xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 5693233823b09b527376e94ce24df71918632dfa)
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_glue.h>
31 #include <mlx5_devx_cmds.h>
32 #include <mlx5_prm.h>
33 
34 #include "mlx5_defs.h"
35 #include "mlx5.h"
36 #include "mlx5_flow.h"
37 #include "mlx5_rxtx.h"
38 
39 /* Dev ops structure defined in mlx5.c */
40 extern const struct eth_dev_ops mlx5_dev_ops;
41 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
42 
43 /** Device flow drivers. */
44 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
45 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
46 #endif
47 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
48 
49 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
50 
51 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
52 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
53 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
54 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
55 #endif
56 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
58 };
59 
60 enum mlx5_expansion {
61 	MLX5_EXPANSION_ROOT,
62 	MLX5_EXPANSION_ROOT_OUTER,
63 	MLX5_EXPANSION_ROOT_ETH_VLAN,
64 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_ETH,
66 	MLX5_EXPANSION_OUTER_ETH_VLAN,
67 	MLX5_EXPANSION_OUTER_VLAN,
68 	MLX5_EXPANSION_OUTER_IPV4,
69 	MLX5_EXPANSION_OUTER_IPV4_UDP,
70 	MLX5_EXPANSION_OUTER_IPV4_TCP,
71 	MLX5_EXPANSION_OUTER_IPV6,
72 	MLX5_EXPANSION_OUTER_IPV6_UDP,
73 	MLX5_EXPANSION_OUTER_IPV6_TCP,
74 	MLX5_EXPANSION_VXLAN,
75 	MLX5_EXPANSION_VXLAN_GPE,
76 	MLX5_EXPANSION_GRE,
77 	MLX5_EXPANSION_MPLS,
78 	MLX5_EXPANSION_ETH,
79 	MLX5_EXPANSION_ETH_VLAN,
80 	MLX5_EXPANSION_VLAN,
81 	MLX5_EXPANSION_IPV4,
82 	MLX5_EXPANSION_IPV4_UDP,
83 	MLX5_EXPANSION_IPV4_TCP,
84 	MLX5_EXPANSION_IPV6,
85 	MLX5_EXPANSION_IPV6_UDP,
86 	MLX5_EXPANSION_IPV6_TCP,
87 };
88 
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 	[MLX5_EXPANSION_ROOT] = {
92 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 						 MLX5_EXPANSION_IPV4,
94 						 MLX5_EXPANSION_IPV6),
95 		.type = RTE_FLOW_ITEM_TYPE_END,
96 	},
97 	[MLX5_EXPANSION_ROOT_OUTER] = {
98 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 						 MLX5_EXPANSION_OUTER_IPV4,
100 						 MLX5_EXPANSION_OUTER_IPV6),
101 		.type = RTE_FLOW_ITEM_TYPE_END,
102 	},
103 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 		.type = RTE_FLOW_ITEM_TYPE_END,
106 	},
107 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 		.type = RTE_FLOW_ITEM_TYPE_END,
110 	},
111 	[MLX5_EXPANSION_OUTER_ETH] = {
112 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 						 MLX5_EXPANSION_OUTER_IPV6,
114 						 MLX5_EXPANSION_MPLS),
115 		.type = RTE_FLOW_ITEM_TYPE_ETH,
116 		.rss_types = 0,
117 	},
118 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 		.type = RTE_FLOW_ITEM_TYPE_ETH,
121 		.rss_types = 0,
122 	},
123 	[MLX5_EXPANSION_OUTER_VLAN] = {
124 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 						 MLX5_EXPANSION_OUTER_IPV6),
126 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
127 	},
128 	[MLX5_EXPANSION_OUTER_IPV4] = {
129 		.next = RTE_FLOW_EXPAND_RSS_NEXT
130 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
131 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 			 MLX5_EXPANSION_GRE,
133 			 MLX5_EXPANSION_IPV4,
134 			 MLX5_EXPANSION_IPV6),
135 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
136 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
137 			ETH_RSS_NONFRAG_IPV4_OTHER,
138 	},
139 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
140 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
141 						 MLX5_EXPANSION_VXLAN_GPE),
142 		.type = RTE_FLOW_ITEM_TYPE_UDP,
143 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
144 	},
145 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
146 		.type = RTE_FLOW_ITEM_TYPE_TCP,
147 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
148 	},
149 	[MLX5_EXPANSION_OUTER_IPV6] = {
150 		.next = RTE_FLOW_EXPAND_RSS_NEXT
151 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
152 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
153 			 MLX5_EXPANSION_IPV4,
154 			 MLX5_EXPANSION_IPV6),
155 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
156 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
157 			ETH_RSS_NONFRAG_IPV6_OTHER,
158 	},
159 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
160 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
161 						 MLX5_EXPANSION_VXLAN_GPE),
162 		.type = RTE_FLOW_ITEM_TYPE_UDP,
163 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
164 	},
165 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
166 		.type = RTE_FLOW_ITEM_TYPE_TCP,
167 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
168 	},
169 	[MLX5_EXPANSION_VXLAN] = {
170 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
171 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
172 	},
173 	[MLX5_EXPANSION_VXLAN_GPE] = {
174 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
175 						 MLX5_EXPANSION_IPV4,
176 						 MLX5_EXPANSION_IPV6),
177 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
178 	},
179 	[MLX5_EXPANSION_GRE] = {
180 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
181 		.type = RTE_FLOW_ITEM_TYPE_GRE,
182 	},
183 	[MLX5_EXPANSION_MPLS] = {
184 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
185 						 MLX5_EXPANSION_IPV6),
186 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
187 	},
188 	[MLX5_EXPANSION_ETH] = {
189 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
190 						 MLX5_EXPANSION_IPV6),
191 		.type = RTE_FLOW_ITEM_TYPE_ETH,
192 	},
193 	[MLX5_EXPANSION_ETH_VLAN] = {
194 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
195 		.type = RTE_FLOW_ITEM_TYPE_ETH,
196 	},
197 	[MLX5_EXPANSION_VLAN] = {
198 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
199 						 MLX5_EXPANSION_IPV6),
200 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
201 	},
202 	[MLX5_EXPANSION_IPV4] = {
203 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
204 						 MLX5_EXPANSION_IPV4_TCP),
205 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
206 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
207 			ETH_RSS_NONFRAG_IPV4_OTHER,
208 	},
209 	[MLX5_EXPANSION_IPV4_UDP] = {
210 		.type = RTE_FLOW_ITEM_TYPE_UDP,
211 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
212 	},
213 	[MLX5_EXPANSION_IPV4_TCP] = {
214 		.type = RTE_FLOW_ITEM_TYPE_TCP,
215 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
216 	},
217 	[MLX5_EXPANSION_IPV6] = {
218 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
219 						 MLX5_EXPANSION_IPV6_TCP),
220 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
221 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
222 			ETH_RSS_NONFRAG_IPV6_OTHER,
223 	},
224 	[MLX5_EXPANSION_IPV6_UDP] = {
225 		.type = RTE_FLOW_ITEM_TYPE_UDP,
226 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
227 	},
228 	[MLX5_EXPANSION_IPV6_TCP] = {
229 		.type = RTE_FLOW_ITEM_TYPE_TCP,
230 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
231 	},
232 };
233 
234 static const struct rte_flow_ops mlx5_flow_ops = {
235 	.validate = mlx5_flow_validate,
236 	.create = mlx5_flow_create,
237 	.destroy = mlx5_flow_destroy,
238 	.flush = mlx5_flow_flush,
239 	.isolate = mlx5_flow_isolate,
240 	.query = mlx5_flow_query,
241 	.dev_dump = mlx5_flow_dev_dump,
242 };
243 
244 /* Convert FDIR request to Generic flow. */
245 struct mlx5_fdir {
246 	struct rte_flow_attr attr;
247 	struct rte_flow_item items[4];
248 	struct rte_flow_item_eth l2;
249 	struct rte_flow_item_eth l2_mask;
250 	union {
251 		struct rte_flow_item_ipv4 ipv4;
252 		struct rte_flow_item_ipv6 ipv6;
253 	} l3;
254 	union {
255 		struct rte_flow_item_ipv4 ipv4;
256 		struct rte_flow_item_ipv6 ipv6;
257 	} l3_mask;
258 	union {
259 		struct rte_flow_item_udp udp;
260 		struct rte_flow_item_tcp tcp;
261 	} l4;
262 	union {
263 		struct rte_flow_item_udp udp;
264 		struct rte_flow_item_tcp tcp;
265 	} l4_mask;
266 	struct rte_flow_action actions[2];
267 	struct rte_flow_action_queue queue;
268 };
269 
270 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
271 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
272 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
273 };
274 
275 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
276 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
277 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
278 	{ 9, 10, 11 }, { 12, 13, 14 },
279 };
280 
281 /* Tunnel information. */
282 struct mlx5_flow_tunnel_info {
283 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
284 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
285 };
286 
287 static struct mlx5_flow_tunnel_info tunnels_info[] = {
288 	{
289 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
290 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
291 	},
292 	{
293 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
294 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
295 	},
296 	{
297 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
298 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
299 	},
300 	{
301 		.tunnel = MLX5_FLOW_LAYER_GRE,
302 		.ptype = RTE_PTYPE_TUNNEL_GRE,
303 	},
304 	{
305 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
306 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
307 	},
308 	{
309 		.tunnel = MLX5_FLOW_LAYER_MPLS,
310 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
311 	},
312 	{
313 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
314 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
315 	},
316 	{
317 		.tunnel = MLX5_FLOW_LAYER_IPIP,
318 		.ptype = RTE_PTYPE_TUNNEL_IP,
319 	},
320 	{
321 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
322 		.ptype = RTE_PTYPE_TUNNEL_IP,
323 	},
324 	{
325 		.tunnel = MLX5_FLOW_LAYER_GTP,
326 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
327 	},
328 };
329 
330 /**
331  * Translate tag ID to register.
332  *
333  * @param[in] dev
334  *   Pointer to the Ethernet device structure.
335  * @param[in] feature
336  *   The feature that request the register.
337  * @param[in] id
338  *   The request register ID.
339  * @param[out] error
340  *   Error description in case of any.
341  *
342  * @return
343  *   The request register on success, a negative errno
344  *   value otherwise and rte_errno is set.
345  */
346 int
347 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
348 		     enum mlx5_feature_name feature,
349 		     uint32_t id,
350 		     struct rte_flow_error *error)
351 {
352 	struct mlx5_priv *priv = dev->data->dev_private;
353 	struct mlx5_dev_config *config = &priv->config;
354 	enum modify_reg start_reg;
355 	bool skip_mtr_reg = false;
356 
357 	switch (feature) {
358 	case MLX5_HAIRPIN_RX:
359 		return REG_B;
360 	case MLX5_HAIRPIN_TX:
361 		return REG_A;
362 	case MLX5_METADATA_RX:
363 		switch (config->dv_xmeta_en) {
364 		case MLX5_XMETA_MODE_LEGACY:
365 			return REG_B;
366 		case MLX5_XMETA_MODE_META16:
367 			return REG_C_0;
368 		case MLX5_XMETA_MODE_META32:
369 			return REG_C_1;
370 		}
371 		break;
372 	case MLX5_METADATA_TX:
373 		return REG_A;
374 	case MLX5_METADATA_FDB:
375 		switch (config->dv_xmeta_en) {
376 		case MLX5_XMETA_MODE_LEGACY:
377 			return REG_NONE;
378 		case MLX5_XMETA_MODE_META16:
379 			return REG_C_0;
380 		case MLX5_XMETA_MODE_META32:
381 			return REG_C_1;
382 		}
383 		break;
384 	case MLX5_FLOW_MARK:
385 		switch (config->dv_xmeta_en) {
386 		case MLX5_XMETA_MODE_LEGACY:
387 			return REG_NONE;
388 		case MLX5_XMETA_MODE_META16:
389 			return REG_C_1;
390 		case MLX5_XMETA_MODE_META32:
391 			return REG_C_0;
392 		}
393 		break;
394 	case MLX5_MTR_SFX:
395 		/*
396 		 * If meter color and flow match share one register, flow match
397 		 * should use the meter color register for match.
398 		 */
399 		if (priv->mtr_reg_share)
400 			return priv->mtr_color_reg;
401 		else
402 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
403 			       REG_C_3;
404 	case MLX5_MTR_COLOR:
405 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
406 		return priv->mtr_color_reg;
407 	case MLX5_COPY_MARK:
408 		/*
409 		 * Metadata COPY_MARK register using is in meter suffix sub
410 		 * flow while with meter. It's safe to share the same register.
411 		 */
412 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
413 	case MLX5_APP_TAG:
414 		/*
415 		 * If meter is enable, it will engage the register for color
416 		 * match and flow match. If meter color match is not using the
417 		 * REG_C_2, need to skip the REG_C_x be used by meter color
418 		 * match.
419 		 * If meter is disable, free to use all available registers.
420 		 */
421 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
422 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
423 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
424 		if (id > (REG_C_7 - start_reg))
425 			return rte_flow_error_set(error, EINVAL,
426 						  RTE_FLOW_ERROR_TYPE_ITEM,
427 						  NULL, "invalid tag id");
428 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
429 			return rte_flow_error_set(error, ENOTSUP,
430 						  RTE_FLOW_ERROR_TYPE_ITEM,
431 						  NULL, "unsupported tag id");
432 		/*
433 		 * This case means meter is using the REG_C_x great than 2.
434 		 * Take care not to conflict with meter color REG_C_x.
435 		 * If the available index REG_C_y >= REG_C_x, skip the
436 		 * color register.
437 		 */
438 		if (skip_mtr_reg && config->flow_mreg_c
439 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
440 			if (config->flow_mreg_c
441 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
442 				return config->flow_mreg_c
443 					       [id + 1 + start_reg - REG_C_0];
444 			return rte_flow_error_set(error, ENOTSUP,
445 						  RTE_FLOW_ERROR_TYPE_ITEM,
446 						  NULL, "unsupported tag id");
447 		}
448 		return config->flow_mreg_c[id + start_reg - REG_C_0];
449 	}
450 	MLX5_ASSERT(false);
451 	return rte_flow_error_set(error, EINVAL,
452 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
453 				  NULL, "invalid feature name");
454 }
455 
456 /**
457  * Check extensive flow metadata register support.
458  *
459  * @param dev
460  *   Pointer to rte_eth_dev structure.
461  *
462  * @return
463  *   True if device supports extensive flow metadata register, otherwise false.
464  */
465 bool
466 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
467 {
468 	struct mlx5_priv *priv = dev->data->dev_private;
469 	struct mlx5_dev_config *config = &priv->config;
470 
471 	/*
472 	 * Having available reg_c can be regarded inclusively as supporting
473 	 * extensive flow metadata register, which could mean,
474 	 * - metadata register copy action by modify header.
475 	 * - 16 modify header actions is supported.
476 	 * - reg_c's are preserved across different domain (FDB and NIC) on
477 	 *   packet loopback by flow lookup miss.
478 	 */
479 	return config->flow_mreg_c[2] != REG_NONE;
480 }
481 
482 /**
483  * Discover the maximum number of priority available.
484  *
485  * @param[in] dev
486  *   Pointer to the Ethernet device structure.
487  *
488  * @return
489  *   number of supported flow priority on success, a negative errno
490  *   value otherwise and rte_errno is set.
491  */
492 int
493 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
494 {
495 	struct mlx5_priv *priv = dev->data->dev_private;
496 	struct {
497 		struct ibv_flow_attr attr;
498 		struct ibv_flow_spec_eth eth;
499 		struct ibv_flow_spec_action_drop drop;
500 	} flow_attr = {
501 		.attr = {
502 			.num_of_specs = 2,
503 			.port = (uint8_t)priv->ibv_port,
504 		},
505 		.eth = {
506 			.type = IBV_FLOW_SPEC_ETH,
507 			.size = sizeof(struct ibv_flow_spec_eth),
508 		},
509 		.drop = {
510 			.size = sizeof(struct ibv_flow_spec_action_drop),
511 			.type = IBV_FLOW_SPEC_ACTION_DROP,
512 		},
513 	};
514 	struct ibv_flow *flow;
515 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
516 	uint16_t vprio[] = { 8, 16 };
517 	int i;
518 	int priority = 0;
519 
520 	if (!drop) {
521 		rte_errno = ENOTSUP;
522 		return -rte_errno;
523 	}
524 	for (i = 0; i != RTE_DIM(vprio); i++) {
525 		flow_attr.attr.priority = vprio[i] - 1;
526 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
527 		if (!flow)
528 			break;
529 		claim_zero(mlx5_glue->destroy_flow(flow));
530 		priority = vprio[i];
531 	}
532 	mlx5_hrxq_drop_release(dev);
533 	switch (priority) {
534 	case 8:
535 		priority = RTE_DIM(priority_map_3);
536 		break;
537 	case 16:
538 		priority = RTE_DIM(priority_map_5);
539 		break;
540 	default:
541 		rte_errno = ENOTSUP;
542 		DRV_LOG(ERR,
543 			"port %u verbs maximum priority: %d expected 8/16",
544 			dev->data->port_id, priority);
545 		return -rte_errno;
546 	}
547 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
548 		dev->data->port_id, priority);
549 	return priority;
550 }
551 
552 /**
553  * Adjust flow priority based on the highest layer and the request priority.
554  *
555  * @param[in] dev
556  *   Pointer to the Ethernet device structure.
557  * @param[in] priority
558  *   The rule base priority.
559  * @param[in] subpriority
560  *   The priority based on the items.
561  *
562  * @return
563  *   The new priority.
564  */
565 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
566 				   uint32_t subpriority)
567 {
568 	uint32_t res = 0;
569 	struct mlx5_priv *priv = dev->data->dev_private;
570 
571 	switch (priv->config.flow_prio) {
572 	case RTE_DIM(priority_map_3):
573 		res = priority_map_3[priority][subpriority];
574 		break;
575 	case RTE_DIM(priority_map_5):
576 		res = priority_map_5[priority][subpriority];
577 		break;
578 	}
579 	return  res;
580 }
581 
582 /**
583  * Verify the @p item specifications (spec, last, mask) are compatible with the
584  * NIC capabilities.
585  *
586  * @param[in] item
587  *   Item specification.
588  * @param[in] mask
589  *   @p item->mask or flow default bit-masks.
590  * @param[in] nic_mask
591  *   Bit-masks covering supported fields by the NIC to compare with user mask.
592  * @param[in] size
593  *   Bit-masks size in bytes.
594  * @param[out] error
595  *   Pointer to error structure.
596  *
597  * @return
598  *   0 on success, a negative errno value otherwise and rte_errno is set.
599  */
600 int
601 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
602 			  const uint8_t *mask,
603 			  const uint8_t *nic_mask,
604 			  unsigned int size,
605 			  struct rte_flow_error *error)
606 {
607 	unsigned int i;
608 
609 	MLX5_ASSERT(nic_mask);
610 	for (i = 0; i < size; ++i)
611 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
612 			return rte_flow_error_set(error, ENOTSUP,
613 						  RTE_FLOW_ERROR_TYPE_ITEM,
614 						  item,
615 						  "mask enables non supported"
616 						  " bits");
617 	if (!item->spec && (item->mask || item->last))
618 		return rte_flow_error_set(error, EINVAL,
619 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
620 					  "mask/last without a spec is not"
621 					  " supported");
622 	if (item->spec && item->last) {
623 		uint8_t spec[size];
624 		uint8_t last[size];
625 		unsigned int i;
626 		int ret;
627 
628 		for (i = 0; i < size; ++i) {
629 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
630 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
631 		}
632 		ret = memcmp(spec, last, size);
633 		if (ret != 0)
634 			return rte_flow_error_set(error, EINVAL,
635 						  RTE_FLOW_ERROR_TYPE_ITEM,
636 						  item,
637 						  "range is not valid");
638 	}
639 	return 0;
640 }
641 
642 /**
643  * Adjust the hash fields according to the @p flow information.
644  *
645  * @param[in] dev_flow.
646  *   Pointer to the mlx5_flow.
647  * @param[in] tunnel
648  *   1 when the hash field is for a tunnel item.
649  * @param[in] layer_types
650  *   ETH_RSS_* types.
651  * @param[in] hash_fields
652  *   Item hash fields.
653  *
654  * @return
655  *   The hash fields that should be used.
656  */
657 uint64_t
658 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
659 			    int tunnel __rte_unused, uint64_t layer_types,
660 			    uint64_t hash_fields)
661 {
662 	struct rte_flow *flow = dev_flow->flow;
663 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
664 	int rss_request_inner = flow->rss.level >= 2;
665 
666 	/* Check RSS hash level for tunnel. */
667 	if (tunnel && rss_request_inner)
668 		hash_fields |= IBV_RX_HASH_INNER;
669 	else if (tunnel || rss_request_inner)
670 		return 0;
671 #endif
672 	/* Check if requested layer matches RSS hash fields. */
673 	if (!(flow->rss.types & layer_types))
674 		return 0;
675 	return hash_fields;
676 }
677 
678 /**
679  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
680  * if several tunnel rules are used on this queue, the tunnel ptype will be
681  * cleared.
682  *
683  * @param rxq_ctrl
684  *   Rx queue to update.
685  */
686 static void
687 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
688 {
689 	unsigned int i;
690 	uint32_t tunnel_ptype = 0;
691 
692 	/* Look up for the ptype to use. */
693 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
694 		if (!rxq_ctrl->flow_tunnels_n[i])
695 			continue;
696 		if (!tunnel_ptype) {
697 			tunnel_ptype = tunnels_info[i].ptype;
698 		} else {
699 			tunnel_ptype = 0;
700 			break;
701 		}
702 	}
703 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
704 }
705 
706 /**
707  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
708  * flow.
709  *
710  * @param[in] dev
711  *   Pointer to the Ethernet device structure.
712  * @param[in] dev_flow
713  *   Pointer to device flow structure.
714  */
715 static void
716 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
717 {
718 	struct mlx5_priv *priv = dev->data->dev_private;
719 	struct rte_flow *flow = dev_flow->flow;
720 	const int mark = !!(dev_flow->actions &
721 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
722 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
723 	unsigned int i;
724 
725 	for (i = 0; i != flow->rss.queue_num; ++i) {
726 		int idx = (*flow->rss.queue)[i];
727 		struct mlx5_rxq_ctrl *rxq_ctrl =
728 			container_of((*priv->rxqs)[idx],
729 				     struct mlx5_rxq_ctrl, rxq);
730 
731 		/*
732 		 * To support metadata register copy on Tx loopback,
733 		 * this must be always enabled (metadata may arive
734 		 * from other port - not from local flows only.
735 		 */
736 		if (priv->config.dv_flow_en &&
737 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
738 		    mlx5_flow_ext_mreg_supported(dev)) {
739 			rxq_ctrl->rxq.mark = 1;
740 			rxq_ctrl->flow_mark_n = 1;
741 		} else if (mark) {
742 			rxq_ctrl->rxq.mark = 1;
743 			rxq_ctrl->flow_mark_n++;
744 		}
745 		if (tunnel) {
746 			unsigned int j;
747 
748 			/* Increase the counter matching the flow. */
749 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
750 				if ((tunnels_info[j].tunnel &
751 				     dev_flow->layers) ==
752 				    tunnels_info[j].tunnel) {
753 					rxq_ctrl->flow_tunnels_n[j]++;
754 					break;
755 				}
756 			}
757 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
758 		}
759 	}
760 }
761 
762 /**
763  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
764  *
765  * @param[in] dev
766  *   Pointer to the Ethernet device structure.
767  * @param[in] flow
768  *   Pointer to flow structure.
769  */
770 static void
771 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
772 {
773 	struct mlx5_flow *dev_flow;
774 
775 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
776 		flow_drv_rxq_flags_set(dev, dev_flow);
777 }
778 
779 /**
780  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
781  * device flow if no other flow uses it with the same kind of request.
782  *
783  * @param dev
784  *   Pointer to Ethernet device.
785  * @param[in] dev_flow
786  *   Pointer to the device flow.
787  */
788 static void
789 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
790 {
791 	struct mlx5_priv *priv = dev->data->dev_private;
792 	struct rte_flow *flow = dev_flow->flow;
793 	const int mark = !!(dev_flow->actions &
794 			    (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
795 	const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
796 	unsigned int i;
797 
798 	MLX5_ASSERT(dev->data->dev_started);
799 	for (i = 0; i != flow->rss.queue_num; ++i) {
800 		int idx = (*flow->rss.queue)[i];
801 		struct mlx5_rxq_ctrl *rxq_ctrl =
802 			container_of((*priv->rxqs)[idx],
803 				     struct mlx5_rxq_ctrl, rxq);
804 
805 		if (priv->config.dv_flow_en &&
806 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
807 		    mlx5_flow_ext_mreg_supported(dev)) {
808 			rxq_ctrl->rxq.mark = 1;
809 			rxq_ctrl->flow_mark_n = 1;
810 		} else if (mark) {
811 			rxq_ctrl->flow_mark_n--;
812 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
813 		}
814 		if (tunnel) {
815 			unsigned int j;
816 
817 			/* Decrease the counter matching the flow. */
818 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
819 				if ((tunnels_info[j].tunnel &
820 				     dev_flow->layers) ==
821 				    tunnels_info[j].tunnel) {
822 					rxq_ctrl->flow_tunnels_n[j]--;
823 					break;
824 				}
825 			}
826 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
827 		}
828 	}
829 }
830 
831 /**
832  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
833  * @p flow if no other flow uses it with the same kind of request.
834  *
835  * @param dev
836  *   Pointer to Ethernet device.
837  * @param[in] flow
838  *   Pointer to the flow.
839  */
840 static void
841 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
842 {
843 	struct mlx5_flow *dev_flow;
844 
845 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
846 		flow_drv_rxq_flags_trim(dev, dev_flow);
847 }
848 
849 /**
850  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
851  *
852  * @param dev
853  *   Pointer to Ethernet device.
854  */
855 static void
856 flow_rxq_flags_clear(struct rte_eth_dev *dev)
857 {
858 	struct mlx5_priv *priv = dev->data->dev_private;
859 	unsigned int i;
860 
861 	for (i = 0; i != priv->rxqs_n; ++i) {
862 		struct mlx5_rxq_ctrl *rxq_ctrl;
863 		unsigned int j;
864 
865 		if (!(*priv->rxqs)[i])
866 			continue;
867 		rxq_ctrl = container_of((*priv->rxqs)[i],
868 					struct mlx5_rxq_ctrl, rxq);
869 		rxq_ctrl->flow_mark_n = 0;
870 		rxq_ctrl->rxq.mark = 0;
871 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
872 			rxq_ctrl->flow_tunnels_n[j] = 0;
873 		rxq_ctrl->rxq.tunnel = 0;
874 	}
875 }
876 
877 /*
878  * return a pointer to the desired action in the list of actions.
879  *
880  * @param[in] actions
881  *   The list of actions to search the action in.
882  * @param[in] action
883  *   The action to find.
884  *
885  * @return
886  *   Pointer to the action in the list, if found. NULL otherwise.
887  */
888 const struct rte_flow_action *
889 mlx5_flow_find_action(const struct rte_flow_action *actions,
890 		      enum rte_flow_action_type action)
891 {
892 	if (actions == NULL)
893 		return NULL;
894 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
895 		if (actions->type == action)
896 			return actions;
897 	return NULL;
898 }
899 
900 /*
901  * Validate the flag action.
902  *
903  * @param[in] action_flags
904  *   Bit-fields that holds the actions detected until now.
905  * @param[in] attr
906  *   Attributes of flow that includes this action.
907  * @param[out] error
908  *   Pointer to error structure.
909  *
910  * @return
911  *   0 on success, a negative errno value otherwise and rte_errno is set.
912  */
913 int
914 mlx5_flow_validate_action_flag(uint64_t action_flags,
915 			       const struct rte_flow_attr *attr,
916 			       struct rte_flow_error *error)
917 {
918 	if (action_flags & MLX5_FLOW_ACTION_MARK)
919 		return rte_flow_error_set(error, EINVAL,
920 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
921 					  "can't mark and flag in same flow");
922 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
923 		return rte_flow_error_set(error, EINVAL,
924 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
925 					  "can't have 2 flag"
926 					  " actions in same flow");
927 	if (attr->egress)
928 		return rte_flow_error_set(error, ENOTSUP,
929 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
930 					  "flag action not supported for "
931 					  "egress");
932 	return 0;
933 }
934 
935 /*
936  * Validate the mark action.
937  *
938  * @param[in] action
939  *   Pointer to the queue action.
940  * @param[in] action_flags
941  *   Bit-fields that holds the actions detected until now.
942  * @param[in] attr
943  *   Attributes of flow that includes this action.
944  * @param[out] error
945  *   Pointer to error structure.
946  *
947  * @return
948  *   0 on success, a negative errno value otherwise and rte_errno is set.
949  */
950 int
951 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
952 			       uint64_t action_flags,
953 			       const struct rte_flow_attr *attr,
954 			       struct rte_flow_error *error)
955 {
956 	const struct rte_flow_action_mark *mark = action->conf;
957 
958 	if (!mark)
959 		return rte_flow_error_set(error, EINVAL,
960 					  RTE_FLOW_ERROR_TYPE_ACTION,
961 					  action,
962 					  "configuration cannot be null");
963 	if (mark->id >= MLX5_FLOW_MARK_MAX)
964 		return rte_flow_error_set(error, EINVAL,
965 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
966 					  &mark->id,
967 					  "mark id must in 0 <= id < "
968 					  RTE_STR(MLX5_FLOW_MARK_MAX));
969 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
970 		return rte_flow_error_set(error, EINVAL,
971 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
972 					  "can't flag and mark in same flow");
973 	if (action_flags & MLX5_FLOW_ACTION_MARK)
974 		return rte_flow_error_set(error, EINVAL,
975 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 					  "can't have 2 mark actions in same"
977 					  " flow");
978 	if (attr->egress)
979 		return rte_flow_error_set(error, ENOTSUP,
980 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
981 					  "mark action not supported for "
982 					  "egress");
983 	return 0;
984 }
985 
986 /*
987  * Validate the drop action.
988  *
989  * @param[in] action_flags
990  *   Bit-fields that holds the actions detected until now.
991  * @param[in] attr
992  *   Attributes of flow that includes this action.
993  * @param[out] error
994  *   Pointer to error structure.
995  *
996  * @return
997  *   0 on success, a negative errno value otherwise and rte_errno is set.
998  */
999 int
1000 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1001 			       const struct rte_flow_attr *attr,
1002 			       struct rte_flow_error *error)
1003 {
1004 	if (attr->egress)
1005 		return rte_flow_error_set(error, ENOTSUP,
1006 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007 					  "drop action not supported for "
1008 					  "egress");
1009 	return 0;
1010 }
1011 
1012 /*
1013  * Validate the queue action.
1014  *
1015  * @param[in] action
1016  *   Pointer to the queue action.
1017  * @param[in] action_flags
1018  *   Bit-fields that holds the actions detected until now.
1019  * @param[in] dev
1020  *   Pointer to the Ethernet device structure.
1021  * @param[in] attr
1022  *   Attributes of flow that includes this action.
1023  * @param[out] error
1024  *   Pointer to error structure.
1025  *
1026  * @return
1027  *   0 on success, a negative errno value otherwise and rte_errno is set.
1028  */
1029 int
1030 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1031 				uint64_t action_flags,
1032 				struct rte_eth_dev *dev,
1033 				const struct rte_flow_attr *attr,
1034 				struct rte_flow_error *error)
1035 {
1036 	struct mlx5_priv *priv = dev->data->dev_private;
1037 	const struct rte_flow_action_queue *queue = action->conf;
1038 
1039 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1040 		return rte_flow_error_set(error, EINVAL,
1041 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1042 					  "can't have 2 fate actions in"
1043 					  " same flow");
1044 	if (!priv->rxqs_n)
1045 		return rte_flow_error_set(error, EINVAL,
1046 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1047 					  NULL, "No Rx queues configured");
1048 	if (queue->index >= priv->rxqs_n)
1049 		return rte_flow_error_set(error, EINVAL,
1050 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1051 					  &queue->index,
1052 					  "queue index out of range");
1053 	if (!(*priv->rxqs)[queue->index])
1054 		return rte_flow_error_set(error, EINVAL,
1055 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1056 					  &queue->index,
1057 					  "queue is not configured");
1058 	if (attr->egress)
1059 		return rte_flow_error_set(error, ENOTSUP,
1060 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1061 					  "queue action not supported for "
1062 					  "egress");
1063 	return 0;
1064 }
1065 
1066 /*
1067  * Validate the rss action.
1068  *
1069  * @param[in] action
1070  *   Pointer to the queue action.
1071  * @param[in] action_flags
1072  *   Bit-fields that holds the actions detected until now.
1073  * @param[in] dev
1074  *   Pointer to the Ethernet device structure.
1075  * @param[in] attr
1076  *   Attributes of flow that includes this action.
1077  * @param[in] item_flags
1078  *   Items that were detected.
1079  * @param[out] error
1080  *   Pointer to error structure.
1081  *
1082  * @return
1083  *   0 on success, a negative errno value otherwise and rte_errno is set.
1084  */
1085 int
1086 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1087 			      uint64_t action_flags,
1088 			      struct rte_eth_dev *dev,
1089 			      const struct rte_flow_attr *attr,
1090 			      uint64_t item_flags,
1091 			      struct rte_flow_error *error)
1092 {
1093 	struct mlx5_priv *priv = dev->data->dev_private;
1094 	const struct rte_flow_action_rss *rss = action->conf;
1095 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1096 	unsigned int i;
1097 
1098 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1099 		return rte_flow_error_set(error, EINVAL,
1100 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1101 					  "can't have 2 fate actions"
1102 					  " in same flow");
1103 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1104 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1105 		return rte_flow_error_set(error, ENOTSUP,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &rss->func,
1108 					  "RSS hash function not supported");
1109 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1110 	if (rss->level > 2)
1111 #else
1112 	if (rss->level > 1)
1113 #endif
1114 		return rte_flow_error_set(error, ENOTSUP,
1115 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 					  &rss->level,
1117 					  "tunnel RSS is not supported");
1118 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1119 	if (rss->key_len == 0 && rss->key != NULL)
1120 		return rte_flow_error_set(error, ENOTSUP,
1121 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1122 					  &rss->key_len,
1123 					  "RSS hash key length 0");
1124 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1125 		return rte_flow_error_set(error, ENOTSUP,
1126 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1127 					  &rss->key_len,
1128 					  "RSS hash key too small");
1129 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1130 		return rte_flow_error_set(error, ENOTSUP,
1131 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1132 					  &rss->key_len,
1133 					  "RSS hash key too large");
1134 	if (rss->queue_num > priv->config.ind_table_max_size)
1135 		return rte_flow_error_set(error, ENOTSUP,
1136 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1137 					  &rss->queue_num,
1138 					  "number of queues too large");
1139 	if (rss->types & MLX5_RSS_HF_MASK)
1140 		return rte_flow_error_set(error, ENOTSUP,
1141 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1142 					  &rss->types,
1143 					  "some RSS protocols are not"
1144 					  " supported");
1145 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1146 	    !(rss->types & ETH_RSS_IP))
1147 		return rte_flow_error_set(error, EINVAL,
1148 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1149 					  "L3 partial RSS requested but L3 RSS"
1150 					  " type not specified");
1151 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1152 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1153 		return rte_flow_error_set(error, EINVAL,
1154 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1155 					  "L4 partial RSS requested but L4 RSS"
1156 					  " type not specified");
1157 	if (!priv->rxqs_n)
1158 		return rte_flow_error_set(error, EINVAL,
1159 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1160 					  NULL, "No Rx queues configured");
1161 	if (!rss->queue_num)
1162 		return rte_flow_error_set(error, EINVAL,
1163 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1164 					  NULL, "No queues configured");
1165 	for (i = 0; i != rss->queue_num; ++i) {
1166 		if (rss->queue[i] >= priv->rxqs_n)
1167 			return rte_flow_error_set
1168 				(error, EINVAL,
1169 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 				 &rss->queue[i], "queue index out of range");
1171 		if (!(*priv->rxqs)[rss->queue[i]])
1172 			return rte_flow_error_set
1173 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1174 				 &rss->queue[i], "queue is not configured");
1175 	}
1176 	if (attr->egress)
1177 		return rte_flow_error_set(error, ENOTSUP,
1178 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1179 					  "rss action not supported for "
1180 					  "egress");
1181 	if (rss->level > 1 &&  !tunnel)
1182 		return rte_flow_error_set(error, EINVAL,
1183 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1184 					  "inner RSS is not supported for "
1185 					  "non-tunnel flows");
1186 	return 0;
1187 }
1188 
1189 /*
1190  * Validate the count action.
1191  *
1192  * @param[in] dev
1193  *   Pointer to the Ethernet device structure.
1194  * @param[in] attr
1195  *   Attributes of flow that includes this action.
1196  * @param[out] error
1197  *   Pointer to error structure.
1198  *
1199  * @return
1200  *   0 on success, a negative errno value otherwise and rte_errno is set.
1201  */
1202 int
1203 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1204 				const struct rte_flow_attr *attr,
1205 				struct rte_flow_error *error)
1206 {
1207 	if (attr->egress)
1208 		return rte_flow_error_set(error, ENOTSUP,
1209 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1210 					  "count action not supported for "
1211 					  "egress");
1212 	return 0;
1213 }
1214 
1215 /**
1216  * Verify the @p attributes will be correctly understood by the NIC and store
1217  * them in the @p flow if everything is correct.
1218  *
1219  * @param[in] dev
1220  *   Pointer to the Ethernet device structure.
1221  * @param[in] attributes
1222  *   Pointer to flow attributes
1223  * @param[out] error
1224  *   Pointer to error structure.
1225  *
1226  * @return
1227  *   0 on success, a negative errno value otherwise and rte_errno is set.
1228  */
1229 int
1230 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1231 			      const struct rte_flow_attr *attributes,
1232 			      struct rte_flow_error *error)
1233 {
1234 	struct mlx5_priv *priv = dev->data->dev_private;
1235 	uint32_t priority_max = priv->config.flow_prio - 1;
1236 
1237 	if (attributes->group)
1238 		return rte_flow_error_set(error, ENOTSUP,
1239 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1240 					  NULL, "groups is not supported");
1241 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1242 	    attributes->priority >= priority_max)
1243 		return rte_flow_error_set(error, ENOTSUP,
1244 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1245 					  NULL, "priority out of range");
1246 	if (attributes->egress)
1247 		return rte_flow_error_set(error, ENOTSUP,
1248 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1249 					  "egress is not supported");
1250 	if (attributes->transfer && !priv->config.dv_esw_en)
1251 		return rte_flow_error_set(error, ENOTSUP,
1252 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1253 					  NULL, "transfer is not supported");
1254 	if (!attributes->ingress)
1255 		return rte_flow_error_set(error, EINVAL,
1256 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1257 					  NULL,
1258 					  "ingress attribute is mandatory");
1259 	return 0;
1260 }
1261 
1262 /**
1263  * Validate ICMP6 item.
1264  *
1265  * @param[in] item
1266  *   Item specification.
1267  * @param[in] item_flags
1268  *   Bit-fields that holds the items detected until now.
1269  * @param[out] error
1270  *   Pointer to error structure.
1271  *
1272  * @return
1273  *   0 on success, a negative errno value otherwise and rte_errno is set.
1274  */
1275 int
1276 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1277 			       uint64_t item_flags,
1278 			       uint8_t target_protocol,
1279 			       struct rte_flow_error *error)
1280 {
1281 	const struct rte_flow_item_icmp6 *mask = item->mask;
1282 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1283 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1284 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1285 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1286 				      MLX5_FLOW_LAYER_OUTER_L4;
1287 	int ret;
1288 
1289 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1290 		return rte_flow_error_set(error, EINVAL,
1291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 					  "protocol filtering not compatible"
1293 					  " with ICMP6 layer");
1294 	if (!(item_flags & l3m))
1295 		return rte_flow_error_set(error, EINVAL,
1296 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1297 					  "IPv6 is mandatory to filter on"
1298 					  " ICMP6");
1299 	if (item_flags & l4m)
1300 		return rte_flow_error_set(error, EINVAL,
1301 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1302 					  "multiple L4 layers not supported");
1303 	if (!mask)
1304 		mask = &rte_flow_item_icmp6_mask;
1305 	ret = mlx5_flow_item_acceptable
1306 		(item, (const uint8_t *)mask,
1307 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1308 		 sizeof(struct rte_flow_item_icmp6), error);
1309 	if (ret < 0)
1310 		return ret;
1311 	return 0;
1312 }
1313 
1314 /**
1315  * Validate ICMP item.
1316  *
1317  * @param[in] item
1318  *   Item specification.
1319  * @param[in] item_flags
1320  *   Bit-fields that holds the items detected until now.
1321  * @param[out] error
1322  *   Pointer to error structure.
1323  *
1324  * @return
1325  *   0 on success, a negative errno value otherwise and rte_errno is set.
1326  */
1327 int
1328 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1329 			     uint64_t item_flags,
1330 			     uint8_t target_protocol,
1331 			     struct rte_flow_error *error)
1332 {
1333 	const struct rte_flow_item_icmp *mask = item->mask;
1334 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1335 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1336 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1337 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1338 				      MLX5_FLOW_LAYER_OUTER_L4;
1339 	int ret;
1340 
1341 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1342 		return rte_flow_error_set(error, EINVAL,
1343 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1344 					  "protocol filtering not compatible"
1345 					  " with ICMP layer");
1346 	if (!(item_flags & l3m))
1347 		return rte_flow_error_set(error, EINVAL,
1348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1349 					  "IPv4 is mandatory to filter"
1350 					  " on ICMP");
1351 	if (item_flags & l4m)
1352 		return rte_flow_error_set(error, EINVAL,
1353 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1354 					  "multiple L4 layers not supported");
1355 	if (!mask)
1356 		mask = &rte_flow_item_icmp_mask;
1357 	ret = mlx5_flow_item_acceptable
1358 		(item, (const uint8_t *)mask,
1359 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1360 		 sizeof(struct rte_flow_item_icmp), error);
1361 	if (ret < 0)
1362 		return ret;
1363 	return 0;
1364 }
1365 
1366 /**
1367  * Validate Ethernet item.
1368  *
1369  * @param[in] item
1370  *   Item specification.
1371  * @param[in] item_flags
1372  *   Bit-fields that holds the items detected until now.
1373  * @param[out] error
1374  *   Pointer to error structure.
1375  *
1376  * @return
1377  *   0 on success, a negative errno value otherwise and rte_errno is set.
1378  */
1379 int
1380 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1381 			    uint64_t item_flags,
1382 			    struct rte_flow_error *error)
1383 {
1384 	const struct rte_flow_item_eth *mask = item->mask;
1385 	const struct rte_flow_item_eth nic_mask = {
1386 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1387 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1388 		.type = RTE_BE16(0xffff),
1389 	};
1390 	int ret;
1391 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1392 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1393 				       MLX5_FLOW_LAYER_OUTER_L2;
1394 
1395 	if (item_flags & ethm)
1396 		return rte_flow_error_set(error, ENOTSUP,
1397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 					  "multiple L2 layers not supported");
1399 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1400 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1401 		return rte_flow_error_set(error, EINVAL,
1402 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 					  "L2 layer should not follow "
1404 					  "L3 layers");
1405 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1406 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1407 		return rte_flow_error_set(error, EINVAL,
1408 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1409 					  "L2 layer should not follow VLAN");
1410 	if (!mask)
1411 		mask = &rte_flow_item_eth_mask;
1412 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1413 					(const uint8_t *)&nic_mask,
1414 					sizeof(struct rte_flow_item_eth),
1415 					error);
1416 	return ret;
1417 }
1418 
1419 /**
1420  * Validate VLAN item.
1421  *
1422  * @param[in] item
1423  *   Item specification.
1424  * @param[in] item_flags
1425  *   Bit-fields that holds the items detected until now.
1426  * @param[in] dev
1427  *   Ethernet device flow is being created on.
1428  * @param[out] error
1429  *   Pointer to error structure.
1430  *
1431  * @return
1432  *   0 on success, a negative errno value otherwise and rte_errno is set.
1433  */
1434 int
1435 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1436 			     uint64_t item_flags,
1437 			     struct rte_eth_dev *dev,
1438 			     struct rte_flow_error *error)
1439 {
1440 	const struct rte_flow_item_vlan *spec = item->spec;
1441 	const struct rte_flow_item_vlan *mask = item->mask;
1442 	const struct rte_flow_item_vlan nic_mask = {
1443 		.tci = RTE_BE16(UINT16_MAX),
1444 		.inner_type = RTE_BE16(UINT16_MAX),
1445 	};
1446 	uint16_t vlan_tag = 0;
1447 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1448 	int ret;
1449 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1450 					MLX5_FLOW_LAYER_INNER_L4) :
1451 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1452 					MLX5_FLOW_LAYER_OUTER_L4);
1453 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1454 					MLX5_FLOW_LAYER_OUTER_VLAN;
1455 
1456 	if (item_flags & vlanm)
1457 		return rte_flow_error_set(error, EINVAL,
1458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1459 					  "multiple VLAN layers not supported");
1460 	else if ((item_flags & l34m) != 0)
1461 		return rte_flow_error_set(error, EINVAL,
1462 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1463 					  "VLAN cannot follow L3/L4 layer");
1464 	if (!mask)
1465 		mask = &rte_flow_item_vlan_mask;
1466 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1467 					(const uint8_t *)&nic_mask,
1468 					sizeof(struct rte_flow_item_vlan),
1469 					error);
1470 	if (ret)
1471 		return ret;
1472 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1473 		struct mlx5_priv *priv = dev->data->dev_private;
1474 
1475 		if (priv->vmwa_context) {
1476 			/*
1477 			 * Non-NULL context means we have a virtual machine
1478 			 * and SR-IOV enabled, we have to create VLAN interface
1479 			 * to make hypervisor to setup E-Switch vport
1480 			 * context correctly. We avoid creating the multiple
1481 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1482 			 */
1483 			return rte_flow_error_set(error, EINVAL,
1484 						  RTE_FLOW_ERROR_TYPE_ITEM,
1485 						  item,
1486 						  "VLAN tag mask is not"
1487 						  " supported in virtual"
1488 						  " environment");
1489 		}
1490 	}
1491 	if (spec) {
1492 		vlan_tag = spec->tci;
1493 		vlan_tag &= mask->tci;
1494 	}
1495 	/*
1496 	 * From verbs perspective an empty VLAN is equivalent
1497 	 * to a packet without VLAN layer.
1498 	 */
1499 	if (!vlan_tag)
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1502 					  item->spec,
1503 					  "VLAN cannot be empty");
1504 	return 0;
1505 }
1506 
1507 /**
1508  * Validate IPV4 item.
1509  *
1510  * @param[in] item
1511  *   Item specification.
1512  * @param[in] item_flags
1513  *   Bit-fields that holds the items detected until now.
1514  * @param[in] acc_mask
1515  *   Acceptable mask, if NULL default internal default mask
1516  *   will be used to check whether item fields are supported.
1517  * @param[out] error
1518  *   Pointer to error structure.
1519  *
1520  * @return
1521  *   0 on success, a negative errno value otherwise and rte_errno is set.
1522  */
1523 int
1524 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1525 			     uint64_t item_flags,
1526 			     uint64_t last_item,
1527 			     uint16_t ether_type,
1528 			     const struct rte_flow_item_ipv4 *acc_mask,
1529 			     struct rte_flow_error *error)
1530 {
1531 	const struct rte_flow_item_ipv4 *mask = item->mask;
1532 	const struct rte_flow_item_ipv4 *spec = item->spec;
1533 	const struct rte_flow_item_ipv4 nic_mask = {
1534 		.hdr = {
1535 			.src_addr = RTE_BE32(0xffffffff),
1536 			.dst_addr = RTE_BE32(0xffffffff),
1537 			.type_of_service = 0xff,
1538 			.next_proto_id = 0xff,
1539 		},
1540 	};
1541 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1542 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1543 				      MLX5_FLOW_LAYER_OUTER_L3;
1544 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1545 				      MLX5_FLOW_LAYER_OUTER_L4;
1546 	int ret;
1547 	uint8_t next_proto = 0xFF;
1548 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1549 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1550 				  MLX5_FLOW_LAYER_INNER_VLAN);
1551 
1552 	if ((last_item & l2_vlan) && ether_type &&
1553 	    ether_type != RTE_ETHER_TYPE_IPV4)
1554 		return rte_flow_error_set(error, EINVAL,
1555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1556 					  "IPv4 cannot follow L2/VLAN layer "
1557 					  "which ether type is not IPv4");
1558 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1559 		if (mask && spec)
1560 			next_proto = mask->hdr.next_proto_id &
1561 				     spec->hdr.next_proto_id;
1562 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1563 			return rte_flow_error_set(error, EINVAL,
1564 						  RTE_FLOW_ERROR_TYPE_ITEM,
1565 						  item,
1566 						  "multiple tunnel "
1567 						  "not supported");
1568 	}
1569 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1570 		return rte_flow_error_set(error, EINVAL,
1571 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1572 					  "wrong tunnel type - IPv6 specified "
1573 					  "but IPv4 item provided");
1574 	if (item_flags & l3m)
1575 		return rte_flow_error_set(error, ENOTSUP,
1576 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1577 					  "multiple L3 layers not supported");
1578 	else if (item_flags & l4m)
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 					  "L3 cannot follow an L4 layer.");
1582 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1583 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1584 		return rte_flow_error_set(error, EINVAL,
1585 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1586 					  "L3 cannot follow an NVGRE layer.");
1587 	if (!mask)
1588 		mask = &rte_flow_item_ipv4_mask;
1589 	else if (mask->hdr.next_proto_id != 0 &&
1590 		 mask->hdr.next_proto_id != 0xff)
1591 		return rte_flow_error_set(error, EINVAL,
1592 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1593 					  "partial mask is not supported"
1594 					  " for protocol");
1595 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1596 					acc_mask ? (const uint8_t *)acc_mask
1597 						 : (const uint8_t *)&nic_mask,
1598 					sizeof(struct rte_flow_item_ipv4),
1599 					error);
1600 	if (ret < 0)
1601 		return ret;
1602 	return 0;
1603 }
1604 
1605 /**
1606  * Validate IPV6 item.
1607  *
1608  * @param[in] item
1609  *   Item specification.
1610  * @param[in] item_flags
1611  *   Bit-fields that holds the items detected until now.
1612  * @param[in] acc_mask
1613  *   Acceptable mask, if NULL default internal default mask
1614  *   will be used to check whether item fields are supported.
1615  * @param[out] error
1616  *   Pointer to error structure.
1617  *
1618  * @return
1619  *   0 on success, a negative errno value otherwise and rte_errno is set.
1620  */
1621 int
1622 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1623 			     uint64_t item_flags,
1624 			     uint64_t last_item,
1625 			     uint16_t ether_type,
1626 			     const struct rte_flow_item_ipv6 *acc_mask,
1627 			     struct rte_flow_error *error)
1628 {
1629 	const struct rte_flow_item_ipv6 *mask = item->mask;
1630 	const struct rte_flow_item_ipv6 *spec = item->spec;
1631 	const struct rte_flow_item_ipv6 nic_mask = {
1632 		.hdr = {
1633 			.src_addr =
1634 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1635 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1636 			.dst_addr =
1637 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1638 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1639 			.vtc_flow = RTE_BE32(0xffffffff),
1640 			.proto = 0xff,
1641 			.hop_limits = 0xff,
1642 		},
1643 	};
1644 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1645 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1646 				      MLX5_FLOW_LAYER_OUTER_L3;
1647 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1648 				      MLX5_FLOW_LAYER_OUTER_L4;
1649 	int ret;
1650 	uint8_t next_proto = 0xFF;
1651 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1652 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1653 				  MLX5_FLOW_LAYER_INNER_VLAN);
1654 
1655 	if ((last_item & l2_vlan) && ether_type &&
1656 	    ether_type != RTE_ETHER_TYPE_IPV6)
1657 		return rte_flow_error_set(error, EINVAL,
1658 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1659 					  "IPv6 cannot follow L2/VLAN layer "
1660 					  "which ether type is not IPv6");
1661 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1662 		if (mask && spec)
1663 			next_proto = mask->hdr.proto & spec->hdr.proto;
1664 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1665 			return rte_flow_error_set(error, EINVAL,
1666 						  RTE_FLOW_ERROR_TYPE_ITEM,
1667 						  item,
1668 						  "multiple tunnel "
1669 						  "not supported");
1670 	}
1671 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1672 		return rte_flow_error_set(error, EINVAL,
1673 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1674 					  "wrong tunnel type - IPv4 specified "
1675 					  "but IPv6 item provided");
1676 	if (item_flags & l3m)
1677 		return rte_flow_error_set(error, ENOTSUP,
1678 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 					  "multiple L3 layers not supported");
1680 	else if (item_flags & l4m)
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 					  "L3 cannot follow an L4 layer.");
1684 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1685 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1688 					  "L3 cannot follow an NVGRE layer.");
1689 	if (!mask)
1690 		mask = &rte_flow_item_ipv6_mask;
1691 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1692 					acc_mask ? (const uint8_t *)acc_mask
1693 						 : (const uint8_t *)&nic_mask,
1694 					sizeof(struct rte_flow_item_ipv6),
1695 					error);
1696 	if (ret < 0)
1697 		return ret;
1698 	return 0;
1699 }
1700 
1701 /**
1702  * Validate UDP item.
1703  *
1704  * @param[in] item
1705  *   Item specification.
1706  * @param[in] item_flags
1707  *   Bit-fields that holds the items detected until now.
1708  * @param[in] target_protocol
1709  *   The next protocol in the previous item.
1710  * @param[in] flow_mask
1711  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1712  * @param[out] error
1713  *   Pointer to error structure.
1714  *
1715  * @return
1716  *   0 on success, a negative errno value otherwise and rte_errno is set.
1717  */
1718 int
1719 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1720 			    uint64_t item_flags,
1721 			    uint8_t target_protocol,
1722 			    struct rte_flow_error *error)
1723 {
1724 	const struct rte_flow_item_udp *mask = item->mask;
1725 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1726 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1727 				      MLX5_FLOW_LAYER_OUTER_L3;
1728 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1729 				      MLX5_FLOW_LAYER_OUTER_L4;
1730 	int ret;
1731 
1732 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1733 		return rte_flow_error_set(error, EINVAL,
1734 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1735 					  "protocol filtering not compatible"
1736 					  " with UDP layer");
1737 	if (!(item_flags & l3m))
1738 		return rte_flow_error_set(error, EINVAL,
1739 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1740 					  "L3 is mandatory to filter on L4");
1741 	if (item_flags & l4m)
1742 		return rte_flow_error_set(error, EINVAL,
1743 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1744 					  "multiple L4 layers not supported");
1745 	if (!mask)
1746 		mask = &rte_flow_item_udp_mask;
1747 	ret = mlx5_flow_item_acceptable
1748 		(item, (const uint8_t *)mask,
1749 		 (const uint8_t *)&rte_flow_item_udp_mask,
1750 		 sizeof(struct rte_flow_item_udp), error);
1751 	if (ret < 0)
1752 		return ret;
1753 	return 0;
1754 }
1755 
1756 /**
1757  * Validate TCP item.
1758  *
1759  * @param[in] item
1760  *   Item specification.
1761  * @param[in] item_flags
1762  *   Bit-fields that holds the items detected until now.
1763  * @param[in] target_protocol
1764  *   The next protocol in the previous item.
1765  * @param[out] error
1766  *   Pointer to error structure.
1767  *
1768  * @return
1769  *   0 on success, a negative errno value otherwise and rte_errno is set.
1770  */
1771 int
1772 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1773 			    uint64_t item_flags,
1774 			    uint8_t target_protocol,
1775 			    const struct rte_flow_item_tcp *flow_mask,
1776 			    struct rte_flow_error *error)
1777 {
1778 	const struct rte_flow_item_tcp *mask = item->mask;
1779 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1780 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1781 				      MLX5_FLOW_LAYER_OUTER_L3;
1782 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1783 				      MLX5_FLOW_LAYER_OUTER_L4;
1784 	int ret;
1785 
1786 	MLX5_ASSERT(flow_mask);
1787 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1788 		return rte_flow_error_set(error, EINVAL,
1789 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1790 					  "protocol filtering not compatible"
1791 					  " with TCP layer");
1792 	if (!(item_flags & l3m))
1793 		return rte_flow_error_set(error, EINVAL,
1794 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1795 					  "L3 is mandatory to filter on L4");
1796 	if (item_flags & l4m)
1797 		return rte_flow_error_set(error, EINVAL,
1798 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1799 					  "multiple L4 layers not supported");
1800 	if (!mask)
1801 		mask = &rte_flow_item_tcp_mask;
1802 	ret = mlx5_flow_item_acceptable
1803 		(item, (const uint8_t *)mask,
1804 		 (const uint8_t *)flow_mask,
1805 		 sizeof(struct rte_flow_item_tcp), error);
1806 	if (ret < 0)
1807 		return ret;
1808 	return 0;
1809 }
1810 
1811 /**
1812  * Validate VXLAN item.
1813  *
1814  * @param[in] item
1815  *   Item specification.
1816  * @param[in] item_flags
1817  *   Bit-fields that holds the items detected until now.
1818  * @param[in] target_protocol
1819  *   The next protocol in the previous item.
1820  * @param[out] error
1821  *   Pointer to error structure.
1822  *
1823  * @return
1824  *   0 on success, a negative errno value otherwise and rte_errno is set.
1825  */
1826 int
1827 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1828 			      uint64_t item_flags,
1829 			      struct rte_flow_error *error)
1830 {
1831 	const struct rte_flow_item_vxlan *spec = item->spec;
1832 	const struct rte_flow_item_vxlan *mask = item->mask;
1833 	int ret;
1834 	union vni {
1835 		uint32_t vlan_id;
1836 		uint8_t vni[4];
1837 	} id = { .vlan_id = 0, };
1838 	uint32_t vlan_id = 0;
1839 
1840 
1841 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1842 		return rte_flow_error_set(error, ENOTSUP,
1843 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1844 					  "multiple tunnel layers not"
1845 					  " supported");
1846 	/*
1847 	 * Verify only UDPv4 is present as defined in
1848 	 * https://tools.ietf.org/html/rfc7348
1849 	 */
1850 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1851 		return rte_flow_error_set(error, EINVAL,
1852 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 					  "no outer UDP layer found");
1854 	if (!mask)
1855 		mask = &rte_flow_item_vxlan_mask;
1856 	ret = mlx5_flow_item_acceptable
1857 		(item, (const uint8_t *)mask,
1858 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1859 		 sizeof(struct rte_flow_item_vxlan),
1860 		 error);
1861 	if (ret < 0)
1862 		return ret;
1863 	if (spec) {
1864 		memcpy(&id.vni[1], spec->vni, 3);
1865 		vlan_id = id.vlan_id;
1866 		memcpy(&id.vni[1], mask->vni, 3);
1867 		vlan_id &= id.vlan_id;
1868 	}
1869 	/*
1870 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1871 	 * only this layer is defined in the Verbs specification it is
1872 	 * interpreted as wildcard and all packets will match this
1873 	 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1874 	 * udp), all packets matching the layers before will also
1875 	 * match this rule.  To avoid such situation, VNI 0 is
1876 	 * currently refused.
1877 	 */
1878 	if (!vlan_id)
1879 		return rte_flow_error_set(error, ENOTSUP,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "VXLAN vni cannot be 0");
1882 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1883 		return rte_flow_error_set(error, ENOTSUP,
1884 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1885 					  "VXLAN tunnel must be fully defined");
1886 	return 0;
1887 }
1888 
1889 /**
1890  * Validate VXLAN_GPE item.
1891  *
1892  * @param[in] item
1893  *   Item specification.
1894  * @param[in] item_flags
1895  *   Bit-fields that holds the items detected until now.
1896  * @param[in] priv
1897  *   Pointer to the private data structure.
1898  * @param[in] target_protocol
1899  *   The next protocol in the previous item.
1900  * @param[out] error
1901  *   Pointer to error structure.
1902  *
1903  * @return
1904  *   0 on success, a negative errno value otherwise and rte_errno is set.
1905  */
1906 int
1907 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1908 				  uint64_t item_flags,
1909 				  struct rte_eth_dev *dev,
1910 				  struct rte_flow_error *error)
1911 {
1912 	struct mlx5_priv *priv = dev->data->dev_private;
1913 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1914 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1915 	int ret;
1916 	union vni {
1917 		uint32_t vlan_id;
1918 		uint8_t vni[4];
1919 	} id = { .vlan_id = 0, };
1920 	uint32_t vlan_id = 0;
1921 
1922 	if (!priv->config.l3_vxlan_en)
1923 		return rte_flow_error_set(error, ENOTSUP,
1924 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1925 					  "L3 VXLAN is not enabled by device"
1926 					  " parameter and/or not configured in"
1927 					  " firmware");
1928 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1929 		return rte_flow_error_set(error, ENOTSUP,
1930 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1931 					  "multiple tunnel layers not"
1932 					  " supported");
1933 	/*
1934 	 * Verify only UDPv4 is present as defined in
1935 	 * https://tools.ietf.org/html/rfc7348
1936 	 */
1937 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1938 		return rte_flow_error_set(error, EINVAL,
1939 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1940 					  "no outer UDP layer found");
1941 	if (!mask)
1942 		mask = &rte_flow_item_vxlan_gpe_mask;
1943 	ret = mlx5_flow_item_acceptable
1944 		(item, (const uint8_t *)mask,
1945 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1946 		 sizeof(struct rte_flow_item_vxlan_gpe),
1947 		 error);
1948 	if (ret < 0)
1949 		return ret;
1950 	if (spec) {
1951 		if (spec->protocol)
1952 			return rte_flow_error_set(error, ENOTSUP,
1953 						  RTE_FLOW_ERROR_TYPE_ITEM,
1954 						  item,
1955 						  "VxLAN-GPE protocol"
1956 						  " not supported");
1957 		memcpy(&id.vni[1], spec->vni, 3);
1958 		vlan_id = id.vlan_id;
1959 		memcpy(&id.vni[1], mask->vni, 3);
1960 		vlan_id &= id.vlan_id;
1961 	}
1962 	/*
1963 	 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1964 	 * layer is defined in the Verbs specification it is interpreted as
1965 	 * wildcard and all packets will match this rule, if it follows a full
1966 	 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1967 	 * before will also match this rule.  To avoid such situation, VNI 0
1968 	 * is currently refused.
1969 	 */
1970 	if (!vlan_id)
1971 		return rte_flow_error_set(error, ENOTSUP,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "VXLAN-GPE vni cannot be 0");
1974 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1975 		return rte_flow_error_set(error, ENOTSUP,
1976 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 					  "VXLAN-GPE tunnel must be fully"
1978 					  " defined");
1979 	return 0;
1980 }
1981 /**
1982  * Validate GRE Key item.
1983  *
1984  * @param[in] item
1985  *   Item specification.
1986  * @param[in] item_flags
1987  *   Bit flags to mark detected items.
1988  * @param[in] gre_item
1989  *   Pointer to gre_item
1990  * @param[out] error
1991  *   Pointer to error structure.
1992  *
1993  * @return
1994  *   0 on success, a negative errno value otherwise and rte_errno is set.
1995  */
1996 int
1997 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1998 				uint64_t item_flags,
1999 				const struct rte_flow_item *gre_item,
2000 				struct rte_flow_error *error)
2001 {
2002 	const rte_be32_t *mask = item->mask;
2003 	int ret = 0;
2004 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2005 	const struct rte_flow_item_gre *gre_spec;
2006 	const struct rte_flow_item_gre *gre_mask;
2007 
2008 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2009 		return rte_flow_error_set(error, ENOTSUP,
2010 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2011 					  "Multiple GRE key not support");
2012 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2013 		return rte_flow_error_set(error, ENOTSUP,
2014 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2015 					  "No preceding GRE header");
2016 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2017 		return rte_flow_error_set(error, ENOTSUP,
2018 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2019 					  "GRE key following a wrong item");
2020 	gre_mask = gre_item->mask;
2021 	if (!gre_mask)
2022 		gre_mask = &rte_flow_item_gre_mask;
2023 	gre_spec = gre_item->spec;
2024 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2025 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2026 		return rte_flow_error_set(error, EINVAL,
2027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 					  "Key bit must be on");
2029 
2030 	if (!mask)
2031 		mask = &gre_key_default_mask;
2032 	ret = mlx5_flow_item_acceptable
2033 		(item, (const uint8_t *)mask,
2034 		 (const uint8_t *)&gre_key_default_mask,
2035 		 sizeof(rte_be32_t), error);
2036 	return ret;
2037 }
2038 
2039 /**
2040  * Validate GRE item.
2041  *
2042  * @param[in] item
2043  *   Item specification.
2044  * @param[in] item_flags
2045  *   Bit flags to mark detected items.
2046  * @param[in] target_protocol
2047  *   The next protocol in the previous item.
2048  * @param[out] error
2049  *   Pointer to error structure.
2050  *
2051  * @return
2052  *   0 on success, a negative errno value otherwise and rte_errno is set.
2053  */
2054 int
2055 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2056 			    uint64_t item_flags,
2057 			    uint8_t target_protocol,
2058 			    struct rte_flow_error *error)
2059 {
2060 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2061 	const struct rte_flow_item_gre *mask = item->mask;
2062 	int ret;
2063 	const struct rte_flow_item_gre nic_mask = {
2064 		.c_rsvd0_ver = RTE_BE16(0xB000),
2065 		.protocol = RTE_BE16(UINT16_MAX),
2066 	};
2067 
2068 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2069 		return rte_flow_error_set(error, EINVAL,
2070 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2071 					  "protocol filtering not compatible"
2072 					  " with this GRE layer");
2073 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2074 		return rte_flow_error_set(error, ENOTSUP,
2075 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 					  "multiple tunnel layers not"
2077 					  " supported");
2078 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2079 		return rte_flow_error_set(error, ENOTSUP,
2080 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2081 					  "L3 Layer is missing");
2082 	if (!mask)
2083 		mask = &rte_flow_item_gre_mask;
2084 	ret = mlx5_flow_item_acceptable
2085 		(item, (const uint8_t *)mask,
2086 		 (const uint8_t *)&nic_mask,
2087 		 sizeof(struct rte_flow_item_gre), error);
2088 	if (ret < 0)
2089 		return ret;
2090 #ifndef HAVE_MLX5DV_DR
2091 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2092 	if (spec && (spec->protocol & mask->protocol))
2093 		return rte_flow_error_set(error, ENOTSUP,
2094 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 					  "without MPLS support the"
2096 					  " specification cannot be used for"
2097 					  " filtering");
2098 #endif
2099 #endif
2100 	return 0;
2101 }
2102 
2103 /**
2104  * Validate Geneve item.
2105  *
2106  * @param[in] item
2107  *   Item specification.
2108  * @param[in] itemFlags
2109  *   Bit-fields that holds the items detected until now.
2110  * @param[in] enPriv
2111  *   Pointer to the private data structure.
2112  * @param[out] error
2113  *   Pointer to error structure.
2114  *
2115  * @return
2116  *   0 on success, a negative errno value otherwise and rte_errno is set.
2117  */
2118 
2119 int
2120 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2121 			       uint64_t item_flags,
2122 			       struct rte_eth_dev *dev,
2123 			       struct rte_flow_error *error)
2124 {
2125 	struct mlx5_priv *priv = dev->data->dev_private;
2126 	const struct rte_flow_item_geneve *spec = item->spec;
2127 	const struct rte_flow_item_geneve *mask = item->mask;
2128 	int ret;
2129 	uint16_t gbhdr;
2130 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2131 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2132 	const struct rte_flow_item_geneve nic_mask = {
2133 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2134 		.vni = "\xff\xff\xff",
2135 		.protocol = RTE_BE16(UINT16_MAX),
2136 	};
2137 
2138 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2139 		return rte_flow_error_set(error, ENOTSUP,
2140 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2141 					  "L3 Geneve is not enabled by device"
2142 					  " parameter and/or not configured in"
2143 					  " firmware");
2144 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2145 		return rte_flow_error_set(error, ENOTSUP,
2146 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2147 					  "multiple tunnel layers not"
2148 					  " supported");
2149 	/*
2150 	 * Verify only UDPv4 is present as defined in
2151 	 * https://tools.ietf.org/html/rfc7348
2152 	 */
2153 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2154 		return rte_flow_error_set(error, EINVAL,
2155 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2156 					  "no outer UDP layer found");
2157 	if (!mask)
2158 		mask = &rte_flow_item_geneve_mask;
2159 	ret = mlx5_flow_item_acceptable
2160 				  (item, (const uint8_t *)mask,
2161 				   (const uint8_t *)&nic_mask,
2162 				   sizeof(struct rte_flow_item_geneve), error);
2163 	if (ret)
2164 		return ret;
2165 	if (spec) {
2166 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2167 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2168 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2169 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2170 			return rte_flow_error_set(error, ENOTSUP,
2171 						  RTE_FLOW_ERROR_TYPE_ITEM,
2172 						  item,
2173 						  "Geneve protocol unsupported"
2174 						  " fields are being used");
2175 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2176 			return rte_flow_error_set
2177 					(error, ENOTSUP,
2178 					 RTE_FLOW_ERROR_TYPE_ITEM,
2179 					 item,
2180 					 "Unsupported Geneve options length");
2181 	}
2182 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2183 		return rte_flow_error_set
2184 				    (error, ENOTSUP,
2185 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2186 				     "Geneve tunnel must be fully defined");
2187 	return 0;
2188 }
2189 
2190 /**
2191  * Validate MPLS item.
2192  *
2193  * @param[in] dev
2194  *   Pointer to the rte_eth_dev structure.
2195  * @param[in] item
2196  *   Item specification.
2197  * @param[in] item_flags
2198  *   Bit-fields that holds the items detected until now.
2199  * @param[in] prev_layer
2200  *   The protocol layer indicated in previous item.
2201  * @param[out] error
2202  *   Pointer to error structure.
2203  *
2204  * @return
2205  *   0 on success, a negative errno value otherwise and rte_errno is set.
2206  */
2207 int
2208 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2209 			     const struct rte_flow_item *item __rte_unused,
2210 			     uint64_t item_flags __rte_unused,
2211 			     uint64_t prev_layer __rte_unused,
2212 			     struct rte_flow_error *error)
2213 {
2214 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2215 	const struct rte_flow_item_mpls *mask = item->mask;
2216 	struct mlx5_priv *priv = dev->data->dev_private;
2217 	int ret;
2218 
2219 	if (!priv->config.mpls_en)
2220 		return rte_flow_error_set(error, ENOTSUP,
2221 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 					  "MPLS not supported or"
2223 					  " disabled in firmware"
2224 					  " configuration.");
2225 	/* MPLS over IP, UDP, GRE is allowed */
2226 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2227 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2228 			    MLX5_FLOW_LAYER_GRE)))
2229 		return rte_flow_error_set(error, EINVAL,
2230 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2231 					  "protocol filtering not compatible"
2232 					  " with MPLS layer");
2233 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2234 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2235 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2236 		return rte_flow_error_set(error, ENOTSUP,
2237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 					  "multiple tunnel layers not"
2239 					  " supported");
2240 	if (!mask)
2241 		mask = &rte_flow_item_mpls_mask;
2242 	ret = mlx5_flow_item_acceptable
2243 		(item, (const uint8_t *)mask,
2244 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2245 		 sizeof(struct rte_flow_item_mpls), error);
2246 	if (ret < 0)
2247 		return ret;
2248 	return 0;
2249 #endif
2250 	return rte_flow_error_set(error, ENOTSUP,
2251 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2252 				  "MPLS is not supported by Verbs, please"
2253 				  " update.");
2254 }
2255 
2256 /**
2257  * Validate NVGRE item.
2258  *
2259  * @param[in] item
2260  *   Item specification.
2261  * @param[in] item_flags
2262  *   Bit flags to mark detected items.
2263  * @param[in] target_protocol
2264  *   The next protocol in the previous item.
2265  * @param[out] error
2266  *   Pointer to error structure.
2267  *
2268  * @return
2269  *   0 on success, a negative errno value otherwise and rte_errno is set.
2270  */
2271 int
2272 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2273 			      uint64_t item_flags,
2274 			      uint8_t target_protocol,
2275 			      struct rte_flow_error *error)
2276 {
2277 	const struct rte_flow_item_nvgre *mask = item->mask;
2278 	int ret;
2279 
2280 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2281 		return rte_flow_error_set(error, EINVAL,
2282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2283 					  "protocol filtering not compatible"
2284 					  " with this GRE layer");
2285 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2286 		return rte_flow_error_set(error, ENOTSUP,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "multiple tunnel layers not"
2289 					  " supported");
2290 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2291 		return rte_flow_error_set(error, ENOTSUP,
2292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 					  "L3 Layer is missing");
2294 	if (!mask)
2295 		mask = &rte_flow_item_nvgre_mask;
2296 	ret = mlx5_flow_item_acceptable
2297 		(item, (const uint8_t *)mask,
2298 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2299 		 sizeof(struct rte_flow_item_nvgre), error);
2300 	if (ret < 0)
2301 		return ret;
2302 	return 0;
2303 }
2304 
2305 /* Allocate unique ID for the split Q/RSS subflows. */
2306 static uint32_t
2307 flow_qrss_get_id(struct rte_eth_dev *dev)
2308 {
2309 	struct mlx5_priv *priv = dev->data->dev_private;
2310 	uint32_t qrss_id, ret;
2311 
2312 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2313 	if (ret)
2314 		return 0;
2315 	MLX5_ASSERT(qrss_id);
2316 	return qrss_id;
2317 }
2318 
2319 /* Free unique ID for the split Q/RSS subflows. */
2320 static void
2321 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2322 {
2323 	struct mlx5_priv *priv = dev->data->dev_private;
2324 
2325 	if (qrss_id)
2326 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2327 }
2328 
2329 /**
2330  * Release resource related QUEUE/RSS action split.
2331  *
2332  * @param dev
2333  *   Pointer to Ethernet device.
2334  * @param flow
2335  *   Flow to release id's from.
2336  */
2337 static void
2338 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2339 			     struct rte_flow *flow)
2340 {
2341 	struct mlx5_flow *dev_flow;
2342 
2343 	LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2344 		if (dev_flow->qrss_id)
2345 			flow_qrss_free_id(dev, dev_flow->qrss_id);
2346 }
2347 
2348 static int
2349 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2350 		   const struct rte_flow_attr *attr __rte_unused,
2351 		   const struct rte_flow_item items[] __rte_unused,
2352 		   const struct rte_flow_action actions[] __rte_unused,
2353 		   bool external __rte_unused,
2354 		   struct rte_flow_error *error)
2355 {
2356 	return rte_flow_error_set(error, ENOTSUP,
2357 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2358 }
2359 
2360 static struct mlx5_flow *
2361 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2362 		  const struct rte_flow_item items[] __rte_unused,
2363 		  const struct rte_flow_action actions[] __rte_unused,
2364 		  struct rte_flow_error *error)
2365 {
2366 	rte_flow_error_set(error, ENOTSUP,
2367 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2368 	return NULL;
2369 }
2370 
2371 static int
2372 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2373 		    struct mlx5_flow *dev_flow __rte_unused,
2374 		    const struct rte_flow_attr *attr __rte_unused,
2375 		    const struct rte_flow_item items[] __rte_unused,
2376 		    const struct rte_flow_action actions[] __rte_unused,
2377 		    struct rte_flow_error *error)
2378 {
2379 	return rte_flow_error_set(error, ENOTSUP,
2380 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2381 }
2382 
2383 static int
2384 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2385 		struct rte_flow *flow __rte_unused,
2386 		struct rte_flow_error *error)
2387 {
2388 	return rte_flow_error_set(error, ENOTSUP,
2389 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2390 }
2391 
2392 static void
2393 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2394 		 struct rte_flow *flow __rte_unused)
2395 {
2396 }
2397 
2398 static void
2399 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2400 		  struct rte_flow *flow __rte_unused)
2401 {
2402 }
2403 
2404 static int
2405 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2406 		struct rte_flow *flow __rte_unused,
2407 		const struct rte_flow_action *actions __rte_unused,
2408 		void *data __rte_unused,
2409 		struct rte_flow_error *error)
2410 {
2411 	return rte_flow_error_set(error, ENOTSUP,
2412 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2413 }
2414 
2415 /* Void driver to protect from null pointer reference. */
2416 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2417 	.validate = flow_null_validate,
2418 	.prepare = flow_null_prepare,
2419 	.translate = flow_null_translate,
2420 	.apply = flow_null_apply,
2421 	.remove = flow_null_remove,
2422 	.destroy = flow_null_destroy,
2423 	.query = flow_null_query,
2424 };
2425 
2426 /**
2427  * Select flow driver type according to flow attributes and device
2428  * configuration.
2429  *
2430  * @param[in] dev
2431  *   Pointer to the dev structure.
2432  * @param[in] attr
2433  *   Pointer to the flow attributes.
2434  *
2435  * @return
2436  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2437  */
2438 static enum mlx5_flow_drv_type
2439 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2440 {
2441 	struct mlx5_priv *priv = dev->data->dev_private;
2442 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2443 
2444 	if (attr->transfer && priv->config.dv_esw_en)
2445 		type = MLX5_FLOW_TYPE_DV;
2446 	if (!attr->transfer)
2447 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2448 						 MLX5_FLOW_TYPE_VERBS;
2449 	return type;
2450 }
2451 
2452 #define flow_get_drv_ops(type) flow_drv_ops[type]
2453 
2454 /**
2455  * Flow driver validation API. This abstracts calling driver specific functions.
2456  * The type of flow driver is determined according to flow attributes.
2457  *
2458  * @param[in] dev
2459  *   Pointer to the dev structure.
2460  * @param[in] attr
2461  *   Pointer to the flow attributes.
2462  * @param[in] items
2463  *   Pointer to the list of items.
2464  * @param[in] actions
2465  *   Pointer to the list of actions.
2466  * @param[in] external
2467  *   This flow rule is created by request external to PMD.
2468  * @param[out] error
2469  *   Pointer to the error structure.
2470  *
2471  * @return
2472  *   0 on success, a negative errno value otherwise and rte_errno is set.
2473  */
2474 static inline int
2475 flow_drv_validate(struct rte_eth_dev *dev,
2476 		  const struct rte_flow_attr *attr,
2477 		  const struct rte_flow_item items[],
2478 		  const struct rte_flow_action actions[],
2479 		  bool external, struct rte_flow_error *error)
2480 {
2481 	const struct mlx5_flow_driver_ops *fops;
2482 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2483 
2484 	fops = flow_get_drv_ops(type);
2485 	return fops->validate(dev, attr, items, actions, external, error);
2486 }
2487 
2488 /**
2489  * Flow driver preparation API. This abstracts calling driver specific
2490  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2491  * calculates the size of memory required for device flow, allocates the memory,
2492  * initializes the device flow and returns the pointer.
2493  *
2494  * @note
2495  *   This function initializes device flow structure such as dv or verbs in
2496  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2497  *   rest. For example, adding returning device flow to flow->dev_flow list and
2498  *   setting backward reference to the flow should be done out of this function.
2499  *   layers field is not filled either.
2500  *
2501  * @param[in] attr
2502  *   Pointer to the flow attributes.
2503  * @param[in] items
2504  *   Pointer to the list of items.
2505  * @param[in] actions
2506  *   Pointer to the list of actions.
2507  * @param[out] error
2508  *   Pointer to the error structure.
2509  *
2510  * @return
2511  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2512  */
2513 static inline struct mlx5_flow *
2514 flow_drv_prepare(const struct rte_flow *flow,
2515 		 const struct rte_flow_attr *attr,
2516 		 const struct rte_flow_item items[],
2517 		 const struct rte_flow_action actions[],
2518 		 struct rte_flow_error *error)
2519 {
2520 	const struct mlx5_flow_driver_ops *fops;
2521 	enum mlx5_flow_drv_type type = flow->drv_type;
2522 
2523 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2524 	fops = flow_get_drv_ops(type);
2525 	return fops->prepare(attr, items, actions, error);
2526 }
2527 
2528 /**
2529  * Flow driver translation API. This abstracts calling driver specific
2530  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2531  * translates a generic flow into a driver flow. flow_drv_prepare() must
2532  * precede.
2533  *
2534  * @note
2535  *   dev_flow->layers could be filled as a result of parsing during translation
2536  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2537  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2538  *   flow->actions could be overwritten even though all the expanded dev_flows
2539  *   have the same actions.
2540  *
2541  * @param[in] dev
2542  *   Pointer to the rte dev structure.
2543  * @param[in, out] dev_flow
2544  *   Pointer to the mlx5 flow.
2545  * @param[in] attr
2546  *   Pointer to the flow attributes.
2547  * @param[in] items
2548  *   Pointer to the list of items.
2549  * @param[in] actions
2550  *   Pointer to the list of actions.
2551  * @param[out] error
2552  *   Pointer to the error structure.
2553  *
2554  * @return
2555  *   0 on success, a negative errno value otherwise and rte_errno is set.
2556  */
2557 static inline int
2558 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2559 		   const struct rte_flow_attr *attr,
2560 		   const struct rte_flow_item items[],
2561 		   const struct rte_flow_action actions[],
2562 		   struct rte_flow_error *error)
2563 {
2564 	const struct mlx5_flow_driver_ops *fops;
2565 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2566 
2567 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2568 	fops = flow_get_drv_ops(type);
2569 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2570 }
2571 
2572 /**
2573  * Flow driver apply API. This abstracts calling driver specific functions.
2574  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2575  * translated driver flows on to device. flow_drv_translate() must precede.
2576  *
2577  * @param[in] dev
2578  *   Pointer to Ethernet device structure.
2579  * @param[in, out] flow
2580  *   Pointer to flow structure.
2581  * @param[out] error
2582  *   Pointer to error structure.
2583  *
2584  * @return
2585  *   0 on success, a negative errno value otherwise and rte_errno is set.
2586  */
2587 static inline int
2588 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2589 	       struct rte_flow_error *error)
2590 {
2591 	const struct mlx5_flow_driver_ops *fops;
2592 	enum mlx5_flow_drv_type type = flow->drv_type;
2593 
2594 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2595 	fops = flow_get_drv_ops(type);
2596 	return fops->apply(dev, flow, error);
2597 }
2598 
2599 /**
2600  * Flow driver remove API. This abstracts calling driver specific functions.
2601  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2602  * on device. All the resources of the flow should be freed by calling
2603  * flow_drv_destroy().
2604  *
2605  * @param[in] dev
2606  *   Pointer to Ethernet device.
2607  * @param[in, out] flow
2608  *   Pointer to flow structure.
2609  */
2610 static inline void
2611 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2612 {
2613 	const struct mlx5_flow_driver_ops *fops;
2614 	enum mlx5_flow_drv_type type = flow->drv_type;
2615 
2616 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2617 	fops = flow_get_drv_ops(type);
2618 	fops->remove(dev, flow);
2619 }
2620 
2621 /**
2622  * Flow driver destroy API. This abstracts calling driver specific functions.
2623  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2624  * on device and releases resources of the flow.
2625  *
2626  * @param[in] dev
2627  *   Pointer to Ethernet device.
2628  * @param[in, out] flow
2629  *   Pointer to flow structure.
2630  */
2631 static inline void
2632 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2633 {
2634 	const struct mlx5_flow_driver_ops *fops;
2635 	enum mlx5_flow_drv_type type = flow->drv_type;
2636 
2637 	flow_mreg_split_qrss_release(dev, flow);
2638 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2639 	fops = flow_get_drv_ops(type);
2640 	fops->destroy(dev, flow);
2641 }
2642 
2643 /**
2644  * Validate a flow supported by the NIC.
2645  *
2646  * @see rte_flow_validate()
2647  * @see rte_flow_ops
2648  */
2649 int
2650 mlx5_flow_validate(struct rte_eth_dev *dev,
2651 		   const struct rte_flow_attr *attr,
2652 		   const struct rte_flow_item items[],
2653 		   const struct rte_flow_action actions[],
2654 		   struct rte_flow_error *error)
2655 {
2656 	int ret;
2657 
2658 	ret = flow_drv_validate(dev, attr, items, actions, true, error);
2659 	if (ret < 0)
2660 		return ret;
2661 	return 0;
2662 }
2663 
2664 /**
2665  * Get port id item from the item list.
2666  *
2667  * @param[in] item
2668  *   Pointer to the list of items.
2669  *
2670  * @return
2671  *   Pointer to the port id item if exist, else return NULL.
2672  */
2673 static const struct rte_flow_item *
2674 find_port_id_item(const struct rte_flow_item *item)
2675 {
2676 	MLX5_ASSERT(item);
2677 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2678 		if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2679 			return item;
2680 	}
2681 	return NULL;
2682 }
2683 
2684 /**
2685  * Get RSS action from the action list.
2686  *
2687  * @param[in] actions
2688  *   Pointer to the list of actions.
2689  *
2690  * @return
2691  *   Pointer to the RSS action if exist, else return NULL.
2692  */
2693 static const struct rte_flow_action_rss*
2694 flow_get_rss_action(const struct rte_flow_action actions[])
2695 {
2696 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2697 		switch (actions->type) {
2698 		case RTE_FLOW_ACTION_TYPE_RSS:
2699 			return (const struct rte_flow_action_rss *)
2700 			       actions->conf;
2701 		default:
2702 			break;
2703 		}
2704 	}
2705 	return NULL;
2706 }
2707 
2708 static unsigned int
2709 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2710 {
2711 	const struct rte_flow_item *item;
2712 	unsigned int has_vlan = 0;
2713 
2714 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2715 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2716 			has_vlan = 1;
2717 			break;
2718 		}
2719 	}
2720 	if (has_vlan)
2721 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2722 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2723 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2724 			       MLX5_EXPANSION_ROOT_OUTER;
2725 }
2726 
2727 /**
2728  * Get QUEUE/RSS action from the action list.
2729  *
2730  * @param[in] actions
2731  *   Pointer to the list of actions.
2732  * @param[out] qrss
2733  *   Pointer to the return pointer.
2734  * @param[out] qrss_type
2735  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2736  *   if no QUEUE/RSS is found.
2737  *
2738  * @return
2739  *   Total number of actions.
2740  */
2741 static int
2742 flow_parse_qrss_action(const struct rte_flow_action actions[],
2743 		       const struct rte_flow_action **qrss)
2744 {
2745 	int actions_n = 0;
2746 
2747 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2748 		switch (actions->type) {
2749 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2750 		case RTE_FLOW_ACTION_TYPE_RSS:
2751 			*qrss = actions;
2752 			break;
2753 		default:
2754 			break;
2755 		}
2756 		actions_n++;
2757 	}
2758 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2759 	return actions_n + 1;
2760 }
2761 
2762 /**
2763  * Check meter action from the action list.
2764  *
2765  * @param[in] actions
2766  *   Pointer to the list of actions.
2767  * @param[out] mtr
2768  *   Pointer to the meter exist flag.
2769  *
2770  * @return
2771  *   Total number of actions.
2772  */
2773 static int
2774 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2775 {
2776 	int actions_n = 0;
2777 
2778 	MLX5_ASSERT(mtr);
2779 	*mtr = 0;
2780 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2781 		switch (actions->type) {
2782 		case RTE_FLOW_ACTION_TYPE_METER:
2783 			*mtr = 1;
2784 			break;
2785 		default:
2786 			break;
2787 		}
2788 		actions_n++;
2789 	}
2790 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2791 	return actions_n + 1;
2792 }
2793 
2794 /**
2795  * Check if the flow should be splited due to hairpin.
2796  * The reason for the split is that in current HW we can't
2797  * support encap on Rx, so if a flow have encap we move it
2798  * to Tx.
2799  *
2800  * @param dev
2801  *   Pointer to Ethernet device.
2802  * @param[in] attr
2803  *   Flow rule attributes.
2804  * @param[in] actions
2805  *   Associated actions (list terminated by the END action).
2806  *
2807  * @return
2808  *   > 0 the number of actions and the flow should be split,
2809  *   0 when no split required.
2810  */
2811 static int
2812 flow_check_hairpin_split(struct rte_eth_dev *dev,
2813 			 const struct rte_flow_attr *attr,
2814 			 const struct rte_flow_action actions[])
2815 {
2816 	int queue_action = 0;
2817 	int action_n = 0;
2818 	int encap = 0;
2819 	const struct rte_flow_action_queue *queue;
2820 	const struct rte_flow_action_rss *rss;
2821 	const struct rte_flow_action_raw_encap *raw_encap;
2822 
2823 	if (!attr->ingress)
2824 		return 0;
2825 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2826 		switch (actions->type) {
2827 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2828 			queue = actions->conf;
2829 			if (queue == NULL)
2830 				return 0;
2831 			if (mlx5_rxq_get_type(dev, queue->index) !=
2832 			    MLX5_RXQ_TYPE_HAIRPIN)
2833 				return 0;
2834 			queue_action = 1;
2835 			action_n++;
2836 			break;
2837 		case RTE_FLOW_ACTION_TYPE_RSS:
2838 			rss = actions->conf;
2839 			if (rss == NULL || rss->queue_num == 0)
2840 				return 0;
2841 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2842 			    MLX5_RXQ_TYPE_HAIRPIN)
2843 				return 0;
2844 			queue_action = 1;
2845 			action_n++;
2846 			break;
2847 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2848 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2849 			encap = 1;
2850 			action_n++;
2851 			break;
2852 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2853 			raw_encap = actions->conf;
2854 			if (raw_encap->size >
2855 			    (sizeof(struct rte_flow_item_eth) +
2856 			     sizeof(struct rte_flow_item_ipv4)))
2857 				encap = 1;
2858 			action_n++;
2859 			break;
2860 		default:
2861 			action_n++;
2862 			break;
2863 		}
2864 	}
2865 	if (encap == 1 && queue_action)
2866 		return action_n;
2867 	return 0;
2868 }
2869 
2870 /* Declare flow create/destroy prototype in advance. */
2871 static struct rte_flow *
2872 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2873 		 const struct rte_flow_attr *attr,
2874 		 const struct rte_flow_item items[],
2875 		 const struct rte_flow_action actions[],
2876 		 bool external, struct rte_flow_error *error);
2877 
2878 static void
2879 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2880 		  struct rte_flow *flow);
2881 
2882 /**
2883  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2884  *
2885  * As mark_id is unique, if there's already a registered flow for the mark_id,
2886  * return by increasing the reference counter of the resource. Otherwise, create
2887  * the resource (mcp_res) and flow.
2888  *
2889  * Flow looks like,
2890  *   - If ingress port is ANY and reg_c[1] is mark_id,
2891  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2892  *
2893  * For default flow (zero mark_id), flow is like,
2894  *   - If ingress port is ANY,
2895  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2896  *
2897  * @param dev
2898  *   Pointer to Ethernet device.
2899  * @param mark_id
2900  *   ID of MARK action, zero means default flow for META.
2901  * @param[out] error
2902  *   Perform verbose error reporting if not NULL.
2903  *
2904  * @return
2905  *   Associated resource on success, NULL otherwise and rte_errno is set.
2906  */
2907 static struct mlx5_flow_mreg_copy_resource *
2908 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2909 			  struct rte_flow_error *error)
2910 {
2911 	struct mlx5_priv *priv = dev->data->dev_private;
2912 	struct rte_flow_attr attr = {
2913 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2914 		.ingress = 1,
2915 	};
2916 	struct mlx5_rte_flow_item_tag tag_spec = {
2917 		.data = mark_id,
2918 	};
2919 	struct rte_flow_item items[] = {
2920 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2921 	};
2922 	struct rte_flow_action_mark ftag = {
2923 		.id = mark_id,
2924 	};
2925 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2926 		.dst = REG_B,
2927 		.src = 0,
2928 	};
2929 	struct rte_flow_action_jump jump = {
2930 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2931 	};
2932 	struct rte_flow_action actions[] = {
2933 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2934 	};
2935 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2936 	int ret;
2937 
2938 	/* Fill the register fileds in the flow. */
2939 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2940 	if (ret < 0)
2941 		return NULL;
2942 	tag_spec.id = ret;
2943 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2944 	if (ret < 0)
2945 		return NULL;
2946 	cp_mreg.src = ret;
2947 	/* Check if already registered. */
2948 	MLX5_ASSERT(priv->mreg_cp_tbl);
2949 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2950 	if (mcp_res) {
2951 		/* For non-default rule. */
2952 		if (mark_id != MLX5_DEFAULT_COPY_ID)
2953 			mcp_res->refcnt++;
2954 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2955 			    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 	MLX5_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 		MLX5_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 	MLX5_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 	MLX5_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 	MLX5_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 	MLX5_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 	MLX5_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 = NULL;
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 	/* Prepare the actions for prefix and suffix flow. */
3485 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3486 		switch (actions->type) {
3487 		case RTE_FLOW_ACTION_TYPE_METER:
3488 			/* Add the extra tag action first. */
3489 			tag_action = actions_pre;
3490 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3491 			actions_pre++;
3492 			memcpy(actions_pre, actions,
3493 			       sizeof(struct rte_flow_action));
3494 			actions_pre++;
3495 			break;
3496 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3497 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3498 			memcpy(actions_pre, actions,
3499 			       sizeof(struct rte_flow_action));
3500 			actions_pre++;
3501 			break;
3502 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3503 			raw_encap = actions->conf;
3504 			if (raw_encap->size >
3505 			    (sizeof(struct rte_flow_item_eth) +
3506 			     sizeof(struct rte_flow_item_ipv4))) {
3507 				memcpy(actions_sfx, actions,
3508 				       sizeof(struct rte_flow_action));
3509 				actions_sfx++;
3510 			} else {
3511 				rte_memcpy(actions_pre, actions,
3512 					   sizeof(struct rte_flow_action));
3513 				actions_pre++;
3514 			}
3515 			break;
3516 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3517 			raw_decap = actions->conf;
3518 			/* Size 0 decap means 50 bytes as vxlan decap. */
3519 			if (raw_decap->size && (raw_decap->size <
3520 			    (sizeof(struct rte_flow_item_eth) +
3521 			     sizeof(struct rte_flow_item_ipv4)))) {
3522 				memcpy(actions_sfx, actions,
3523 				       sizeof(struct rte_flow_action));
3524 				actions_sfx++;
3525 			} else {
3526 				rte_memcpy(actions_pre, actions,
3527 					   sizeof(struct rte_flow_action));
3528 				actions_pre++;
3529 			}
3530 			break;
3531 		default:
3532 			memcpy(actions_sfx, actions,
3533 				sizeof(struct rte_flow_action));
3534 			actions_sfx++;
3535 			break;
3536 		}
3537 	}
3538 	/* Add end action to the actions. */
3539 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3540 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3541 	actions_pre++;
3542 	/* Set the tag. */
3543 	set_tag = (void *)actions_pre;
3544 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3545 	/*
3546 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3547 	 */
3548 	tag_id = flow_qrss_get_id(dev);
3549 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3550 	assert(tag_action);
3551 	tag_action->conf = set_tag;
3552 	return tag_id;
3553 }
3554 
3555 /**
3556  * Split action list having QUEUE/RSS for metadata register copy.
3557  *
3558  * Once Q/RSS action is detected in user's action list, the flow action
3559  * should be split in order to copy metadata registers, which will happen in
3560  * RX_CP_TBL like,
3561  *   - CQE->flow_tag := reg_c[1] (MARK)
3562  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3563  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3564  * This is because the last action of each flow must be a terminal action
3565  * (QUEUE, RSS or DROP).
3566  *
3567  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3568  * stored and kept in the mlx5_flow structure per each sub_flow.
3569  *
3570  * The Q/RSS action is replaced with,
3571  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3572  * And the following JUMP action is added at the end,
3573  *   - JUMP, to RX_CP_TBL.
3574  *
3575  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3576  * flow_create_split_metadata() routine. The flow will look like,
3577  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3578  *
3579  * @param dev
3580  *   Pointer to Ethernet device.
3581  * @param[out] split_actions
3582  *   Pointer to store split actions to jump to CP_TBL.
3583  * @param[in] actions
3584  *   Pointer to the list of original flow actions.
3585  * @param[in] qrss
3586  *   Pointer to the Q/RSS action.
3587  * @param[in] actions_n
3588  *   Number of original actions.
3589  * @param[out] error
3590  *   Perform verbose error reporting if not NULL.
3591  *
3592  * @return
3593  *   non-zero unique flow_id on success, otherwise 0 and
3594  *   error/rte_error are set.
3595  */
3596 static uint32_t
3597 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3598 			  struct rte_flow_action *split_actions,
3599 			  const struct rte_flow_action *actions,
3600 			  const struct rte_flow_action *qrss,
3601 			  int actions_n, struct rte_flow_error *error)
3602 {
3603 	struct mlx5_rte_flow_action_set_tag *set_tag;
3604 	struct rte_flow_action_jump *jump;
3605 	const int qrss_idx = qrss - actions;
3606 	uint32_t flow_id = 0;
3607 	int ret = 0;
3608 
3609 	/*
3610 	 * Given actions will be split
3611 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3612 	 * - Add jump to mreg CP_TBL.
3613 	 * As a result, there will be one more action.
3614 	 */
3615 	++actions_n;
3616 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3617 	set_tag = (void *)(split_actions + actions_n);
3618 	/*
3619 	 * If tag action is not set to void(it means we are not the meter
3620 	 * suffix flow), add the tag action. Since meter suffix flow already
3621 	 * has the tag added.
3622 	 */
3623 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3624 		/*
3625 		 * Allocate the new subflow ID. This one is unique within
3626 		 * device and not shared with representors. Otherwise,
3627 		 * we would have to resolve multi-thread access synch
3628 		 * issue. Each flow on the shared device is appended
3629 		 * with source vport identifier, so the resulting
3630 		 * flows will be unique in the shared (by master and
3631 		 * representors) domain even if they have coinciding
3632 		 * IDs.
3633 		 */
3634 		flow_id = flow_qrss_get_id(dev);
3635 		if (!flow_id)
3636 			return rte_flow_error_set(error, ENOMEM,
3637 						  RTE_FLOW_ERROR_TYPE_ACTION,
3638 						  NULL, "can't allocate id "
3639 						  "for split Q/RSS subflow");
3640 		/* Internal SET_TAG action to set flow ID. */
3641 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3642 			.data = flow_id,
3643 		};
3644 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3645 		if (ret < 0)
3646 			return ret;
3647 		set_tag->id = ret;
3648 		/* Construct new actions array. */
3649 		/* Replace QUEUE/RSS action. */
3650 		split_actions[qrss_idx] = (struct rte_flow_action){
3651 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3652 			.conf = set_tag,
3653 		};
3654 	}
3655 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3656 	jump = (void *)(set_tag + 1);
3657 	*jump = (struct rte_flow_action_jump){
3658 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3659 	};
3660 	split_actions[actions_n - 2] = (struct rte_flow_action){
3661 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3662 		.conf = jump,
3663 	};
3664 	split_actions[actions_n - 1] = (struct rte_flow_action){
3665 		.type = RTE_FLOW_ACTION_TYPE_END,
3666 	};
3667 	return flow_id;
3668 }
3669 
3670 /**
3671  * Extend the given action list for Tx metadata copy.
3672  *
3673  * Copy the given action list to the ext_actions and add flow metadata register
3674  * copy action in order to copy reg_a set by WQE to reg_c[0].
3675  *
3676  * @param[out] ext_actions
3677  *   Pointer to the extended action list.
3678  * @param[in] actions
3679  *   Pointer to the list of actions.
3680  * @param[in] actions_n
3681  *   Number of actions in the list.
3682  * @param[out] error
3683  *   Perform verbose error reporting if not NULL.
3684  *
3685  * @return
3686  *   0 on success, negative value otherwise
3687  */
3688 static int
3689 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3690 		       struct rte_flow_action *ext_actions,
3691 		       const struct rte_flow_action *actions,
3692 		       int actions_n, struct rte_flow_error *error)
3693 {
3694 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3695 		(struct mlx5_flow_action_copy_mreg *)
3696 			(ext_actions + actions_n + 1);
3697 	int ret;
3698 
3699 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3700 	if (ret < 0)
3701 		return ret;
3702 	cp_mreg->dst = ret;
3703 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3704 	if (ret < 0)
3705 		return ret;
3706 	cp_mreg->src = ret;
3707 	memcpy(ext_actions, actions,
3708 			sizeof(*ext_actions) * actions_n);
3709 	ext_actions[actions_n - 1] = (struct rte_flow_action){
3710 		.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3711 		.conf = cp_mreg,
3712 	};
3713 	ext_actions[actions_n] = (struct rte_flow_action){
3714 		.type = RTE_FLOW_ACTION_TYPE_END,
3715 	};
3716 	return 0;
3717 }
3718 
3719 /**
3720  * The splitting for metadata feature.
3721  *
3722  * - Q/RSS action on NIC Rx should be split in order to pass by
3723  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3724  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3725  *
3726  * - All the actions on NIC Tx should have a mreg copy action to
3727  *   copy reg_a from WQE to reg_c[0].
3728  *
3729  * @param dev
3730  *   Pointer to Ethernet device.
3731  * @param[in] flow
3732  *   Parent flow structure pointer.
3733  * @param[in] attr
3734  *   Flow rule attributes.
3735  * @param[in] items
3736  *   Pattern specification (list terminated by the END pattern item).
3737  * @param[in] actions
3738  *   Associated actions (list terminated by the END action).
3739  * @param[in] external
3740  *   This flow rule is created by request external to PMD.
3741  * @param[out] error
3742  *   Perform verbose error reporting if not NULL.
3743  * @return
3744  *   0 on success, negative value otherwise
3745  */
3746 static int
3747 flow_create_split_metadata(struct rte_eth_dev *dev,
3748 			   struct rte_flow *flow,
3749 			   const struct rte_flow_attr *attr,
3750 			   const struct rte_flow_item items[],
3751 			   const struct rte_flow_action actions[],
3752 			   bool external, struct rte_flow_error *error)
3753 {
3754 	struct mlx5_priv *priv = dev->data->dev_private;
3755 	struct mlx5_dev_config *config = &priv->config;
3756 	const struct rte_flow_action *qrss = NULL;
3757 	struct rte_flow_action *ext_actions = NULL;
3758 	struct mlx5_flow *dev_flow = NULL;
3759 	uint32_t qrss_id = 0;
3760 	int mtr_sfx = 0;
3761 	size_t act_size;
3762 	int actions_n;
3763 	int ret;
3764 
3765 	/* Check whether extensive metadata feature is engaged. */
3766 	if (!config->dv_flow_en ||
3767 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3768 	    !mlx5_flow_ext_mreg_supported(dev))
3769 		return flow_create_split_inner(dev, flow, NULL, attr, items,
3770 					       actions, external, error);
3771 	actions_n = flow_parse_qrss_action(actions, &qrss);
3772 	if (qrss) {
3773 		/* Exclude hairpin flows from splitting. */
3774 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3775 			const struct rte_flow_action_queue *queue;
3776 
3777 			queue = qrss->conf;
3778 			if (mlx5_rxq_get_type(dev, queue->index) ==
3779 			    MLX5_RXQ_TYPE_HAIRPIN)
3780 				qrss = NULL;
3781 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3782 			const struct rte_flow_action_rss *rss;
3783 
3784 			rss = qrss->conf;
3785 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3786 			    MLX5_RXQ_TYPE_HAIRPIN)
3787 				qrss = NULL;
3788 		}
3789 	}
3790 	if (qrss) {
3791 		/* Check if it is in meter suffix table. */
3792 		mtr_sfx = attr->group == (attr->transfer ?
3793 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3794 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3795 		/*
3796 		 * Q/RSS action on NIC Rx should be split in order to pass by
3797 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3798 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3799 		 */
3800 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3801 			   sizeof(struct rte_flow_action_set_tag) +
3802 			   sizeof(struct rte_flow_action_jump);
3803 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3804 		if (!ext_actions)
3805 			return rte_flow_error_set(error, ENOMEM,
3806 						  RTE_FLOW_ERROR_TYPE_ACTION,
3807 						  NULL, "no memory to split "
3808 						  "metadata flow");
3809 		/*
3810 		 * If we are the suffix flow of meter, tag already exist.
3811 		 * Set the tag action to void.
3812 		 */
3813 		if (mtr_sfx)
3814 			ext_actions[qrss - actions].type =
3815 						RTE_FLOW_ACTION_TYPE_VOID;
3816 		else
3817 			ext_actions[qrss - actions].type =
3818 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3819 		/*
3820 		 * Create the new actions list with removed Q/RSS action
3821 		 * and appended set tag and jump to register copy table
3822 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3823 		 * in advance, because it is needed for set tag action.
3824 		 */
3825 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3826 						    qrss, actions_n, error);
3827 		if (!mtr_sfx && !qrss_id) {
3828 			ret = -rte_errno;
3829 			goto exit;
3830 		}
3831 	} else if (attr->egress && !attr->transfer) {
3832 		/*
3833 		 * All the actions on NIC Tx should have a metadata register
3834 		 * copy action to copy reg_a from WQE to reg_c[meta]
3835 		 */
3836 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3837 			   sizeof(struct mlx5_flow_action_copy_mreg);
3838 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3839 		if (!ext_actions)
3840 			return rte_flow_error_set(error, ENOMEM,
3841 						  RTE_FLOW_ERROR_TYPE_ACTION,
3842 						  NULL, "no memory to split "
3843 						  "metadata flow");
3844 		/* Create the action list appended with copy register. */
3845 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3846 					     actions_n, error);
3847 		if (ret < 0)
3848 			goto exit;
3849 	}
3850 	/* Add the unmodified original or prefix subflow. */
3851 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3852 				      ext_actions ? ext_actions : actions,
3853 				      external, error);
3854 	if (ret < 0)
3855 		goto exit;
3856 	MLX5_ASSERT(dev_flow);
3857 	if (qrss) {
3858 		const struct rte_flow_attr q_attr = {
3859 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3860 			.ingress = 1,
3861 		};
3862 		/* Internal PMD action to set register. */
3863 		struct mlx5_rte_flow_item_tag q_tag_spec = {
3864 			.data = qrss_id,
3865 			.id = 0,
3866 		};
3867 		struct rte_flow_item q_items[] = {
3868 			{
3869 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3870 				.spec = &q_tag_spec,
3871 				.last = NULL,
3872 				.mask = NULL,
3873 			},
3874 			{
3875 				.type = RTE_FLOW_ITEM_TYPE_END,
3876 			},
3877 		};
3878 		struct rte_flow_action q_actions[] = {
3879 			{
3880 				.type = qrss->type,
3881 				.conf = qrss->conf,
3882 			},
3883 			{
3884 				.type = RTE_FLOW_ACTION_TYPE_END,
3885 			},
3886 		};
3887 		uint64_t hash_fields = dev_flow->hash_fields;
3888 
3889 		/*
3890 		 * Configure the tag item only if there is no meter subflow.
3891 		 * Since tag is already marked in the meter suffix subflow
3892 		 * we can just use the meter suffix items as is.
3893 		 */
3894 		if (qrss_id) {
3895 			/* Not meter subflow. */
3896 			MLX5_ASSERT(!mtr_sfx);
3897 			/*
3898 			 * Put unique id in prefix flow due to it is destroyed
3899 			 * after suffix flow and id will be freed after there
3900 			 * is no actual flows with this id and identifier
3901 			 * reallocation becomes possible (for example, for
3902 			 * other flows in other threads).
3903 			 */
3904 			dev_flow->qrss_id = qrss_id;
3905 			qrss_id = 0;
3906 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3907 						   error);
3908 			if (ret < 0)
3909 				goto exit;
3910 			q_tag_spec.id = ret;
3911 		}
3912 		dev_flow = NULL;
3913 		/* Add suffix subflow to execute Q/RSS. */
3914 		ret = flow_create_split_inner(dev, flow, &dev_flow,
3915 					      &q_attr, mtr_sfx ? items :
3916 					      q_items, q_actions,
3917 					      external, error);
3918 		if (ret < 0)
3919 			goto exit;
3920 		MLX5_ASSERT(dev_flow);
3921 		dev_flow->hash_fields = hash_fields;
3922 	}
3923 
3924 exit:
3925 	/*
3926 	 * We do not destroy the partially created sub_flows in case of error.
3927 	 * These ones are included into parent flow list and will be destroyed
3928 	 * by flow_drv_destroy.
3929 	 */
3930 	flow_qrss_free_id(dev, qrss_id);
3931 	rte_free(ext_actions);
3932 	return ret;
3933 }
3934 
3935 /**
3936  * The splitting for meter feature.
3937  *
3938  * - The meter flow will be split to two flows as prefix and
3939  *   suffix flow. The packets make sense only it pass the prefix
3940  *   meter action.
3941  *
3942  * - Reg_C_5 is used for the packet to match betweend prefix and
3943  *   suffix flow.
3944  *
3945  * @param dev
3946  *   Pointer to Ethernet device.
3947  * @param[in] flow
3948  *   Parent flow structure pointer.
3949  * @param[in] attr
3950  *   Flow rule attributes.
3951  * @param[in] items
3952  *   Pattern specification (list terminated by the END pattern item).
3953  * @param[in] actions
3954  *   Associated actions (list terminated by the END action).
3955  * @param[in] external
3956  *   This flow rule is created by request external to PMD.
3957  * @param[out] error
3958  *   Perform verbose error reporting if not NULL.
3959  * @return
3960  *   0 on success, negative value otherwise
3961  */
3962 static int
3963 flow_create_split_meter(struct rte_eth_dev *dev,
3964 			   struct rte_flow *flow,
3965 			   const struct rte_flow_attr *attr,
3966 			   const struct rte_flow_item items[],
3967 			   const struct rte_flow_action actions[],
3968 			   bool external, struct rte_flow_error *error)
3969 {
3970 	struct mlx5_priv *priv = dev->data->dev_private;
3971 	struct rte_flow_action *sfx_actions = NULL;
3972 	struct rte_flow_action *pre_actions = NULL;
3973 	struct rte_flow_item *sfx_items = NULL;
3974 	const  struct rte_flow_item *sfx_port_id_item;
3975 	struct mlx5_flow *dev_flow = NULL;
3976 	struct rte_flow_attr sfx_attr = *attr;
3977 	uint32_t mtr = 0;
3978 	uint32_t mtr_tag_id = 0;
3979 	size_t act_size;
3980 	size_t item_size;
3981 	int actions_n = 0;
3982 	int ret;
3983 
3984 	if (priv->mtr_en)
3985 		actions_n = flow_check_meter_action(actions, &mtr);
3986 	if (mtr) {
3987 		struct mlx5_rte_flow_item_tag *tag_spec;
3988 		struct mlx5_rte_flow_item_tag *tag_mask;
3989 		/* The five prefix actions: meter, decap, encap, tag, end. */
3990 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3991 			   sizeof(struct rte_flow_action_set_tag);
3992 		/* tag, end. */
3993 #define METER_SUFFIX_ITEM 3
3994 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3995 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
3996 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3997 		if (!sfx_actions)
3998 			return rte_flow_error_set(error, ENOMEM,
3999 						  RTE_FLOW_ERROR_TYPE_ACTION,
4000 						  NULL, "no memory to split "
4001 						  "meter flow");
4002 		pre_actions = sfx_actions + actions_n;
4003 		mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4004 						     pre_actions);
4005 		if (!mtr_tag_id) {
4006 			ret = -rte_errno;
4007 			goto exit;
4008 		}
4009 		/* Add the prefix subflow. */
4010 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4011 						  pre_actions, external, error);
4012 		if (ret) {
4013 			ret = -rte_errno;
4014 			goto exit;
4015 		}
4016 		dev_flow->mtr_flow_id = mtr_tag_id;
4017 		/* Prepare the suffix flow match pattern. */
4018 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4019 			     act_size);
4020 		tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4021 			    METER_SUFFIX_ITEM);
4022 		tag_spec->data = dev_flow->mtr_flow_id << MLX5_MTR_COLOR_BITS;
4023 		tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4024 						    error);
4025 		tag_mask = tag_spec + 1;
4026 		tag_mask->data = 0xffffff00;
4027 		sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4028 		sfx_items->spec = tag_spec;
4029 		sfx_items->last = NULL;
4030 		sfx_items->mask = tag_mask;
4031 		sfx_items++;
4032 		sfx_port_id_item = find_port_id_item(items);
4033 		if (sfx_port_id_item) {
4034 			memcpy(sfx_items, sfx_port_id_item,
4035 			       sizeof(*sfx_items));
4036 			sfx_items++;
4037 		}
4038 		sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4039 		sfx_items -= sfx_port_id_item ? 2 : 1;
4040 		/* Setting the sfx group atrr. */
4041 		sfx_attr.group = sfx_attr.transfer ?
4042 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4043 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4044 	}
4045 	/* Add the prefix subflow. */
4046 	ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4047 					 sfx_items ? sfx_items : items,
4048 					 sfx_actions ? sfx_actions : actions,
4049 					 external, error);
4050 exit:
4051 	if (sfx_actions)
4052 		rte_free(sfx_actions);
4053 	return ret;
4054 }
4055 
4056 /**
4057  * Split the flow to subflow set. The splitters might be linked
4058  * in the chain, like this:
4059  * flow_create_split_outer() calls:
4060  *   flow_create_split_meter() calls:
4061  *     flow_create_split_metadata(meter_subflow_0) calls:
4062  *       flow_create_split_inner(metadata_subflow_0)
4063  *       flow_create_split_inner(metadata_subflow_1)
4064  *       flow_create_split_inner(metadata_subflow_2)
4065  *     flow_create_split_metadata(meter_subflow_1) calls:
4066  *       flow_create_split_inner(metadata_subflow_0)
4067  *       flow_create_split_inner(metadata_subflow_1)
4068  *       flow_create_split_inner(metadata_subflow_2)
4069  *
4070  * This provide flexible way to add new levels of flow splitting.
4071  * The all of successfully created subflows are included to the
4072  * parent flow dev_flow list.
4073  *
4074  * @param dev
4075  *   Pointer to Ethernet device.
4076  * @param[in] flow
4077  *   Parent flow structure pointer.
4078  * @param[in] attr
4079  *   Flow rule attributes.
4080  * @param[in] items
4081  *   Pattern specification (list terminated by the END pattern item).
4082  * @param[in] actions
4083  *   Associated actions (list terminated by the END action).
4084  * @param[in] external
4085  *   This flow rule is created by request external to PMD.
4086  * @param[out] error
4087  *   Perform verbose error reporting if not NULL.
4088  * @return
4089  *   0 on success, negative value otherwise
4090  */
4091 static int
4092 flow_create_split_outer(struct rte_eth_dev *dev,
4093 			struct rte_flow *flow,
4094 			const struct rte_flow_attr *attr,
4095 			const struct rte_flow_item items[],
4096 			const struct rte_flow_action actions[],
4097 			bool external, struct rte_flow_error *error)
4098 {
4099 	int ret;
4100 
4101 	ret = flow_create_split_meter(dev, flow, attr, items,
4102 					 actions, external, error);
4103 	MLX5_ASSERT(ret <= 0);
4104 	return ret;
4105 }
4106 
4107 /**
4108  * Create a flow and add it to @p list.
4109  *
4110  * @param dev
4111  *   Pointer to Ethernet device.
4112  * @param list
4113  *   Pointer to a TAILQ flow list. If this parameter NULL,
4114  *   no list insertion occurred, flow is just created,
4115  *   this is caller's responsibility to track the
4116  *   created flow.
4117  * @param[in] attr
4118  *   Flow rule attributes.
4119  * @param[in] items
4120  *   Pattern specification (list terminated by the END pattern item).
4121  * @param[in] actions
4122  *   Associated actions (list terminated by the END action).
4123  * @param[in] external
4124  *   This flow rule is created by request external to PMD.
4125  * @param[out] error
4126  *   Perform verbose error reporting if not NULL.
4127  *
4128  * @return
4129  *   A flow on success, NULL otherwise and rte_errno is set.
4130  */
4131 static struct rte_flow *
4132 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4133 		 const struct rte_flow_attr *attr,
4134 		 const struct rte_flow_item items[],
4135 		 const struct rte_flow_action actions[],
4136 		 bool external, struct rte_flow_error *error)
4137 {
4138 	struct mlx5_priv *priv = dev->data->dev_private;
4139 	struct rte_flow *flow = NULL;
4140 	struct mlx5_flow *dev_flow;
4141 	const struct rte_flow_action_rss *rss;
4142 	union {
4143 		struct rte_flow_expand_rss buf;
4144 		uint8_t buffer[2048];
4145 	} expand_buffer;
4146 	union {
4147 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4148 		uint8_t buffer[2048];
4149 	} actions_rx;
4150 	union {
4151 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4152 		uint8_t buffer[2048];
4153 	} actions_hairpin_tx;
4154 	union {
4155 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4156 		uint8_t buffer[2048];
4157 	} items_tx;
4158 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4159 	const struct rte_flow_action *p_actions_rx = actions;
4160 	uint32_t i;
4161 	uint32_t flow_size;
4162 	int hairpin_flow = 0;
4163 	uint32_t hairpin_id = 0;
4164 	struct rte_flow_attr attr_tx = { .priority = 0 };
4165 	int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4166 				    error);
4167 
4168 	if (ret < 0)
4169 		return NULL;
4170 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4171 	if (hairpin_flow > 0) {
4172 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4173 			rte_errno = EINVAL;
4174 			return NULL;
4175 		}
4176 		flow_hairpin_split(dev, actions, actions_rx.actions,
4177 				   actions_hairpin_tx.actions, items_tx.items,
4178 				   &hairpin_id);
4179 		p_actions_rx = actions_rx.actions;
4180 	}
4181 	flow_size = sizeof(struct rte_flow);
4182 	rss = flow_get_rss_action(p_actions_rx);
4183 	if (rss)
4184 		flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4185 					    sizeof(void *));
4186 	else
4187 		flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4188 	flow = rte_calloc(__func__, 1, flow_size, 0);
4189 	if (!flow) {
4190 		rte_errno = ENOMEM;
4191 		goto error_before_flow;
4192 	}
4193 	flow->drv_type = flow_get_drv_type(dev, attr);
4194 	if (hairpin_id != 0)
4195 		flow->hairpin_flow_id = hairpin_id;
4196 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4197 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4198 	flow->rss.queue = (void *)(flow + 1);
4199 	if (rss) {
4200 		/*
4201 		 * The following information is required by
4202 		 * mlx5_flow_hashfields_adjust() in advance.
4203 		 */
4204 		flow->rss.level = rss->level;
4205 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4206 		flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4207 	}
4208 	LIST_INIT(&flow->dev_flows);
4209 	if (rss && rss->types) {
4210 		unsigned int graph_root;
4211 
4212 		graph_root = find_graph_root(items, rss->level);
4213 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4214 					  items, rss->types,
4215 					  mlx5_support_expansion,
4216 					  graph_root);
4217 		MLX5_ASSERT(ret > 0 &&
4218 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4219 	} else {
4220 		buf->entries = 1;
4221 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4222 	}
4223 	for (i = 0; i < buf->entries; ++i) {
4224 		/*
4225 		 * The splitter may create multiple dev_flows,
4226 		 * depending on configuration. In the simplest
4227 		 * case it just creates unmodified original flow.
4228 		 */
4229 		ret = flow_create_split_outer(dev, flow, attr,
4230 					      buf->entry[i].pattern,
4231 					      p_actions_rx, external,
4232 					      error);
4233 		if (ret < 0)
4234 			goto error;
4235 	}
4236 	/* Create the tx flow. */
4237 	if (hairpin_flow) {
4238 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4239 		attr_tx.ingress = 0;
4240 		attr_tx.egress = 1;
4241 		dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4242 					    actions_hairpin_tx.actions, error);
4243 		if (!dev_flow)
4244 			goto error;
4245 		dev_flow->flow = flow;
4246 		dev_flow->external = 0;
4247 		LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4248 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4249 					 items_tx.items,
4250 					 actions_hairpin_tx.actions, error);
4251 		if (ret < 0)
4252 			goto error;
4253 	}
4254 	/*
4255 	 * Update the metadata register copy table. If extensive
4256 	 * metadata feature is enabled and registers are supported
4257 	 * we might create the extra rte_flow for each unique
4258 	 * MARK/FLAG action ID.
4259 	 *
4260 	 * The table is updated for ingress Flows only, because
4261 	 * the egress Flows belong to the different device and
4262 	 * copy table should be updated in peer NIC Rx domain.
4263 	 */
4264 	if (attr->ingress &&
4265 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4266 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4267 		if (ret)
4268 			goto error;
4269 	}
4270 	if (dev->data->dev_started) {
4271 		ret = flow_drv_apply(dev, flow, error);
4272 		if (ret < 0)
4273 			goto error;
4274 	}
4275 	if (list)
4276 		TAILQ_INSERT_TAIL(list, flow, next);
4277 	flow_rxq_flags_set(dev, flow);
4278 	return flow;
4279 error_before_flow:
4280 	if (hairpin_id)
4281 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4282 				     hairpin_id);
4283 	return NULL;
4284 error:
4285 	MLX5_ASSERT(flow);
4286 	flow_mreg_del_copy_action(dev, flow);
4287 	ret = rte_errno; /* Save rte_errno before cleanup. */
4288 	if (flow->hairpin_flow_id)
4289 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4290 				     flow->hairpin_flow_id);
4291 	MLX5_ASSERT(flow);
4292 	flow_drv_destroy(dev, flow);
4293 	rte_free(flow);
4294 	rte_errno = ret; /* Restore rte_errno. */
4295 	return NULL;
4296 }
4297 
4298 /**
4299  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4300  * incoming packets to table 1.
4301  *
4302  * Other flow rules, requested for group n, will be created in
4303  * e-switch table n+1.
4304  * Jump action to e-switch group n will be created to group n+1.
4305  *
4306  * Used when working in switchdev mode, to utilise advantages of table 1
4307  * and above.
4308  *
4309  * @param dev
4310  *   Pointer to Ethernet device.
4311  *
4312  * @return
4313  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4314  */
4315 struct rte_flow *
4316 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4317 {
4318 	const struct rte_flow_attr attr = {
4319 		.group = 0,
4320 		.priority = 0,
4321 		.ingress = 1,
4322 		.egress = 0,
4323 		.transfer = 1,
4324 	};
4325 	const struct rte_flow_item pattern = {
4326 		.type = RTE_FLOW_ITEM_TYPE_END,
4327 	};
4328 	struct rte_flow_action_jump jump = {
4329 		.group = 1,
4330 	};
4331 	const struct rte_flow_action actions[] = {
4332 		{
4333 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4334 			.conf = &jump,
4335 		},
4336 		{
4337 			.type = RTE_FLOW_ACTION_TYPE_END,
4338 		},
4339 	};
4340 	struct mlx5_priv *priv = dev->data->dev_private;
4341 	struct rte_flow_error error;
4342 
4343 	return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4344 				actions, false, &error);
4345 }
4346 
4347 /**
4348  * Create a flow.
4349  *
4350  * @see rte_flow_create()
4351  * @see rte_flow_ops
4352  */
4353 struct rte_flow *
4354 mlx5_flow_create(struct rte_eth_dev *dev,
4355 		 const struct rte_flow_attr *attr,
4356 		 const struct rte_flow_item items[],
4357 		 const struct rte_flow_action actions[],
4358 		 struct rte_flow_error *error)
4359 {
4360 	struct mlx5_priv *priv = dev->data->dev_private;
4361 
4362 	return flow_list_create(dev, &priv->flows,
4363 				attr, items, actions, true, error);
4364 }
4365 
4366 /**
4367  * Destroy a flow in a list.
4368  *
4369  * @param dev
4370  *   Pointer to Ethernet device.
4371  * @param list
4372  *   Pointer to a TAILQ flow list. If this parameter NULL,
4373  *   there is no flow removal from the list.
4374  * @param[in] flow
4375  *   Flow to destroy.
4376  */
4377 static void
4378 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4379 		  struct rte_flow *flow)
4380 {
4381 	struct mlx5_priv *priv = dev->data->dev_private;
4382 
4383 	/*
4384 	 * Update RX queue flags only if port is started, otherwise it is
4385 	 * already clean.
4386 	 */
4387 	if (dev->data->dev_started)
4388 		flow_rxq_flags_trim(dev, flow);
4389 	if (flow->hairpin_flow_id)
4390 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4391 				     flow->hairpin_flow_id);
4392 	flow_drv_destroy(dev, flow);
4393 	if (list)
4394 		TAILQ_REMOVE(list, flow, next);
4395 	flow_mreg_del_copy_action(dev, flow);
4396 	rte_free(flow->fdir);
4397 	rte_free(flow);
4398 }
4399 
4400 /**
4401  * Destroy all flows.
4402  *
4403  * @param dev
4404  *   Pointer to Ethernet device.
4405  * @param list
4406  *   Pointer to a TAILQ flow list.
4407  */
4408 void
4409 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4410 {
4411 	while (!TAILQ_EMPTY(list)) {
4412 		struct rte_flow *flow;
4413 
4414 		flow = TAILQ_FIRST(list);
4415 		flow_list_destroy(dev, list, flow);
4416 	}
4417 }
4418 
4419 /**
4420  * Remove all flows.
4421  *
4422  * @param dev
4423  *   Pointer to Ethernet device.
4424  * @param list
4425  *   Pointer to a TAILQ flow list.
4426  */
4427 void
4428 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4429 {
4430 	struct rte_flow *flow;
4431 
4432 	TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4433 		flow_drv_remove(dev, flow);
4434 		flow_mreg_stop_copy_action(dev, flow);
4435 	}
4436 	flow_mreg_del_default_copy_action(dev);
4437 	flow_rxq_flags_clear(dev);
4438 }
4439 
4440 /**
4441  * Add all flows.
4442  *
4443  * @param dev
4444  *   Pointer to Ethernet device.
4445  * @param list
4446  *   Pointer to a TAILQ flow list.
4447  *
4448  * @return
4449  *   0 on success, a negative errno value otherwise and rte_errno is set.
4450  */
4451 int
4452 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4453 {
4454 	struct rte_flow *flow;
4455 	struct rte_flow_error error;
4456 	int ret = 0;
4457 
4458 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4459 	ret = flow_mreg_add_default_copy_action(dev, &error);
4460 	if (ret < 0)
4461 		return -rte_errno;
4462 	/* Apply Flows created by application. */
4463 	TAILQ_FOREACH(flow, list, next) {
4464 		ret = flow_mreg_start_copy_action(dev, flow);
4465 		if (ret < 0)
4466 			goto error;
4467 		ret = flow_drv_apply(dev, flow, &error);
4468 		if (ret < 0)
4469 			goto error;
4470 		flow_rxq_flags_set(dev, flow);
4471 	}
4472 	return 0;
4473 error:
4474 	ret = rte_errno; /* Save rte_errno before cleanup. */
4475 	mlx5_flow_stop(dev, list);
4476 	rte_errno = ret; /* Restore rte_errno. */
4477 	return -rte_errno;
4478 }
4479 
4480 /**
4481  * Verify the flow list is empty
4482  *
4483  * @param dev
4484  *  Pointer to Ethernet device.
4485  *
4486  * @return the number of flows not released.
4487  */
4488 int
4489 mlx5_flow_verify(struct rte_eth_dev *dev)
4490 {
4491 	struct mlx5_priv *priv = dev->data->dev_private;
4492 	struct rte_flow *flow;
4493 	int ret = 0;
4494 
4495 	TAILQ_FOREACH(flow, &priv->flows, next) {
4496 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4497 			dev->data->port_id, (void *)flow);
4498 		++ret;
4499 	}
4500 	return ret;
4501 }
4502 
4503 /**
4504  * Enable default hairpin egress flow.
4505  *
4506  * @param dev
4507  *   Pointer to Ethernet device.
4508  * @param queue
4509  *   The queue index.
4510  *
4511  * @return
4512  *   0 on success, a negative errno value otherwise and rte_errno is set.
4513  */
4514 int
4515 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4516 			    uint32_t queue)
4517 {
4518 	struct mlx5_priv *priv = dev->data->dev_private;
4519 	const struct rte_flow_attr attr = {
4520 		.egress = 1,
4521 		.priority = 0,
4522 	};
4523 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4524 		.queue = queue,
4525 	};
4526 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4527 		.queue = UINT32_MAX,
4528 	};
4529 	struct rte_flow_item items[] = {
4530 		{
4531 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4532 			.spec = &queue_spec,
4533 			.last = NULL,
4534 			.mask = &queue_mask,
4535 		},
4536 		{
4537 			.type = RTE_FLOW_ITEM_TYPE_END,
4538 		},
4539 	};
4540 	struct rte_flow_action_jump jump = {
4541 		.group = MLX5_HAIRPIN_TX_TABLE,
4542 	};
4543 	struct rte_flow_action actions[2];
4544 	struct rte_flow *flow;
4545 	struct rte_flow_error error;
4546 
4547 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4548 	actions[0].conf = &jump;
4549 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4550 	flow = flow_list_create(dev, &priv->ctrl_flows,
4551 				&attr, items, actions, false, &error);
4552 	if (!flow) {
4553 		DRV_LOG(DEBUG,
4554 			"Failed to create ctrl flow: rte_errno(%d),"
4555 			" type(%d), message(%s)",
4556 			rte_errno, error.type,
4557 			error.message ? error.message : " (no stated reason)");
4558 		return -rte_errno;
4559 	}
4560 	return 0;
4561 }
4562 
4563 /**
4564  * Enable a control flow configured from the control plane.
4565  *
4566  * @param dev
4567  *   Pointer to Ethernet device.
4568  * @param eth_spec
4569  *   An Ethernet flow spec to apply.
4570  * @param eth_mask
4571  *   An Ethernet flow mask to apply.
4572  * @param vlan_spec
4573  *   A VLAN flow spec to apply.
4574  * @param vlan_mask
4575  *   A VLAN flow mask to apply.
4576  *
4577  * @return
4578  *   0 on success, a negative errno value otherwise and rte_errno is set.
4579  */
4580 int
4581 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4582 		    struct rte_flow_item_eth *eth_spec,
4583 		    struct rte_flow_item_eth *eth_mask,
4584 		    struct rte_flow_item_vlan *vlan_spec,
4585 		    struct rte_flow_item_vlan *vlan_mask)
4586 {
4587 	struct mlx5_priv *priv = dev->data->dev_private;
4588 	const struct rte_flow_attr attr = {
4589 		.ingress = 1,
4590 		.priority = MLX5_FLOW_PRIO_RSVD,
4591 	};
4592 	struct rte_flow_item items[] = {
4593 		{
4594 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4595 			.spec = eth_spec,
4596 			.last = NULL,
4597 			.mask = eth_mask,
4598 		},
4599 		{
4600 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4601 					      RTE_FLOW_ITEM_TYPE_END,
4602 			.spec = vlan_spec,
4603 			.last = NULL,
4604 			.mask = vlan_mask,
4605 		},
4606 		{
4607 			.type = RTE_FLOW_ITEM_TYPE_END,
4608 		},
4609 	};
4610 	uint16_t queue[priv->reta_idx_n];
4611 	struct rte_flow_action_rss action_rss = {
4612 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4613 		.level = 0,
4614 		.types = priv->rss_conf.rss_hf,
4615 		.key_len = priv->rss_conf.rss_key_len,
4616 		.queue_num = priv->reta_idx_n,
4617 		.key = priv->rss_conf.rss_key,
4618 		.queue = queue,
4619 	};
4620 	struct rte_flow_action actions[] = {
4621 		{
4622 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4623 			.conf = &action_rss,
4624 		},
4625 		{
4626 			.type = RTE_FLOW_ACTION_TYPE_END,
4627 		},
4628 	};
4629 	struct rte_flow *flow;
4630 	struct rte_flow_error error;
4631 	unsigned int i;
4632 
4633 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4634 		return 0;
4635 	}
4636 	for (i = 0; i != priv->reta_idx_n; ++i)
4637 		queue[i] = (*priv->reta_idx)[i];
4638 	flow = flow_list_create(dev, &priv->ctrl_flows,
4639 				&attr, items, actions, false, &error);
4640 	if (!flow)
4641 		return -rte_errno;
4642 	return 0;
4643 }
4644 
4645 /**
4646  * Enable a flow control configured from the control plane.
4647  *
4648  * @param dev
4649  *   Pointer to Ethernet device.
4650  * @param eth_spec
4651  *   An Ethernet flow spec to apply.
4652  * @param eth_mask
4653  *   An Ethernet flow mask to apply.
4654  *
4655  * @return
4656  *   0 on success, a negative errno value otherwise and rte_errno is set.
4657  */
4658 int
4659 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4660 	       struct rte_flow_item_eth *eth_spec,
4661 	       struct rte_flow_item_eth *eth_mask)
4662 {
4663 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4664 }
4665 
4666 /**
4667  * Destroy a flow.
4668  *
4669  * @see rte_flow_destroy()
4670  * @see rte_flow_ops
4671  */
4672 int
4673 mlx5_flow_destroy(struct rte_eth_dev *dev,
4674 		  struct rte_flow *flow,
4675 		  struct rte_flow_error *error __rte_unused)
4676 {
4677 	struct mlx5_priv *priv = dev->data->dev_private;
4678 
4679 	flow_list_destroy(dev, &priv->flows, flow);
4680 	return 0;
4681 }
4682 
4683 /**
4684  * Destroy all flows.
4685  *
4686  * @see rte_flow_flush()
4687  * @see rte_flow_ops
4688  */
4689 int
4690 mlx5_flow_flush(struct rte_eth_dev *dev,
4691 		struct rte_flow_error *error __rte_unused)
4692 {
4693 	struct mlx5_priv *priv = dev->data->dev_private;
4694 
4695 	mlx5_flow_list_flush(dev, &priv->flows);
4696 	return 0;
4697 }
4698 
4699 /**
4700  * Isolated mode.
4701  *
4702  * @see rte_flow_isolate()
4703  * @see rte_flow_ops
4704  */
4705 int
4706 mlx5_flow_isolate(struct rte_eth_dev *dev,
4707 		  int enable,
4708 		  struct rte_flow_error *error)
4709 {
4710 	struct mlx5_priv *priv = dev->data->dev_private;
4711 
4712 	if (dev->data->dev_started) {
4713 		rte_flow_error_set(error, EBUSY,
4714 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4715 				   NULL,
4716 				   "port must be stopped first");
4717 		return -rte_errno;
4718 	}
4719 	priv->isolated = !!enable;
4720 	if (enable)
4721 		dev->dev_ops = &mlx5_dev_ops_isolate;
4722 	else
4723 		dev->dev_ops = &mlx5_dev_ops;
4724 	return 0;
4725 }
4726 
4727 /**
4728  * Query a flow.
4729  *
4730  * @see rte_flow_query()
4731  * @see rte_flow_ops
4732  */
4733 static int
4734 flow_drv_query(struct rte_eth_dev *dev,
4735 	       struct rte_flow *flow,
4736 	       const struct rte_flow_action *actions,
4737 	       void *data,
4738 	       struct rte_flow_error *error)
4739 {
4740 	const struct mlx5_flow_driver_ops *fops;
4741 	enum mlx5_flow_drv_type ftype = flow->drv_type;
4742 
4743 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4744 	fops = flow_get_drv_ops(ftype);
4745 
4746 	return fops->query(dev, flow, actions, data, error);
4747 }
4748 
4749 /**
4750  * Query a flow.
4751  *
4752  * @see rte_flow_query()
4753  * @see rte_flow_ops
4754  */
4755 int
4756 mlx5_flow_query(struct rte_eth_dev *dev,
4757 		struct rte_flow *flow,
4758 		const struct rte_flow_action *actions,
4759 		void *data,
4760 		struct rte_flow_error *error)
4761 {
4762 	int ret;
4763 
4764 	ret = flow_drv_query(dev, flow, actions, data, error);
4765 	if (ret < 0)
4766 		return ret;
4767 	return 0;
4768 }
4769 
4770 /**
4771  * Convert a flow director filter to a generic flow.
4772  *
4773  * @param dev
4774  *   Pointer to Ethernet device.
4775  * @param fdir_filter
4776  *   Flow director filter to add.
4777  * @param attributes
4778  *   Generic flow parameters structure.
4779  *
4780  * @return
4781  *   0 on success, a negative errno value otherwise and rte_errno is set.
4782  */
4783 static int
4784 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4785 			 const struct rte_eth_fdir_filter *fdir_filter,
4786 			 struct mlx5_fdir *attributes)
4787 {
4788 	struct mlx5_priv *priv = dev->data->dev_private;
4789 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
4790 	const struct rte_eth_fdir_masks *mask =
4791 		&dev->data->dev_conf.fdir_conf.mask;
4792 
4793 	/* Validate queue number. */
4794 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4795 		DRV_LOG(ERR, "port %u invalid queue number %d",
4796 			dev->data->port_id, fdir_filter->action.rx_queue);
4797 		rte_errno = EINVAL;
4798 		return -rte_errno;
4799 	}
4800 	attributes->attr.ingress = 1;
4801 	attributes->items[0] = (struct rte_flow_item) {
4802 		.type = RTE_FLOW_ITEM_TYPE_ETH,
4803 		.spec = &attributes->l2,
4804 		.mask = &attributes->l2_mask,
4805 	};
4806 	switch (fdir_filter->action.behavior) {
4807 	case RTE_ETH_FDIR_ACCEPT:
4808 		attributes->actions[0] = (struct rte_flow_action){
4809 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
4810 			.conf = &attributes->queue,
4811 		};
4812 		break;
4813 	case RTE_ETH_FDIR_REJECT:
4814 		attributes->actions[0] = (struct rte_flow_action){
4815 			.type = RTE_FLOW_ACTION_TYPE_DROP,
4816 		};
4817 		break;
4818 	default:
4819 		DRV_LOG(ERR, "port %u invalid behavior %d",
4820 			dev->data->port_id,
4821 			fdir_filter->action.behavior);
4822 		rte_errno = ENOTSUP;
4823 		return -rte_errno;
4824 	}
4825 	attributes->queue.index = fdir_filter->action.rx_queue;
4826 	/* Handle L3. */
4827 	switch (fdir_filter->input.flow_type) {
4828 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4829 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4830 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4831 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4832 			.src_addr = input->flow.ip4_flow.src_ip,
4833 			.dst_addr = input->flow.ip4_flow.dst_ip,
4834 			.time_to_live = input->flow.ip4_flow.ttl,
4835 			.type_of_service = input->flow.ip4_flow.tos,
4836 		};
4837 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4838 			.src_addr = mask->ipv4_mask.src_ip,
4839 			.dst_addr = mask->ipv4_mask.dst_ip,
4840 			.time_to_live = mask->ipv4_mask.ttl,
4841 			.type_of_service = mask->ipv4_mask.tos,
4842 			.next_proto_id = mask->ipv4_mask.proto,
4843 		};
4844 		attributes->items[1] = (struct rte_flow_item){
4845 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
4846 			.spec = &attributes->l3,
4847 			.mask = &attributes->l3_mask,
4848 		};
4849 		break;
4850 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4851 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4852 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4853 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4854 			.hop_limits = input->flow.ipv6_flow.hop_limits,
4855 			.proto = input->flow.ipv6_flow.proto,
4856 		};
4857 
4858 		memcpy(attributes->l3.ipv6.hdr.src_addr,
4859 		       input->flow.ipv6_flow.src_ip,
4860 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4861 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
4862 		       input->flow.ipv6_flow.dst_ip,
4863 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4864 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4865 		       mask->ipv6_mask.src_ip,
4866 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4867 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4868 		       mask->ipv6_mask.dst_ip,
4869 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4870 		attributes->items[1] = (struct rte_flow_item){
4871 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
4872 			.spec = &attributes->l3,
4873 			.mask = &attributes->l3_mask,
4874 		};
4875 		break;
4876 	default:
4877 		DRV_LOG(ERR, "port %u invalid flow type%d",
4878 			dev->data->port_id, fdir_filter->input.flow_type);
4879 		rte_errno = ENOTSUP;
4880 		return -rte_errno;
4881 	}
4882 	/* Handle L4. */
4883 	switch (fdir_filter->input.flow_type) {
4884 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4885 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4886 			.src_port = input->flow.udp4_flow.src_port,
4887 			.dst_port = input->flow.udp4_flow.dst_port,
4888 		};
4889 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4890 			.src_port = mask->src_port_mask,
4891 			.dst_port = mask->dst_port_mask,
4892 		};
4893 		attributes->items[2] = (struct rte_flow_item){
4894 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4895 			.spec = &attributes->l4,
4896 			.mask = &attributes->l4_mask,
4897 		};
4898 		break;
4899 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4900 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4901 			.src_port = input->flow.tcp4_flow.src_port,
4902 			.dst_port = input->flow.tcp4_flow.dst_port,
4903 		};
4904 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4905 			.src_port = mask->src_port_mask,
4906 			.dst_port = mask->dst_port_mask,
4907 		};
4908 		attributes->items[2] = (struct rte_flow_item){
4909 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4910 			.spec = &attributes->l4,
4911 			.mask = &attributes->l4_mask,
4912 		};
4913 		break;
4914 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4915 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
4916 			.src_port = input->flow.udp6_flow.src_port,
4917 			.dst_port = input->flow.udp6_flow.dst_port,
4918 		};
4919 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4920 			.src_port = mask->src_port_mask,
4921 			.dst_port = mask->dst_port_mask,
4922 		};
4923 		attributes->items[2] = (struct rte_flow_item){
4924 			.type = RTE_FLOW_ITEM_TYPE_UDP,
4925 			.spec = &attributes->l4,
4926 			.mask = &attributes->l4_mask,
4927 		};
4928 		break;
4929 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4930 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4931 			.src_port = input->flow.tcp6_flow.src_port,
4932 			.dst_port = input->flow.tcp6_flow.dst_port,
4933 		};
4934 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4935 			.src_port = mask->src_port_mask,
4936 			.dst_port = mask->dst_port_mask,
4937 		};
4938 		attributes->items[2] = (struct rte_flow_item){
4939 			.type = RTE_FLOW_ITEM_TYPE_TCP,
4940 			.spec = &attributes->l4,
4941 			.mask = &attributes->l4_mask,
4942 		};
4943 		break;
4944 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4945 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4946 		break;
4947 	default:
4948 		DRV_LOG(ERR, "port %u invalid flow type%d",
4949 			dev->data->port_id, fdir_filter->input.flow_type);
4950 		rte_errno = ENOTSUP;
4951 		return -rte_errno;
4952 	}
4953 	return 0;
4954 }
4955 
4956 #define FLOW_FDIR_CMP(f1, f2, fld) \
4957 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4958 
4959 /**
4960  * Compare two FDIR flows. If items and actions are identical, the two flows are
4961  * regarded as same.
4962  *
4963  * @param dev
4964  *   Pointer to Ethernet device.
4965  * @param f1
4966  *   FDIR flow to compare.
4967  * @param f2
4968  *   FDIR flow to compare.
4969  *
4970  * @return
4971  *   Zero on match, 1 otherwise.
4972  */
4973 static int
4974 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4975 {
4976 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
4977 	    FLOW_FDIR_CMP(f1, f2, l2) ||
4978 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4979 	    FLOW_FDIR_CMP(f1, f2, l3) ||
4980 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4981 	    FLOW_FDIR_CMP(f1, f2, l4) ||
4982 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4983 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
4984 		return 1;
4985 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4986 	    FLOW_FDIR_CMP(f1, f2, queue))
4987 		return 1;
4988 	return 0;
4989 }
4990 
4991 /**
4992  * Search device flow list to find out a matched FDIR flow.
4993  *
4994  * @param dev
4995  *   Pointer to Ethernet device.
4996  * @param fdir_flow
4997  *   FDIR flow to lookup.
4998  *
4999  * @return
5000  *   Pointer of flow if found, NULL otherwise.
5001  */
5002 static struct rte_flow *
5003 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5004 {
5005 	struct mlx5_priv *priv = dev->data->dev_private;
5006 	struct rte_flow *flow = NULL;
5007 
5008 	MLX5_ASSERT(fdir_flow);
5009 	TAILQ_FOREACH(flow, &priv->flows, next) {
5010 		if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5011 			DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5012 				dev->data->port_id, (void *)flow);
5013 			break;
5014 		}
5015 	}
5016 	return flow;
5017 }
5018 
5019 /**
5020  * Add new flow director filter and store it in list.
5021  *
5022  * @param dev
5023  *   Pointer to Ethernet device.
5024  * @param fdir_filter
5025  *   Flow director filter to add.
5026  *
5027  * @return
5028  *   0 on success, a negative errno value otherwise and rte_errno is set.
5029  */
5030 static int
5031 flow_fdir_filter_add(struct rte_eth_dev *dev,
5032 		     const struct rte_eth_fdir_filter *fdir_filter)
5033 {
5034 	struct mlx5_priv *priv = dev->data->dev_private;
5035 	struct mlx5_fdir *fdir_flow;
5036 	struct rte_flow *flow;
5037 	int ret;
5038 
5039 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5040 	if (!fdir_flow) {
5041 		rte_errno = ENOMEM;
5042 		return -rte_errno;
5043 	}
5044 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5045 	if (ret)
5046 		goto error;
5047 	flow = flow_fdir_filter_lookup(dev, fdir_flow);
5048 	if (flow) {
5049 		rte_errno = EEXIST;
5050 		goto error;
5051 	}
5052 	flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5053 				fdir_flow->items, fdir_flow->actions, true,
5054 				NULL);
5055 	if (!flow)
5056 		goto error;
5057 	MLX5_ASSERT(!flow->fdir);
5058 	flow->fdir = fdir_flow;
5059 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5060 		dev->data->port_id, (void *)flow);
5061 	return 0;
5062 error:
5063 	rte_free(fdir_flow);
5064 	return -rte_errno;
5065 }
5066 
5067 /**
5068  * Delete specific filter.
5069  *
5070  * @param dev
5071  *   Pointer to Ethernet device.
5072  * @param fdir_filter
5073  *   Filter to be deleted.
5074  *
5075  * @return
5076  *   0 on success, a negative errno value otherwise and rte_errno is set.
5077  */
5078 static int
5079 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5080 			const struct rte_eth_fdir_filter *fdir_filter)
5081 {
5082 	struct mlx5_priv *priv = dev->data->dev_private;
5083 	struct rte_flow *flow;
5084 	struct mlx5_fdir fdir_flow = {
5085 		.attr.group = 0,
5086 	};
5087 	int ret;
5088 
5089 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5090 	if (ret)
5091 		return -rte_errno;
5092 	flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5093 	if (!flow) {
5094 		rte_errno = ENOENT;
5095 		return -rte_errno;
5096 	}
5097 	flow_list_destroy(dev, &priv->flows, flow);
5098 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5099 		dev->data->port_id, (void *)flow);
5100 	return 0;
5101 }
5102 
5103 /**
5104  * Update queue for specific filter.
5105  *
5106  * @param dev
5107  *   Pointer to Ethernet device.
5108  * @param fdir_filter
5109  *   Filter to be updated.
5110  *
5111  * @return
5112  *   0 on success, a negative errno value otherwise and rte_errno is set.
5113  */
5114 static int
5115 flow_fdir_filter_update(struct rte_eth_dev *dev,
5116 			const struct rte_eth_fdir_filter *fdir_filter)
5117 {
5118 	int ret;
5119 
5120 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5121 	if (ret)
5122 		return ret;
5123 	return flow_fdir_filter_add(dev, fdir_filter);
5124 }
5125 
5126 /**
5127  * Flush all filters.
5128  *
5129  * @param dev
5130  *   Pointer to Ethernet device.
5131  */
5132 static void
5133 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5134 {
5135 	struct mlx5_priv *priv = dev->data->dev_private;
5136 
5137 	mlx5_flow_list_flush(dev, &priv->flows);
5138 }
5139 
5140 /**
5141  * Get flow director information.
5142  *
5143  * @param dev
5144  *   Pointer to Ethernet device.
5145  * @param[out] fdir_info
5146  *   Resulting flow director information.
5147  */
5148 static void
5149 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5150 {
5151 	struct rte_eth_fdir_masks *mask =
5152 		&dev->data->dev_conf.fdir_conf.mask;
5153 
5154 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5155 	fdir_info->guarant_spc = 0;
5156 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5157 	fdir_info->max_flexpayload = 0;
5158 	fdir_info->flow_types_mask[0] = 0;
5159 	fdir_info->flex_payload_unit = 0;
5160 	fdir_info->max_flex_payload_segment_num = 0;
5161 	fdir_info->flex_payload_limit = 0;
5162 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5163 }
5164 
5165 /**
5166  * Deal with flow director operations.
5167  *
5168  * @param dev
5169  *   Pointer to Ethernet device.
5170  * @param filter_op
5171  *   Operation to perform.
5172  * @param arg
5173  *   Pointer to operation-specific structure.
5174  *
5175  * @return
5176  *   0 on success, a negative errno value otherwise and rte_errno is set.
5177  */
5178 static int
5179 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5180 		    void *arg)
5181 {
5182 	enum rte_fdir_mode fdir_mode =
5183 		dev->data->dev_conf.fdir_conf.mode;
5184 
5185 	if (filter_op == RTE_ETH_FILTER_NOP)
5186 		return 0;
5187 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5188 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5189 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5190 			dev->data->port_id, fdir_mode);
5191 		rte_errno = EINVAL;
5192 		return -rte_errno;
5193 	}
5194 	switch (filter_op) {
5195 	case RTE_ETH_FILTER_ADD:
5196 		return flow_fdir_filter_add(dev, arg);
5197 	case RTE_ETH_FILTER_UPDATE:
5198 		return flow_fdir_filter_update(dev, arg);
5199 	case RTE_ETH_FILTER_DELETE:
5200 		return flow_fdir_filter_delete(dev, arg);
5201 	case RTE_ETH_FILTER_FLUSH:
5202 		flow_fdir_filter_flush(dev);
5203 		break;
5204 	case RTE_ETH_FILTER_INFO:
5205 		flow_fdir_info_get(dev, arg);
5206 		break;
5207 	default:
5208 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5209 			dev->data->port_id, filter_op);
5210 		rte_errno = EINVAL;
5211 		return -rte_errno;
5212 	}
5213 	return 0;
5214 }
5215 
5216 /**
5217  * Manage filter operations.
5218  *
5219  * @param dev
5220  *   Pointer to Ethernet device structure.
5221  * @param filter_type
5222  *   Filter type.
5223  * @param filter_op
5224  *   Operation to perform.
5225  * @param arg
5226  *   Pointer to operation-specific structure.
5227  *
5228  * @return
5229  *   0 on success, a negative errno value otherwise and rte_errno is set.
5230  */
5231 int
5232 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5233 		     enum rte_filter_type filter_type,
5234 		     enum rte_filter_op filter_op,
5235 		     void *arg)
5236 {
5237 	switch (filter_type) {
5238 	case RTE_ETH_FILTER_GENERIC:
5239 		if (filter_op != RTE_ETH_FILTER_GET) {
5240 			rte_errno = EINVAL;
5241 			return -rte_errno;
5242 		}
5243 		*(const void **)arg = &mlx5_flow_ops;
5244 		return 0;
5245 	case RTE_ETH_FILTER_FDIR:
5246 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5247 	default:
5248 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5249 			dev->data->port_id, filter_type);
5250 		rte_errno = ENOTSUP;
5251 		return -rte_errno;
5252 	}
5253 	return 0;
5254 }
5255 
5256 /**
5257  * Create the needed meter and suffix tables.
5258  *
5259  * @param[in] dev
5260  *   Pointer to Ethernet device.
5261  * @param[in] fm
5262  *   Pointer to the flow meter.
5263  *
5264  * @return
5265  *   Pointer to table set on success, NULL otherwise.
5266  */
5267 struct mlx5_meter_domains_infos *
5268 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5269 			  const struct mlx5_flow_meter *fm)
5270 {
5271 	const struct mlx5_flow_driver_ops *fops;
5272 
5273 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5274 	return fops->create_mtr_tbls(dev, fm);
5275 }
5276 
5277 /**
5278  * Destroy the meter table set.
5279  *
5280  * @param[in] dev
5281  *   Pointer to Ethernet device.
5282  * @param[in] tbl
5283  *   Pointer to the meter table set.
5284  *
5285  * @return
5286  *   0 on success.
5287  */
5288 int
5289 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5290 			   struct mlx5_meter_domains_infos *tbls)
5291 {
5292 	const struct mlx5_flow_driver_ops *fops;
5293 
5294 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5295 	return fops->destroy_mtr_tbls(dev, tbls);
5296 }
5297 
5298 /**
5299  * Create policer rules.
5300  *
5301  * @param[in] dev
5302  *   Pointer to Ethernet device.
5303  * @param[in] fm
5304  *   Pointer to flow meter structure.
5305  * @param[in] attr
5306  *   Pointer to flow attributes.
5307  *
5308  * @return
5309  *   0 on success, -1 otherwise.
5310  */
5311 int
5312 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5313 			       struct mlx5_flow_meter *fm,
5314 			       const struct rte_flow_attr *attr)
5315 {
5316 	const struct mlx5_flow_driver_ops *fops;
5317 
5318 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5319 	return fops->create_policer_rules(dev, fm, attr);
5320 }
5321 
5322 /**
5323  * Destroy policer rules.
5324  *
5325  * @param[in] fm
5326  *   Pointer to flow meter structure.
5327  * @param[in] attr
5328  *   Pointer to flow attributes.
5329  *
5330  * @return
5331  *   0 on success, -1 otherwise.
5332  */
5333 int
5334 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5335 				struct mlx5_flow_meter *fm,
5336 				const struct rte_flow_attr *attr)
5337 {
5338 	const struct mlx5_flow_driver_ops *fops;
5339 
5340 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5341 	return fops->destroy_policer_rules(dev, fm, attr);
5342 }
5343 
5344 /**
5345  * Allocate a counter.
5346  *
5347  * @param[in] dev
5348  *   Pointer to Ethernet device structure.
5349  *
5350  * @return
5351  *   Pointer to allocated counter  on success, NULL otherwise.
5352  */
5353 struct mlx5_flow_counter *
5354 mlx5_counter_alloc(struct rte_eth_dev *dev)
5355 {
5356 	const struct mlx5_flow_driver_ops *fops;
5357 	struct rte_flow_attr attr = { .transfer = 0 };
5358 
5359 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5360 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5361 		return fops->counter_alloc(dev);
5362 	}
5363 	DRV_LOG(ERR,
5364 		"port %u counter allocate is not supported.",
5365 		 dev->data->port_id);
5366 	return NULL;
5367 }
5368 
5369 /**
5370  * Free a counter.
5371  *
5372  * @param[in] dev
5373  *   Pointer to Ethernet device structure.
5374  * @param[in] cnt
5375  *   Pointer to counter to be free.
5376  */
5377 void
5378 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5379 {
5380 	const struct mlx5_flow_driver_ops *fops;
5381 	struct rte_flow_attr attr = { .transfer = 0 };
5382 
5383 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5384 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5385 		fops->counter_free(dev, cnt);
5386 		return;
5387 	}
5388 	DRV_LOG(ERR,
5389 		"port %u counter free is not supported.",
5390 		 dev->data->port_id);
5391 }
5392 
5393 /**
5394  * Query counter statistics.
5395  *
5396  * @param[in] dev
5397  *   Pointer to Ethernet device structure.
5398  * @param[in] cnt
5399  *   Pointer to counter to query.
5400  * @param[in] clear
5401  *   Set to clear counter statistics.
5402  * @param[out] pkts
5403  *   The counter hits packets number to save.
5404  * @param[out] bytes
5405  *   The counter hits bytes number to save.
5406  *
5407  * @return
5408  *   0 on success, a negative errno value otherwise.
5409  */
5410 int
5411 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5412 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5413 {
5414 	const struct mlx5_flow_driver_ops *fops;
5415 	struct rte_flow_attr attr = { .transfer = 0 };
5416 
5417 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5418 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5419 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5420 	}
5421 	DRV_LOG(ERR,
5422 		"port %u counter query is not supported.",
5423 		 dev->data->port_id);
5424 	return -ENOTSUP;
5425 }
5426 
5427 #define MLX5_POOL_QUERY_FREQ_US 1000000
5428 
5429 /**
5430  * Set the periodic procedure for triggering asynchronous batch queries for all
5431  * the counter pools.
5432  *
5433  * @param[in] sh
5434  *   Pointer to mlx5_ibv_shared object.
5435  */
5436 void
5437 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5438 {
5439 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5440 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5441 	uint32_t us;
5442 
5443 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5444 	pools_n += rte_atomic16_read(&cont->n_valid);
5445 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5446 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5447 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5448 		sh->cmng.query_thread_on = 0;
5449 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5450 	} else {
5451 		sh->cmng.query_thread_on = 1;
5452 	}
5453 }
5454 
5455 /**
5456  * The periodic procedure for triggering asynchronous batch queries for all the
5457  * counter pools. This function is probably called by the host thread.
5458  *
5459  * @param[in] arg
5460  *   The parameter for the alarm process.
5461  */
5462 void
5463 mlx5_flow_query_alarm(void *arg)
5464 {
5465 	struct mlx5_ibv_shared *sh = arg;
5466 	struct mlx5_devx_obj *dcs;
5467 	uint16_t offset;
5468 	int ret;
5469 	uint8_t batch = sh->cmng.batch;
5470 	uint16_t pool_index = sh->cmng.pool_index;
5471 	struct mlx5_pools_container *cont;
5472 	struct mlx5_pools_container *mcont;
5473 	struct mlx5_flow_counter_pool *pool;
5474 
5475 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5476 		goto set_alarm;
5477 next_container:
5478 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5479 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5480 	/* Check if resize was done and need to flip a container. */
5481 	if (cont != mcont) {
5482 		if (cont->pools) {
5483 			/* Clean the old container. */
5484 			rte_free(cont->pools);
5485 			memset(cont, 0, sizeof(*cont));
5486 		}
5487 		rte_cio_wmb();
5488 		 /* Flip the host container. */
5489 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5490 		cont = mcont;
5491 	}
5492 	if (!cont->pools) {
5493 		/* 2 empty containers case is unexpected. */
5494 		if (unlikely(batch != sh->cmng.batch))
5495 			goto set_alarm;
5496 		batch ^= 0x1;
5497 		pool_index = 0;
5498 		goto next_container;
5499 	}
5500 	pool = cont->pools[pool_index];
5501 	if (pool->raw_hw)
5502 		/* There is a pool query in progress. */
5503 		goto set_alarm;
5504 	pool->raw_hw =
5505 		LIST_FIRST(&sh->cmng.free_stat_raws);
5506 	if (!pool->raw_hw)
5507 		/* No free counter statistics raw memory. */
5508 		goto set_alarm;
5509 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5510 							      (&pool->a64_dcs);
5511 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5512 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5513 					       offset, NULL, NULL,
5514 					       pool->raw_hw->mem_mng->dm->id,
5515 					       (void *)(uintptr_t)
5516 					       (pool->raw_hw->data + offset),
5517 					       sh->devx_comp,
5518 					       (uint64_t)(uintptr_t)pool);
5519 	if (ret) {
5520 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5521 			" %d", pool->min_dcs->id);
5522 		pool->raw_hw = NULL;
5523 		goto set_alarm;
5524 	}
5525 	pool->raw_hw->min_dcs_id = dcs->id;
5526 	LIST_REMOVE(pool->raw_hw, next);
5527 	sh->cmng.pending_queries++;
5528 	pool_index++;
5529 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5530 		batch ^= 0x1;
5531 		pool_index = 0;
5532 	}
5533 set_alarm:
5534 	sh->cmng.batch = batch;
5535 	sh->cmng.pool_index = pool_index;
5536 	mlx5_set_query_alarm(sh);
5537 }
5538 
5539 /**
5540  * Handler for the HW respond about ready values from an asynchronous batch
5541  * query. This function is probably called by the host thread.
5542  *
5543  * @param[in] sh
5544  *   The pointer to the shared IB device context.
5545  * @param[in] async_id
5546  *   The Devx async ID.
5547  * @param[in] status
5548  *   The status of the completion.
5549  */
5550 void
5551 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5552 				  uint64_t async_id, int status)
5553 {
5554 	struct mlx5_flow_counter_pool *pool =
5555 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5556 	struct mlx5_counter_stats_raw *raw_to_free;
5557 
5558 	if (unlikely(status)) {
5559 		raw_to_free = pool->raw_hw;
5560 	} else {
5561 		raw_to_free = pool->raw;
5562 		rte_spinlock_lock(&pool->sl);
5563 		pool->raw = pool->raw_hw;
5564 		rte_spinlock_unlock(&pool->sl);
5565 		rte_atomic64_add(&pool->query_gen, 1);
5566 		/* Be sure the new raw counters data is updated in memory. */
5567 		rte_cio_wmb();
5568 	}
5569 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5570 	pool->raw_hw = NULL;
5571 	sh->cmng.pending_queries--;
5572 }
5573 
5574 /**
5575  * Translate the rte_flow group index to HW table value.
5576  *
5577  * @param[in] attributes
5578  *   Pointer to flow attributes
5579  * @param[in] external
5580  *   Value is part of flow rule created by request external to PMD.
5581  * @param[in] group
5582  *   rte_flow group index value.
5583  * @param[out] fdb_def_rule
5584  *   Whether fdb jump to table 1 is configured.
5585  * @param[out] table
5586  *   HW table value.
5587  * @param[out] error
5588  *   Pointer to error structure.
5589  *
5590  * @return
5591  *   0 on success, a negative errno value otherwise and rte_errno is set.
5592  */
5593 int
5594 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5595 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5596 			 struct rte_flow_error *error)
5597 {
5598 	if (attributes->transfer && external && fdb_def_rule) {
5599 		if (group == UINT32_MAX)
5600 			return rte_flow_error_set
5601 						(error, EINVAL,
5602 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5603 						 NULL,
5604 						 "group index not supported");
5605 		*table = group + 1;
5606 	} else {
5607 		*table = group;
5608 	}
5609 	return 0;
5610 }
5611 
5612 /**
5613  * Discover availability of metadata reg_c's.
5614  *
5615  * Iteratively use test flows to check availability.
5616  *
5617  * @param[in] dev
5618  *   Pointer to the Ethernet device structure.
5619  *
5620  * @return
5621  *   0 on success, a negative errno value otherwise and rte_errno is set.
5622  */
5623 int
5624 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5625 {
5626 	struct mlx5_priv *priv = dev->data->dev_private;
5627 	struct mlx5_dev_config *config = &priv->config;
5628 	enum modify_reg idx;
5629 	int n = 0;
5630 
5631 	/* reg_c[0] and reg_c[1] are reserved. */
5632 	config->flow_mreg_c[n++] = REG_C_0;
5633 	config->flow_mreg_c[n++] = REG_C_1;
5634 	/* Discover availability of other reg_c's. */
5635 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5636 		struct rte_flow_attr attr = {
5637 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5638 			.priority = MLX5_FLOW_PRIO_RSVD,
5639 			.ingress = 1,
5640 		};
5641 		struct rte_flow_item items[] = {
5642 			[0] = {
5643 				.type = RTE_FLOW_ITEM_TYPE_END,
5644 			},
5645 		};
5646 		struct rte_flow_action actions[] = {
5647 			[0] = {
5648 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5649 				.conf = &(struct mlx5_flow_action_copy_mreg){
5650 					.src = REG_C_1,
5651 					.dst = idx,
5652 				},
5653 			},
5654 			[1] = {
5655 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5656 				.conf = &(struct rte_flow_action_jump){
5657 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5658 				},
5659 			},
5660 			[2] = {
5661 				.type = RTE_FLOW_ACTION_TYPE_END,
5662 			},
5663 		};
5664 		struct rte_flow *flow;
5665 		struct rte_flow_error error;
5666 
5667 		if (!config->dv_flow_en)
5668 			break;
5669 		/* Create internal flow, validation skips copy action. */
5670 		flow = flow_list_create(dev, NULL, &attr, items,
5671 					actions, false, &error);
5672 		if (!flow)
5673 			continue;
5674 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5675 			config->flow_mreg_c[n++] = idx;
5676 		flow_list_destroy(dev, NULL, flow);
5677 	}
5678 	for (; n < MLX5_MREG_C_NUM; ++n)
5679 		config->flow_mreg_c[n] = REG_NONE;
5680 	return 0;
5681 }
5682 
5683 /**
5684  * Dump flow raw hw data to file
5685  *
5686  * @param[in] dev
5687  *    The pointer to Ethernet device.
5688  * @param[in] file
5689  *   A pointer to a file for output.
5690  * @param[out] error
5691  *   Perform verbose error reporting if not NULL. PMDs initialize this
5692  *   structure in case of error only.
5693  * @return
5694  *   0 on success, a nagative value otherwise.
5695  */
5696 int
5697 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5698 		   FILE *file,
5699 		   struct rte_flow_error *error __rte_unused)
5700 {
5701 	struct mlx5_priv *priv = dev->data->dev_private;
5702 	struct mlx5_ibv_shared *sh = priv->sh;
5703 
5704 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5705 				       sh->tx_domain, file);
5706 }
5707