attr;
246 	struct rte_flow_item items[4];
247 	struct rte_flow_item_eth l2;
248 	struct rte_flow_item_eth l2_mask;
249 	union {
250 		struct rte_flow_item_ipv4 ipv4;
251 		struct rte_flow_item_ipv6 ipv6;
252 	} l3;
253 	union {
254 		struct rte_flow_item_ipv4 ipv4;
255 		struct rte_flow_item_ipv6 ipv6;
256 	} l3_mask;
257 	union {
258 		struct rte_flow_item_udp udp;
259 		struct rte_flow_item_tcp tcp;
260 	} l4;
261 	union {
262 		struct rte_flow_item_udp udp;
263 		struct rte_flow_item_tcp tcp;
264 	} l4_mask;
265 	struct rte_flow_action actions[2];
266 	struct rte_flow_action_queue queue;
267 };
268 
269 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
270 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
271 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
272 };
273 
274 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
275 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
276 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
277 	{ 9, 10, 11 }, { 12, 13, 14 },
278 };
279 
280 /* Tunnel information. */
281 struct mlx5_flow_tunnel_info {
282 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
283 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
284 };
285 
286 static struct mlx5_flow_tunnel_info tunnels_info[] = {
287 	{
288 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
289 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
290 	},
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
293 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
297 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_GRE,
301 		.ptype = RTE_PTYPE_TUNNEL_GRE,
302 	},
303 	{
304 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
305 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
306 	},
307 	{
308 		.tunnel = MLX5_FLOW_LAYER_MPLS,
309 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
310 	},
311 	{
312 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
313 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
314 	},
315 	{
316 		.tunnel = MLX5_FLOW_LAYER_IPIP,
317 		.ptype = RTE_PTYPE_TUNNEL_IP,
318 	},
319 	{
320 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
321 		.ptype = RTE_PTYPE_TUNNEL_IP,
322 	},
323 	{
324 		.tunnel = MLX5_FLOW_LAYER_GTP,
325 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
326 	},
327 };
328 
329 /**
330  * Translate tag ID to register.
331  *
332  * @param[in] dev
333  *   Pointer to the Ethernet device structure.
334  * @param[in] feature
335  *   The feature that request the register.
336  * @param[in] id
337  *   The request register ID.
338  * @param[out] error
339  *   Error description in case of any.
340  *
341  * @return
342  *   The request register on success, a negative errno
343  *   value otherwise and rte_errno is set.
344  */
345 int
346 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
347 		     enum mlx5_feature_name feature,
348 		     uint32_t id,
349 		     struct rte_flow_error *error)
350 {
351 	struct mlx5_priv *priv = dev->data->dev_private;
352 	struct mlx5_dev_config *config = &priv->config;
353 	enum modify_reg start_reg;
354 	bool skip_mtr_reg = false;
355 
356 	switch (feature) {
357 	case MLX5_HAIRPIN_RX:
358 		return REG_B;
359 	case MLX5_HAIRPIN_TX:
360 		return REG_A;
361 	case MLX5_METADATA_RX:
362 		switch (config->dv_xmeta_en) {
363 		case MLX5_XMETA_MODE_LEGACY:
364 			return REG_B;
365 		case MLX5_XMETA_MODE_META16:
366 			return REG_C_0;
367 		case MLX5_XMETA_MODE_META32:
368 			return REG_C_1;
369 		}
370 		break;
371 	case MLX5_METADATA_TX:
372 		return REG_A;
373 	case MLX5_METADATA_FDB:
374 		switch (config->dv_xmeta_en) {
375 		case MLX5_XMETA_MODE_LEGACY:
376 			return REG_NONE;
377 		case MLX5_XMETA_MODE_META16:
378 			return REG_C_0;
379 		case MLX5_XMETA_MODE_META32:
380 			return REG_C_1;
381 		}
382 		break;
383 	case MLX5_FLOW_MARK:
384 		switch (config->dv_xmeta_en) {
385 		case MLX5_XMETA_MODE_LEGACY:
386 			return REG_NONE;
387 		case MLX5_XMETA_MODE_META16:
388 			return REG_C_1;
389 		case MLX5_XMETA_MODE_META32:
390 			return REG_C_0;
391 		}
392 		break;
393 	case MLX5_MTR_SFX:
394 		/*
395 		 * If meter color and flow match share one register, flow match
396 		 * should use the meter color register for match.
397 		 */
398 		if (priv->mtr_reg_share)
399 			return priv->mtr_color_reg;
400 		else
401 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
402 			       REG_C_3;
403 	case MLX5_MTR_COLOR:
404 		RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
405 		return priv->mtr_color_reg;
406 	case MLX5_COPY_MARK:
407 		/*
408 		 * Metadata COPY_MARK register using is in meter suffix sub
409 		 * flow while with meter. It's safe to share the same register.
410 		 */
411 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
412 	case MLX5_APP_TAG:
413 		/*
414 		 * If meter is enable, it will engage the register for color
415 		 * match and flow match. If meter color match is not using the
416 		 * REG_C_2, need to skip the REG_C_x be used by meter color
417 		 * match.
418 		 * If meter is disable, free to use all available registers.
419 		 */
420 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
421 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
422 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
423 		if (id > (REG_C_7 - start_reg))
424 			return rte_flow_error_set(error, EINVAL,
425 						  RTE_FLOW_ERROR_TYPE_ITEM,
426 						  NULL, "invalid tag id");
427 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
428 			return rte_flow_error_set(error, ENOTSUP,
429 						  RTE_FLOW_ERROR_TYPE_ITEM,
430 						  NULL, "unsupported tag id");
431 		/*
432 		 * This case means meter is using the REG_C_x great than 2.
433 		 * Take care not to conflict with meter color REG_C_x.
434 		 * If the available index REG_C_y >= REG_C_x, skip the
435 		 * color register.
436 		 */
437 		if (skip_mtr_reg && config->flow_mreg_c
438 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
439 			if (config->flow_mreg_c
440 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
441 				return config->flow_mreg_c
442 					       [id + 1 + start_reg - REG_C_0];
443 			return rte_flow_error_set(error, ENOTSUP,
444 						  RTE_FLOW_ERROR_TYPE_ITEM,
445 						  NULL, "unsupported tag id");
446 		}
447 		return config->flow_mreg_c[id + start_reg - REG_C_0];
448 	}
449 	assert(false);
450 	return rte_flow_error_set(error, EINVAL,
451 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
452 				  NULL, "invalid feature name");
453 }
454 
455 /**
456  * Check extensive flow metadata register support.
457  *
458  * @param dev
459  *   Pointer to rte_eth_dev structure.
460  *
461  * @return
462  *   True if device supports extensive flow metadata register, otherwise false.
463  */
464 bool
465 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
466 {
467 	struct mlx5_priv *priv = dev->data->dev_private;
468 	struct mlx5_dev_config *config = &priv->config;
469 
470 	/*
471 	 * Having available reg_c can be regarded inclusively as supporting
472 	 * extensive flow metadata register, which could mean,
473 	 * - metadata register copy action by modify header.
474 	 * - 16 modify header actions is supported.
475 	 * - reg_c's are preserved across different domain (FDB and NIC) on
476 	 *   packet loopback by flow lookup miss.
477 	 */
478 	return config->flow_mreg_c[2] != REG_NONE;
479 }
480 
481 /**
482  * Discover the maximum number of priority available.
483  *
484  * @param[in] dev
485  *   Pointer to the Ethernet device structure.
486  *
487  * @return
488  *   number of supported flow priority on success, a negative errno
489  *   value otherwise and rte_errno is set.
490  */
491 int
492 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
493 {
494 	struct mlx5_priv *priv = dev->data->dev_private;
495 	struct {
496 		struct ibv_flow_attr attr;
497 		struct ibv_flow_spec_eth eth;
498 		struct ibv_flow_spec_action_drop drop;
499 	} flow_attr = {
500 		.attr = {
501 			.num_of_specs = 2,
502 			.port = (uint8_t)priv->ibv_port,
503 		},
504 		.eth = {
505 			.type = IBV_FLOW_SPEC_ETH,
506 			.size = sizeof(struct ibv_flow_spec_eth),
507 		},
508 		.drop = {
509 			.size = sizeof(struct ibv_flow_spec_action_drop),
510 			.type = IBV_FLOW_SPEC_ACTION_DROP,
511 		},
512 	};
513 	struct ibv_flow *flow;
514 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
515 	uint16_t vprio[] = { 8, 16 };
516 	int i;
517 	int priority = 0;
518 
519 	if (!drop) {
520 		rte_errno = ENOTSUP;
521 		return -rte_errno;
522 	}
523 	for (i = 0; i != RTE_DIM(vprio); i++) {
524 		flow_attr.attr.priority = vprio[i] - 1;
525 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
526 		if (!flow)
527 			break;
528 		claim_zero(mlx5_glue->destroy_flow(flow));
529 		priority = vprio[i];
530 	}
531 	mlx5_hrxq_drop_release(dev);
532 	switch (priority) {
533 	case 8:
534 		priority = RTE_DIM(priority_map_3);
535 		break;
536 	case 16:
537 		priority = RTE_DIM(priority_map_5);
538 		break;
539 	default:
540 		rte_errno = ENOTSUP;
541 		DRV_LOG(ERR,
542 			"port %u verbs maximum priority: %d expected 8/16",
543 			dev->data->port_id, priority);
544 		return -rte_errno;
545 	}
546 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
547 		dev->data->port_id, priority);
548 	return priority;
549 }
550 
551 /**
552  * Adjust flow priority based on the highest layer and the request priority.
553  *
554  * @param[in] dev
555  *   Pointer to the Ethernet device structure.
556  * @param[in] priority
557  *   The rule base priority.
558  * @param[in] subpriority
559  *   The priority based on the items.
560  *
561  * @return
562  *   The new priority.
563  */
564 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
565 				   uint32_t subpriority)
566 {
567 	uint32_t res = 0;
568 	struct mlx5_priv *priv = dev->data->dev_private;
569 
570 	switch (priv->config.flow_prio) {
571 	case RTE_DIM(priority_map_3):
572 		res = priority_map_3[priority][subpriority];
573 		break;
574 	case RTE_DIM(priority_map_5):
575 		res = priority_map_5[priority][subpriority];
576 		break;
577 	}
578 	return  res;
579 }
580 
581 /**
582  * Verify the @p item specifications (spec, last, mask) are compatible with the
583  * NIC capabilities.
584  *
585  * @param[in] item
586  *   Item specification.
587  * @param[in] mask
588  *   @p item->mask or flow default bit-masks.
589  * @param[in] nic_mask
590  *   Bit-masks covering supported fields by the NIC to compare with user mask.
591  * @param[in] size
592  *   Bit-masks size in bytes.
593  * @param[out] error
594  *   Pointer to error structure.
595  *
596  * @return
597  *   0 on success, a negative errno value otherwise and rte_errno is set.
598  */
599 int
600 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
601 			  const uint8_t *mask,
602 			  const uint8_t *nic_mask,
603 			  unsigned int size,
604 			  struct rte_flow_error *error)
605 {
606 	unsigned int i;
607 
608 	assert(nic_mask);
609 	for (i = 0; i < size; ++i)
610 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
611 			return rte_flow_error_set(error, ENOTSUP,
612 						  RTE_FLOW_ERROR_TYPE_ITEM,
613 						  item,
614 						  "mask enables non supported"
615 						  " bits");
616 	if (!item->spec && (item->mask || item->last))
617 		return rte_flow_error_set(error, EINVAL,
618 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
619 					  "mask/last without a spec is not"
620 					  " supported");
621 	if (item->spec && item->last) {
622 		uint8_t spec[size];
623 		uint8_t last[size];
624 		unsigned int i;
625 		int ret;
626 
627 		for (i = 0; i < size; ++i) {
628 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
629 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
630 		}
631 		ret = memcmp(spec, last, size);
632 		if (ret != 0)
633 			return rte_flow_error_set(error, EINVAL,
634 						  RTE_FLOW_ERROR_TYPE_ITEM,
635 						  item,
636 						  "range is not valid");
637 	}
638 	return 0;
639 }
640 
641 /**
642  * Adjust the hash fields according to the @p flow information.
643  *
644  * @param[in] dev_flow.
645  *   Pointer to the mlx5_flow.
646  * @param[in] tunnel
647  *   1 when the hash field is for a tunnel item.
648  * @param[in] layer_types
649  *   ETH_RSS_* types.
650  * @param[in] hash_fields
651  *   Item hash fields.
652  *
653  * @return
654  *   The hash fields that should be used.
655  */
656 uint64_t
657 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
658 			    int tunnel __rte_unused, uint64_t layer_types,
659 			    uint64_t hash_fields)
660 {
661 	struct rte_flow *flow = dev_flow->flow;
662 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
663 	int rss_request_inner = flow->rss.level >= 2;
664 
665 	/* Check RSS hash level for tunnel. */
666 	if (tunnel && rss_request_inner)
667 		hash_fields |= IBV_RX_HASH_INNER;
668 	else if (tunnel || rss_request_inner)
669 		return 0;
670 #endif
671 	/* Check if requested layer matches RSS hash fields. */
672 	if (!(flow->rss.types & layer_types))
673 		return 0;
674 	return hash_fields;
675 }
676 
677 /**
678  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
679  * if several tunnel rules are used on this queue, the tunnel ptype will be
680  * cleared.
681  *
682  * @param rxq_ctrl
683  *   Rx queue to update.
684  */
685 static void
686 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
687 {
688 	unsigned int i;
689 	uint32_t tunnel_ptype = 0;
690 
691 	/* Look up for the ptype to use. */
692 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
693 		if (!rxq_ctrl->flow_tunnels_n[i])
694 			continue;
695 		if (!tunnel_ptype) {
696 			tunnel_ptype = tunnels_info[i].ptype;
697 		} else {
698 			tunnel_ptype = 0;
699 			break;
700 		}
701 	}
702 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
703 }
704 
705 /**
706  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
707  * flow.
708  *
709  * @param[in] dev
710  *   Pointer to the Ethernet device structure.
711  * @param[in] dev_flow
712  *   Pointer to device flow structure.
713  */
714 static void
715 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
716 {
717 	struct mlx5_priv *priv = dev->data->dev_private;
718 	struct rte_flow *flow = dev_flow->flow;
719 	const int mark = !!(dev_flow->actions &
720 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
721 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
722 	unsigned int i;
723 
724 	for (i = 0; i != flow->rss.queue_num; ++i) {
725 		int idx = (*flow->rss.queue)[i];
726 		struct mlx5_rxq_ctrl *rxq_ctrl =
727 			container_of((*priv->rxqs)[idx],
728 				     struct mlx5_rxq_ctrl, rxq);
729 
730 		/*
731 		 * To support metadata register copy on Tx loopback,
732 		 * this must be always enabled (metadata may arive
733 		 * from other port - not from local flows only.
734 		 */
735 		if (priv->config.dv_flow_en &&
736 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
737 		    mlx5_flow_ext_mreg_supported(dev)) {
738 			rxq_ctrl->rxq.mark = 1;
739 			rxq_ctrl->flow_mark_n = 1;
740 		} else if (mark) {
741 			rxq_ctrl->rxq.mark = 1;
742 			rxq_ctrl->flow_mark_n++;
743 		}
744 		if (tunnel) {
745 			unsigned int j;
746 
747 			/* Increase the counter matching the flow. */
748 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
749 				if ((tunnels_info[j].tunnel &
750 				     dev_flow->layers) ==
751 				    tunnels_info[j].tunnel) {
752 					rxq_ctrl->flow_tunnels_n[j]++;
753 					break;
754 				}
755 			}
756 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
757 		}
758 	}
759 }
760 
761 /**
762  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
763  *
764  * @param[in] dev
765  *   Pointer to the Ethernet device structure.
766  * @param[in] flow
767  *   Pointer to flow structure.
768  */
769 static void
770 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
771 {
772 	struct mlx5_flow *dev_flow;
773 
774 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
775 		flow_drv_rxq_flags_set(dev, dev_flow);
776 }
777 
778 /**
779  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
780  * device flow if no other flow uses it with the same kind of request.
781  *
782  * @param dev
783  *   Pointer to Ethernet device.
784  * @param[in] dev_flow
785  *   Pointer to the device flow.
786  */
787 static void
788 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
789 {
790 	struct mlx5_priv *priv = dev->data->dev_private;
791 	struct rte_flow *flow = dev_flow->flow;
792 	const int mark = !!(dev_flow->actions &
793 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
794 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
795 	unsigned int i;
796 
797 	assert(dev->data->dev_started);
798 	for (i = 0; i != flow->rss.queue_num; ++i) {
799 		int idx = (*flow->rss.queue)[i];
800 		struct mlx5_rxq_ctrl *rxq_ctrl =
801 			container_of((*priv->rxqs)[idx],
802 				     struct mlx5_rxq_ctrl, rxq);
803 
804 		if (priv->config.dv_flow_en &&
805 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
806 		    mlx5_flow_ext_mreg_supported(dev)) {
807 			rxq_ctrl->rxq.mark = 1;
808 			rxq_ctrl->flow_mark_n = 1;
809 		} else if (mark) {
810 			rxq_ctrl->flow_mark_n--;
811 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
812 		}
813 		if (tunnel) {
814 			unsigned int j;
815 
816 			/* Decrease the counter matching the flow. */
817 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
818 				if ((tunnels_info[j].tunnel &
819 				     dev_flow->layers) ==
820 				    tunnels_info[j].tunnel) {
821 					rxq_ctrl->flow_tunnels_n[j]--;
822 					break;
823 				}
824 			}
825 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
826 		}
827 	}
828 }
829 
830 /**
831  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
832  * @p flow if no other flow uses it with the same kind of request.
833  *
834  * @param dev
835  *   Pointer to Ethernet device.
836  * @param[in] flow
837  *   Pointer to the flow.
838  */
839 static void
840 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
841 {
842 	struct mlx5_flow *dev_flow;
843 
844 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
845 		flow_drv_rxq_flags_trim(dev, dev_flow);
846 }
847 
848 /**
849  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
850  *
851  * @param dev
852  *   Pointer to Ethernet device.
853  */
854 static void
855 flow_rxq_flags_clear(struct rte_eth_dev *dev)
856 {
857 	struct mlx5_priv *priv = dev->data->dev_private;
858 	unsigned int i;
859 
860 	for (i = 0; i != priv->rxqs_n; ++i) {
861 		struct mlx5_rxq_ctrl *rxq_ctrl;
862 		unsigned int j;
863 
864 		if (!(*priv->rxqs)[i])
865 			continue;
866 		rxq_ctrl = container_of((*priv->rxqs)[i],
867 					struct mlx5_rxq_ctrl, rxq);
868 		rxq_ctrl->flow_mark_n = 0;
869 		rxq_ctrl->rxq.mark = 0;
870 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
871 			rxq_ctrl->flow_tunnels_n[j] = 0;
872 		rxq_ctrl->rxq.tunnel = 0;
873 	}
874 }
875 
876 /*
877  * return a pointer to the desired action in the list of actions.
878  *
879  * @param[in] actions
880  *   The list of actions to search the action in.
881  * @param[in] action
882  *   The action to find.
883  *
884  * @return
885  *   Pointer to the action in the list, if found. NULL otherwise.
886  */
887 const struct rte_flow_action *
888 mlx5_flow_find_action(const struct rte_flow_action *actions,
889 		      enum rte_flow_action_type action)
890 {
891 	if (actions == NULL)
892 		return NULL;
893 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
894 		if (actions->type == action)
895 			return actions;
896 	return NULL;
897 }
898 
899 /*
900  * Validate the flag action.
901  *
902  * @param[in] action_flags
903  *   Bit-fields that holds the actions detected until now.
904  * @param[in] attr
905  *   Attributes of flow that includes this action.
906  * @param[out] error
907  *   Pointer to error structure.
908  *
909  * @return
910  *   0 on success, a negative errno value otherwise and rte_errno is set.
911  */
912 int
913 mlx5_flow_validate_action_flag(uint64_t action_flags,
914 			       const struct rte_flow_attr *attr,
915 			       struct rte_flow_error *error)
916 {
917 	if (action_flags & MLX5_FLOW_ACTION_MARK)
918 		return rte_flow_error_set(error, EINVAL,
919 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
920 					  "can't mark and flag in same flow");
921 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
922 		return rte_flow_error_set(error, EINVAL,
923 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
924 					  "can't have 2 flag"
925 					  " actions in same flow");
926 	if (attr->egress)
927 		return rte_flow_error_set(error, ENOTSUP,
928 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
929 					  "flag action not supported for "
930 					  "egress");
931 	return 0;
932 }
933 
934 /*
935  * Validate the mark action.
936  *
937  * @param[in] action
938  *   Pointer to the queue action.
939  * @param[in] action_flags
940  *   Bit-fields that holds the actions detected until now.
941  * @param[in] attr
942  *   Attributes of flow that includes this action.
943  * @param[out] error
944  *   Pointer to error structure.
945  *
946  * @return
947  *   0 on success, a negative errno value otherwise and rte_errno is set.
948  */
949 int
950 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
951 			       uint64_t action_flags,
952 			       const struct rte_flow_attr *attr,
953 			       struct rte_flow_error *error)
954 {
955 	const struct rte_flow_action_mark *mark = action->conf;
956 
957 	if (!mark)
958 		return rte_flow_error_set(error, EINVAL,
959 					  RTE_FLOW_ERROR_TYPE_ACTION,
960 					  action,
961 					  "configuration cannot be null");
962 	if (mark->id >= MLX5_FLOW_MARK_MAX)
963 		return rte_flow_error_set(error, EINVAL,
964 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
965 					  &mark->id,
966 					  "mark id must in 0 <= id < "
967 					  RTE_STR(MLX5_FLOW_MARK_MAX));
968 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
969 		return rte_flow_error_set(error, EINVAL,
970 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
971 					  "can't flag and mark in same flow");
972 	if (action_flags & MLX5_FLOW_ACTION_MARK)
973 		return rte_flow_error_set(error, EINVAL,
974 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
975 					  "can't have 2 mark actions in same"
976 					  " flow");
977 	if (attr->egress)
978 		return rte_flow_error_set(error, ENOTSUP,
979 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
980 					  "mark action not supported for "
981 					  "egress");
982 	return 0;
983 }
984 
985 /*
986  * Validate the drop action.
987  *
988  * @param[in] action_flags
989  *   Bit-fields that holds the actions detected until now.
990  * @param[in] attr
991  *   Attributes of flow that includes this action.
992  * @param[out] error
993  *   Pointer to error structure.
994  *
995  * @return
996  *   0 on success, a negative errno value otherwise and rte_errno is set.
997  */
998 int
999 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1000 			       const struct rte_flow_attr *attr,
1001 			       struct rte_flow_error *error)
1002 {
1003 	if (attr->egress)
1004 		return rte_flow_error_set(error, ENOTSUP,
1005 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1006 					  "drop action not supported for "
1007 					  "egress");
1008 	return 0;
1009 }
1010 
1011 /*
1012  * Validate the queue action.
1013  *
1014  * @param[in] action
1015  *   Pointer to the queue action.
1016  * @param[in] action_flags
1017  *   Bit-fields that holds the actions detected until now.
1018  * @param[in] dev
1019  *   Pointer to the Ethernet device structure.
1020  * @param[in] attr
1021  *   Attributes of flow that includes this action.
1022  * @param[out] error
1023  *   Pointer to error structure.
1024  *
1025  * @return
1026  *   0 on success, a negative errno value otherwise and rte_errno is set.
1027  */
1028 int
1029 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1030 				uint64_t action_flags,
1031 				struct rte_eth_dev *dev,
1032 				const struct rte_flow_attr *attr,
1033 				struct rte_flow_error *error)
1034 {
1035 	struct mlx5_priv *priv = dev->data->dev_private;
1036 	const struct rte_flow_action_queue *queue = action->conf;
1037 
1038 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1039 		return rte_flow_error_set(error, EINVAL,
1040 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1041 					  "can't have 2 fate actions in"
1042 					  " same flow");
1043 	if (!priv->rxqs_n)
1044 		return rte_flow_error_set(error, EINVAL,
1045 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1046 					  NULL, "No Rx queues configured");
1047 	if (queue->index >= priv->rxqs_n)
1048 		return rte_flow_error_set(error, EINVAL,
1049 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1050 					  &queue->index,
1051 					  "queue index out of range");
1052 	if (!(*priv->rxqs)[queue->index])
1053 		return rte_flow_error_set(error, EINVAL,
1054 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1055 					  &queue->index,
1056 					  "queue is not configured");
1057 	if (attr->egress)
1058 		return rte_flow_error_set(error, ENOTSUP,
1059 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1060 					  "queue action not supported for "
1061 					  "egress");
1062 	return 0;
1063 }
1064 
1065 /*
1066  * Validate the rss action.
1067  *
1068  * @param[in] action
1069  *   Pointer to the queue action.
1070  * @param[in] action_flags
1071  *   Bit-fields that holds the actions detected until now.
1072  * @param[in] dev
1073  *   Pointer to the Ethernet device structure.
1074  * @param[in] attr
1075  *   Attributes of flow that includes this action.
1076  * @param[in] item_flags
1077  *   Items that were detected.
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_rss(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 			      uint64_t item_flags,
1090 			      struct rte_flow_error *error)
1091 {
1092 	struct mlx5_priv *priv = dev->data->dev_private;
1093 	const struct rte_flow_action_rss *rss = action->conf;
1094 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1095 	unsigned int i;
1096 
1097 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1098 		return rte_flow_error_set(error, EINVAL,
1099 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1100 					  "can't have 2 fate actions"
1101 					  " in same flow");
1102 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1103 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1104 		return rte_flow_error_set(error, ENOTSUP,
1105 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1106 					  &rss->func,
1107 					  "RSS hash function not supported");
1108 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1109 	if (rss->level > 2)
1110 #else
1111 	if (rss->level > 1)
1112 #endif
1113 		return rte_flow_error_set(error, ENOTSUP,
1114 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1115 					  &rss->level,
1116 					  "tunnel RSS is not supported");
1117 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1118 	if (rss->key_len == 0 && rss->key != NULL)
1119 		return rte_flow_error_set(error, ENOTSUP,
1120 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1121 					  &rss->key_len,
1122 					  "RSS hash key length 0");
1123 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1124 		return rte_flow_error_set(error, ENOTSUP,
1125 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1126 					  &rss->key_len,
1127 					  "RSS hash key too small");
1128 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1129 		return rte_flow_error_set(error, ENOTSUP,
1130 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1131 					  &rss->key_len,
1132 					  "RSS hash key too large");
1133 	if (rss->queue_num > priv->config.ind_table_max_size)
1134 		return rte_flow_error_set(error, ENOTSUP,
1135 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136 					  &rss->queue_num,
1137 					  "number of queues too large");
1138 	if (rss->types & MLX5_RSS_HF_MASK)
1139 		return rte_flow_error_set(error, ENOTSUP,
1140 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1141 					  &rss->types,
1142 					  "some RSS protocols are not"
1143 					  " supported");
1144 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1145 	    !(rss->types & ETH_RSS_IP))
1146 		return rte_flow_error_set(error, EINVAL,
1147 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1148 					  "L3 partial RSS requested but L3 RSS"
1149 					  " type not specified");
1150 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1151 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1152 		return rte_flow_error_set(error, EINVAL,
1153 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1154 					  "L4 partial RSS requested but L4 RSS"
1155 					  " type not specified");
1156 	if (!priv->rxqs_n)
1157 		return rte_flow_error_set(error, EINVAL,
1158 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1159 					  NULL, "No Rx queues configured");
1160 	if (!rss->queue_num)
1161 		return rte_flow_error_set(error, EINVAL,
1162 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1163 					  NULL, "No queues configured");
1164 	for (i = 0; i != rss->queue_num; ++i) {
1165 		if (rss->queue[i] >= priv->rxqs_n)
1166 			return rte_flow_error_set
1167 				(error, EINVAL,
1168 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1169 				 &rss->queue[i], "queue index out of range");
1170 		if (!(*priv->rxqs)[rss->queue[i]])
1171 			return rte_flow_error_set
1172 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 				 &rss->queue[i], "queue is not configured");
1174 	}
1175 	if (attr->egress)
1176 		return rte_flow_error_set(error, ENOTSUP,
1177 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1178 					  "rss action not supported for "
1179 					  "egress");
1180 	if (rss->level > 1 &&  !tunnel)
1181 		return rte_flow_error_set(error, EINVAL,
1182 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1183 					  "inner RSS is not supported for "
1184 					  "non-tunnel flows");
1185 	return 0;
1186 }
1187 
1188 /*
1189  * Validate the count action.
1190  *
1191  * @param[in] dev
1192  *   Pointer to the Ethernet device structure.
1193  * @param[in] attr
1194  *   Attributes of flow that includes this action.
1195  * @param[out] error
1196  *   Pointer to error structure.
1197  *
1198  * @return
1199  *   0 on success, a negative errno value otherwise and rte_errno is set.
1200  */
1201 int
1202 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1203 				const struct rte_flow_attr *attr,
1204 				struct rte_flow_error *error)
1205 {
1206 	if (attr->egress)
1207 		return rte_flow_error_set(error, ENOTSUP,
1208 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1209 					  "count action not supported for "
1210 					  "egress");
1211 	return 0;
1212 }
1213 
1214 /**
1215  * Verify the @p attributes will be correctly understood by the NIC and store
1216  * them in the @p flow if everything is correct.
1217  *
1218  * @param[in] dev
1219  *   Pointer to the Ethernet device structure.
1220  * @param[in] attributes
1221  *   Pointer to flow attributes
1222  * @param[out] error
1223  *   Pointer to error structure.
1224  *
1225  * @return
1226  *   0 on success, a negative errno value otherwise and rte_errno is set.
1227  */
1228 int
1229 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1230 			      const struct rte_flow_attr *attributes,
1231 			      struct rte_flow_error *error)
1232 {
1233 	struct mlx5_priv *priv = dev->data->dev_private;
1234 	uint32_t priority_max = priv->config.flow_prio - 1;
1235 
1236 	if (attributes->group)
1237 		return rte_flow_error_set(error, ENOTSUP,
1238 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1239 					  NULL, "groups is not supported");
1240 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1241 	    attributes->priority >= priority_max)
1242 		return rte_flow_error_set(error, ENOTSUP,
1243 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1244 					  NULL, "priority out of range");
1245 	if (attributes->egress)
1246 		return rte_flow_error_set(error, ENOTSUP,
1247 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1248 					  "egress is not supported");
1249 	if (attributes->transfer && !priv->config.dv_esw_en)
1250 		return rte_flow_error_set(error, ENOTSUP,
1251 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1252 					  NULL, "transfer is not supported");
1253 	if (!attributes->ingress)
1254 		return rte_flow_error_set(error, EINVAL,
1255 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1256 					  NULL,
1257 					  "ingress attribute is mandatory");
1258 	return 0;
1259 }
1260 
1261 /**
1262  * Validate ICMP6 item.
1263  *
1264  * @param[in] item
1265  *   Item specification.
1266  * @param[in] item_flags
1267  *   Bit-fields that holds the items detected until now.
1268  * @param[out] error
1269  *   Pointer to error structure.
1270  *
1271  * @return
1272  *   0 on success, a negative errno value otherwise and rte_errno is set.
1273  */
1274 int
1275 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1276 			       uint64_t item_flags,
1277 			       uint8_t target_protocol,
1278 			       struct rte_flow_error *error)
1279 {
1280 	const struct rte_flow_item_icmp6 *mask = item->mask;
1281 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1282 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1283 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1284 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1285 				      MLX5_FLOW_LAYER_OUTER_L4;
1286 	int ret;
1287 
1288 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1289 		return rte_flow_error_set(error, EINVAL,
1290 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1291 					  "protocol filtering not compatible"
1292 					  " with ICMP6 layer");
1293 	if (!(item_flags & l3m))
1294 		return rte_flow_error_set(error, EINVAL,
1295 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1296 					  "IPv6 is mandatory to filter on"
1297 					  " ICMP6");
1298 	if (item_flags & l4m)
1299 		return rte_flow_error_set(error, EINVAL,
1300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1301 					  "multiple L4 layers not supported");
1302 	if (!mask)
1303 		mask = &rte_flow_item_icmp6_mask;
1304 	ret = mlx5_flow_item_acceptable
1305 		(item, (const uint8_t *)mask,
1306 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1307 		 sizeof(struct rte_flow_item_icmp6), error);
1308 	if (ret < 0)
1309 		return ret;
1310 	return 0;
1311 }
1312 
1313 /**
1314  * Validate ICMP item.
1315  *
1316  * @param[in] item
1317  *   Item specification.
1318  * @param[in] item_flags
1319  *   Bit-fields that holds the items detected until now.
1320  * @param[out] error
1321  *   Pointer to error structure.
1322  *
1323  * @return
1324  *   0 on success, a negative errno value otherwise and rte_errno is set.
1325  */
1326 int
1327 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1328 			     uint64_t item_flags,
1329 			     uint8_t target_protocol,
1330 			     struct rte_flow_error *error)
1331 {
1332 	const struct rte_flow_item_icmp *mask = item->mask;
1333 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1334 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1335 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1336 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1337 				      MLX5_FLOW_LAYER_OUTER_L4;
1338 	int ret;
1339 
1340 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1341 		return rte_flow_error_set(error, EINVAL,
1342 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1343 					  "protocol filtering not compatible"
1344 					  " with ICMP layer");
1345 	if (!(item_flags & l3m))
1346 		return rte_flow_error_set(error, EINVAL,
1347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1348 					  "IPv4 is mandatory to filter"
1349 					  " on ICMP");
1350 	if (item_flags & l4m)
1351 		return rte_flow_error_set(error, EINVAL,
1352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 					  "multiple L4 layers not supported");
1354 	if (!mask)
1355 		mask = &rte_flow_item_icmp_mask;
1356 	ret = mlx5_flow_item_acceptable
1357 		(item, (const uint8_t *)mask,
1358 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1359 		 sizeof(struct rte_flow_item_icmp), error);
1360 	if (ret < 0)
1361 		return ret;
1362 	return 0;
1363 }
1364 
1365 /**
1366  * Validate Ethernet item.
1367  *
1368  * @param[in] item
1369  *   Item specification.
1370  * @param[in] item_flags
1371  *   Bit-fields that holds the items detected until now.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1380 			    uint64_t item_flags,
1381 			    struct rte_flow_error *error)
1382 {
1383 	const struct rte_flow_item_eth *mask = item->mask;
1384 	const struct rte_flow_item_eth nic_mask = {
1385 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1386 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1387 		.type = RTE_BE16(0xffff),
1388 	};
1389 	int ret;
1390 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1391 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1392 				       MLX5_FLOW_LAYER_OUTER_L2;
1393 
1394 	if (item_flags & ethm)
1395 		return rte_flow_error_set(error, ENOTSUP,
1396 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1397 					  "multiple L2 layers not supported");
1398 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1399 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1400 		return rte_flow_error_set(error, EINVAL,
1401 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1402 					  "L2 layer should not follow "
1403 					  "L3 layers");
1404 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1405 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1406 		return rte_flow_error_set(error, EINVAL,
1407 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1408 					  "L2 layer should not follow VLAN");
1409 	if (!mask)
1410 		mask = &rte_flow_item_eth_mask;
1411 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1412 					(const uint8_t *)&nic_mask,
1413 					sizeof(struct rte_flow_item_eth),
1414 					error);
1415 	return ret;
1416 }
1417 
1418 /**
1419  * Validate VLAN item.
1420  *
1421  * @param[in] item
1422  *   Item specification.
1423  * @param[in] item_flags
1424  *   Bit-fields that holds the items detected until now.
1425  * @param[in] dev
1426  *   Ethernet device flow is being created on.
1427  * @param[out] error
1428  *   Pointer to error structure.
1429  *
1430  * @return
1431  *   0 on success, a negative errno value otherwise and rte_errno is set.
1432  */
1433 int
1434 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1435 			     uint64_t item_flags,
1436 			     struct rte_eth_dev *dev,
1437 			     struct rte_flow_error *error)
1438 {
1439 	const struct rte_flow_item_vlan *spec = item->spec;
1440 	const struct rte_flow_item_vlan *mask = item->mask;
1441 	const struct rte_flow_item_vlan nic_mask = {
1442 		.tci = RTE_BE16(UINT16_MAX),
1443 		.inner_type = RTE_BE16(UINT16_MAX),
1444 	};
1445 	uint16_t vlan_tag = 0;
1446 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1447 	int ret;
1448 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1449 					MLX5_FLOW_LAYER_INNER_L4) :
1450 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1451 					MLX5_FLOW_LAYER_OUTER_L4);
1452 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1453 					MLX5_FLOW_LAYER_OUTER_VLAN;
1454 
1455 	if (item_flags & vlanm)
1456 		return rte_flow_error_set(error, EINVAL,
1457 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1458 					  "multiple VLAN layers not supported");
1459 	else if ((item_flags & l34m) != 0)
1460 		return rte_flow_error_set(error, EINVAL,
1461 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1462 					  "VLAN cannot follow L3/L4 layer");
1463 	if (!mask)
1464 		mask = &rte_flow_item_vlan_mask;
1465 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1466 					(const uint8_t *)&nic_mask,
1467 					sizeof(struct rte_flow_item_vlan),
1468 					error);
1469 	if (ret)
1470 		return ret;
1471 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1472 		struct mlx5_priv *priv = dev->data->dev_private;
1473 
1474 		if (priv->vmwa_context) {
1475 			/*
1476 			 * Non-NULL context means we have a virtual machine
1477 			 * and SR-IOV enabled, we have to create VLAN interface
1478 			 * to make hypervisor to setup E-Switch vport
1479 			 * context correctly. We avoid creating the multiple
1480 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1481 			 */
1482 			return rte_flow_error_set(error, EINVAL,
1483 						  RTE_FLOW_ERROR_TYPE_ITEM,
1484 						  item,
1485 						  "VLAN tag mask is not"
1486 						  " supported in virtual"
1487 						  " environment");
1488 		}
1489 	}
1490 	if (spec) {
1491 		vlan_tag = spec->tci;
1492 		vlan_tag &= mask->tci;
1493 	}
1494 	/*
1495 	 * From verbs perspective an empty VLAN is equivalent
1496 	 * to a packet without VLAN layer.
1497 	 */
1498 	if (!vlan_tag)
1499 		return rte_flow_error_set(error, EINVAL,
1500 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1501 					  item->spec,
1502 					  "VLAN cannot be empty");
1503 	return 0;
1504 }
1505 
1506 /**
1507  * Validate IPV4 item.
1508  *
1509  * @param[in] item
1510  *   Item specification.
1511  * @param[in] item_flags
1512  *   Bit-fields that holds the items detected until now.
1513  * @param[in] acc_mask
1514  *   Acceptable mask, if NULL default internal default mask
1515  *   will be used to check whether item fields are supported.
1516  * @param[out] error
1517  *   Pointer to error structure.
1518  *
1519  * @return
1520  *   0 on success, a negative errno value otherwise and rte_errno is set.
1521  */
1522 int
1523 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1524 			     uint64_t item_flags,
1525 			     uint64_t last_item,
1526 			     uint16_t ether_type,
1527 			     const struct rte_flow_item_ipv4 *acc_mask,
1528 			     struct rte_flow_error *error)
1529 {
1530 	const struct rte_flow_item_ipv4 *mask = item->mask;
1531 	const struct rte_flow_item_ipv4 *spec = item->spec;
1532 	const struct rte_flow_item_ipv4 nic_mask = {
1533 		.hdr = {
1534 			.src_addr = RTE_BE32(0xffffffff),
1535 			.dst_addr = RTE_BE32(0xffffffff),
1536 			.type_of_service = 0xff,
1537 			.next_proto_id = 0xff,
1538 		},
1539 	};
1540 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1541 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1542 				      MLX5_FLOW_LAYER_OUTER_L3;
1543 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1544 				      MLX5_FLOW_LAYER_OUTER_L4;
1545 	int ret;
1546 	uint8_t next_proto = 0xFF;
1547 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1548 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1549 				  MLX5_FLOW_LAYER_INNER_VLAN);
1550 
1551 	if ((last_item & l2_vlan) && ether_type &&
1552 	    ether_type != RTE_ETHER_TYPE_IPV4)
1553 		return rte_flow_error_set(error, EINVAL,
1554 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1555 					  "IPv4 cannot follow L2/VLAN layer "
1556 					  "which ether type is not IPv4");
1557 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1558 		if (mask && spec)
1559 			next_proto = mask->hdr.next_proto_id &
1560 				     spec->hdr.next_proto_id;
1561 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1562 			return rte_flow_error_set(error, EINVAL,
1563 						  RTE_FLOW_ERROR_TYPE_ITEM,
1564 						  item,
1565 						  "multiple tunnel "
1566 						  "not supported");
1567 	}
1568 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1569 		return rte_flow_error_set(error, EINVAL,
1570 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1571 					  "wrong tunnel type - IPv6 specified "
1572 					  "but IPv4 item provided");
1573 	if (item_flags & l3m)
1574 		return rte_flow_error_set(error, ENOTSUP,
1575 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 					  "multiple L3 layers not supported");
1577 	else if (item_flags & l4m)
1578 		return rte_flow_error_set(error, EINVAL,
1579 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1580 					  "L3 cannot follow an L4 layer.");
1581 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1582 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1583 		return rte_flow_error_set(error, EINVAL,
1584 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1585 					  "L3 cannot follow an NVGRE layer.");
1586 	if (!mask)
1587 		mask = &rte_flow_item_ipv4_mask;
1588 	else if (mask->hdr.next_proto_id != 0 &&
1589 		 mask->hdr.next_proto_id != 0xff)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1592 					  "partial mask is not supported"
1593 					  " for protocol");
1594 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1595 					acc_mask ? (const uint8_t *)acc_mask
1596 						 : (const uint8_t *)&nic_mask,
1597 					sizeof(struct rte_flow_item_ipv4),
1598 					error);
1599 	if (ret < 0)
1600 		return ret;
1601 	return 0;
1602 }
1603 
1604 /**
1605  * Validate IPV6 item.
1606  *
1607  * @param[in] item
1608  *   Item specification.
1609  * @param[in] item_flags
1610  *   Bit-fields that holds the items detected until now.
1611  * @param[in] acc_mask
1612  *   Acceptable mask, if NULL default internal default mask
1613  *   will be used to check whether item fields are supported.
1614  * @param[out] error
1615  *   Pointer to error structure.
1616  *
1617  * @return
1618  *   0 on success, a negative errno value otherwise and rte_errno is set.
1619  */
1620 int
1621 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1622 			     uint64_t item_flags,
1623 			     uint64_t last_item,
1624 			     uint16_t ether_type,
1625 			     const struct rte_flow_item_ipv6 *acc_mask,
1626 			     struct rte_flow_error *error)
1627 {
1628 	const struct rte_flow_item_ipv6 *mask = item->mask;
1629 	const struct rte_flow_item_ipv6 *spec = item->spec;
1630 	const struct rte_flow_item_ipv6 nic_mask = {
1631 		.hdr = {
1632 			.src_addr =
1633 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1634 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1635 			.dst_addr =
1636 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1637 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1638 			.vtc_flow = RTE_BE32(0xffffffff),
1639 			.proto = 0xff,
1640 			.hop_limits = 0xff,
1641 		},
1642 	};
1643 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1644 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1645 				      MLX5_FLOW_LAYER_OUTER_L3;
1646 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1647 				      MLX5_FLOW_LAYER_OUTER_L4;
1648 	int ret;
1649 	uint8_t next_proto = 0xFF;
1650 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1651 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1652 				  MLX5_FLOW_LAYER_INNER_VLAN);
1653 
1654 	if ((last_item & l2_vlan) && ether_type &&
1655 	    ether_type != RTE_ETHER_TYPE_IPV6)
1656 		return rte_flow_error_set(error, EINVAL,
1657 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1658 					  "IPv6 cannot follow L2/VLAN layer "
1659 					  "which ether type is not IPv6");
1660 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1661 		if (mask && spec)
1662 			next_proto = mask->hdr.proto & spec->hdr.proto;
1663 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1664 			return rte_flow_error_set(error, EINVAL,
1665 						  RTE_FLOW_ERROR_TYPE_ITEM,
1666 						  item,
1667 						  "multiple tunnel "
1668 						  "not supported");
1669 	}
1670 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1671 		return rte_flow_error_set(error, EINVAL,
1672 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1673 					  "wrong tunnel type - IPv4 specified "
1674 					  "but IPv6 item provided");
1675 	if (item_flags & l3m)
1676 		return rte_flow_error_set(error, ENOTSUP,
1677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1678 					  "multiple L3 layers not supported");
1679 	else if (item_flags & l4m)
1680 		return rte_flow_error_set(error, EINVAL,
1681 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1682 					  "L3 cannot follow an L4 layer.");
1683 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1684 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1685 		return rte_flow_error_set(error, EINVAL,
1686 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1687 					  "L3 cannot follow an NVGRE layer.");
1688 	if (!mask)
1689 		mask = &rte_flow_item_ipv6_mask;
1690 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1691 					acc_mask ? (const uint8_t *)acc_mask
1692 						 : (const uint8_t *)&nic_mask,
1693 					sizeof(struct rte_flow_item_ipv6),
1694 					error);
1695 	if (ret < 0)
1696 		return ret;
1697 	return 0;
1698 }
1699 
1700 /**
1701  * Validate UDP item.
1702  *
1703  * @param[in] item
1704  *   Item specification.
1705  * @param[in] item_flags
1706  *   Bit-fields that holds the items detected until now.
1707  * @param[in] target_protocol
1708  *   The next protocol in the previous item.
1709  * @param[in] flow_mask
1710  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1711  * @param[out] error
1712  *   Pointer to error structure.
1713  *
1714  * @return
1715  *   0 on success, a negative errno value otherwise and rte_errno is set.
1716  */
1717 int
1718 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1719 			    uint64_t item_flags,
1720 			    uint8_t target_protocol,
1721 			    struct rte_flow_error *error)
1722 {
1723 	const struct rte_flow_item_udp *mask = item->mask;
1724 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1725 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1726 				      MLX5_FLOW_LAYER_OUTER_L3;
1727 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1728 				      MLX5_FLOW_LAYER_OUTER_L4;
1729 	int ret;
1730 
1731 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1732 		return rte_flow_error_set(error, EINVAL,
1733 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1734 					  "protocol filtering not compatible"
1735 					  " with UDP layer");
1736 	if (!(item_flags & l3m))
1737 		return rte_flow_error_set(error, EINVAL,
1738 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1739 					  "L3 is mandatory to filter on L4");
1740 	if (item_flags & l4m)
1741 		return rte_flow_error_set(error, EINVAL,
1742 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1743 					  "multiple L4 layers not supported");
1744 	if (!mask)
1745 		mask = &rte_flow_item_udp_mask;
1746 	ret = mlx5_flow_item_acceptable
1747 		(item, (const uint8_t *)mask,
1748 		 (const uint8_t *)&rte_flow_item_udp_mask,
1749 		 sizeof(struct rte_flow_item_udp), error);
1750 	if (ret < 0)
1751 		return ret;
1752 	return 0;
1753 }
1754 
1755 /**
1756  * Validate TCP 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[out] error
1765  *   Pointer to error structure.
1766  *
1767  * @return
1768  *   0 on success, a negative errno value otherwise and rte_errno is set.
1769  */
1770 int
1771 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1772 			    uint64_t item_flags,
1773 			    uint8_t target_protocol,
1774 			    const struct rte_flow_item_tcp *flow_mask,
1775 			    struct rte_flow_error *error)
1776 {
1777 	const struct rte_flow_item_tcp *mask = item->mask;
1778 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1779 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1780 				      MLX5_FLOW_LAYER_OUTER_L3;
1781 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1782 				      MLX5_FLOW_LAYER_OUTER_L4;
1783 	int ret;
1784 
1785 	assert(flow_mask);
1786 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1787 		return rte_flow_error_set(error, EINVAL,
1788 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1789 					  "protocol filtering not compatible"
1790 					  " with TCP 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_tcp_mask;
1801 	ret = mlx5_flow_item_acceptable
1802 		(item, (const uint8_t *)mask,
1803 		 (const uint8_t *)flow_mask,
1804 		 sizeof(struct rte_flow_item_tcp), error);
1805 	if (ret < 0)
1806 		return ret;
1807 	return 0;
1808 }
1809 
1810 /**
1811  * Validate VXLAN 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_vxlan(const struct rte_flow_item *item,
1827 			      uint64_t item_flags,
1828 			      struct rte_flow_error *error)
1829 {
1830 	const struct rte_flow_item_vxlan *spec = item->spec;
1831 	const struct rte_flow_item_vxlan *mask = item->mask;
1832 	int ret;
1833 	union vni {
1834 		uint32_t vlan_id;
1835 		uint8_t vni[4];
1836 	} id = { .vlan_id = 0, };
1837 	uint32_t vlan_id = 0;
1838 
1839 
1840 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1841 		return rte_flow_error_set(error, ENOTSUP,
1842 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1843 					  "multiple tunnel layers not"
1844 					  " supported");
1845 	/*
1846 	 * Verify only UDPv4 is present as defined in
1847 	 * https://tools.ietf.org/html/rfc7348
1848 	 */
1849 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1850 		return rte_flow_error_set(error, EINVAL,
1851 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1852 					  "no outer UDP layer found");
1853 	if (!mask)
1854 		mask = &rte_flow_item_vxlan_mask;
1855 	ret = mlx5_flow_item_acceptable
1856 		(item, (const uint8_t *)mask,
1857 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1858 		 sizeof(struct rte_flow_item_vxlan),
1859 		 error);
1860 	if (ret < 0)
1861 		return ret;
1862 	if (spec) {
1863 		memcpy(&id.vni[1], spec->vni, 3);
1864 		vlan_id = id.vlan_id;
1865 		memcpy(&id.vni[1], mask->vni, 3);
1866 		vlan_id &= id.vlan_id;
1867 	}
1868 	/*
1869 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1870 	 * only this layer is defined in the Verbs specification it is
1871 	 * interpreted as wildcard and all packets will match this
1872 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1873 	 * udp), all packets matching the layers before will also
1874 	 * match this rule.  To avoid such situation, VNI 0 is
1875 	 * currently refused.
1876 	 */
1877 	if (!vlan_id)
1878 		return rte_flow_error_set(error, ENOTSUP,
1879 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1880 					  "VXLAN vni cannot be 0");
1881 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1882 		return rte_flow_error_set(error, ENOTSUP,
1883 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1884 					  "VXLAN tunnel must be fully defined");
1885 	return 0;
1886 }
1887 
1888 /**
1889  * Validate VXLAN_GPE item.
1890  *
1891  * @param[in] item
1892  *   Item specification.
1893  * @param[in] item_flags
1894  *   Bit-fields that holds the items detected until now.
1895  * @param[in] priv
1896  *   Pointer to the private data structure.
1897  * @param[in] target_protocol
1898  *   The next protocol in the previous item.
1899  * @param[out] error
1900  *   Pointer to error structure.
1901  *
1902  * @return
1903  *   0 on success, a negative errno value otherwise and rte_errno is set.
1904  */
1905 int
1906 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1907 				  uint64_t item_flags,
1908 				  struct rte_eth_dev *dev,
1909 				  struct rte_flow_error *error)
1910 {
1911 	struct mlx5_priv *priv = dev->data->dev_private;
1912 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1913 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1914 	int ret;
1915 	union vni {
1916 		uint32_t vlan_id;
1917 		uint8_t vni[4];
1918 	} id = { .vlan_id = 0, };
1919 	uint32_t vlan_id = 0;
1920 
1921 	if (!priv->config.l3_vxlan_en)
1922 		return rte_flow_error_set(error, ENOTSUP,
1923 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1924 					  "L3 VXLAN is not enabled by device"
1925 					  " parameter and/or not configured in"
1926 					  " firmware");
1927 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1928 		return rte_flow_error_set(error, ENOTSUP,
1929 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1930 					  "multiple tunnel layers not"
1931 					  " supported");
1932 	/*
1933 	 * Verify only UDPv4 is present as defined in
1934 	 * https://tools.ietf.org/html/rfc7348
1935 	 */
1936 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1937 		return rte_flow_error_set(error, EINVAL,
1938 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1939 					  "no outer UDP layer found");
1940 	if (!mask)
1941 		mask = &rte_flow_item_vxlan_gpe_mask;
1942 	ret = mlx5_flow_item_acceptable
1943 		(item, (const uint8_t *)mask,
1944 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1945 		 sizeof(struct rte_flow_item_vxlan_gpe),
1946 		 error);
1947 	if (ret < 0)
1948 		return ret;
1949 	if (spec) {
1950 		if (spec->protocol)
1951 			return rte_flow_error_set(error, ENOTSUP,
1952 						  RTE_FLOW_ERROR_TYPE_ITEM,
1953 						  item,
1954 						  "VxLAN-GPE protocol"
1955 						  " not supported");
1956 		memcpy(&id.vni[1], spec->vni, 3);
1957 		vlan_id = id.vlan_id;
1958 		memcpy(&id.vni[1], mask->vni, 3);
1959 		vlan_id &= id.vlan_id;
1960 	}
1961 	/*
1962 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1963 	 * layer is defined in the Verbs specification it is interpreted as
1964 	 * wildcard and all packets will match this rule, if it follows a full
1965 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1966 	 * before will also match this rule.  To avoid such situation, VNI 0
1967 	 * is currently refused.
1968 	 */
1969 	if (!vlan_id)
1970 		return rte_flow_error_set(error, ENOTSUP,
1971 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1972 					  "VXLAN-GPE vni cannot be 0");
1973 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1974 		return rte_flow_error_set(error, ENOTSUP,
1975 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1976 					  "VXLAN-GPE tunnel must be fully"
1977 					  " defined");
1978 	return 0;
1979 }
1980 /**
1981  * Validate GRE Key item.
1982  *
1983  * @param[in] item
1984  *   Item specification.
1985  * @param[in] item_flags
1986  *   Bit flags to mark detected items.
1987  * @param[in] gre_item
1988  *   Pointer to gre_item
1989  * @param[out] error
1990  *   Pointer to error structure.
1991  *
1992  * @return
1993  *   0 on success, a negative errno value otherwise and rte_errno is set.
1994  */
1995 int
1996 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1997 				uint64_t item_flags,
1998 				const struct rte_flow_item *gre_item,
1999 				struct rte_flow_error *error)
2000 {
2001 	const rte_be32_t *mask = item->mask;
2002 	int ret = 0;
2003 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2004 	const struct rte_flow_item_gre *gre_spec;
2005 	const struct rte_flow_item_gre *gre_mask;
2006 
2007 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2008 		return rte_flow_error_set(error, ENOTSUP,
2009 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2010 					  "Multiple GRE key not support");
2011 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2012 		return rte_flow_error_set(error, ENOTSUP,
2013 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2014 					  "No preceding GRE header");
2015 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2016 		return rte_flow_error_set(error, ENOTSUP,
2017 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2018 					  "GRE key following a wrong item");
2019 	gre_mask = gre_item->mask;
2020 	if (!gre_mask)
2021 		gre_mask = &rte_flow_item_gre_mask;
2022 	gre_spec = gre_item->spec;
2023 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2024 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2025 		return rte_flow_error_set(error, EINVAL,
2026 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2027 					  "Key bit must be on");
2028 
2029 	if (!mask)
2030 		mask = &gre_key_default_mask;
2031 	ret = mlx5_flow_item_acceptable
2032 		(item, (const uint8_t *)mask,
2033 		 (const uint8_t *)&gre_key_default_mask,
2034 		 sizeof(rte_be32_t), error);
2035 	return ret;
2036 }
2037 
2038 /**
2039  * Validate GRE item.
2040  *
2041  * @param[in] item
2042  *   Item specification.
2043  * @param[in] item_flags
2044  *   Bit flags to mark detected items.
2045  * @param[in] target_protocol
2046  *   The next protocol in the previous item.
2047  * @param[out] error
2048  *   Pointer to error structure.
2049  *
2050  * @return
2051  *   0 on success, a negative errno value otherwise and rte_errno is set.
2052  */
2053 int
2054 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2055 			    uint64_t item_flags,
2056 			    uint8_t target_protocol,
2057 			    struct rte_flow_error *error)
2058 {
2059 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2060 	const struct rte_flow_item_gre *mask = item->mask;
2061 	int ret;
2062 	const struct rte_flow_item_gre nic_mask = {
2063 		.c_rsvd0_ver = RTE_BE16(0xB000),
2064 		.protocol = RTE_BE16(UINT16_MAX),
2065 	};
2066 
2067 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2068 		return rte_flow_error_set(error, EINVAL,
2069 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2070 					  "protocol filtering not compatible"
2071 					  " with this GRE layer");
2072 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2073 		return rte_flow_error_set(error, ENOTSUP,
2074 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2075 					  "multiple tunnel layers not"
2076 					  " supported");
2077 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2078 		return rte_flow_error_set(error, ENOTSUP,
2079 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2080 					  "L3 Layer is missing");
2081 	if (!mask)
2082 		mask = &rte_flow_item_gre_mask;
2083 	ret = mlx5_flow_item_acceptable
2084 		(item, (const uint8_t *)mask,
2085 		 (const uint8_t *)&nic_mask,
2086 		 sizeof(struct rte_flow_item_gre), error);
2087 	if (ret < 0)
2088 		return ret;
2089 #ifndef HAVE_MLX5DV_DR
2090 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2091 	if (spec && (spec->protocol & mask->protocol))
2092 		return rte_flow_error_set(error, ENOTSUP,
2093 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2094 					  "without MPLS support the"
2095 					  " specification cannot be used for"
2096 					  " filtering");
2097 #endif
2098 #endif
2099 	return 0;
2100 }
2101 
2102 /**
2103  * Validate Geneve item.
2104  *
2105  * @param[in] item
2106  *   Item specification.
2107  * @param[in] itemFlags
2108  *   Bit-fields that holds the items detected until now.
2109  * @param[in] enPriv
2110  *   Pointer to the private data structure.
2111  * @param[out] error
2112  *   Pointer to error structure.
2113  *
2114  * @return
2115  *   0 on success, a negative errno value otherwise and rte_errno is set.
2116  */
2117 
2118 int
2119 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2120 			       uint64_t item_flags,
2121 			       struct rte_eth_dev *dev,
2122 			       struct rte_flow_error *error)
2123 {
2124 	struct mlx5_priv *priv = dev->data->dev_private;
2125 	const struct rte_flow_item_geneve *spec = item->spec;
2126 	const struct rte_flow_item_geneve *mask = item->mask;
2127 	int ret;
2128 	uint16_t gbhdr;
2129 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2130 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2131 	const struct rte_flow_item_geneve nic_mask = {
2132 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2133 		.vni = "\xff\xff\xff",
2134 		.protocol = RTE_BE16(UINT16_MAX),
2135 	};
2136 
2137 	if (!(priv->config.hca_attr.flex_parser_protocols &
2138 	      MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2139 	    !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2140 		return rte_flow_error_set(error, ENOTSUP,
2141 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2142 					  "L3 Geneve is not enabled by device"
2143 					  " parameter and/or not configured in"
2144 					  " firmware");
2145 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2146 		return rte_flow_error_set(error, ENOTSUP,
2147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2148 					  "multiple tunnel layers not"
2149 					  " supported");
2150 	/*
2151 	 * Verify only UDPv4 is present as defined in
2152 	 * https://tools.ietf.org/html/rfc7348
2153 	 */
2154 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2155 		return rte_flow_error_set(error, EINVAL,
2156 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 					  "no outer UDP layer found");
2158 	if (!mask)
2159 		mask = &rte_flow_item_geneve_mask;
2160 	ret = mlx5_flow_item_acceptable
2161 				  (item, (const uint8_t *)mask,
2162 				   (const uint8_t *)&nic_mask,
2163 				   sizeof(struct rte_flow_item_geneve), error);
2164 	if (ret)
2165 		return ret;
2166 	if (spec) {
2167 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2168 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2169 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2170 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2171 			return rte_flow_error_set(error, ENOTSUP,
2172 						  RTE_FLOW_ERROR_TYPE_ITEM,
2173 						  item,
2174 						  "Geneve protocol unsupported"
2175 						  " fields are being used");
2176 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2177 			return rte_flow_error_set
2178 					(error, ENOTSUP,
2179 					 RTE_FLOW_ERROR_TYPE_ITEM,
2180 					 item,
2181 					 "Unsupported Geneve options length");
2182 	}
2183 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2184 		return rte_flow_error_set
2185 				    (error, ENOTSUP,
2186 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2187 				     "Geneve tunnel must be fully defined");
2188 	return 0;
2189 }
2190 
2191 /**
2192  * Validate MPLS item.
2193  *
2194  * @param[in] dev
2195  *   Pointer to the rte_eth_dev structure.
2196  * @param[in] item
2197  *   Item specification.
2198  * @param[in] item_flags
2199  *   Bit-fields that holds the items detected until now.
2200  * @param[in] prev_layer
2201  *   The protocol layer indicated in previous item.
2202  * @param[out] error
2203  *   Pointer to error structure.
2204  *
2205  * @return
2206  *   0 on success, a negative errno value otherwise and rte_errno is set.
2207  */
2208 int
2209 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2210 			     const struct rte_flow_item *item __rte_unused,
2211 			     uint64_t item_flags __rte_unused,
2212 			     uint64_t prev_layer __rte_unused,
2213 			     struct rte_flow_error *error)
2214 {
2215 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2216 	const struct rte_flow_item_mpls *mask = item->mask;
2217 	struct mlx5_priv *priv = dev->data->dev_private;
2218 	int ret;
2219 
2220 	if (!priv->config.mpls_en)
2221 		return rte_flow_error_set(error, ENOTSUP,
2222 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2223 					  "MPLS not supported or"
2224 					  " disabled in firmware"
2225 					  " configuration.");
2226 	/* MPLS over IP, UDP, GRE is allowed */
2227 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2228 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2229 			    MLX5_FLOW_LAYER_GRE)))
2230 		return rte_flow_error_set(error, EINVAL,
2231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2232 					  "protocol filtering not compatible"
2233 					  " with MPLS layer");
2234 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2235 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2236 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2237 		return rte_flow_error_set(error, ENOTSUP,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "multiple tunnel layers not"
2240 					  " supported");
2241 	if (!mask)
2242 		mask = &rte_flow_item_mpls_mask;
2243 	ret = mlx5_flow_item_acceptable
2244 		(item, (const uint8_t *)mask,
2245 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2246 		 sizeof(struct rte_flow_item_mpls), error);
2247 	if (ret < 0)
2248 		return ret;
2249 	return 0;
2250 #endif
2251 	return rte_flow_error_set(error, ENOTSUP,
2252 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2253 				  "MPLS is not supported by Verbs, please"
2254 				  " update.");
2255 }
2256 
2257 /**
2258  * Validate NVGRE item.
2259  *
2260  * @param[in] item
2261  *   Item specification.
2262  * @param[in] item_flags
2263  *   Bit flags to mark detected items.
2264  * @param[in] target_protocol
2265  *   The next protocol in the previous item.
2266  * @param[out] error
2267  *   Pointer to error structure.
2268  *
2269  * @return
2270  *   0 on success, a negative errno value otherwise and rte_errno is set.
2271  */
2272 int
2273 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2274 			      uint64_t item_flags,
2275 			      uint8_t target_protocol,
2276 			      struct rte_flow_error *error)
2277 {
2278 	const struct rte_flow_item_nvgre *mask = item->mask;
2279 	int ret;
2280 
2281 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2282 		return rte_flow_error_set(error, EINVAL,
2283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2284 					  "protocol filtering not compatible"
2285 					  " with this GRE layer");
2286 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2287 		return rte_flow_error_set(error, ENOTSUP,
2288 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2289 					  "multiple tunnel layers not"
2290 					  " supported");
2291 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2292 		return rte_flow_error_set(error, ENOTSUP,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "L3 Layer is missing");
2295 	if (!mask)
2296 		mask = &rte_flow_item_nvgre_mask;
2297 	ret = mlx5_flow_item_acceptable
2298 		(item, (const uint8_t *)mask,
2299 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2300 		 sizeof(struct rte_flow_item_nvgre), error);
2301 	if (ret < 0)
2302 		return ret;
2303 	return 0;
2304 }
2305 
2306 /* Allocate unique ID for the split Q/RSS subflows. */
2307 static uint32_t
2308 flow_qrss_get_id(struct rte_eth_dev *dev)
2309 {
2310 	struct mlx5_priv *priv = dev->data->dev_private;
2311 	uint32_t qrss_id, ret;
2312 
2313 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2314 	if (ret)
2315 		return 0;
2316 	assert(qrss_id);
2317 	return qrss_id;
2318 }
2319 
2320 /* Free unique ID for the split Q/RSS subflows. */
2321 static void
2322 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2323 {
2324 	struct mlx5_priv *priv = dev->data->dev_private;
2325 
2326 	if (qrss_id)
2327 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2328 }
2329 
2330 /**
2331  * Release resource related QUEUE/RSS action split.
2332  *
2333  * @param dev
2334  *   Pointer to Ethernet device.
2335  * @param flow
2336  *   Flow to release id's from.
2337  */
2338 static void
2339 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2340 			     struct rte_flow *flow)
2341 {
2342 	struct mlx5_flow *dev_flow;
2343 
2344 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2345 		if (dev_flow->qrss_id)
2346 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2347 }
2348 
2349 static int
2350 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2351 		   const struct rte_flow_attr *attr __rte_unused,
2352 		   const struct rte_flow_item items[] __rte_unused,
2353 		   const struct rte_flow_action actions[] __rte_unused,
2354 		   bool external __rte_unused,
2355 		   struct rte_flow_error *error)
2356 {
2357 	return rte_flow_error_set(error, ENOTSUP,
2358 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2359 }
2360 
2361 static struct mlx5_flow *
2362 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2363 		  const struct rte_flow_item items[] __rte_unused,
2364 		  const struct rte_flow_action actions[] __rte_unused,
2365 		  struct rte_flow_error *error)
2366 {
2367 	rte_flow_error_set(error, ENOTSUP,
2368 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2369 	return NULL;
2370 }
2371 
2372 static int
2373 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2374 		    struct mlx5_flow *dev_flow __rte_unused,
2375 		    const struct rte_flow_attr *attr __rte_unused,
2376 		    const struct rte_flow_item items[] __rte_unused,
2377 		    const struct rte_flow_action actions[] __rte_unused,
2378 		    struct rte_flow_error *error)
2379 {
2380 	return rte_flow_error_set(error, ENOTSUP,
2381 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2382 }
2383 
2384 static int
2385 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2386 		struct rte_flow *flow __rte_unused,
2387 		struct rte_flow_error *error)
2388 {
2389 	return rte_flow_error_set(error, ENOTSUP,
2390 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2391 }
2392 
2393 static void
2394 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2395 		 struct rte_flow *flow __rte_unused)
2396 {
2397 }
2398 
2399 static void
2400 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2401 		  struct rte_flow *flow __rte_unused)
2402 {
2403 }
2404 
2405 static int
2406 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2407 		struct rte_flow *flow __rte_unused,
2408 		const struct rte_flow_action *actions __rte_unused,
2409 		void *data __rte_unused,
2410 		struct rte_flow_error *error)
2411 {
2412 	return rte_flow_error_set(error, ENOTSUP,
2413 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2414 }
2415 
2416 /* Void driver to protect from null pointer reference. */
2417 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2418 	.validate = flow_null_validate,
2419 	.prepare = flow_null_prepare,
2420 	.translate = flow_null_translate,
2421 	.apply = flow_null_apply,
2422 	.remove = flow_null_remove,
2423 	.destroy = flow_null_destroy,
2424 	.query = flow_null_query,
2425 };
2426 
2427 /**
2428  * Select flow driver type according to flow attributes and device
2429  * configuration.
2430  *
2431  * @param[in] dev
2432  *   Pointer to the dev structure.
2433  * @param[in] attr
2434  *   Pointer to the flow attributes.
2435  *
2436  * @return
2437  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2438  */
2439 static enum mlx5_flow_drv_type
2440 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2441 {
2442 	struct mlx5_priv *priv = dev->data->dev_private;
2443 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2444 
2445 	if (attr->transfer && priv->config.dv_esw_en)
2446 		type = MLX5_FLOW_TYPE_DV;
2447 	if (!attr->transfer)
2448 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2449 						 MLX5_FLOW_TYPE_VERBS;
2450 	return type;
2451 }
2452 
2453 #define flow_get_drv_ops(type) flow_drv_ops[type]
2454 
2455 /**
2456  * Flow driver validation API. This abstracts calling driver specific functions.
2457  * The type of flow driver is determined according to flow attributes.
2458  *
2459  * @param[in] dev
2460  *   Pointer to the dev structure.
2461  * @param[in] attr
2462  *   Pointer to the flow attributes.
2463  * @param[in] items
2464  *   Pointer to the list of items.
2465  * @param[in] actions
2466  *   Pointer to the list of actions.
2467  * @param[in] external
2468  *   This flow rule is created by request external to PMD.
2469  * @param[out] error
2470  *   Pointer to the error structure.
2471  *
2472  * @return
2473  *   0 on success, a negative errno value otherwise and rte_errno is set.
2474  */
2475 static inline int
2476 flow_drv_validate(struct rte_eth_dev *dev,
2477 		  const struct rte_flow_attr *attr,
2478 		  const struct rte_flow_item items[],
2479 		  const struct rte_flow_action actions[],
2480 		  bool external, struct rte_flow_error *error)
2481 {
2482 	const struct mlx5_flow_driver_ops *fops;
2483 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2484 
2485 	fops = flow_get_drv_ops(type);
2486 	return fops->validate(dev, attr, items, actions, external, error);
2487 }
2488 
2489 /**
2490  * Flow driver preparation API. This abstracts calling driver specific
2491  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2492  * calculates the size of memory required for device flow, allocates the memory,
2493  * initializes the device flow and returns the pointer.
2494  *
2495  * @note
2496  *   This function initializes device flow structure such as dv or verbs in
2497  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2498  *   rest. For example, adding returning device flow to flow->dev_flow list and
2499  *   setting backward reference to the flow should be done out of this function.
2500  *   layers field is not filled either.
2501  *
2502  * @param[in] attr
2503  *   Pointer to the flow attributes.
2504  * @param[in] items
2505  *   Pointer to the list of items.
2506  * @param[in] actions
2507  *   Pointer to the list of actions.
2508  * @param[out] error
2509  *   Pointer to the error structure.
2510  *
2511  * @return
2512  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2513  */
2514 static inline struct mlx5_flow *
2515 flow_drv_prepare(const struct rte_flow *flow,
2516 		 const struct rte_flow_attr *attr,
2517 		 const struct rte_flow_item items[],
2518 		 const struct rte_flow_action actions[],
2519 		 struct rte_flow_error *error)
2520 {
2521 	const struct mlx5_flow_driver_ops *fops;
2522 	enum mlx5_flow_drv_type type = flow->drv_type;
2523 
2524 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2525 	fops = flow_get_drv_ops(type);
2526 	return fops->prepare(attr, items, actions, error);
2527 }
2528 
2529 /**
2530  * Flow driver translation API. This abstracts calling driver specific
2531  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2532  * translates a generic flow into a driver flow. flow_drv_prepare() must
2533  * precede.
2534  *
2535  * @note
2536  *   dev_flow->layers could be filled as a result of parsing during translation
2537  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2538  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2539  *   flow->actions could be overwritten even though all the expanded dev_flows
2540  *   have the same actions.
2541  *
2542  * @param[in] dev
2543  *   Pointer to the rte dev structure.
2544  * @param[in, out] dev_flow
2545  *   Pointer to the mlx5 flow.
2546  * @param[in] attr
2547  *   Pointer to the flow attributes.
2548  * @param[in] items
2549  *   Pointer to the list of items.
2550  * @param[in] actions
2551  *   Pointer to the list of actions.
2552  * @param[out] error
2553  *   Pointer to the error structure.
2554  *
2555  * @return
2556  *   0 on success, a negative errno value otherwise and rte_errno is set.
2557  */
2558 static inline int
2559 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2560 		   const struct rte_flow_attr *attr,
2561 		   const struct rte_flow_item items[],
2562 		   const struct rte_flow_action actions[],
2563 		   struct rte_flow_error *error)
2564 {
2565 	const struct mlx5_flow_driver_ops *fops;
2566 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2567 
2568 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2569 	fops = flow_get_drv_ops(type);
2570 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2571 }
2572 
2573 /**
2574  * Flow driver apply API. This abstracts calling driver specific functions.
2575  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2576  * translated driver flows on to device. flow_drv_translate() must precede.
2577  *
2578  * @param[in] dev
2579  *   Pointer to Ethernet device structure.
2580  * @param[in, out] flow
2581  *   Pointer to flow structure.
2582  * @param[out] error
2583  *   Pointer to error structure.
2584  *
2585  * @return
2586  *   0 on success, a negative errno value otherwise and rte_errno is set.
2587  */
2588 static inline int
2589 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2590 	       struct rte_flow_error *error)
2591 {
2592 	const struct mlx5_flow_driver_ops *fops;
2593 	enum mlx5_flow_drv_type type = flow->drv_type;
2594 
2595 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2596 	fops = flow_get_drv_ops(type);
2597 	return fops->apply(dev, flow, error);
2598 }
2599 
2600 /**
2601  * Flow driver remove API. This abstracts calling driver specific functions.
2602  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2603  * on device. All the resources of the flow should be freed by calling
2604  * flow_drv_destroy().
2605  *
2606  * @param[in] dev
2607  *   Pointer to Ethernet device.
2608  * @param[in, out] flow
2609  *   Pointer to flow structure.
2610  */
2611 static inline void
2612 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2613 {
2614 	const struct mlx5_flow_driver_ops *fops;
2615 	enum mlx5_flow_drv_type type = flow->drv_type;
2616 
2617 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2618 	fops = flow_get_drv_ops(type);
2619 	fops->remove(dev, flow);
2620 }
2621 
2622 /**
2623  * Flow driver destroy API. This abstracts calling driver specific functions.
2624  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2625  * on device and releases resources of the flow.
2626  *
2627  * @param[in] dev
2628  *   Pointer to Ethernet device.
2629  * @param[in, out] flow
2630  *   Pointer to flow structure.
2631  */
2632 static inline void
2633 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2634 {
2635 	const struct mlx5_flow_driver_ops *fops;
2636 	enum mlx5_flow_drv_type type = flow->drv_type;
2637 
2638 	flow_mreg_split_qrss_release(dev, flow);
2639 	assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2640 	fops = flow_get_drv_ops(type);
2641 	fops->destroy(dev, flow);
2642 }
2643 
2644 /**
2645  * Validate a flow supported by the NIC.
2646  *
2647  * @see rte_flow_validate()
2648  * @see rte_flow_ops
2649  */
2650 int
2651 mlx5_flow_validate(struct rte_eth_dev *dev,
2652 		   const struct rte_flow_attr *attr,
2653 		   const struct rte_flow_item items[],
2654 		   const struct rte_flow_action actions[],
2655 		   struct rte_flow_error *error)
2656 {
2657 	int ret;
2658 
2659 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2660 	if (ret < 0)
2661 		return ret;
2662 	return 0;
2663 }
2664 
2665 /**
2666  * Get port id item from the item list.
2667  *
2668  * @param[in] item
2669  *   Pointer to the list of items.
2670  *
2671  * @return
2672  *   Pointer to the port id item if exist, else return NULL.
2673  */
2674 static const struct rte_flow_item *
2675 find_port_id_item(const struct rte_flow_item *item)
2676 {
2677 	assert(item);
2678 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2679 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2680 			return item;
2681 	}
2682 	return NULL;
2683 }
2684 
2685 /**
2686  * Get RSS action from the action list.
2687  *
2688  * @param[in] actions
2689  *   Pointer to the list of actions.
2690  *
2691  * @return
2692  *   Pointer to the RSS action if exist, else return NULL.
2693  */
2694 static const struct rte_flow_action_rss*
2695 flow_get_rss_action(const struct rte_flow_action actions[])
2696 {
2697 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2698 		switch (actions->type) {
2699 		case RTE_FLOW_ACTION_TYPE_RSS:
2700 			return (const struct rte_flow_action_rss *)
2701 			       actions->conf;
2702 		default:
2703 			break;
2704 		}
2705 	}
2706 	return NULL;
2707 }
2708 
2709 static unsigned int
2710 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2711 {
2712 	const struct rte_flow_item *item;
2713 	unsigned int has_vlan = 0;
2714 
2715 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2716 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2717 			has_vlan = 1;
2718 			break;
2719 		}
2720 	}
2721 	if (has_vlan)
2722 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2723 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2724 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2725 			       MLX5_EXPANSION_ROOT_OUTER;
2726 }
2727 
2728 /**
2729  * Get QUEUE/RSS action from the action list.
2730  *
2731  * @param[in] actions
2732  *   Pointer to the list of actions.
2733  * @param[out] qrss
2734  *   Pointer to the return pointer.
2735  * @param[out] qrss_type
2736  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2737  *   if no QUEUE/RSS is found.
2738  *
2739  * @return
2740  *   Total number of actions.
2741  */
2742 static int
2743 flow_parse_qrss_action(const struct rte_flow_action actions[],
2744 		       const struct rte_flow_action **qrss)
2745 {
2746 	int actions_n = 0;
2747 
2748 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2749 		switch (actions->type) {
2750 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2751 		case RTE_FLOW_ACTION_TYPE_RSS:
2752 			*qrss = actions;
2753 			break;
2754 		default:
2755 			break;
2756 		}
2757 		actions_n++;
2758 	}
2759 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2760 	return actions_n + 1;
2761 }
2762 
2763 /**
2764  * Check meter action from the action list.
2765  *
2766  * @param[in] actions
2767  *   Pointer to the list of actions.
2768  * @param[out] mtr
2769  *   Pointer to the meter exist flag.
2770  *
2771  * @return
2772  *   Total number of actions.
2773  */
2774 static int
2775 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2776 {
2777 	int actions_n = 0;
2778 
2779 	assert(mtr);
2780 	*mtr = 0;
2781 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2782 		switch (actions->type) {
2783 		case RTE_FLOW_ACTION_TYPE_METER:
2784 			*mtr = 1;
2785 			break;
2786 		default:
2787 			break;
2788 		}
2789 		actions_n++;
2790 	}
2791 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2792 	return actions_n + 1;
2793 }
2794 
2795 /**
2796  * Check if the flow should be splited due to hairpin.
2797  * The reason for the split is that in current HW we can't
2798  * support encap on Rx, so if a flow have encap we move it
2799  * to Tx.
2800  *
2801  * @param dev
2802  *   Pointer to Ethernet device.
2803  * @param[in] attr
2804  *   Flow rule attributes.
2805  * @param[in] actions
2806  *   Associated actions (list terminated by the END action).
2807  *
2808  * @return
2809  *   > 0 the number of actions and the flow should be split,
2810  *   0 when no split required.
2811  */
2812 static int
2813 flow_check_hairpin_split(struct rte_eth_dev *dev,
2814 			 const struct rte_flow_attr *attr,
2815 			 const struct rte_flow_action actions[])
2816 {
2817 	int queue_action = 0;
2818 	int action_n = 0;
2819 	int encap = 0;
2820 	const struct rte_flow_action_queue *queue;
2821 	const struct rte_flow_action_rss *rss;
2822 	const struct rte_flow_action_raw_encap *raw_encap;
2823 
2824 	if (!attr->ingress)
2825 		return 0;
2826 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2827 		switch (actions->type) {
2828 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2829 			queue = actions->conf;
2830 			if (queue == NULL)
2831 				return 0;
2832 			if (mlx5_rxq_get_type(dev, queue->index) !=
2833 			    MLX5_RXQ_TYPE_HAIRPIN)
2834 				return 0;
2835 			queue_action = 1;
2836 			action_n++;
2837 			break;
2838 		case RTE_FLOW_ACTION_TYPE_RSS:
2839 			rss = actions->conf;
2840 			if (rss == NULL || rss->queue_num == 0)
2841 				return 0;
2842 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2843 			    MLX5_RXQ_TYPE_HAIRPIN)
2844 				return 0;
2845 			queue_action = 1;
2846 			action_n++;
2847 			break;
2848 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2849 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2850 			encap = 1;
2851 			action_n++;
2852 			break;
2853 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2854 			raw_encap = actions->conf;
2855 			if (raw_encap->size >
2856 			    (sizeof(struct rte_flow_item_eth) +
2857 			     sizeof(struct rte_flow_item_ipv4)))
2858 				encap = 1;
2859 			action_n++;
2860 			break;
2861 		default:
2862 			action_n++;
2863 			break;
2864 		}
2865 	}
2866 	if (encap == 1 && queue_action)
2867 		return action_n;
2868 	return 0;
2869 }
2870 
2871 /* Declare flow create/destroy prototype in advance. */
2872 static struct rte_flow *
2873 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2874 		 const struct rte_flow_attr *attr,
2875 		 const struct rte_flow_item items[],
2876 		 const struct rte_flow_action actions[],
2877 		 bool external, struct rte_flow_error *error);
2878 
2879 static void
2880 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2881 		  struct rte_flow *flow);
2882 
2883 /**
2884  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2885  *
2886  * As mark_id is unique, if there's already a registered flow for the mark_id,
2887  * return by increasing the reference counter of the resource. Otherwise, create
2888  * the resource (mcp_res) and flow.
2889  *
2890  * Flow looks like,
2891  *   - If ingress port is ANY and reg_c[1] is mark_id,
2892  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2893  *
2894  * For default flow (zero mark_id), flow is like,
2895  *   - If ingress port is ANY,
2896  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2897  *
2898  * @param dev
2899  *   Pointer to Ethernet device.
2900  * @param mark_id
2901  *   ID of MARK action, zero means default flow for META.
2902  * @param[out] error
2903  *   Perform verbose error reporting if not NULL.
2904  *
2905  * @return
2906  *   Associated resource on success, NULL otherwise and rte_errno is set.
2907  */
2908 static struct mlx5_flow_mreg_copy_resource *
2909 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2910 			  struct rte_flow_error *error)
2911 {
2912 	struct mlx5_priv *priv = dev->data->dev_private;
2913 	struct rte_flow_attr attr = {
2914 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2915 		.ingress = 1,
2916 	};
2917 	struct mlx5_rte_flow_item_tag tag_spec = {
2918 		.data = mark_id,
2919 	};
2920 	struct rte_flow_item items[] = {
2921 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2922 	};
2923 	struct rte_flow_action_mark ftag = {
2924 		.id = mark_id,
2925 	};
2926 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2927 		.dst = REG_B,
2928 		.src = 0,
2929 	};
2930 	struct rte_flow_action_jump jump = {
2931 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2932 	};
2933 	struct rte_flow_action actions[] = {
2934 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2935 	};
2936 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2937 	int ret;
2938 
2939 	/* Fill the register fileds in the flow. */
2940 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2941 	if (ret < 0)
2942 		return NULL;
2943 	tag_spec.id = ret;
2944 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2945 	if (ret < 0)
2946 		return NULL;
2947 	cp_mreg.src = ret;
2948 	/* Check if already registered. */
2949 	assert(priv->mreg_cp_tbl);
2950 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2951 	if (mcp_res) {
2952 		/* For non-default rule. */
2953 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2954 			mcp_res->refcnt++;
2955 		assert(mark_id != MLX5_DEFAULT_COPY_ID || mcp_res->refcnt == 1);
2956 		return mcp_res;
2957 	}
2958 	/* Provide the full width of FLAG specific value. */
2959 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2960 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2961 	/* Build a new flow. */
2962 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
2963 		items[0] = (struct rte_flow_item){
2964 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2965 			.spec = &tag_spec,
2966 		};
2967 		items[1] = (struct rte_flow_item){
2968 			.type = RTE_FLOW_ITEM_TYPE_END,
2969 		};
2970 		actions[0] = (struct rte_flow_action){
2971 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2972 			.conf = &ftag,
2973 		};
2974 		actions[1] = (struct rte_flow_action){
2975 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2976 			.conf = &cp_mreg,
2977 		};
2978 		actions[2] = (struct rte_flow_action){
2979 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2980 			.conf = &jump,
2981 		};
2982 		actions[3] = (struct rte_flow_action){
2983 			.type = RTE_FLOW_ACTION_TYPE_END,
2984 		};
2985 	} else {
2986 		/* Default rule, wildcard match. */
2987 		attr.priority = MLX5_FLOW_PRIO_RSVD;
2988 		items[0] = (struct rte_flow_item){
2989 			.type = RTE_FLOW_ITEM_TYPE_END,
2990 		};
2991 		actions[0] = (struct rte_flow_action){
2992 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2993 			.conf = &cp_mreg,
2994 		};
2995 		actions[1] = (struct rte_flow_action){
2996 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
2997 			.conf = &jump,
2998 		};
2999 		actions[2] = (struct rte_flow_action){
3000 			.type = RTE_FLOW_ACTION_TYPE_END,
3001 		};
3002 	}
3003 	/* Build a new entry. */
3004 	mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3005 	if (!mcp_res) {
3006 		rte_errno = ENOMEM;
3007 		return NULL;
3008 	}
3009 	/*
3010 	 * The copy Flows are not included in any list. There
3011 	 * ones are referenced from other Flows and can not
3012 	 * be applied, removed, deleted in ardbitrary order
3013 	 * by list traversing.
3014 	 */
3015 	mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3016 					 actions, false, error);
3017 	if (!mcp_res->flow)
3018 		goto error;
3019 	mcp_res->refcnt++;
3020 	mcp_res->hlist_ent.key = mark_id;
3021 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3022 				&mcp_res->hlist_ent);
3023 	assert(!ret);
3024 	if (ret)
3025 		goto error;
3026 	return mcp_res;
3027 error:
3028 	if (mcp_res->flow)
3029 		flow_list_destroy(dev, NULL, mcp_res->flow);
3030 	rte_free(mcp_res);
3031 	return NULL;
3032 }
3033 
3034 /**
3035  * Release flow in RX_CP_TBL.
3036  *
3037  * @param dev
3038  *   Pointer to Ethernet device.
3039  * @flow
3040  *   Parent flow for wich copying is provided.
3041  */
3042 static void
3043 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3044 			  struct rte_flow *flow)
3045 {
3046 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3047 	struct mlx5_priv *priv = dev->data->dev_private;
3048 
3049 	if (!mcp_res || !priv->mreg_cp_tbl)
3050 		return;
3051 	if (flow->copy_applied) {
3052 		assert(mcp_res->appcnt);
3053 		flow->copy_applied = 0;
3054 		--mcp_res->appcnt;
3055 		if (!mcp_res->appcnt)
3056 			flow_drv_remove(dev, mcp_res->flow);
3057 	}
3058 	/*
3059 	 * We do not check availability of metadata registers here,
3060 	 * because copy resources are not allocated in this case.
3061 	 */
3062 	if (--mcp_res->refcnt)
3063 		return;
3064 	assert(mcp_res->flow);
3065 	flow_list_destroy(dev, NULL, mcp_res->flow);
3066 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3067 	rte_free(mcp_res);
3068 	flow->mreg_copy = NULL;
3069 }
3070 
3071 /**
3072  * Start flow in RX_CP_TBL.
3073  *
3074  * @param dev
3075  *   Pointer to Ethernet device.
3076  * @flow
3077  *   Parent flow for wich copying is provided.
3078  *
3079  * @return
3080  *   0 on success, a negative errno value otherwise and rte_errno is set.
3081  */
3082 static int
3083 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3084 			    struct rte_flow *flow)
3085 {
3086 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3087 	int ret;
3088 
3089 	if (!mcp_res || flow->copy_applied)
3090 		return 0;
3091 	if (!mcp_res->appcnt) {
3092 		ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3093 		if (ret)
3094 			return ret;
3095 	}
3096 	++mcp_res->appcnt;
3097 	flow->copy_applied = 1;
3098 	return 0;
3099 }
3100 
3101 /**
3102  * Stop flow in RX_CP_TBL.
3103  *
3104  * @param dev
3105  *   Pointer to Ethernet device.
3106  * @flow
3107  *   Parent flow for wich copying is provided.
3108  */
3109 static void
3110 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3111 			   struct rte_flow *flow)
3112 {
3113 	struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3114 
3115 	if (!mcp_res || !flow->copy_applied)
3116 		return;
3117 	assert(mcp_res->appcnt);
3118 	--mcp_res->appcnt;
3119 	flow->copy_applied = 0;
3120 	if (!mcp_res->appcnt)
3121 		flow_drv_remove(dev, mcp_res->flow);
3122 }
3123 
3124 /**
3125  * Remove the default copy action from RX_CP_TBL.
3126  *
3127  * @param dev
3128  *   Pointer to Ethernet device.
3129  */
3130 static void
3131 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3132 {
3133 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3134 	struct mlx5_priv *priv = dev->data->dev_private;
3135 
3136 	/* Check if default flow is registered. */
3137 	if (!priv->mreg_cp_tbl)
3138 		return;
3139 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3140 					    MLX5_DEFAULT_COPY_ID);
3141 	if (!mcp_res)
3142 		return;
3143 	assert(mcp_res->flow);
3144 	flow_list_destroy(dev, NULL, mcp_res->flow);
3145 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3146 	rte_free(mcp_res);
3147 }
3148 
3149 /**
3150  * Add the default copy action in in RX_CP_TBL.
3151  *
3152  * @param dev
3153  *   Pointer to Ethernet device.
3154  * @param[out] error
3155  *   Perform verbose error reporting if not NULL.
3156  *
3157  * @return
3158  *   0 for success, negative value otherwise and rte_errno is set.
3159  */
3160 static int
3161 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3162 				  struct rte_flow_error *error)
3163 {
3164 	struct mlx5_priv *priv = dev->data->dev_private;
3165 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3166 
3167 	/* Check whether extensive metadata feature is engaged. */
3168 	if (!priv->config.dv_flow_en ||
3169 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3170 	    !mlx5_flow_ext_mreg_supported(dev) ||
3171 	    !priv->sh->dv_regc0_mask)
3172 		return 0;
3173 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3174 	if (!mcp_res)
3175 		return -rte_errno;
3176 	return 0;
3177 }
3178 
3179 /**
3180  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3181  *
3182  * All the flow having Q/RSS action should be split by
3183  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3184  * performs the following,
3185  *   - CQE->flow_tag := reg_c[1] (MARK)
3186  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3187  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3188  * but there should be a flow per each MARK ID set by MARK action.
3189  *
3190  * For the aforementioned reason, if there's a MARK action in flow's action
3191  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3192  * the MARK ID to CQE's flow_tag like,
3193  *   - If reg_c[1] is mark_id,
3194  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3195  *
3196  * For SET_META action which stores value in reg_c[0], as the destination is
3197  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3198  * MARK ID means the default flow. The default flow looks like,
3199  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3200  *
3201  * @param dev
3202  *   Pointer to Ethernet device.
3203  * @param flow
3204  *   Pointer to flow structure.
3205  * @param[in] actions
3206  *   Pointer to the list of actions.
3207  * @param[out] error
3208  *   Perform verbose error reporting if not NULL.
3209  *
3210  * @return
3211  *   0 on success, negative value otherwise and rte_errno is set.
3212  */
3213 static int
3214 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3215 			    struct rte_flow *flow,
3216 			    const struct rte_flow_action *actions,
3217 			    struct rte_flow_error *error)
3218 {
3219 	struct mlx5_priv *priv = dev->data->dev_private;
3220 	struct mlx5_dev_config *config = &priv->config;
3221 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3222 	const struct rte_flow_action_mark *mark;
3223 
3224 	/* Check whether extensive metadata feature is engaged. */
3225 	if (!config->dv_flow_en ||
3226 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3227 	    !mlx5_flow_ext_mreg_supported(dev) ||
3228 	    !priv->sh->dv_regc0_mask)
3229 		return 0;
3230 	/* Find MARK action. */
3231 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3232 		switch (actions->type) {
3233 		case RTE_FLOW_ACTION_TYPE_FLAG:
3234 			mcp_res = flow_mreg_add_copy_action
3235 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3236 			if (!mcp_res)
3237 				return -rte_errno;
3238 			flow->mreg_copy = mcp_res;
3239 			if (dev->data->dev_started) {
3240 				mcp_res->appcnt++;
3241 				flow->copy_applied = 1;
3242 			}
3243 			return 0;
3244 		case RTE_FLOW_ACTION_TYPE_MARK:
3245 			mark = (const struct rte_flow_action_mark *)
3246 				actions->conf;
3247 			mcp_res =
3248 				flow_mreg_add_copy_action(dev, mark->id, error);
3249 			if (!mcp_res)
3250 				return -rte_errno;
3251 			flow->mreg_copy = mcp_res;
3252 			if (dev->data->dev_started) {
3253 				mcp_res->appcnt++;
3254 				flow->copy_applied = 1;
3255 			}
3256 			return 0;
3257 		default:
3258 			break;
3259 		}
3260 	}
3261 	return 0;
3262 }
3263 
3264 #define MLX5_MAX_SPLIT_ACTIONS 24
3265 #define MLX5_MAX_SPLIT_ITEMS 24
3266 
3267 /**
3268  * Split the hairpin flow.
3269  * Since HW can't support encap on Rx we move the encap to Tx.
3270  * If the count action is after the encap then we also
3271  * move the count action. in this case the count will also measure
3272  * the outer bytes.
3273  *
3274  * @param dev
3275  *   Pointer to Ethernet device.
3276  * @param[in] actions
3277  *   Associated actions (list terminated by the END action).
3278  * @param[out] actions_rx
3279  *   Rx flow actions.
3280  * @param[out] actions_tx
3281  *   Tx flow actions..
3282  * @param[out] pattern_tx
3283  *   The pattern items for the Tx flow.
3284  * @param[out] flow_id
3285  *   The flow ID connected to this flow.
3286  *
3287  * @return
3288  *   0 on success.
3289  */
3290 static int
3291 flow_hairpin_split(struct rte_eth_dev *dev,
3292 		   const struct rte_flow_action actions[],
3293 		   struct rte_flow_action actions_rx[],
3294 		   struct rte_flow_action actions_tx[],
3295 		   struct rte_flow_item pattern_tx[],
3296 		   uint32_t *flow_id)
3297 {
3298 	struct mlx5_priv *priv = dev->data->dev_private;
3299 	const struct rte_flow_action_raw_encap *raw_encap;
3300 	const struct rte_flow_action_raw_decap *raw_decap;
3301 	struct mlx5_rte_flow_action_set_tag *set_tag;
3302 	struct rte_flow_action *tag_action;
3303 	struct mlx5_rte_flow_item_tag *tag_item;
3304 	struct rte_flow_item *item;
3305 	char *addr;
3306 	int encap = 0;
3307 
3308 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3309 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3310 		switch (actions->type) {
3311 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3312 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3313 			rte_memcpy(actions_tx, actions,
3314 			       sizeof(struct rte_flow_action));
3315 			actions_tx++;
3316 			break;
3317 		case RTE_FLOW_ACTION_TYPE_COUNT:
3318 			if (encap) {
3319 				rte_memcpy(actions_tx, actions,
3320 					   sizeof(struct rte_flow_action));
3321 				actions_tx++;
3322 			} else {
3323 				rte_memcpy(actions_rx, actions,
3324 					   sizeof(struct rte_flow_action));
3325 				actions_rx++;
3326 			}
3327 			break;
3328 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3329 			raw_encap = actions->conf;
3330 			if (raw_encap->size >
3331 			    (sizeof(struct rte_flow_item_eth) +
3332 			     sizeof(struct rte_flow_item_ipv4))) {
3333 				memcpy(actions_tx, actions,
3334 				       sizeof(struct rte_flow_action));
3335 				actions_tx++;
3336 				encap = 1;
3337 			} else {
3338 				rte_memcpy(actions_rx, actions,
3339 					   sizeof(struct rte_flow_action));
3340 				actions_rx++;
3341 			}
3342 			break;
3343 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3344 			raw_decap = actions->conf;
3345 			if (raw_decap->size <
3346 			    (sizeof(struct rte_flow_item_eth) +
3347 			     sizeof(struct rte_flow_item_ipv4))) {
3348 				memcpy(actions_tx, actions,
3349 				       sizeof(struct rte_flow_action));
3350 				actions_tx++;
3351 			} else {
3352 				rte_memcpy(actions_rx, actions,
3353 					   sizeof(struct rte_flow_action));
3354 				actions_rx++;
3355 			}
3356 			break;
3357 		default:
3358 			rte_memcpy(actions_rx, actions,
3359 				   sizeof(struct rte_flow_action));
3360 			actions_rx++;
3361 			break;
3362 		}
3363 	}
3364 	/* Add set meta action and end action for the Rx flow. */
3365 	tag_action = actions_rx;
3366 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3367 	actions_rx++;
3368 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3369 	actions_rx++;
3370 	set_tag = (void *)actions_rx;
3371 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3372 	assert(set_tag->id > REG_NONE);
3373 	set_tag->data = *flow_id;
3374 	tag_action->conf = set_tag;
3375 	/* Create Tx item list. */
3376 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3377 	addr = (void *)&pattern_tx[2];
3378 	item = pattern_tx;
3379 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3380 	tag_item = (void *)addr;
3381 	tag_item->data = *flow_id;
3382 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3383 	assert(set_tag->id > REG_NONE);
3384 	item->spec = tag_item;
3385 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3386 	tag_item = (void *)addr;
3387 	tag_item->data = UINT32_MAX;
3388 	tag_item->id = UINT16_MAX;
3389 	item->mask = tag_item;
3390 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3391 	item->last = NULL;
3392 	item++;
3393 	item->type = RTE_FLOW_ITEM_TYPE_END;
3394 	return 0;
3395 }
3396 
3397 /**
3398  * The last stage of splitting chain, just creates the subflow
3399  * without any modification.
3400  *
3401  * @param dev
3402  *   Pointer to Ethernet device.
3403  * @param[in] flow
3404  *   Parent flow structure pointer.
3405  * @param[in, out] sub_flow
3406  *   Pointer to return the created subflow, may be NULL.
3407  * @param[in] attr
3408  *   Flow rule attributes.
3409  * @param[in] items
3410  *   Pattern specification (list terminated by the END pattern item).
3411  * @param[in] actions
3412  *   Associated actions (list terminated by the END action).
3413  * @param[in] external
3414  *   This flow rule is created by request external to PMD.
3415  * @param[out] error
3416  *   Perform verbose error reporting if not NULL.
3417  * @return
3418  *   0 on success, negative value otherwise
3419  */
3420 static int
3421 flow_create_split_inner(struct rte_eth_dev *dev,
3422 			struct rte_flow *flow,
3423 			struct mlx5_flow **sub_flow,
3424 			const struct rte_flow_attr *attr,
3425 			const struct rte_flow_item items[],
3426 			const struct rte_flow_action actions[],
3427 			bool external, struct rte_flow_error *error)
3428 {
3429 	struct mlx5_flow *dev_flow;
3430 
3431 	dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3432 	if (!dev_flow)
3433 		return -rte_errno;
3434 	dev_flow->flow = flow;
3435 	dev_flow->external = external;
3436 	/* Subflow object was created, we must include one in the list. */
3437 	LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3438 	if (sub_flow)
3439 		*sub_flow = dev_flow;
3440 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3441 }
3442 
3443 /**
3444  * Split the meter flow.
3445  *
3446  * As meter flow will split to three sub flow, other than meter
3447  * action, the other actions make sense to only meter accepts
3448  * the packet. If it need to be dropped, no other additional
3449  * actions should be take.
3450  *
3451  * One kind of special action which decapsulates the L3 tunnel
3452  * header will be in the prefix sub flow, as not to take the
3453  * L3 tunnel header into account.
3454  *
3455  * @param dev
3456  *   Pointer to Ethernet device.
3457  * @param[in] actions
3458  *   Associated actions (list terminated by the END action).
3459  * @param[out] actions_sfx
3460  *   Suffix flow actions.
3461  * @param[out] actions_pre
3462  *   Prefix flow actions.
3463  * @param[out] pattern_sfx
3464  *   The pattern items for the suffix flow.
3465  * @param[out] tag_sfx
3466  *   Pointer to suffix flow tag.
3467  *
3468  * @return
3469  *   0 on success.
3470  */
3471 static int
3472 flow_meter_split_prep(struct rte_eth_dev *dev,
3473 		 const struct rte_flow_action actions[],
3474 		 struct rte_flow_action actions_sfx[],
3475 		 struct rte_flow_action actions_pre[])
3476 {
3477 	struct rte_flow_action *tag_action;
3478 	struct mlx5_rte_flow_action_set_tag *set_tag;
3479 	struct rte_flow_error error;
3480 	const struct rte_flow_action_raw_encap *raw_encap;
3481 	const struct rte_flow_action_raw_decap *raw_decap;
3482 	uint32_t tag_id;
3483 
3484 	/* Add the extra tag action first. */
3485 	tag_action = actions_pre;
3486 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3487 	actions_pre++;
3488 	/* Prepare the actions for prefix and suffix flow. */
3489 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3490 		switch (actions->type) {
3491 		case RTE_FLOW_ACTION_TYPE_METER:
3492 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3493 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3494 			memcpy(actions_pre, actions,
3495 			       sizeof(struct rte_flow_action));
3496 			actions_pre++;
3497 			break;
3498 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3499 			raw_encap = actions->conf;
3500 			if (raw_encap->size >
3501 			    (sizeof(struct rte_flow_item_eth) +
3502 			     sizeof(struct rte_flow_item_ipv4))) {
3503 				memcpy(actions_sfx, actions,
3504 				       sizeof(struct rte_flow_action));
3505 				actions_sfx++;
3506 			} else {
3507 				rte_memcpy(actions_pre, actions,
3508 					   sizeof(struct rte_flow_action));
3509 				actions_pre++;
3510 			}
3511 			break;
3512 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3513 			raw_decap = actions->conf;
3514 			/* Size 0 decap means 50 bytes as vxlan decap. */
3515 			if (raw_decap->size && (raw_decap->size <
3516 			    (sizeof(struct rte_flow_item_eth) +
3517 			     sizeof(struct rte_flow_item_ipv4)))) {
3518 				memcpy(actions_sfx, actions,
3519 				       sizeof(struct rte_flow_action));
3520 				actions_sfx++;
3521 			} else {
3522 				rte_memcpy(actions_pre, actions,
3523 					   sizeof(struct rte_flow_action));
3524 				actions_pre++;
3525 			}
3526 			break;
3527 		default:
3528 			memcpy(actions_sfx, actions,
3529 				sizeof(struct rte_flow_action));
3530 			actions_sfx++;
3531 			break;
3532 		}
3533 	}
3534 	/* Add end action to the actions. */
3535 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3536 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3537 	actions_pre++;
3538 	/* Set the tag. */
3539 	set_tag = (void *)actions_pre;
3540 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3541 	/*
3542 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3543 	 */
3544 	tag_id = flow_qrss_get_id(dev);
3545 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3546 	tag_action->conf = set_tag;
3547 	return tag_id;
3548 }
3549 
3550 /**
3551  * Split action list having QUEUE/RSS for metadata register copy.
3552  *
3553  * Once Q/RSS action is detected in user's action list, the flow action
3554  * should be split in order to copy metadata registers, which will happen in
3555  * RX_CP_TBL like,
3556  *   - CQE->flow_tag := reg_c[1] (MARK)
3557  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3558  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3559  * This is because the last action of each flow must be a terminal action
3560  * (QUEUE, RSS or DROP).
3561  *
3562  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3563  * stored and kept in the mlx5_flow structure per each sub_flow.
3564  *
3565  * The Q/RSS action is replaced with,
3566  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3567  * And the following JUMP action is added at the end,
3568  *   - JUMP, to RX_CP_TBL.
3569  *
3570  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3571  * flow_create_split_metadata() routine. The flow will look like,
3572  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3573  *
3574  * @param dev
3575  *   Pointer to Ethernet device.
3576  * @param[out] split_actions
3577  *   Pointer to store split actions to jump to CP_TBL.
3578  * @param[in] actions
3579  *   Pointer to the list of original flow actions.
3580  * @param[in] qrss
3581  *   Pointer to the Q/RSS action.
3582  * @param[in] actions_n
3583  *   Number of original actions.
3584  * @param[out] error
3585  *   Perform verbose error reporting if not NULL.
3586  *
3587  * @return
3588  *   non-zero unique flow_id on success, otherwise 0 and
3589  *   error/rte_error are set.
3590  */
3591 static uint32_t
3592 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3593 			  struct rte_flow_action *split_actions,
3594 			  const struct rte_flow_action *actions,
3595 			  const struct rte_flow_action *qrss,
3596 			  int actions_n, struct rte_flow_error *error)
3597 {
3598 	struct mlx5_rte_flow_action_set_tag *set_tag;
3599 	struct rte_flow_action_jump *jump;
3600 	const int qrss_idx = qrss - actions;
3601 	uint32_t flow_id = 0;
3602 	int ret = 0;
3603 
3604 	/*
3605 	 * Given actions will be split
3606 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3607 	 * - Add jump to mreg CP_TBL.
3608 	 * As a result, there will be one more action.
3609 	 */
3610 	++actions_n;
3611 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3612 	set_tag = (void *)(split_actions + actions_n);
3613 	/*
3614 	 * If tag action is not set to void(it means we are not the meter
3615 	 * suffix flow), add the tag action. Since meter suffix flow already
3616 	 * has the tag added.
3617 	 */
3618 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3619 		/*
3620 		 * Allocate the new subflow ID. This one is unique within
3621 		 * device and not shared with representors. Otherwise,
3622 		 * we would have to resolve multi-thread access synch
3623 		 * issue. Each flow on the shared device is appended
3624 		 * with source vport identifier, so the resulting
3625 		 * flows will be unique in the shared (by master and
3626 		 * representors) domain even if they have coinciding
3627 		 * IDs.
3628 		 */
3629 		flow_id = flow_qrss_get_id(dev);
3630 		if (!flow_id)
3631 			return rte_flow_error_set(error, ENOMEM,
3632 						  RTE_FLOW_ERROR_TYPE_ACTION,
3633 						  NULL, "can't allocate id "
3634 						  "for split Q/RSS subflow");
3635 		/* Internal SET_TAG action to set flow ID. */
3636 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3637 			.data = flow_id,
3638 		};
3639 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3640 		if (ret < 0)
3641 			return ret;
3642 		set_tag->id = ret;
3643 		/* Construct new actions array. */
3644 		/* Replace QUEUE/RSS action. */
3645 		split_actions[qrss_idx] = (struct rte_flow_action){
3646 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3647 			.conf = set_tag,
3648 		};
3649 	}
3650 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3651 	jump = (void *)(set_tag + 1);
3652 	*jump = (struct rte_flow_action_jump){
3653 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3654 	};
3655 	split_actions[actions_n - 2] = (struct rte_flow_action){
3656 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3657 		.conf = jump,
3658 	};
3659 	split_actions[actions_n - 1] = (struct rte_flow_action){
3660 		.type = RTE_FLOW_ACTION_TYPE_END,
3661 	};
3662 	return flow_id;
3663 }
3664 
3665 /**
3666  * Extend the given action list for Tx metadata copy.
3667  *
3668  * Copy the given action list to the ext_actions and add flow metadata register
3669  * copy action in order to copy reg_a set by WQE to reg_c[0].
3670  *
3671  * @param[out] ext_actions
3672  *   Pointer to the extended action list.
3673  * @param[in] actions
3674  *   Pointer to the list of actions.
3675  * @param[in] actions_n
3676  *   Number of actions in the list.
3677  * @param[out] error
3678  *   Perform verbose error reporting if not NULL.
3679  *
3680  * @return
3681  *   0 on success, negative value otherwise
3682  */
3683 static int
3684 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3685 		       struct rte_flow_action *ext_actions,
3686 		       const struct rte_flow_action *actions,
3687 		       int actions_n, struct rte_flow_error *error)
3688 {
3689 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3690 		(struct mlx5_flow_action_copy_mreg *)
3691 			(ext_actions + actions_n + 1);
3692 	int ret;
3693 
3694 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3695 	if (ret < 0)
3696 		return ret;
3697 	cp_mreg->dst = ret;
3698 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3699 	if (ret < 0)
3700 		return ret;
3701 	cp_mreg->src = ret;
3702 	memcpy(ext_actions, actions,
3703 			sizeof(*ext_actions) * actions_n);
3704 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3705 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3706 		.conf = cp_mreg,
3707 	};
3708 	ext_actions[actions_n] = (struct rte_flow_action){
3709 		.type = RTE_FLOW_ACTION_TYPE_END,
3710 	};
3711 	return 0;
3712 }
3713 
3714 /**
3715  * The splitting for metadata feature.
3716  *
3717  * - Q/RSS action on NIC Rx should be split in order to pass by
3718  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3719  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3720  *
3721  * - All the actions on NIC Tx should have a mreg copy action to
3722  *   copy reg_a from WQE to reg_c[0].
3723  *
3724  * @param dev
3725  *   Pointer to Ethernet device.
3726  * @param[in] flow
3727  *   Parent flow structure pointer.
3728  * @param[in] attr
3729  *   Flow rule attributes.
3730  * @param[in] items
3731  *   Pattern specification (list terminated by the END pattern item).
3732  * @param[in] actions
3733  *   Associated actions (list terminated by the END action).
3734  * @param[in] external
3735  *   This flow rule is created by request external to PMD.
3736  * @param[out] error
3737  *   Perform verbose error reporting if not NULL.
3738  * @return
3739  *   0 on success, negative value otherwise
3740  */
3741 static int
3742 flow_create_split_metadata(struct rte_eth_dev *dev,
3743 			   struct rte_flow *flow,
3744 			   const struct rte_flow_attr *attr,
3745 			   const struct rte_flow_item items[],
3746 			   const struct rte_flow_action actions[],
3747 			   bool external, struct rte_flow_error *error)
3748 {
3749 	struct mlx5_priv *priv = dev->data->dev_private;
3750 	struct mlx5_dev_config *config = &priv->config;
3751 	const struct rte_flow_action *qrss = NULL;
3752 	struct rte_flow_action *ext_actions = NULL;
3753 	struct mlx5_flow *dev_flow = NULL;
3754 	uint32_t qrss_id = 0;
3755 	int mtr_sfx = 0;
3756 	size_t act_size;
3757 	int actions_n;
3758 	int ret;
3759 
3760 	/* Check whether extensive metadata feature is engaged. */
3761 	if (!config->dv_flow_en ||
3762 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3763 	    !mlx5_flow_ext_mreg_supported(dev))
3764 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3765 					       actions, external, error);
3766 	actions_n = flow_parse_qrss_action(actions, &qrss);
3767 	if (qrss) {
3768 		/* Exclude hairpin flows from splitting. */
3769 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3770 			const struct rte_flow_action_queue *queue;
3771 
3772 			queue = qrss->conf;
3773 			if (mlx5_rxq_get_type(dev, queue->index) ==
3774 			    MLX5_RXQ_TYPE_HAIRPIN)
3775 				qrss = NULL;
3776 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3777 			const struct rte_flow_action_rss *rss;
3778 
3779 			rss = qrss->conf;
3780 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3781 			    MLX5_RXQ_TYPE_HAIRPIN)
3782 				qrss = NULL;
3783 		}
3784 	}
3785 	if (qrss) {
3786 		/* Check if it is in meter suffix table. */
3787 		mtr_sfx = attr->group == (attr->transfer ?
3788 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3789 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3790 		/*
3791 		 * Q/RSS action on NIC Rx should be split in order to pass by
3792 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3793 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3794 		 */
3795 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3796 			   sizeof(struct rte_flow_action_set_tag) +
3797 			   sizeof(struct rte_flow_action_jump);
3798 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3799 		if (!ext_actions)
3800 			return rte_flow_error_set(error, ENOMEM,
3801 						  RTE_FLOW_ERROR_TYPE_ACTION,
3802 						  NULL, "no memory to split "
3803 						  "metadata flow");
3804 		/*
3805 		 * If we are the suffix flow of meter, tag already exist.
3806 		 * Set the tag action to void.
3807 		 */
3808 		if (mtr_sfx)
3809 			ext_actions[qrss - actions].type =
3810 						RTE_FLOW_ACTION_TYPE_VOID;
3811 		else
3812 			ext_actions[qrss - actions].type =
3813 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3814 		/*
3815 		 * Create the new actions list with removed Q/RSS action
3816 		 * and appended set tag and jump to register copy table
3817 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3818 		 * in advance, because it is needed for set tag action.
3819 		 */
3820 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3821 						    qrss, actions_n, error);
3822 		if (!mtr_sfx && !qrss_id) {
3823 			ret = -rte_errno;
3824 			goto exit;
3825 		}
3826 	} else if (attr->egress && !attr->transfer) {
3827 		/*
3828 		 * All the actions on NIC Tx should have a metadata register
3829 		 * copy action to copy reg_a from WQE to reg_c[meta]
3830 		 */
3831 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3832 			   sizeof(struct mlx5_flow_action_copy_mreg);
3833 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3834 		if (!ext_actions)
3835 			return rte_flow_error_set(error, ENOMEM,
3836 						  RTE_FLOW_ERROR_TYPE_ACTION,
3837 						  NULL, "no memory to split "
3838 						  "metadata flow");
3839 		/* Create the action list appended with copy register. */
3840 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3841 					     actions_n, error);
3842 		if (ret < 0)
3843 			goto exit;
3844 	}
3845 	/* Add the unmodified original or prefix subflow. */
3846 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3847 				      ext_actions ? ext_actions : actions,
3848 				      external, error);
3849 	if (ret < 0)
3850 		goto exit;
3851 	assert(dev_flow);
3852 	if (qrss) {
3853 		const struct rte_flow_attr q_attr = {
3854 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3855 			.ingress = 1,
3856 		};
3857 		/* Internal PMD action to set register. */
3858 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3859 			.data = qrss_id,
3860 			.id = 0,
3861 		};
3862 		struct rte_flow_item q_items[] = {
3863 			{
3864 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3865 				.spec = &q_tag_spec,
3866 				.last = NULL,
3867 				.mask = NULL,
3868 			},
3869 			{
3870 				.type = RTE_FLOW_ITEM_TYPE_END,
3871 			},
3872 		};
3873 		struct rte_flow_action q_actions[] = {
3874 			{
3875 				.type = qrss->type,
3876 				.conf = qrss->conf,
3877 			},
3878 			{
3879 				.type = RTE_FLOW_ACTION_TYPE_END,
3880 			},
3881 		};
3882 		uint64_t hash_fields = dev_flow->hash_fields;
3883 
3884 		/*
3885 		 * Configure the tag item only if there is no meter subflow.
3886 		 * Since tag is already marked in the meter suffix subflow
3887 		 * we can just use the meter suffix items as is.
3888 		 */
3889 		if (qrss_id) {
3890 			/* Not meter subflow. */
3891 			assert(!mtr_sfx);
3892 			/*
3893 			 * Put unique id in prefix flow due to it is destroyed
3894 			 * after suffix flow and id will be freed after there
3895 			 * is no actual flows with this id and identifier
3896 			 * reallocation becomes possible (for example, for
3897 			 * other flows in other threads).
3898 			 */
3899 			dev_flow->qrss_id = qrss_id;
3900 			qrss_id = 0;
3901 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3902 						   error);
3903 			if (ret < 0)
3904 				goto exit;
3905 			q_tag_spec.id = ret;
3906 		}
3907 		dev_flow = NULL;
3908 		/* Add suffix subflow to execute Q/RSS. */
3909 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3910 					      &q_attr, mtr_sfx ? items :
3911 					      q_items, q_actions,
3912 					      external, error);
3913 		if (ret < 0)
3914 			goto exit;
3915 		assert(dev_flow);
3916 		dev_flow->hash_fields = hash_fields;
3917 	}
3918 
3919 exit:
3920 	/*
3921 	 * We do not destroy the partially created sub_flows in case of error.
3922 	 * These ones are included into parent flow list and will be destroyed
3923 	 * by flow_drv_destroy.
3924 	 */
3925 	flow_qrss_free_id(dev, qrss_id);
3926 	rte_free(ext_actions);
3927 	return ret;
3928 }
3929 
3930 /**
3931  * The splitting for meter feature.
3932  *
3933  * - The meter flow will be split to two flows as prefix and
3934  *   suffix flow. The packets make sense only it pass the prefix
3935  *   meter action.
3936  *
3937  * - Reg_C_5 is used for the packet to match betweend prefix and
3938  *   suffix flow.
3939  *
3940  * @param dev
3941  *   Pointer to Ethernet device.
3942  * @param[in] flow
3943  *   Parent flow structure pointer.
3944  * @param[in] attr
3945  *   Flow rule attributes.
3946  * @param[in] items
3947  *   Pattern specification (list terminated by the END pattern item).
3948  * @param[in] actions
3949  *   Associated actions (list terminated by the END action).
3950  * @param[in] external
3951  *   This flow rule is created by request external to PMD.
3952  * @param[out] error
3953  *   Perform verbose error reporting if not NULL.
3954  * @return
3955  *   0 on success, negative value otherwise
3956  */
3957 static int
3958 flow_create_split_meter(struct rte_eth_dev *dev,
3959 			   struct rte_flow *flow,
3960 			   const struct rte_flow_attr *attr,
3961 			   const struct rte_flow_item items[],
3962 			   const struct rte_flow_action actions[],
3963 			   bool external, struct rte_flow_error *error)
3964 {
3965 	struct mlx5_priv *priv = dev->data->dev_private;
3966 	struct rte_flow_action *sfx_actions = NULL;
3967 	struct rte_flow_action *pre_actions = NULL;
3968 	struct rte_flow_item *sfx_items = NULL;
3969 	const  struct rte_flow_item *sfx_port_id_item;
3970 	struct mlx5_flow *dev_flow = NULL;
3971 	struct rte_flow_attr sfx_attr = *attr;
3972 	uint32_t mtr = 0;
3973 	uint32_t mtr_tag_id = 0;
3974 	size_t act_size;
3975 	size_t item_size;
3976 	int actions_n = 0;
3977 	int ret;
3978 
3979 	if (priv->mtr_en)
3980 		actions_n = flow_check_meter_action(actions, &mtr);
3981 	if (mtr) {
3982 		struct mlx5_rte_flow_item_tag *tag_spec;
3983 		struct mlx5_rte_flow_item_tag *tag_mask;
3984 		/* The five prefix actions: meter, decap, encap, tag, end. */
3985 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3986 			   sizeof(struct rte_flow_action_set_tag);
3987 		/* tag, end. */
3988 #define METER_SUFFIX_ITEM 3
3989 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3990 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
3991 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3992 		if (!sfx_actions)
3993 			return rte_flow_error_set(error, ENOMEM,
3994 						  RTE_FLOW_ERROR_TYPE_ACTION,
3995 						  NULL, "no memory to split "
3996 						  "meter flow");
3997 		pre_actions = sfx_actions + actions_n;
3998 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
3999 						     pre_actions);
4000 		if (!mtr_tag_id) {
4001 			ret = -rte_errno;
4002 			goto exit;
4003 		}
4004 		/* Add the prefix subflow. */
4005 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4006 						  pre_actions, external, error);
4007 		if (ret) {
4008 			ret = -rte_errno;
4009 			goto exit;
4010 		}
4011 		dev_flow->mtr_flow_id = mtr_tag_id;
4012 		/* Prepare the suffix flow match pattern. */
4013 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4014 			     act_size);
4015 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4016 			    METER_SUFFIX_ITEM);
4017 		tag_spec->data = dev_flow->mtr_flow_id << MLX5_MTR_COLOR_BITS;
4018 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4019 						    error);
4020 		tag_mask = tag_spec + 1;
4021 		tag_mask->data = 0xffffff00;
4022 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4023 		sfx_items->spec = tag_spec;
4024 		sfx_items->last = NULL;
4025 		sfx_items->mask = tag_mask;
4026 		sfx_items++;
4027 		sfx_port_id_item = find_port_id_item(items);
4028 		if (sfx_port_id_item) {
4029 			memcpy(sfx_items, sfx_port_id_item,
4030 			       sizeof(*sfx_items));
4031 			sfx_items++;
4032 		}
4033 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4034 		sfx_items -= sfx_port_id_item ? 2 : 1;
4035 		/* Setting the sfx group atrr. */
4036 		sfx_attr.group = sfx_attr.transfer ?
4037 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4038 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4039 	}
4040 	/* Add the prefix subflow. */
4041 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4042 					 sfx_items ? sfx_items : items,
4043 					 sfx_actions ? sfx_actions : actions,
4044 					 external, error);
4045 exit:
4046 	if (sfx_actions)
4047 		rte_free(sfx_actions);
4048 	return ret;
4049 }
4050 
4051 /**
4052  * Split the flow to subflow set. The splitters might be linked
4053  * in the chain, like this:
4054  * flow_create_split_outer() calls:
4055  *   flow_create_split_meter() calls:
4056  *     flow_create_split_metadata(meter_subflow_0) calls:
4057  *       flow_create_split_inner(metadata_subflow_0)
4058  *       flow_create_split_inner(metadata_subflow_1)
4059  *       flow_create_split_inner(metadata_subflow_2)
4060  *     flow_create_split_metadata(meter_subflow_1) calls:
4061  *       flow_create_split_inner(metadata_subflow_0)
4062  *       flow_create_split_inner(metadata_subflow_1)
4063  *       flow_create_split_inner(metadata_subflow_2)
4064  *
4065  * This provide flexible way to add new levels of flow splitting.
4066  * The all of successfully created subflows are included to the
4067  * parent flow dev_flow list.
4068  *
4069  * @param dev
4070  *   Pointer to Ethernet device.
4071  * @param[in] flow
4072  *   Parent flow structure pointer.
4073  * @param[in] attr
4074  *   Flow rule attributes.
4075  * @param[in] items
4076  *   Pattern specification (list terminated by the END pattern item).
4077  * @param[in] actions
4078  *   Associated actions (list terminated by the END action).
4079  * @param[in] external
4080  *   This flow rule is created by request external to PMD.
4081  * @param[out] error
4082  *   Perform verbose error reporting if not NULL.
4083  * @return
4084  *   0 on success, negative value otherwise
4085  */
4086 static int
4087 flow_create_split_outer(struct rte_eth_dev *dev,
4088 			struct rte_flow *flow,
4089 			const struct rte_flow_attr *attr,
4090 			const struct rte_flow_item items[],
4091 			const struct rte_flow_action actions[],
4092 			bool external, struct rte_flow_error *error)
4093 {
4094 	int ret;
4095 
4096 	ret = flow_create_split_meter(dev, flow, attr, items,
4097 					 actions, external, error);
4098 	assert(ret <= 0);
4099 	return ret;
4100 }
4101 
4102 /**
4103  * Create a flow and add it to @p list.
4104  *
4105  * @param dev
4106  *   Pointer to Ethernet device.
4107  * @param list
4108  *   Pointer to a TAILQ flow list. If this parameter NULL,
4109  *   no list insertion occurred, flow is just created,
4110  *   this is caller's responsibility to track the
4111  *   created flow.
4112  * @param[in] attr
4113  *   Flow rule attributes.
4114  * @param[in] items
4115  *   Pattern specification (list terminated by the END pattern item).
4116  * @param[in] actions
4117  *   Associated actions (list terminated by the END action).
4118  * @param[in] external
4119  *   This flow rule is created by request external to PMD.
4120  * @param[out] error
4121  *   Perform verbose error reporting if not NULL.
4122  *
4123  * @return
4124  *   A flow on success, NULL otherwise and rte_errno is set.
4125  */
4126 static struct rte_flow *
4127 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4128 		 const struct rte_flow_attr *attr,
4129 		 const struct rte_flow_item items[],
4130 		 const struct rte_flow_action actions[],
4131 		 bool external, struct rte_flow_error *error)
4132 {
4133 	struct mlx5_priv *priv = dev->data->dev_private;
4134 	struct rte_flow *flow = NULL;
4135 	struct mlx5_flow *dev_flow;
4136 	const struct rte_flow_action_rss *rss;
4137 	union {
4138 		struct rte_flow_expand_rss buf;
4139 		uint8_t buffer[2048];
4140 	} expand_buffer;
4141 	union {
4142 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4143 		uint8_t buffer[2048];
4144 	} actions_rx;
4145 	union {
4146 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4147 		uint8_t buffer[2048];
4148 	} actions_hairpin_tx;
4149 	union {
4150 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4151 		uint8_t buffer[2048];
4152 	} items_tx;
4153 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4154 	const struct rte_flow_action *p_actions_rx = actions;
4155 	int ret;
4156 	uint32_t i;
4157 	uint32_t flow_size;
4158 	int hairpin_flow = 0;
4159 	uint32_t hairpin_id = 0;
4160 	struct rte_flow_attr attr_tx = { .priority = 0 };
4161 
4162 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4163 	if (hairpin_flow > 0) {
4164 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4165 			rte_errno = EINVAL;
4166 			return NULL;
4167 		}
4168 		flow_hairpin_split(dev, actions, actions_rx.actions,
4169 				   actions_hairpin_tx.actions, items_tx.items,
4170 				   &hairpin_id);
4171 		p_actions_rx = actions_rx.actions;
4172 	}
4173 	ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4174 				error);
4175 	if (ret < 0)
4176 		goto error_before_flow;
4177 	flow_size = sizeof(struct rte_flow);
4178 	rss = flow_get_rss_action(p_actions_rx);
4179 	if (rss)
4180 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4181 					    sizeof(void *));
4182 	else
4183 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4184 	flow = rte_calloc(__func__, 1, flow_size, 0);
4185 	if (!flow) {
4186 		rte_errno = ENOMEM;
4187 		goto error_before_flow;
4188 	}
4189 	flow->drv_type = flow_get_drv_type(dev, attr);
4190 	if (hairpin_id != 0)
4191 		flow->hairpin_flow_id = hairpin_id;
4192 	assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4193 	       flow->drv_type < MLX5_FLOW_TYPE_MAX);
4194 	flow->rss.queue = (void *)(flow + 1);
4195 	if (rss) {
4196 		/*
4197 		 * The following information is required by
4198 		 * mlx5_flow_hashfields_adjust() in advance.
4199 		 */
4200 		flow->rss.level = rss->level;
4201 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4202 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4203 	}
4204 	LIST_INIT(&flow->dev_flows);
4205 	if (rss && rss->types) {
4206 		unsigned int graph_root;
4207 
4208 		graph_root = find_graph_root(items, rss->level);
4209 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4210 					  items, rss->types,
4211 					  mlx5_support_expansion,
4212 					  graph_root);
4213 		assert(ret > 0 &&
4214 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4215 	} else {
4216 		buf->entries = 1;
4217 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4218 	}
4219 	for (i = 0; i < buf->entries; ++i) {
4220 		/*
4221 		 * The splitter may create multiple dev_flows,
4222 		 * depending on configuration. In the simplest
4223 		 * case it just creates unmodified original flow.
4224 		 */
4225 		ret = flow_create_split_outer(dev, flow, attr,
4226 					      buf->entry[i].pattern,
4227 					      p_actions_rx, external,
4228 					      error);
4229 		if (ret < 0)
4230 			goto error;
4231 	}
4232 	/* Create the tx flow. */
4233 	if (hairpin_flow) {
4234 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4235 		attr_tx.ingress = 0;
4236 		attr_tx.egress = 1;
4237 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4238 					    actions_hairpin_tx.actions, error);
4239 		if (!dev_flow)
4240 			goto error;
4241 		dev_flow->flow = flow;
4242 		dev_flow->external = 0;
4243 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4244 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4245 					 items_tx.items,
4246 					 actions_hairpin_tx.actions, error);
4247 		if (ret < 0)
4248 			goto error;
4249 	}
4250 	/*
4251 	 * Update the metadata register copy table. If extensive
4252 	 * metadata feature is enabled and registers are supported
4253 	 * we might create the extra rte_flow for each unique
4254 	 * MARK/FLAG action ID.
4255 	 *
4256 	 * The table is updated for ingress Flows only, because
4257 	 * the egress Flows belong to the different device and
4258 	 * copy table should be updated in peer NIC Rx domain.
4259 	 */
4260 	if (attr->ingress &&
4261 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4262 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4263 		if (ret)
4264 			goto error;
4265 	}
4266 	if (dev->data->dev_started) {
4267 		ret = flow_drv_apply(dev, flow, error);
4268 		if (ret < 0)
4269 			goto error;
4270 	}
4271 	if (list)
4272 		TAILQ_INSERT_TAIL(list, flow, next);
4273 	flow_rxq_flags_set(dev, flow);
4274 	return flow;
4275 error_before_flow:
4276 	if (hairpin_id)
4277 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4278 				     hairpin_id);
4279 	return NULL;
4280 error:
4281 	assert(flow);
4282 	flow_mreg_del_copy_action(dev, flow);
4283 	ret = rte_errno; /* Save rte_errno before cleanup. */
4284 	if (flow->hairpin_flow_id)
4285 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4286 				     flow->hairpin_flow_id);
4287 	assert(flow);
4288 	flow_drv_destroy(dev, flow);
4289 	rte_free(flow);
4290 	rte_errno = ret; /* Restore rte_errno. */
4291 	return NULL;
4292 }
4293 
4294 /**
4295  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4296  * incoming packets to table 1.
4297  *
4298  * Other flow rules, requested for group n, will be created in
4299  * e-switch table n+1.
4300  * Jump action to e-switch group n will be created to group n+1.
4301  *
4302  * Used when working in switchdev mode, to utilise advantages of table 1
4303  * and above.
4304  *
4305  * @param dev
4306  *   Pointer to Ethernet device.
4307  *
4308  * @return
4309  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4310  */
4311 struct rte_flow *
4312 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4313 {
4314 	const struct rte_flow_attr attr = {
4315 		.group = 0,
4316 		.priority = 0,
4317 		.ingress = 1,
4318 		.egress = 0,
4319 		.transfer = 1,
4320 	};
4321 	const struct rte_flow_item pattern = {
4322 		.type = RTE_FLOW_ITEM_TYPE_END,
4323 	};
4324 	struct rte_flow_action_jump jump = {
4325 		.group = 1,
4326 	};
4327 	const struct rte_flow_action actions[] = {
4328 		{
4329 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4330 			.conf = &jump,
4331 		},
4332 		{
4333 			.type = RTE_FLOW_ACTION_TYPE_END,
4334 		},
4335 	};
4336 	struct mlx5_priv *priv = dev->data->dev_private;
4337 	struct rte_flow_error error;
4338 
4339 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4340 				actions, false, &error);
4341 }
4342 
4343 /**
4344  * Create a flow.
4345  *
4346  * @see rte_flow_create()
4347  * @see rte_flow_ops
4348  */
4349 struct rte_flow *
4350 mlx5_flow_create(struct rte_eth_dev *dev,
4351 		 const struct rte_flow_attr *attr,
4352 		 const struct rte_flow_item items[],
4353 		 const struct rte_flow_action actions[],
4354 		 struct rte_flow_error *error)
4355 {
4356 	struct mlx5_priv *priv = dev->data->dev_private;
4357 
4358 	return flow_list_create(dev, &priv->flows,
4359 				attr, items, actions, true, error);
4360 }
4361 
4362 /**
4363  * Destroy a flow in a list.
4364  *
4365  * @param dev
4366  *   Pointer to Ethernet device.
4367  * @param list
4368  *   Pointer to a TAILQ flow list. If this parameter NULL,
4369  *   there is no flow removal from the list.
4370  * @param[in] flow
4371  *   Flow to destroy.
4372  */
4373 static void
4374 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4375 		  struct rte_flow *flow)
4376 {
4377 	struct mlx5_priv *priv = dev->data->dev_private;
4378 
4379 	/*
4380 	 * Update RX queue flags only if port is started, otherwise it is
4381 	 * already clean.
4382 	 */
4383 	if (dev->data->dev_started)
4384 		flow_rxq_flags_trim(dev, flow);
4385 	if (flow->hairpin_flow_id)
4386 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4387 				     flow->hairpin_flow_id);
4388 	flow_drv_destroy(dev, flow);
4389 	if (list)
4390 		TAILQ_REMOVE(list, flow, next);
4391 	flow_mreg_del_copy_action(dev, flow);
4392 	rte_free(flow->fdir);
4393 	rte_free(flow);
4394 }
4395 
4396 /**
4397  * Destroy all flows.
4398  *
4399  * @param dev
4400  *   Pointer to Ethernet device.
4401  * @param list
4402  *   Pointer to a TAILQ flow list.
4403  */
4404 void
4405 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4406 {
4407 	while (!TAILQ_EMPTY(list)) {
4408 		struct rte_flow *flow;
4409 
4410 		flow = TAILQ_FIRST(list);
4411 		flow_list_destroy(dev, list, flow);
4412 	}
4413 }
4414 
4415 /**
4416  * Remove all flows.
4417  *
4418  * @param dev
4419  *   Pointer to Ethernet device.
4420  * @param list
4421  *   Pointer to a TAILQ flow list.
4422  */
4423 void
4424 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4425 {
4426 	struct rte_flow *flow;
4427 
4428 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4429 		flow_drv_remove(dev, flow);
4430 		flow_mreg_stop_copy_action(dev, flow);
4431 	}
4432 	flow_mreg_del_default_copy_action(dev);
4433 	flow_rxq_flags_clear(dev);
4434 }
4435 
4436 /**
4437  * Add all flows.
4438  *
4439  * @param dev
4440  *   Pointer to Ethernet device.
4441  * @param list
4442  *   Pointer to a TAILQ flow list.
4443  *
4444  * @return
4445  *   0 on success, a negative errno value otherwise and rte_errno is set.
4446  */
4447 int
4448 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4449 {
4450 	struct rte_flow *flow;
4451 	struct rte_flow_error error;
4452 	int ret = 0;
4453 
4454 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4455 	ret = flow_mreg_add_default_copy_action(dev, &error);
4456 	if (ret < 0)
4457 		return -rte_errno;
4458 	/* Apply Flows created by application. */
4459 	TAILQ_FOREACH(flow, list, next) {
4460 		ret = flow_mreg_start_copy_action(dev, flow);
4461 		if (ret < 0)
4462 			goto error;
4463 		ret = flow_drv_apply(dev, flow, &error);
4464 		if (ret < 0)
4465 			goto error;
4466 		flow_rxq_flags_set(dev, flow);
4467 	}
4468 	return 0;
4469 error:
4470 	ret = rte_errno; /* Save rte_errno before cleanup. */
4471 	mlx5_flow_stop(dev, list);
4472 	rte_errno = ret; /* Restore rte_errno. */
4473 	return -rte_errno;
4474 }
4475 
4476 /**
4477  * Verify the flow list is empty
4478  *
4479  * @param dev
4480  *  Pointer to Ethernet device.
4481  *
4482  * @return the number of flows not released.
4483  */
4484 int
4485 mlx5_flow_verify(struct rte_eth_dev *dev)
4486 {
4487 	struct mlx5_priv *priv = dev->data->dev_private;
4488 	struct rte_flow *flow;
4489 	int ret = 0;
4490 
4491 	TAILQ_FOREACH(flow, &priv->flows, next) {
4492 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4493 			dev->data->port_id, (void *)flow);
4494 		++ret;
4495 	}
4496 	return ret;
4497 }
4498 
4499 /**
4500  * Enable default hairpin egress flow.
4501  *
4502  * @param dev
4503  *   Pointer to Ethernet device.
4504  * @param queue
4505  *   The queue index.
4506  *
4507  * @return
4508  *   0 on success, a negative errno value otherwise and rte_errno is set.
4509  */
4510 int
4511 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4512 			    uint32_t queue)
4513 {
4514 	struct mlx5_priv *priv = dev->data->dev_private;
4515 	const struct rte_flow_attr attr = {
4516 		.egress = 1,
4517 		.priority = 0,
4518 	};
4519 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4520 		.queue = queue,
4521 	};
4522 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4523 		.queue = UINT32_MAX,
4524 	};
4525 	struct rte_flow_item items[] = {
4526 		{
4527 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4528 			.spec = &queue_spec,
4529 			.last = NULL,
4530 			.mask = &queue_mask,
4531 		},
4532 		{
4533 			.type = RTE_FLOW_ITEM_TYPE_END,
4534 		},
4535 	};
4536 	struct rte_flow_action_jump jump = {
4537 		.group = MLX5_HAIRPIN_TX_TABLE,
4538 	};
4539 	struct rte_flow_action actions[2];
4540 	struct rte_flow *flow;
4541 	struct rte_flow_error error;
4542 
4543 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4544 	actions[0].conf = &jump;
4545 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4546 	flow = flow_list_create(dev, &priv->ctrl_flows,
4547 				&attr, items, actions, false, &error);
4548 	if (!flow) {
4549 		DRV_LOG(DEBUG,
4550 			"Failed to create ctrl flow: rte_errno(%d),"
4551 			" type(%d), message(%s)",
4552 			rte_errno, error.type,
4553 			error.message ? error.message : " (no stated reason)");
4554 		return -rte_errno;
4555 	}
4556 	return 0;
4557 }
4558 
4559 /**
4560  * Enable a control flow configured from the control plane.
4561  *
4562  * @param dev
4563  *   Pointer to Ethernet device.
4564  * @param eth_spec
4565  *   An Ethernet flow spec to apply.
4566  * @param eth_mask
4567  *   An Ethernet flow mask to apply.
4568  * @param vlan_spec
4569  *   A VLAN flow spec to apply.
4570  * @param vlan_mask
4571  *   A VLAN flow mask to apply.
4572  *
4573  * @return
4574  *   0 on success, a negative errno value otherwise and rte_errno is set.
4575  */
4576 int
4577 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4578 		    struct rte_flow_item_eth *eth_spec,
4579 		    struct rte_flow_item_eth *eth_mask,
4580 		    struct rte_flow_item_vlan *vlan_spec,
4581 		    struct rte_flow_item_vlan *vlan_mask)
4582 {
4583 	struct mlx5_priv *priv = dev->data->dev_private;
4584 	const struct rte_flow_attr attr = {
4585 		.ingress = 1,
4586 		.priority = MLX5_FLOW_PRIO_RSVD,
4587 	};
4588 	struct rte_flow_item items[] = {
4589 		{
4590 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4591 			.spec = eth_spec,
4592 			.last = NULL,
4593 			.mask = eth_mask,
4594 		},
4595 		{
4596 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4597 					      RTE_FLOW_ITEM_TYPE_END,
4598 			.spec = vlan_spec,
4599 			.last = NULL,
4600 			.mask = vlan_mask,
4601 		},
4602 		{
4603 			.type = RTE_FLOW_ITEM_TYPE_END,
4604 		},
4605 	};
4606 	uint16_t queue[priv->reta_idx_n];
4607 	struct rte_flow_action_rss action_rss = {
4608 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4609 		.level = 0,
4610 		.types = priv->rss_conf.rss_hf,
4611 		.key_len = priv->rss_conf.rss_key_len,
4612 		.queue_num = priv->reta_idx_n,
4613 		.key = priv->rss_conf.rss_key,
4614 		.queue = queue,
4615 	};
4616 	struct rte_flow_action actions[] = {
4617 		{
4618 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4619 			.conf = &action_rss,
4620 		},
4621 		{
4622 			.type = RTE_FLOW_ACTION_TYPE_END,
4623 		},
4624 	};
4625 	struct rte_flow *flow;
4626 	struct rte_flow_error error;
4627 	unsigned int i;
4628 
4629 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4630 		return 0;
4631 	}
4632 	for (i = 0; i != priv->reta_idx_n; ++i)
4633 		queue[i] = (*priv->reta_idx)[i];
4634 	flow = flow_list_create(dev, &priv->ctrl_flows,
4635 				&attr, items, actions, false, &error);
4636 	if (!flow)
4637 		return -rte_errno;
4638 	return 0;
4639 }
4640 
4641 /**
4642  * Enable a flow control configured from the control plane.
4643  *
4644  * @param dev
4645  *   Pointer to Ethernet device.
4646  * @param eth_spec
4647  *   An Ethernet flow spec to apply.
4648  * @param eth_mask
4649  *   An Ethernet flow mask to apply.
4650  *
4651  * @return
4652  *   0 on success, a negative errno value otherwise and rte_errno is set.
4653  */
4654 int
4655 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4656 	       struct rte_flow_item_eth *eth_spec,
4657 	       struct rte_flow_item_eth *eth_mask)
4658 {
4659 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4660 }
4661 
4662 /**
4663  * Destroy a flow.
4664  *
4665  * @see rte_flow_destroy()
4666  * @see rte_flow_ops
4667  */
4668 int
4669 mlx5_flow_destroy(struct rte_eth_dev *dev,
4670 		  struct rte_flow *flow,
4671 		  struct rte_flow_error *error __rte_unused)
4672 {
4673 	struct mlx5_priv *priv = dev->data->dev_private;
4674 
4675 	flow_list_destroy(dev, &priv->flows, flow);
4676 	return 0;
4677 }
4678 
4679 /**
4680  * Destroy all flows.
4681  *
4682  * @see rte_flow_flush()
4683  * @see rte_flow_ops
4684  */
4685 int
4686 mlx5_flow_flush(struct rte_eth_dev *dev,
4687 		struct rte_flow_error *error __rte_unused)
4688 {
4689 	struct mlx5_priv *priv = dev->data->dev_private;
4690 
4691 	mlx5_flow_list_flush(dev, &priv->flows);
4692 	return 0;
4693 }
4694 
4695 /**
4696  * Isolated mode.
4697  *
4698  * @see rte_flow_isolate()
4699  * @see rte_flow_ops
4700  */
4701 int
4702 mlx5_flow_isolate(struct rte_eth_dev *dev,
4703 		  int enable,
4704 		  struct rte_flow_error *error)
4705 {
4706 	struct mlx5_priv *priv = dev->data->dev_private;
4707 
4708 	if (dev->data->dev_started) {
4709 		rte_flow_error_set(error, EBUSY,
4710 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4711 				   NULL,
4712 				   "port must be stopped first");
4713 		return -rte_errno;
4714 	}
4715 	priv->isolated = !!enable;
4716 	if (enable)
4717 		dev->dev_ops = &mlx5_dev_ops_isolate;
4718 	else
4719 		dev->dev_ops = &mlx5_dev_ops;
4720 	return 0;
4721 }
4722 
4723 /**
4724  * Query a flow.
4725  *
4726  * @see rte_flow_query()
4727  * @see rte_flow_ops
4728  */
4729 static int
4730 flow_drv_query(struct rte_eth_dev *dev,
4731 	       struct rte_flow *flow,
4732 	       const struct rte_flow_action *actions,
4733 	       void *data,
4734 	       struct rte_flow_error *error)
4735 {
4736 	const struct mlx5_flow_driver_ops *fops;
4737 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4738 
4739 	assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4740 	fops = flow_get_drv_ops(ftype);
4741 
4742 	return fops->query(dev, flow, actions, data, error);
4743 }
4744 
4745 /**
4746  * Query a flow.
4747  *
4748  * @see rte_flow_query()
4749  * @see rte_flow_ops
4750  */
4751 int
4752 mlx5_flow_query(struct rte_eth_dev *dev,
4753 		struct rte_flow *flow,
4754 		const struct rte_flow_action *actions,
4755 		void *data,
4756 		struct rte_flow_error *error)
4757 {
4758 	int ret;
4759 
4760 	ret = flow_drv_query(dev, flow, actions, data, error);
4761 	if (ret < 0)
4762 		return ret;
4763 	return 0;
4764 }
4765 
4766 /**
4767  * Convert a flow director filter to a generic flow.
4768  *
4769  * @param dev
4770  *   Pointer to Ethernet device.
4771  * @param fdir_filter
4772  *   Flow director filter to add.
4773  * @param attributes
4774  *   Generic flow parameters structure.
4775  *
4776  * @return
4777  *   0 on success, a negative errno value otherwise and rte_errno is set.
4778  */
4779 static int
4780 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4781 			 const struct rte_eth_fdir_filter *fdir_filter,
4782 			 struct mlx5_fdir *attributes)
4783 {
4784 	struct mlx5_priv *priv = dev->data->dev_private;
4785 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4786 	const struct rte_eth_fdir_masks *mask =
4787 		&dev->data->dev_conf.fdir_conf.mask;
4788 
4789 	/* Validate queue number. */
4790 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4791 		DRV_LOG(ERR, "port %u invalid queue number %d",
4792 			dev->data->port_id, fdir_filter->action.rx_queue);
4793 		rte_errno = EINVAL;
4794 		return -rte_errno;
4795 	}
4796 	attributes->attr.ingress = 1;
4797 	attributes->items[0] = (struct rte_flow_item) {
4798 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4799 		.spec = &attributes->l2,
4800 		.mask = &attributes->l2_mask,
4801 	};
4802 	switch (fdir_filter->action.behavior) {
4803 	case RTE_ETH_FDIR_ACCEPT:
4804 		attributes->actions[0] = (struct rte_flow_action){
4805 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4806 			.conf = &attributes->queue,
4807 		};
4808 		break;
4809 	case RTE_ETH_FDIR_REJECT:
4810 		attributes->actions[0] = (struct rte_flow_action){
4811 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4812 		};
4813 		break;
4814 	default:
4815 		DRV_LOG(ERR, "port %u invalid behavior %d",
4816 			dev->data->port_id,
4817 			fdir_filter->action.behavior);
4818 		rte_errno = ENOTSUP;
4819 		return -rte_errno;
4820 	}
4821 	attributes->queue.index = fdir_filter->action.rx_queue;
4822 	/* Handle L3. */
4823 	switch (fdir_filter->input.flow_type) {
4824 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4825 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4826 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4827 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4828 			.src_addr = input->flow.ip4_flow.src_ip,
4829 			.dst_addr = input->flow.ip4_flow.dst_ip,
4830 			.time_to_live = input->flow.ip4_flow.ttl,
4831 			.type_of_service = input->flow.ip4_flow.tos,
4832 		};
4833 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4834 			.src_addr = mask->ipv4_mask.src_ip,
4835 			.dst_addr = mask->ipv4_mask.dst_ip,
4836 			.time_to_live = mask->ipv4_mask.ttl,
4837 			.type_of_service = mask->ipv4_mask.tos,
4838 			.next_proto_id = mask->ipv4_mask.proto,
4839 		};
4840 		attributes->items[1] = (struct rte_flow_item){
4841 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4842 			.spec = &attributes->l3,
4843 			.mask = &attributes->l3_mask,
4844 		};
4845 		break;
4846 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4847 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4848 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4849 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4850 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4851 			.proto = input->flow.ipv6_flow.proto,
4852 		};
4853 
4854 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4855 		       input->flow.ipv6_flow.src_ip,
4856 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4857 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4858 		       input->flow.ipv6_flow.dst_ip,
4859 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4860 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4861 		       mask->ipv6_mask.src_ip,
4862 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4863 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4864 		       mask->ipv6_mask.dst_ip,
4865 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4866 		attributes->items[1] = (struct rte_flow_item){
4867 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4868 			.spec = &attributes->l3,
4869 			.mask = &attributes->l3_mask,
4870 		};
4871 		break;
4872 	default:
4873 		DRV_LOG(ERR, "port %u invalid flow type%d",
4874 			dev->data->port_id, fdir_filter->input.flow_type);
4875 		rte_errno = ENOTSUP;
4876 		return -rte_errno;
4877 	}
4878 	/* Handle L4. */
4879 	switch (fdir_filter->input.flow_type) {
4880 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4881 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4882 			.src_port = input->flow.udp4_flow.src_port,
4883 			.dst_port = input->flow.udp4_flow.dst_port,
4884 		};
4885 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4886 			.src_port = mask->src_port_mask,
4887 			.dst_port = mask->dst_port_mask,
4888 		};
4889 		attributes->items[2] = (struct rte_flow_item){
4890 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4891 			.spec = &attributes->l4,
4892 			.mask = &attributes->l4_mask,
4893 		};
4894 		break;
4895 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4896 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4897 			.src_port = input->flow.tcp4_flow.src_port,
4898 			.dst_port = input->flow.tcp4_flow.dst_port,
4899 		};
4900 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4901 			.src_port = mask->src_port_mask,
4902 			.dst_port = mask->dst_port_mask,
4903 		};
4904 		attributes->items[2] = (struct rte_flow_item){
4905 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4906 			.spec = &attributes->l4,
4907 			.mask = &attributes->l4_mask,
4908 		};
4909 		break;
4910 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4911 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4912 			.src_port = input->flow.udp6_flow.src_port,
4913 			.dst_port = input->flow.udp6_flow.dst_port,
4914 		};
4915 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4916 			.src_port = mask->src_port_mask,
4917 			.dst_port = mask->dst_port_mask,
4918 		};
4919 		attributes->items[2] = (struct rte_flow_item){
4920 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4921 			.spec = &attributes->l4,
4922 			.mask = &attributes->l4_mask,
4923 		};
4924 		break;
4925 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4926 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4927 			.src_port = input->flow.tcp6_flow.src_port,
4928 			.dst_port = input->flow.tcp6_flow.dst_port,
4929 		};
4930 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4931 			.src_port = mask->src_port_mask,
4932 			.dst_port = mask->dst_port_mask,
4933 		};
4934 		attributes->items[2] = (struct rte_flow_item){
4935 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4936 			.spec = &attributes->l4,
4937 			.mask = &attributes->l4_mask,
4938 		};
4939 		break;
4940 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4941 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4942 		break;
4943 	default:
4944 		DRV_LOG(ERR, "port %u invalid flow type%d",
4945 			dev->data->port_id, fdir_filter->input.flow_type);
4946 		rte_errno = ENOTSUP;
4947 		return -rte_errno;
4948 	}
4949 	return 0;
4950 }
4951 
4952 #define FLOW_FDIR_CMP(f1, f2, fld) \
4953 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4954 
4955 /**
4956  * Compare two FDIR flows. If items and actions are identical, the two flows are
4957  * regarded as same.
4958  *
4959  * @param dev
4960  *   Pointer to Ethernet device.
4961  * @param f1
4962  *   FDIR flow to compare.
4963  * @param f2
4964  *   FDIR flow to compare.
4965  *
4966  * @return
4967  *   Zero on match, 1 otherwise.
4968  */
4969 static int
4970 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4971 {
4972 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4973 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4974 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4975 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4976 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4977 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4978 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4979 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4980 		return 1;
4981 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4982 	    FLOW_FDIR_CMP(f1, f2, queue))
4983 		return 1;
4984 	return 0;
4985 }
4986 
4987 /**
4988  * Search device flow list to find out a matched FDIR flow.
4989  *
4990  * @param dev
4991  *   Pointer to Ethernet device.
4992  * @param fdir_flow
4993  *   FDIR flow to lookup.
4994  *
4995  * @return
4996  *   Pointer of flow if found, NULL otherwise.
4997  */
4998 static struct rte_flow *
4999 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5000 {
5001 	struct mlx5_priv *priv = dev->data->dev_private;
5002 	struct rte_flow *flow = NULL;
5003 
5004 	assert(fdir_flow);
5005 	TAILQ_FOREACH(flow, &priv->flows, next) {
5006 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5007 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5008 				dev->data->port_id, (void *)flow);
5009 			break;
5010 		}
5011 	}
5012 	return flow;
5013 }
5014 
5015 /**
5016  * Add new flow director filter and store it in list.
5017  *
5018  * @param dev
5019  *   Pointer to Ethernet device.
5020  * @param fdir_filter
5021  *   Flow director filter to add.
5022  *
5023  * @return
5024  *   0 on success, a negative errno value otherwise and rte_errno is set.
5025  */
5026 static int
5027 flow_fdir_filter_add(struct rte_eth_dev *dev,
5028 		     const struct rte_eth_fdir_filter *fdir_filter)
5029 {
5030 	struct mlx5_priv *priv = dev->data->dev_private;
5031 	struct mlx5_fdir *fdir_flow;
5032 	struct rte_flow *flow;
5033 	int ret;
5034 
5035 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5036 	if (!fdir_flow) {
5037 		rte_errno = ENOMEM;
5038 		return -rte_errno;
5039 	}
5040 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5041 	if (ret)
5042 		goto error;
5043 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5044 	if (flow) {
5045 		rte_errno = EEXIST;
5046 		goto error;
5047 	}
5048 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5049 				fdir_flow->items, fdir_flow->actions, true,
5050 				NULL);
5051 	if (!flow)
5052 		goto error;
5053 	assert(!flow->fdir);
5054 	flow->fdir = fdir_flow;
5055 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5056 		dev->data->port_id, (void *)flow);
5057 	return 0;
5058 error:
5059 	rte_free(fdir_flow);
5060 	return -rte_errno;
5061 }
5062 
5063 /**
5064  * Delete specific filter.
5065  *
5066  * @param dev
5067  *   Pointer to Ethernet device.
5068  * @param fdir_filter
5069  *   Filter to be deleted.
5070  *
5071  * @return
5072  *   0 on success, a negative errno value otherwise and rte_errno is set.
5073  */
5074 static int
5075 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5076 			const struct rte_eth_fdir_filter *fdir_filter)
5077 {
5078 	struct mlx5_priv *priv = dev->data->dev_private;
5079 	struct rte_flow *flow;
5080 	struct mlx5_fdir fdir_flow = {
5081 		.attr.group = 0,
5082 	};
5083 	int ret;
5084 
5085 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5086 	if (ret)
5087 		return -rte_errno;
5088 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5089 	if (!flow) {
5090 		rte_errno = ENOENT;
5091 		return -rte_errno;
5092 	}
5093 	flow_list_destroy(dev, &priv->flows, flow);
5094 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5095 		dev->data->port_id, (void *)flow);
5096 	return 0;
5097 }
5098 
5099 /**
5100  * Update queue for specific filter.
5101  *
5102  * @param dev
5103  *   Pointer to Ethernet device.
5104  * @param fdir_filter
5105  *   Filter to be updated.
5106  *
5107  * @return
5108  *   0 on success, a negative errno value otherwise and rte_errno is set.
5109  */
5110 static int
5111 flow_fdir_filter_update(struct rte_eth_dev *dev,
5112 			const struct rte_eth_fdir_filter *fdir_filter)
5113 {
5114 	int ret;
5115 
5116 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5117 	if (ret)
5118 		return ret;
5119 	return flow_fdir_filter_add(dev, fdir_filter);
5120 }
5121 
5122 /**
5123  * Flush all filters.
5124  *
5125  * @param dev
5126  *   Pointer to Ethernet device.
5127  */
5128 static void
5129 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5130 {
5131 	struct mlx5_priv *priv = dev->data->dev_private;
5132 
5133 	mlx5_flow_list_flush(dev, &priv->flows);
5134 }
5135 
5136 /**
5137  * Get flow director information.
5138  *
5139  * @param dev
5140  *   Pointer to Ethernet device.
5141  * @param[out] fdir_info
5142  *   Resulting flow director information.
5143  */
5144 static void
5145 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5146 {
5147 	struct rte_eth_fdir_masks *mask =
5148 		&dev->data->dev_conf.fdir_conf.mask;
5149 
5150 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5151 	fdir_info->guarant_spc = 0;
5152 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5153 	fdir_info->max_flexpayload = 0;
5154 	fdir_info->flow_types_mask[0] = 0;
5155 	fdir_info->flex_payload_unit = 0;
5156 	fdir_info->max_flex_payload_segment_num = 0;
5157 	fdir_info->flex_payload_limit = 0;
5158 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5159 }
5160 
5161 /**
5162  * Deal with flow director operations.
5163  *
5164  * @param dev
5165  *   Pointer to Ethernet device.
5166  * @param filter_op
5167  *   Operation to perform.
5168  * @param arg
5169  *   Pointer to operation-specific structure.
5170  *
5171  * @return
5172  *   0 on success, a negative errno value otherwise and rte_errno is set.
5173  */
5174 static int
5175 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5176 		    void *arg)
5177 {
5178 	enum rte_fdir_mode fdir_mode =
5179 		dev->data->dev_conf.fdir_conf.mode;
5180 
5181 	if (filter_op == RTE_ETH_FILTER_NOP)
5182 		return 0;
5183 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5184 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5185 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5186 			dev->data->port_id, fdir_mode);
5187 		rte_errno = EINVAL;
5188 		return -rte_errno;
5189 	}
5190 	switch (filter_op) {
5191 	case RTE_ETH_FILTER_ADD:
5192 		return flow_fdir_filter_add(dev, arg);
5193 	case RTE_ETH_FILTER_UPDATE:
5194 		return flow_fdir_filter_update(dev, arg);
5195 	case RTE_ETH_FILTER_DELETE:
5196 		return flow_fdir_filter_delete(dev, arg);
5197 	case RTE_ETH_FILTER_FLUSH:
5198 		flow_fdir_filter_flush(dev);
5199 		break;
5200 	case RTE_ETH_FILTER_INFO:
5201 		flow_fdir_info_get(dev, arg);
5202 		break;
5203 	default:
5204 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5205 			dev->data->port_id, filter_op);
5206 		rte_errno = EINVAL;
5207 		return -rte_errno;
5208 	}
5209 	return 0;
5210 }
5211 
5212 /**
5213  * Manage filter operations.
5214  *
5215  * @param dev
5216  *   Pointer to Ethernet device structure.
5217  * @param filter_type
5218  *   Filter type.
5219  * @param filter_op
5220  *   Operation to perform.
5221  * @param arg
5222  *   Pointer to operation-specific structure.
5223  *
5224  * @return
5225  *   0 on success, a negative errno value otherwise and rte_errno is set.
5226  */
5227 int
5228 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5229 		     enum rte_filter_type filter_type,
5230 		     enum rte_filter_op filter_op,
5231 		     void *arg)
5232 {
5233 	switch (filter_type) {
5234 	case RTE_ETH_FILTER_GENERIC:
5235 		if (filter_op != RTE_ETH_FILTER_GET) {
5236 			rte_errno = EINVAL;
5237 			return -rte_errno;
5238 		}
5239 		*(const void **)arg = &mlx5_flow_ops;
5240 		return 0;
5241 	case RTE_ETH_FILTER_FDIR:
5242 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5243 	default:
5244 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5245 			dev->data->port_id, filter_type);
5246 		rte_errno = ENOTSUP;
5247 		return -rte_errno;
5248 	}
5249 	return 0;
5250 }
5251 
5252 /**
5253  * Create the needed meter and suffix tables.
5254  *
5255  * @param[in] dev
5256  *   Pointer to Ethernet device.
5257  * @param[in] fm
5258  *   Pointer to the flow meter.
5259  *
5260  * @return
5261  *   Pointer to table set on success, NULL otherwise.
5262  */
5263 struct mlx5_meter_domains_infos *
5264 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5265 			  const struct mlx5_flow_meter *fm)
5266 {
5267 	const struct mlx5_flow_driver_ops *fops;
5268 
5269 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5270 	return fops->create_mtr_tbls(dev, fm);
5271 }
5272 
5273 /**
5274  * Destroy the meter table set.
5275  *
5276  * @param[in] dev
5277  *   Pointer to Ethernet device.
5278  * @param[in] tbl
5279  *   Pointer to the meter table set.
5280  *
5281  * @return
5282  *   0 on success.
5283  */
5284 int
5285 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5286 			   struct mlx5_meter_domains_infos *tbls)
5287 {
5288 	const struct mlx5_flow_driver_ops *fops;
5289 
5290 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5291 	return fops->destroy_mtr_tbls(dev, tbls);
5292 }
5293 
5294 /**
5295  * Create policer rules.
5296  *
5297  * @param[in] dev
5298  *   Pointer to Ethernet device.
5299  * @param[in] fm
5300  *   Pointer to flow meter structure.
5301  * @param[in] attr
5302  *   Pointer to flow attributes.
5303  *
5304  * @return
5305  *   0 on success, -1 otherwise.
5306  */
5307 int
5308 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5309 			       struct mlx5_flow_meter *fm,
5310 			       const struct rte_flow_attr *attr)
5311 {
5312 	const struct mlx5_flow_driver_ops *fops;
5313 
5314 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5315 	return fops->create_policer_rules(dev, fm, attr);
5316 }
5317 
5318 /**
5319  * Destroy policer rules.
5320  *
5321  * @param[in] fm
5322  *   Pointer to flow meter structure.
5323  * @param[in] attr
5324  *   Pointer to flow attributes.
5325  *
5326  * @return
5327  *   0 on success, -1 otherwise.
5328  */
5329 int
5330 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5331 				struct mlx5_flow_meter *fm,
5332 				const struct rte_flow_attr *attr)
5333 {
5334 	const struct mlx5_flow_driver_ops *fops;
5335 
5336 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5337 	return fops->destroy_policer_rules(dev, fm, attr);
5338 }
5339 
5340 /**
5341  * Allocate a counter.
5342  *
5343  * @param[in] dev
5344  *   Pointer to Ethernet device structure.
5345  *
5346  * @return
5347  *   Pointer to allocated counter  on success, NULL otherwise.
5348  */
5349 struct mlx5_flow_counter *
5350 mlx5_counter_alloc(struct rte_eth_dev *dev)
5351 {
5352 	const struct mlx5_flow_driver_ops *fops;
5353 	struct rte_flow_attr attr = { .transfer = 0 };
5354 
5355 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5356 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5357 		return fops->counter_alloc(dev);
5358 	}
5359 	DRV_LOG(ERR,
5360 		"port %u counter allocate is not supported.",
5361 		 dev->data->port_id);
5362 	return NULL;
5363 }
5364 
5365 /**
5366  * Free a counter.
5367  *
5368  * @param[in] dev
5369  *   Pointer to Ethernet device structure.
5370  * @param[in] cnt
5371  *   Pointer to counter to be free.
5372  */
5373 void
5374 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5375 {
5376 	const struct mlx5_flow_driver_ops *fops;
5377 	struct rte_flow_attr attr = { .transfer = 0 };
5378 
5379 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5380 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5381 		fops->counter_free(dev, cnt);
5382 		return;
5383 	}
5384 	DRV_LOG(ERR,
5385 		"port %u counter free is not supported.",
5386 		 dev->data->port_id);
5387 }
5388 
5389 /**
5390  * Query counter statistics.
5391  *
5392  * @param[in] dev
5393  *   Pointer to Ethernet device structure.
5394  * @param[in] cnt
5395  *   Pointer to counter to query.
5396  * @param[in] clear
5397  *   Set to clear counter statistics.
5398  * @param[out] pkts
5399  *   The counter hits packets number to save.
5400  * @param[out] bytes
5401  *   The counter hits bytes number to save.
5402  *
5403  * @return
5404  *   0 on success, a negative errno value otherwise.
5405  */
5406 int
5407 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5408 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5409 {
5410 	const struct mlx5_flow_driver_ops *fops;
5411 	struct rte_flow_attr attr = { .transfer = 0 };
5412 
5413 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5414 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5415 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5416 	}
5417 	DRV_LOG(ERR,
5418 		"port %u counter query is not supported.",
5419 		 dev->data->port_id);
5420 	return -ENOTSUP;
5421 }
5422 
5423 #define MLX5_POOL_QUERY_FREQ_US 1000000
5424 
5425 /**
5426  * Set the periodic procedure for triggering asynchronous batch queries for all
5427  * the counter pools.
5428  *
5429  * @param[in] sh
5430  *   Pointer to mlx5_ibv_shared object.
5431  */
5432 void
5433 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5434 {
5435 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5436 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5437 	uint32_t us;
5438 
5439 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5440 	pools_n += rte_atomic16_read(&cont->n_valid);
5441 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5442 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5443 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5444 		sh->cmng.query_thread_on = 0;
5445 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5446 	} else {
5447 		sh->cmng.query_thread_on = 1;
5448 	}
5449 }
5450 
5451 /**
5452  * The periodic procedure for triggering asynchronous batch queries for all the
5453  * counter pools. This function is probably called by the host thread.
5454  *
5455  * @param[in] arg
5456  *   The parameter for the alarm process.
5457  */
5458 void
5459 mlx5_flow_query_alarm(void *arg)
5460 {
5461 	struct mlx5_ibv_shared *sh = arg;
5462 	struct mlx5_devx_obj *dcs;
5463 	uint16_t offset;
5464 	int ret;
5465 	uint8_t batch = sh->cmng.batch;
5466 	uint16_t pool_index = sh->cmng.pool_index;
5467 	struct mlx5_pools_container *cont;
5468 	struct mlx5_pools_container *mcont;
5469 	struct mlx5_flow_counter_pool *pool;
5470 
5471 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5472 		goto set_alarm;
5473 next_container:
5474 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5475 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5476 	/* Check if resize was done and need to flip a container. */
5477 	if (cont != mcont) {
5478 		if (cont->pools) {
5479 			/* Clean the old container. */
5480 			rte_free(cont->pools);
5481 			memset(cont, 0, sizeof(*cont));
5482 		}
5483 		rte_cio_wmb();
5484 		 /* Flip the host container. */
5485 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5486 		cont = mcont;
5487 	}
5488 	if (!cont->pools) {
5489 		/* 2 empty containers case is unexpected. */
5490 		if (unlikely(batch != sh->cmng.batch))
5491 			goto set_alarm;
5492 		batch ^= 0x1;
5493 		pool_index = 0;
5494 		goto next_container;
5495 	}
5496 	pool = cont->pools[pool_index];
5497 	if (pool->raw_hw)
5498 		/* There is a pool query in progress. */
5499 		goto set_alarm;
5500 	pool->raw_hw =
5501 		LIST_FIRST(&sh->cmng.free_stat_raws);
5502 	if (!pool->raw_hw)
5503 		/* No free counter statistics raw memory. */
5504 		goto set_alarm;
5505 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5506 							      (&pool->a64_dcs);
5507 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5508 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5509 					       offset, NULL, NULL,
5510 					       pool->raw_hw->mem_mng->dm->id,
5511 					       (void *)(uintptr_t)
5512 					       (pool->raw_hw->data + offset),
5513 					       sh->devx_comp,
5514 					       (uint64_t)(uintptr_t)pool);
5515 	if (ret) {
5516 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5517 			" %d", pool->min_dcs->id);
5518 		pool->raw_hw = NULL;
5519 		goto set_alarm;
5520 	}
5521 	pool->raw_hw->min_dcs_id = dcs->id;
5522 	LIST_REMOVE(pool->raw_hw, next);
5523 	sh->cmng.pending_queries++;
5524 	pool_index++;
5525 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5526 		batch ^= 0x1;
5527 		pool_index = 0;
5528 	}
5529 set_alarm:
5530 	sh->cmng.batch = batch;
5531 	sh->cmng.pool_index = pool_index;
5532 	mlx5_set_query_alarm(sh);
5533 }
5534 
5535 /**
5536  * Handler for the HW respond about ready values from an asynchronous batch
5537  * query. This function is probably called by the host thread.
5538  *
5539  * @param[in] sh
5540  *   The pointer to the shared IB device context.
5541  * @param[in] async_id
5542  *   The Devx async ID.
5543  * @param[in] status
5544  *   The status of the completion.
5545  */
5546 void
5547 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5548 				  uint64_t async_id, int status)
5549 {
5550 	struct mlx5_flow_counter_pool *pool =
5551 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5552 	struct mlx5_counter_stats_raw *raw_to_free;
5553 
5554 	if (unlikely(status)) {
5555 		raw_to_free = pool->raw_hw;
5556 	} else {
5557 		raw_to_free = pool->raw;
5558 		rte_spinlock_lock(&pool->sl);
5559 		pool->raw = pool->raw_hw;
5560 		rte_spinlock_unlock(&pool->sl);
5561 		rte_atomic64_add(&pool->query_gen, 1);
5562 		/* Be sure the new raw counters data is updated in memory. */
5563 		rte_cio_wmb();
5564 	}
5565 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5566 	pool->raw_hw = NULL;
5567 	sh->cmng.pending_queries--;
5568 }
5569 
5570 /**
5571  * Translate the rte_flow group index to HW table value.
5572  *
5573  * @param[in] attributes
5574  *   Pointer to flow attributes
5575  * @param[in] external
5576  *   Value is part of flow rule created by request external to PMD.
5577  * @param[in] group
5578  *   rte_flow group index value.
5579  * @param[out] table
5580  *   HW table value.
5581  * @param[out] error
5582  *   Pointer to error structure.
5583  *
5584  * @return
5585  *   0 on success, a negative errno value otherwise and rte_errno is set.
5586  */
5587 int
5588 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5589 			 uint32_t group, uint32_t *table,
5590 			 struct rte_flow_error *error)
5591 {
5592 	if (attributes->transfer && external) {
5593 		if (group == UINT32_MAX)
5594 			return rte_flow_error_set
5595 						(error, EINVAL,
5596 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5597 						 NULL,
5598 						 "group index not supported");
5599 		*table = group + 1;
5600 	} else {
5601 		*table = group;
5602 	}
5603 	return 0;
5604 }
5605 
5606 /**
5607  * Discover availability of metadata reg_c's.
5608  *
5609  * Iteratively use test flows to check availability.
5610  *
5611  * @param[in] dev
5612  *   Pointer to the Ethernet device structure.
5613  *
5614  * @return
5615  *   0 on success, a negative errno value otherwise and rte_errno is set.
5616  */
5617 int
5618 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5619 {
5620 	struct mlx5_priv *priv = dev->data->dev_private;
5621 	struct mlx5_dev_config *config = &priv->config;
5622 	enum modify_reg idx;
5623 	int n = 0;
5624 
5625 	/* reg_c[0] and reg_c[1] are reserved. */
5626 	config->flow_mreg_c[n++] = REG_C_0;
5627 	config->flow_mreg_c[n++] = REG_C_1;
5628 	/* Discover availability of other reg_c's. */
5629 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5630 		struct rte_flow_attr attr = {
5631 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5632 			.priority = MLX5_FLOW_PRIO_RSVD,
5633 			.ingress = 1,
5634 		};
5635 		struct rte_flow_item items[] = {
5636 			[0] = {
5637 				.type = RTE_FLOW_ITEM_TYPE_END,
5638 			},
5639 		};
5640 		struct rte_flow_action actions[] = {
5641 			[0] = {
5642 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5643 				.conf = &(struct mlx5_flow_action_copy_mreg){
5644 					.src = REG_C_1,
5645 					.dst = idx,
5646 				},
5647 			},
5648 			[1] = {
5649 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5650 				.conf = &(struct rte_flow_action_jump){
5651 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5652 				},
5653 			},
5654 			[2] = {
5655 				.type = RTE_FLOW_ACTION_TYPE_END,
5656 			},
5657 		};
5658 		struct rte_flow *flow;
5659 		struct rte_flow_error error;
5660 
5661 		if (!config->dv_flow_en)
5662 			break;
5663 		/* Create internal flow, validation skips copy action. */
5664 		flow = flow_list_create(dev, NULL, &attr, items,
5665 					actions, false, &error);
5666 		if (!flow)
5667 			continue;
5668 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5669 			config->flow_mreg_c[n++] = idx;
5670 		flow_list_destroy(dev, NULL, flow);
5671 	}
5672 	for (; n < MLX5_MREG_C_NUM; ++n)
5673 		config->flow_mreg_c[n] = REG_NONE;
5674 	return 0;
5675 }
5676 
5677 /**
5678  * Dump flow raw hw data to file
5679  *
5680  * @param[in] dev
5681  *    The pointer to Ethernet device.
5682  * @param[in] file
5683  *   A pointer to a file for output.
5684  * @param[out] error
5685  *   Perform verbose error reporting if not NULL. PMDs initialize this
5686  *   structure in case of error only.
5687  * @return
5688  *   0 on success, a nagative value otherwise.
5689  */
5690 int
5691 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5692 		   FILE *file,
5693 		   struct rte_flow_error *error __rte_unused)
5694 {
5695 	struct mlx5_priv *priv = dev->data->dev_private;
5696 	struct mlx5_ibv_shared *sh = priv->sh;
5697 
5698 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5699 				       sh->tx_domain, file);
5700 }
5701