xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 042f5c94fd3a49e4fe85e5ae2405a39ca65a2436)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 /* Verbs header. */
14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
15 #ifdef PEDANTIC
16 #pragma GCC diagnostic ignored "-Wpedantic"
17 #endif
18 #include <infiniband/verbs.h>
19 #ifdef PEDANTIC
20 #pragma GCC diagnostic error "-Wpedantic"
21 #endif
22 
23 #include <rte_common.h>
24 #include <rte_ether.h>
25 #include <rte_ethdev_driver.h>
26 #include <rte_flow.h>
27 #include <rte_cycles.h>
28 #include <rte_flow_driver.h>
29 #include <rte_malloc.h>
30 #include <rte_ip.h>
31 
32 #include <mlx5_glue.h>
33 #include <mlx5_devx_cmds.h>
34 #include <mlx5_prm.h>
35 
36 #include "mlx5_defs.h"
37 #include "mlx5.h"
38 #include "mlx5_flow.h"
39 #include "mlx5_rxtx.h"
40 
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
44 #endif
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
46 
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
48 
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
53 #endif
54 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
56 };
57 
58 enum mlx5_expansion {
59 	MLX5_EXPANSION_ROOT,
60 	MLX5_EXPANSION_ROOT_OUTER,
61 	MLX5_EXPANSION_ROOT_ETH_VLAN,
62 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 	MLX5_EXPANSION_OUTER_ETH,
64 	MLX5_EXPANSION_OUTER_ETH_VLAN,
65 	MLX5_EXPANSION_OUTER_VLAN,
66 	MLX5_EXPANSION_OUTER_IPV4,
67 	MLX5_EXPANSION_OUTER_IPV4_UDP,
68 	MLX5_EXPANSION_OUTER_IPV4_TCP,
69 	MLX5_EXPANSION_OUTER_IPV6,
70 	MLX5_EXPANSION_OUTER_IPV6_UDP,
71 	MLX5_EXPANSION_OUTER_IPV6_TCP,
72 	MLX5_EXPANSION_VXLAN,
73 	MLX5_EXPANSION_VXLAN_GPE,
74 	MLX5_EXPANSION_GRE,
75 	MLX5_EXPANSION_MPLS,
76 	MLX5_EXPANSION_ETH,
77 	MLX5_EXPANSION_ETH_VLAN,
78 	MLX5_EXPANSION_VLAN,
79 	MLX5_EXPANSION_IPV4,
80 	MLX5_EXPANSION_IPV4_UDP,
81 	MLX5_EXPANSION_IPV4_TCP,
82 	MLX5_EXPANSION_IPV6,
83 	MLX5_EXPANSION_IPV6_UDP,
84 	MLX5_EXPANSION_IPV6_TCP,
85 };
86 
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 	[MLX5_EXPANSION_ROOT] = {
90 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
91 						 MLX5_EXPANSION_IPV4,
92 						 MLX5_EXPANSION_IPV6),
93 		.type = RTE_FLOW_ITEM_TYPE_END,
94 	},
95 	[MLX5_EXPANSION_ROOT_OUTER] = {
96 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 						 MLX5_EXPANSION_OUTER_IPV4,
98 						 MLX5_EXPANSION_OUTER_IPV6),
99 		.type = RTE_FLOW_ITEM_TYPE_END,
100 	},
101 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 		.type = RTE_FLOW_ITEM_TYPE_END,
104 	},
105 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 		.type = RTE_FLOW_ITEM_TYPE_END,
108 	},
109 	[MLX5_EXPANSION_OUTER_ETH] = {
110 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 						 MLX5_EXPANSION_OUTER_IPV6,
112 						 MLX5_EXPANSION_MPLS),
113 		.type = RTE_FLOW_ITEM_TYPE_ETH,
114 		.rss_types = 0,
115 	},
116 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 		.type = RTE_FLOW_ITEM_TYPE_ETH,
119 		.rss_types = 0,
120 	},
121 	[MLX5_EXPANSION_OUTER_VLAN] = {
122 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 						 MLX5_EXPANSION_OUTER_IPV6),
124 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
125 	},
126 	[MLX5_EXPANSION_OUTER_IPV4] = {
127 		.next = RTE_FLOW_EXPAND_RSS_NEXT
128 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
129 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
130 			 MLX5_EXPANSION_GRE,
131 			 MLX5_EXPANSION_IPV4,
132 			 MLX5_EXPANSION_IPV6),
133 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
134 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135 			ETH_RSS_NONFRAG_IPV4_OTHER,
136 	},
137 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139 						 MLX5_EXPANSION_VXLAN_GPE),
140 		.type = RTE_FLOW_ITEM_TYPE_UDP,
141 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
142 	},
143 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144 		.type = RTE_FLOW_ITEM_TYPE_TCP,
145 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
146 	},
147 	[MLX5_EXPANSION_OUTER_IPV6] = {
148 		.next = RTE_FLOW_EXPAND_RSS_NEXT
149 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
150 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
151 			 MLX5_EXPANSION_IPV4,
152 			 MLX5_EXPANSION_IPV6),
153 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
154 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155 			ETH_RSS_NONFRAG_IPV6_OTHER,
156 	},
157 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159 						 MLX5_EXPANSION_VXLAN_GPE),
160 		.type = RTE_FLOW_ITEM_TYPE_UDP,
161 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
162 	},
163 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164 		.type = RTE_FLOW_ITEM_TYPE_TCP,
165 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
166 	},
167 	[MLX5_EXPANSION_VXLAN] = {
168 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
169 						 MLX5_EXPANSION_IPV4,
170 						 MLX5_EXPANSION_IPV6),
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 	.get_aged_flows = mlx5_flow_get_aged_flows,
243 };
244 
245 /* Convert FDIR request to Generic flow. */
246 struct mlx5_fdir {
247 	struct rte_flow_attr attr;
248 	struct rte_flow_item items[4];
249 	struct rte_flow_item_eth l2;
250 	struct rte_flow_item_eth l2_mask;
251 	union {
252 		struct rte_flow_item_ipv4 ipv4;
253 		struct rte_flow_item_ipv6 ipv6;
254 	} l3;
255 	union {
256 		struct rte_flow_item_ipv4 ipv4;
257 		struct rte_flow_item_ipv6 ipv6;
258 	} l3_mask;
259 	union {
260 		struct rte_flow_item_udp udp;
261 		struct rte_flow_item_tcp tcp;
262 	} l4;
263 	union {
264 		struct rte_flow_item_udp udp;
265 		struct rte_flow_item_tcp tcp;
266 	} l4_mask;
267 	struct rte_flow_action actions[2];
268 	struct rte_flow_action_queue queue;
269 };
270 
271 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
272 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
273 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
274 };
275 
276 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
277 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
278 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
279 	{ 9, 10, 11 }, { 12, 13, 14 },
280 };
281 
282 /* Tunnel information. */
283 struct mlx5_flow_tunnel_info {
284 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
285 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
286 };
287 
288 static struct mlx5_flow_tunnel_info tunnels_info[] = {
289 	{
290 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
291 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
292 	},
293 	{
294 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
295 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
296 	},
297 	{
298 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
299 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
300 	},
301 	{
302 		.tunnel = MLX5_FLOW_LAYER_GRE,
303 		.ptype = RTE_PTYPE_TUNNEL_GRE,
304 	},
305 	{
306 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
307 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
308 	},
309 	{
310 		.tunnel = MLX5_FLOW_LAYER_MPLS,
311 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
312 	},
313 	{
314 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
315 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
316 	},
317 	{
318 		.tunnel = MLX5_FLOW_LAYER_IPIP,
319 		.ptype = RTE_PTYPE_TUNNEL_IP,
320 	},
321 	{
322 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
323 		.ptype = RTE_PTYPE_TUNNEL_IP,
324 	},
325 	{
326 		.tunnel = MLX5_FLOW_LAYER_GTP,
327 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
328 	},
329 };
330 
331 /**
332  * Translate tag ID to register.
333  *
334  * @param[in] dev
335  *   Pointer to the Ethernet device structure.
336  * @param[in] feature
337  *   The feature that request the register.
338  * @param[in] id
339  *   The request register ID.
340  * @param[out] error
341  *   Error description in case of any.
342  *
343  * @return
344  *   The request register on success, a negative errno
345  *   value otherwise and rte_errno is set.
346  */
347 int
348 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
349 		     enum mlx5_feature_name feature,
350 		     uint32_t id,
351 		     struct rte_flow_error *error)
352 {
353 	struct mlx5_priv *priv = dev->data->dev_private;
354 	struct mlx5_dev_config *config = &priv->config;
355 	enum modify_reg start_reg;
356 	bool skip_mtr_reg = false;
357 
358 	switch (feature) {
359 	case MLX5_HAIRPIN_RX:
360 		return REG_B;
361 	case MLX5_HAIRPIN_TX:
362 		return REG_A;
363 	case MLX5_METADATA_RX:
364 		switch (config->dv_xmeta_en) {
365 		case MLX5_XMETA_MODE_LEGACY:
366 			return REG_B;
367 		case MLX5_XMETA_MODE_META16:
368 			return REG_C_0;
369 		case MLX5_XMETA_MODE_META32:
370 			return REG_C_1;
371 		}
372 		break;
373 	case MLX5_METADATA_TX:
374 		return REG_A;
375 	case MLX5_METADATA_FDB:
376 		switch (config->dv_xmeta_en) {
377 		case MLX5_XMETA_MODE_LEGACY:
378 			return REG_NONE;
379 		case MLX5_XMETA_MODE_META16:
380 			return REG_C_0;
381 		case MLX5_XMETA_MODE_META32:
382 			return REG_C_1;
383 		}
384 		break;
385 	case MLX5_FLOW_MARK:
386 		switch (config->dv_xmeta_en) {
387 		case MLX5_XMETA_MODE_LEGACY:
388 			return REG_NONE;
389 		case MLX5_XMETA_MODE_META16:
390 			return REG_C_1;
391 		case MLX5_XMETA_MODE_META32:
392 			return REG_C_0;
393 		}
394 		break;
395 	case MLX5_MTR_SFX:
396 		/*
397 		 * If meter color and flow match share one register, flow match
398 		 * should use the meter color register for match.
399 		 */
400 		if (priv->mtr_reg_share)
401 			return priv->mtr_color_reg;
402 		else
403 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
404 			       REG_C_3;
405 	case MLX5_MTR_COLOR:
406 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
407 		return priv->mtr_color_reg;
408 	case MLX5_COPY_MARK:
409 		/*
410 		 * Metadata COPY_MARK register using is in meter suffix sub
411 		 * flow while with meter. It's safe to share the same register.
412 		 */
413 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
414 	case MLX5_APP_TAG:
415 		/*
416 		 * If meter is enable, it will engage the register for color
417 		 * match and flow match. If meter color match is not using the
418 		 * REG_C_2, need to skip the REG_C_x be used by meter color
419 		 * match.
420 		 * If meter is disable, free to use all available registers.
421 		 */
422 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
423 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
424 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
425 		if (id > (REG_C_7 - start_reg))
426 			return rte_flow_error_set(error, EINVAL,
427 						  RTE_FLOW_ERROR_TYPE_ITEM,
428 						  NULL, "invalid tag id");
429 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
430 			return rte_flow_error_set(error, ENOTSUP,
431 						  RTE_FLOW_ERROR_TYPE_ITEM,
432 						  NULL, "unsupported tag id");
433 		/*
434 		 * This case means meter is using the REG_C_x great than 2.
435 		 * Take care not to conflict with meter color REG_C_x.
436 		 * If the available index REG_C_y >= REG_C_x, skip the
437 		 * color register.
438 		 */
439 		if (skip_mtr_reg && config->flow_mreg_c
440 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
441 			if (id >= (REG_C_7 - start_reg))
442 				return rte_flow_error_set(error, EINVAL,
443 						       RTE_FLOW_ERROR_TYPE_ITEM,
444 							NULL, "invalid tag id");
445 			if (config->flow_mreg_c
446 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
447 				return config->flow_mreg_c
448 					       [id + 1 + start_reg - REG_C_0];
449 			return rte_flow_error_set(error, ENOTSUP,
450 						  RTE_FLOW_ERROR_TYPE_ITEM,
451 						  NULL, "unsupported tag id");
452 		}
453 		return config->flow_mreg_c[id + start_reg - REG_C_0];
454 	}
455 	MLX5_ASSERT(false);
456 	return rte_flow_error_set(error, EINVAL,
457 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
458 				  NULL, "invalid feature name");
459 }
460 
461 /**
462  * Check extensive flow metadata register support.
463  *
464  * @param dev
465  *   Pointer to rte_eth_dev structure.
466  *
467  * @return
468  *   True if device supports extensive flow metadata register, otherwise false.
469  */
470 bool
471 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
472 {
473 	struct mlx5_priv *priv = dev->data->dev_private;
474 	struct mlx5_dev_config *config = &priv->config;
475 
476 	/*
477 	 * Having available reg_c can be regarded inclusively as supporting
478 	 * extensive flow metadata register, which could mean,
479 	 * - metadata register copy action by modify header.
480 	 * - 16 modify header actions is supported.
481 	 * - reg_c's are preserved across different domain (FDB and NIC) on
482 	 *   packet loopback by flow lookup miss.
483 	 */
484 	return config->flow_mreg_c[2] != REG_NONE;
485 }
486 
487 /**
488  * Discover the maximum number of priority available.
489  *
490  * @param[in] dev
491  *   Pointer to the Ethernet device structure.
492  *
493  * @return
494  *   number of supported flow priority on success, a negative errno
495  *   value otherwise and rte_errno is set.
496  */
497 int
498 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
499 {
500 	struct mlx5_priv *priv = dev->data->dev_private;
501 	struct {
502 		struct ibv_flow_attr attr;
503 		struct ibv_flow_spec_eth eth;
504 		struct ibv_flow_spec_action_drop drop;
505 	} flow_attr = {
506 		.attr = {
507 			.num_of_specs = 2,
508 			.port = (uint8_t)priv->dev_port,
509 		},
510 		.eth = {
511 			.type = IBV_FLOW_SPEC_ETH,
512 			.size = sizeof(struct ibv_flow_spec_eth),
513 		},
514 		.drop = {
515 			.size = sizeof(struct ibv_flow_spec_action_drop),
516 			.type = IBV_FLOW_SPEC_ACTION_DROP,
517 		},
518 	};
519 	struct ibv_flow *flow;
520 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
521 	uint16_t vprio[] = { 8, 16 };
522 	int i;
523 	int priority = 0;
524 
525 	if (!drop) {
526 		rte_errno = ENOTSUP;
527 		return -rte_errno;
528 	}
529 	for (i = 0; i != RTE_DIM(vprio); i++) {
530 		flow_attr.attr.priority = vprio[i] - 1;
531 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
532 		if (!flow)
533 			break;
534 		claim_zero(mlx5_glue->destroy_flow(flow));
535 		priority = vprio[i];
536 	}
537 	mlx5_hrxq_drop_release(dev);
538 	switch (priority) {
539 	case 8:
540 		priority = RTE_DIM(priority_map_3);
541 		break;
542 	case 16:
543 		priority = RTE_DIM(priority_map_5);
544 		break;
545 	default:
546 		rte_errno = ENOTSUP;
547 		DRV_LOG(ERR,
548 			"port %u verbs maximum priority: %d expected 8/16",
549 			dev->data->port_id, priority);
550 		return -rte_errno;
551 	}
552 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
553 		dev->data->port_id, priority);
554 	return priority;
555 }
556 
557 /**
558  * Adjust flow priority based on the highest layer and the request priority.
559  *
560  * @param[in] dev
561  *   Pointer to the Ethernet device structure.
562  * @param[in] priority
563  *   The rule base priority.
564  * @param[in] subpriority
565  *   The priority based on the items.
566  *
567  * @return
568  *   The new priority.
569  */
570 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
571 				   uint32_t subpriority)
572 {
573 	uint32_t res = 0;
574 	struct mlx5_priv *priv = dev->data->dev_private;
575 
576 	switch (priv->config.flow_prio) {
577 	case RTE_DIM(priority_map_3):
578 		res = priority_map_3[priority][subpriority];
579 		break;
580 	case RTE_DIM(priority_map_5):
581 		res = priority_map_5[priority][subpriority];
582 		break;
583 	}
584 	return  res;
585 }
586 
587 /**
588  * Verify the @p item specifications (spec, last, mask) are compatible with the
589  * NIC capabilities.
590  *
591  * @param[in] item
592  *   Item specification.
593  * @param[in] mask
594  *   @p item->mask or flow default bit-masks.
595  * @param[in] nic_mask
596  *   Bit-masks covering supported fields by the NIC to compare with user mask.
597  * @param[in] size
598  *   Bit-masks size in bytes.
599  * @param[out] error
600  *   Pointer to error structure.
601  *
602  * @return
603  *   0 on success, a negative errno value otherwise and rte_errno is set.
604  */
605 int
606 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
607 			  const uint8_t *mask,
608 			  const uint8_t *nic_mask,
609 			  unsigned int size,
610 			  struct rte_flow_error *error)
611 {
612 	unsigned int i;
613 
614 	MLX5_ASSERT(nic_mask);
615 	for (i = 0; i < size; ++i)
616 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
617 			return rte_flow_error_set(error, ENOTSUP,
618 						  RTE_FLOW_ERROR_TYPE_ITEM,
619 						  item,
620 						  "mask enables non supported"
621 						  " bits");
622 	if (!item->spec && (item->mask || item->last))
623 		return rte_flow_error_set(error, EINVAL,
624 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
625 					  "mask/last without a spec is not"
626 					  " supported");
627 	if (item->spec && item->last) {
628 		uint8_t spec[size];
629 		uint8_t last[size];
630 		unsigned int i;
631 		int ret;
632 
633 		for (i = 0; i < size; ++i) {
634 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
635 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
636 		}
637 		ret = memcmp(spec, last, size);
638 		if (ret != 0)
639 			return rte_flow_error_set(error, EINVAL,
640 						  RTE_FLOW_ERROR_TYPE_ITEM,
641 						  item,
642 						  "range is not valid");
643 	}
644 	return 0;
645 }
646 
647 /**
648  * Adjust the hash fields according to the @p flow information.
649  *
650  * @param[in] dev_flow.
651  *   Pointer to the mlx5_flow.
652  * @param[in] tunnel
653  *   1 when the hash field is for a tunnel item.
654  * @param[in] layer_types
655  *   ETH_RSS_* types.
656  * @param[in] hash_fields
657  *   Item hash fields.
658  *
659  * @return
660  *   The hash fields that should be used.
661  */
662 uint64_t
663 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
664 			    int tunnel __rte_unused, uint64_t layer_types,
665 			    uint64_t hash_fields)
666 {
667 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
668 	int rss_request_inner = rss_desc->level >= 2;
669 
670 	/* Check RSS hash level for tunnel. */
671 	if (tunnel && rss_request_inner)
672 		hash_fields |= IBV_RX_HASH_INNER;
673 	else if (tunnel || rss_request_inner)
674 		return 0;
675 #endif
676 	/* Check if requested layer matches RSS hash fields. */
677 	if (!(rss_desc->types & layer_types))
678 		return 0;
679 	return hash_fields;
680 }
681 
682 /**
683  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
684  * if several tunnel rules are used on this queue, the tunnel ptype will be
685  * cleared.
686  *
687  * @param rxq_ctrl
688  *   Rx queue to update.
689  */
690 static void
691 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
692 {
693 	unsigned int i;
694 	uint32_t tunnel_ptype = 0;
695 
696 	/* Look up for the ptype to use. */
697 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
698 		if (!rxq_ctrl->flow_tunnels_n[i])
699 			continue;
700 		if (!tunnel_ptype) {
701 			tunnel_ptype = tunnels_info[i].ptype;
702 		} else {
703 			tunnel_ptype = 0;
704 			break;
705 		}
706 	}
707 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
708 }
709 
710 /**
711  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
712  * flow.
713  *
714  * @param[in] dev
715  *   Pointer to the Ethernet device structure.
716  * @param[in] dev_handle
717  *   Pointer to device flow handle structure.
718  */
719 static void
720 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
721 		       struct mlx5_flow_handle *dev_handle)
722 {
723 	struct mlx5_priv *priv = dev->data->dev_private;
724 	const int mark = dev_handle->mark;
725 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
726 	struct mlx5_hrxq *hrxq;
727 	unsigned int i;
728 
729 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
730 		return;
731 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
732 			      dev_handle->rix_hrxq);
733 	if (!hrxq)
734 		return;
735 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
736 		int idx = hrxq->ind_table->queues[i];
737 		struct mlx5_rxq_ctrl *rxq_ctrl =
738 			container_of((*priv->rxqs)[idx],
739 				     struct mlx5_rxq_ctrl, rxq);
740 
741 		/*
742 		 * To support metadata register copy on Tx loopback,
743 		 * this must be always enabled (metadata may arive
744 		 * from other port - not from local flows only.
745 		 */
746 		if (priv->config.dv_flow_en &&
747 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
748 		    mlx5_flow_ext_mreg_supported(dev)) {
749 			rxq_ctrl->rxq.mark = 1;
750 			rxq_ctrl->flow_mark_n = 1;
751 		} else if (mark) {
752 			rxq_ctrl->rxq.mark = 1;
753 			rxq_ctrl->flow_mark_n++;
754 		}
755 		if (tunnel) {
756 			unsigned int j;
757 
758 			/* Increase the counter matching the flow. */
759 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
760 				if ((tunnels_info[j].tunnel &
761 				     dev_handle->layers) ==
762 				    tunnels_info[j].tunnel) {
763 					rxq_ctrl->flow_tunnels_n[j]++;
764 					break;
765 				}
766 			}
767 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
768 		}
769 	}
770 }
771 
772 /**
773  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
774  *
775  * @param[in] dev
776  *   Pointer to the Ethernet device structure.
777  * @param[in] flow
778  *   Pointer to flow structure.
779  */
780 static void
781 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
782 {
783 	struct mlx5_priv *priv = dev->data->dev_private;
784 	uint32_t handle_idx;
785 	struct mlx5_flow_handle *dev_handle;
786 
787 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
788 		       handle_idx, dev_handle, next)
789 		flow_drv_rxq_flags_set(dev, dev_handle);
790 }
791 
792 /**
793  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
794  * device flow if no other flow uses it with the same kind of request.
795  *
796  * @param dev
797  *   Pointer to Ethernet device.
798  * @param[in] dev_handle
799  *   Pointer to the device flow handle structure.
800  */
801 static void
802 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
803 			struct mlx5_flow_handle *dev_handle)
804 {
805 	struct mlx5_priv *priv = dev->data->dev_private;
806 	const int mark = dev_handle->mark;
807 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
808 	struct mlx5_hrxq *hrxq;
809 	unsigned int i;
810 
811 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
812 		return;
813 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
814 			      dev_handle->rix_hrxq);
815 	if (!hrxq)
816 		return;
817 	MLX5_ASSERT(dev->data->dev_started);
818 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
819 		int idx = hrxq->ind_table->queues[i];
820 		struct mlx5_rxq_ctrl *rxq_ctrl =
821 			container_of((*priv->rxqs)[idx],
822 				     struct mlx5_rxq_ctrl, rxq);
823 
824 		if (priv->config.dv_flow_en &&
825 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
826 		    mlx5_flow_ext_mreg_supported(dev)) {
827 			rxq_ctrl->rxq.mark = 1;
828 			rxq_ctrl->flow_mark_n = 1;
829 		} else if (mark) {
830 			rxq_ctrl->flow_mark_n--;
831 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
832 		}
833 		if (tunnel) {
834 			unsigned int j;
835 
836 			/* Decrease the counter matching the flow. */
837 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
838 				if ((tunnels_info[j].tunnel &
839 				     dev_handle->layers) ==
840 				    tunnels_info[j].tunnel) {
841 					rxq_ctrl->flow_tunnels_n[j]--;
842 					break;
843 				}
844 			}
845 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
846 		}
847 	}
848 }
849 
850 /**
851  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
852  * @p flow if no other flow uses it with the same kind of request.
853  *
854  * @param dev
855  *   Pointer to Ethernet device.
856  * @param[in] flow
857  *   Pointer to the flow.
858  */
859 static void
860 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
861 {
862 	struct mlx5_priv *priv = dev->data->dev_private;
863 	uint32_t handle_idx;
864 	struct mlx5_flow_handle *dev_handle;
865 
866 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
867 		       handle_idx, dev_handle, next)
868 		flow_drv_rxq_flags_trim(dev, dev_handle);
869 }
870 
871 /**
872  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
873  *
874  * @param dev
875  *   Pointer to Ethernet device.
876  */
877 static void
878 flow_rxq_flags_clear(struct rte_eth_dev *dev)
879 {
880 	struct mlx5_priv *priv = dev->data->dev_private;
881 	unsigned int i;
882 
883 	for (i = 0; i != priv->rxqs_n; ++i) {
884 		struct mlx5_rxq_ctrl *rxq_ctrl;
885 		unsigned int j;
886 
887 		if (!(*priv->rxqs)[i])
888 			continue;
889 		rxq_ctrl = container_of((*priv->rxqs)[i],
890 					struct mlx5_rxq_ctrl, rxq);
891 		rxq_ctrl->flow_mark_n = 0;
892 		rxq_ctrl->rxq.mark = 0;
893 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
894 			rxq_ctrl->flow_tunnels_n[j] = 0;
895 		rxq_ctrl->rxq.tunnel = 0;
896 	}
897 }
898 
899 /**
900  * Set the Rx queue dynamic metadata (mask and offset) for a flow
901  *
902  * @param[in] dev
903  *   Pointer to the Ethernet device structure.
904  */
905 void
906 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
907 {
908 	struct mlx5_priv *priv = dev->data->dev_private;
909 	struct mlx5_rxq_data *data;
910 	unsigned int i;
911 
912 	for (i = 0; i != priv->rxqs_n; ++i) {
913 		if (!(*priv->rxqs)[i])
914 			continue;
915 		data = (*priv->rxqs)[i];
916 		if (!rte_flow_dynf_metadata_avail()) {
917 			data->dynf_meta = 0;
918 			data->flow_meta_mask = 0;
919 			data->flow_meta_offset = -1;
920 		} else {
921 			data->dynf_meta = 1;
922 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
923 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
924 		}
925 	}
926 }
927 
928 /*
929  * return a pointer to the desired action in the list of actions.
930  *
931  * @param[in] actions
932  *   The list of actions to search the action in.
933  * @param[in] action
934  *   The action to find.
935  *
936  * @return
937  *   Pointer to the action in the list, if found. NULL otherwise.
938  */
939 const struct rte_flow_action *
940 mlx5_flow_find_action(const struct rte_flow_action *actions,
941 		      enum rte_flow_action_type action)
942 {
943 	if (actions == NULL)
944 		return NULL;
945 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
946 		if (actions->type == action)
947 			return actions;
948 	return NULL;
949 }
950 
951 /*
952  * Validate the flag action.
953  *
954  * @param[in] action_flags
955  *   Bit-fields that holds the actions detected until now.
956  * @param[in] attr
957  *   Attributes of flow that includes this action.
958  * @param[out] error
959  *   Pointer to error structure.
960  *
961  * @return
962  *   0 on success, a negative errno value otherwise and rte_errno is set.
963  */
964 int
965 mlx5_flow_validate_action_flag(uint64_t action_flags,
966 			       const struct rte_flow_attr *attr,
967 			       struct rte_flow_error *error)
968 {
969 	if (action_flags & MLX5_FLOW_ACTION_MARK)
970 		return rte_flow_error_set(error, EINVAL,
971 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
972 					  "can't mark and flag in same flow");
973 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
974 		return rte_flow_error_set(error, EINVAL,
975 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 					  "can't have 2 flag"
977 					  " actions in same flow");
978 	if (attr->egress)
979 		return rte_flow_error_set(error, ENOTSUP,
980 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
981 					  "flag action not supported for "
982 					  "egress");
983 	return 0;
984 }
985 
986 /*
987  * Validate the mark action.
988  *
989  * @param[in] action
990  *   Pointer to the queue action.
991  * @param[in] action_flags
992  *   Bit-fields that holds the actions detected until now.
993  * @param[in] attr
994  *   Attributes of flow that includes this action.
995  * @param[out] error
996  *   Pointer to error structure.
997  *
998  * @return
999  *   0 on success, a negative errno value otherwise and rte_errno is set.
1000  */
1001 int
1002 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1003 			       uint64_t action_flags,
1004 			       const struct rte_flow_attr *attr,
1005 			       struct rte_flow_error *error)
1006 {
1007 	const struct rte_flow_action_mark *mark = action->conf;
1008 
1009 	if (!mark)
1010 		return rte_flow_error_set(error, EINVAL,
1011 					  RTE_FLOW_ERROR_TYPE_ACTION,
1012 					  action,
1013 					  "configuration cannot be null");
1014 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1015 		return rte_flow_error_set(error, EINVAL,
1016 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1017 					  &mark->id,
1018 					  "mark id must in 0 <= id < "
1019 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1020 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1021 		return rte_flow_error_set(error, EINVAL,
1022 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1023 					  "can't flag and mark in same flow");
1024 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1025 		return rte_flow_error_set(error, EINVAL,
1026 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1027 					  "can't have 2 mark actions in same"
1028 					  " flow");
1029 	if (attr->egress)
1030 		return rte_flow_error_set(error, ENOTSUP,
1031 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1032 					  "mark action not supported for "
1033 					  "egress");
1034 	return 0;
1035 }
1036 
1037 /*
1038  * Validate the drop action.
1039  *
1040  * @param[in] action_flags
1041  *   Bit-fields that holds the actions detected until now.
1042  * @param[in] attr
1043  *   Attributes of flow that includes this action.
1044  * @param[out] error
1045  *   Pointer to error structure.
1046  *
1047  * @return
1048  *   0 on success, a negative errno value otherwise and rte_errno is set.
1049  */
1050 int
1051 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1052 			       const struct rte_flow_attr *attr,
1053 			       struct rte_flow_error *error)
1054 {
1055 	if (attr->egress)
1056 		return rte_flow_error_set(error, ENOTSUP,
1057 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1058 					  "drop action not supported for "
1059 					  "egress");
1060 	return 0;
1061 }
1062 
1063 /*
1064  * Validate the queue action.
1065  *
1066  * @param[in] action
1067  *   Pointer to the queue action.
1068  * @param[in] action_flags
1069  *   Bit-fields that holds the actions detected until now.
1070  * @param[in] dev
1071  *   Pointer to the Ethernet device structure.
1072  * @param[in] attr
1073  *   Attributes of flow that includes this action.
1074  * @param[out] error
1075  *   Pointer to error structure.
1076  *
1077  * @return
1078  *   0 on success, a negative errno value otherwise and rte_errno is set.
1079  */
1080 int
1081 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1082 				uint64_t action_flags,
1083 				struct rte_eth_dev *dev,
1084 				const struct rte_flow_attr *attr,
1085 				struct rte_flow_error *error)
1086 {
1087 	struct mlx5_priv *priv = dev->data->dev_private;
1088 	const struct rte_flow_action_queue *queue = action->conf;
1089 
1090 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1091 		return rte_flow_error_set(error, EINVAL,
1092 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1093 					  "can't have 2 fate actions in"
1094 					  " same flow");
1095 	if (!priv->rxqs_n)
1096 		return rte_flow_error_set(error, EINVAL,
1097 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1098 					  NULL, "No Rx queues configured");
1099 	if (queue->index >= priv->rxqs_n)
1100 		return rte_flow_error_set(error, EINVAL,
1101 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1102 					  &queue->index,
1103 					  "queue index out of range");
1104 	if (!(*priv->rxqs)[queue->index])
1105 		return rte_flow_error_set(error, EINVAL,
1106 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1107 					  &queue->index,
1108 					  "queue is not configured");
1109 	if (attr->egress)
1110 		return rte_flow_error_set(error, ENOTSUP,
1111 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1112 					  "queue action not supported for "
1113 					  "egress");
1114 	return 0;
1115 }
1116 
1117 /*
1118  * Validate the rss action.
1119  *
1120  * @param[in] action
1121  *   Pointer to the queue action.
1122  * @param[in] action_flags
1123  *   Bit-fields that holds the actions detected until now.
1124  * @param[in] dev
1125  *   Pointer to the Ethernet device structure.
1126  * @param[in] attr
1127  *   Attributes of flow that includes this action.
1128  * @param[in] item_flags
1129  *   Items that were detected.
1130  * @param[out] error
1131  *   Pointer to error structure.
1132  *
1133  * @return
1134  *   0 on success, a negative errno value otherwise and rte_errno is set.
1135  */
1136 int
1137 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1138 			      uint64_t action_flags,
1139 			      struct rte_eth_dev *dev,
1140 			      const struct rte_flow_attr *attr,
1141 			      uint64_t item_flags,
1142 			      struct rte_flow_error *error)
1143 {
1144 	struct mlx5_priv *priv = dev->data->dev_private;
1145 	const struct rte_flow_action_rss *rss = action->conf;
1146 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1147 	unsigned int i;
1148 
1149 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1152 					  "can't have 2 fate actions"
1153 					  " in same flow");
1154 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1155 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1156 		return rte_flow_error_set(error, ENOTSUP,
1157 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1158 					  &rss->func,
1159 					  "RSS hash function not supported");
1160 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1161 	if (rss->level > 2)
1162 #else
1163 	if (rss->level > 1)
1164 #endif
1165 		return rte_flow_error_set(error, ENOTSUP,
1166 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1167 					  &rss->level,
1168 					  "tunnel RSS is not supported");
1169 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1170 	if (rss->key_len == 0 && rss->key != NULL)
1171 		return rte_flow_error_set(error, ENOTSUP,
1172 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 					  &rss->key_len,
1174 					  "RSS hash key length 0");
1175 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1176 		return rte_flow_error_set(error, ENOTSUP,
1177 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1178 					  &rss->key_len,
1179 					  "RSS hash key too small");
1180 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1181 		return rte_flow_error_set(error, ENOTSUP,
1182 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183 					  &rss->key_len,
1184 					  "RSS hash key too large");
1185 	if (rss->queue_num > priv->config.ind_table_max_size)
1186 		return rte_flow_error_set(error, ENOTSUP,
1187 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1188 					  &rss->queue_num,
1189 					  "number of queues too large");
1190 	if (rss->types & MLX5_RSS_HF_MASK)
1191 		return rte_flow_error_set(error, ENOTSUP,
1192 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1193 					  &rss->types,
1194 					  "some RSS protocols are not"
1195 					  " supported");
1196 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1197 	    !(rss->types & ETH_RSS_IP))
1198 		return rte_flow_error_set(error, EINVAL,
1199 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1200 					  "L3 partial RSS requested but L3 RSS"
1201 					  " type not specified");
1202 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1203 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1204 		return rte_flow_error_set(error, EINVAL,
1205 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1206 					  "L4 partial RSS requested but L4 RSS"
1207 					  " type not specified");
1208 	if (!priv->rxqs_n)
1209 		return rte_flow_error_set(error, EINVAL,
1210 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1211 					  NULL, "No Rx queues configured");
1212 	if (!rss->queue_num)
1213 		return rte_flow_error_set(error, EINVAL,
1214 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1215 					  NULL, "No queues configured");
1216 	for (i = 0; i != rss->queue_num; ++i) {
1217 		if (rss->queue[i] >= priv->rxqs_n)
1218 			return rte_flow_error_set
1219 				(error, EINVAL,
1220 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1221 				 &rss->queue[i], "queue index out of range");
1222 		if (!(*priv->rxqs)[rss->queue[i]])
1223 			return rte_flow_error_set
1224 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1225 				 &rss->queue[i], "queue is not configured");
1226 	}
1227 	if (attr->egress)
1228 		return rte_flow_error_set(error, ENOTSUP,
1229 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230 					  "rss action not supported for "
1231 					  "egress");
1232 	if (rss->level > 1 &&  !tunnel)
1233 		return rte_flow_error_set(error, EINVAL,
1234 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1235 					  "inner RSS is not supported for "
1236 					  "non-tunnel flows");
1237 	return 0;
1238 }
1239 
1240 /*
1241  * Validate the count action.
1242  *
1243  * @param[in] dev
1244  *   Pointer to the Ethernet device structure.
1245  * @param[in] attr
1246  *   Attributes of flow that includes this action.
1247  * @param[out] error
1248  *   Pointer to error structure.
1249  *
1250  * @return
1251  *   0 on success, a negative errno value otherwise and rte_errno is set.
1252  */
1253 int
1254 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1255 				const struct rte_flow_attr *attr,
1256 				struct rte_flow_error *error)
1257 {
1258 	if (attr->egress)
1259 		return rte_flow_error_set(error, ENOTSUP,
1260 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1261 					  "count action not supported for "
1262 					  "egress");
1263 	return 0;
1264 }
1265 
1266 /**
1267  * Verify the @p attributes will be correctly understood by the NIC and store
1268  * them in the @p flow if everything is correct.
1269  *
1270  * @param[in] dev
1271  *   Pointer to the Ethernet device structure.
1272  * @param[in] attributes
1273  *   Pointer to flow attributes
1274  * @param[out] error
1275  *   Pointer to error structure.
1276  *
1277  * @return
1278  *   0 on success, a negative errno value otherwise and rte_errno is set.
1279  */
1280 int
1281 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1282 			      const struct rte_flow_attr *attributes,
1283 			      struct rte_flow_error *error)
1284 {
1285 	struct mlx5_priv *priv = dev->data->dev_private;
1286 	uint32_t priority_max = priv->config.flow_prio - 1;
1287 
1288 	if (attributes->group)
1289 		return rte_flow_error_set(error, ENOTSUP,
1290 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1291 					  NULL, "groups is not supported");
1292 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1293 	    attributes->priority >= priority_max)
1294 		return rte_flow_error_set(error, ENOTSUP,
1295 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1296 					  NULL, "priority out of range");
1297 	if (attributes->egress)
1298 		return rte_flow_error_set(error, ENOTSUP,
1299 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1300 					  "egress is not supported");
1301 	if (attributes->transfer && !priv->config.dv_esw_en)
1302 		return rte_flow_error_set(error, ENOTSUP,
1303 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1304 					  NULL, "transfer is not supported");
1305 	if (!attributes->ingress)
1306 		return rte_flow_error_set(error, EINVAL,
1307 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1308 					  NULL,
1309 					  "ingress attribute is mandatory");
1310 	return 0;
1311 }
1312 
1313 /**
1314  * Validate ICMP6 item.
1315  *
1316  * @param[in] item
1317  *   Item specification.
1318  * @param[in] item_flags
1319  *   Bit-fields that holds the items detected until now.
1320  * @param[out] error
1321  *   Pointer to error structure.
1322  *
1323  * @return
1324  *   0 on success, a negative errno value otherwise and rte_errno is set.
1325  */
1326 int
1327 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1328 			       uint64_t item_flags,
1329 			       uint8_t target_protocol,
1330 			       struct rte_flow_error *error)
1331 {
1332 	const struct rte_flow_item_icmp6 *mask = item->mask;
1333 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1334 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1335 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1336 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1337 				      MLX5_FLOW_LAYER_OUTER_L4;
1338 	int ret;
1339 
1340 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1341 		return rte_flow_error_set(error, EINVAL,
1342 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1343 					  "protocol filtering not compatible"
1344 					  " with ICMP6 layer");
1345 	if (!(item_flags & l3m))
1346 		return rte_flow_error_set(error, EINVAL,
1347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1348 					  "IPv6 is mandatory to filter on"
1349 					  " ICMP6");
1350 	if (item_flags & l4m)
1351 		return rte_flow_error_set(error, EINVAL,
1352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1353 					  "multiple L4 layers not supported");
1354 	if (!mask)
1355 		mask = &rte_flow_item_icmp6_mask;
1356 	ret = mlx5_flow_item_acceptable
1357 		(item, (const uint8_t *)mask,
1358 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1359 		 sizeof(struct rte_flow_item_icmp6), error);
1360 	if (ret < 0)
1361 		return ret;
1362 	return 0;
1363 }
1364 
1365 /**
1366  * Validate ICMP item.
1367  *
1368  * @param[in] item
1369  *   Item specification.
1370  * @param[in] item_flags
1371  *   Bit-fields that holds the items detected until now.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1380 			     uint64_t item_flags,
1381 			     uint8_t target_protocol,
1382 			     struct rte_flow_error *error)
1383 {
1384 	const struct rte_flow_item_icmp *mask = item->mask;
1385 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1386 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1387 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1388 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1389 				      MLX5_FLOW_LAYER_OUTER_L4;
1390 	int ret;
1391 
1392 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1393 		return rte_flow_error_set(error, EINVAL,
1394 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1395 					  "protocol filtering not compatible"
1396 					  " with ICMP layer");
1397 	if (!(item_flags & l3m))
1398 		return rte_flow_error_set(error, EINVAL,
1399 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1400 					  "IPv4 is mandatory to filter"
1401 					  " on ICMP");
1402 	if (item_flags & l4m)
1403 		return rte_flow_error_set(error, EINVAL,
1404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 					  "multiple L4 layers not supported");
1406 	if (!mask)
1407 		mask = &rte_flow_item_icmp_mask;
1408 	ret = mlx5_flow_item_acceptable
1409 		(item, (const uint8_t *)mask,
1410 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1411 		 sizeof(struct rte_flow_item_icmp), error);
1412 	if (ret < 0)
1413 		return ret;
1414 	return 0;
1415 }
1416 
1417 /**
1418  * Validate Ethernet item.
1419  *
1420  * @param[in] item
1421  *   Item specification.
1422  * @param[in] item_flags
1423  *   Bit-fields that holds the items detected until now.
1424  * @param[out] error
1425  *   Pointer to error structure.
1426  *
1427  * @return
1428  *   0 on success, a negative errno value otherwise and rte_errno is set.
1429  */
1430 int
1431 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1432 			    uint64_t item_flags,
1433 			    struct rte_flow_error *error)
1434 {
1435 	const struct rte_flow_item_eth *mask = item->mask;
1436 	const struct rte_flow_item_eth nic_mask = {
1437 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1438 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1439 		.type = RTE_BE16(0xffff),
1440 	};
1441 	int ret;
1442 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1443 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1444 				       MLX5_FLOW_LAYER_OUTER_L2;
1445 
1446 	if (item_flags & ethm)
1447 		return rte_flow_error_set(error, ENOTSUP,
1448 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1449 					  "multiple L2 layers not supported");
1450 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1451 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1452 		return rte_flow_error_set(error, EINVAL,
1453 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1454 					  "L2 layer should not follow "
1455 					  "L3 layers");
1456 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1457 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1458 		return rte_flow_error_set(error, EINVAL,
1459 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1460 					  "L2 layer should not follow VLAN");
1461 	if (!mask)
1462 		mask = &rte_flow_item_eth_mask;
1463 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1464 					(const uint8_t *)&nic_mask,
1465 					sizeof(struct rte_flow_item_eth),
1466 					error);
1467 	return ret;
1468 }
1469 
1470 /**
1471  * Validate VLAN item.
1472  *
1473  * @param[in] item
1474  *   Item specification.
1475  * @param[in] item_flags
1476  *   Bit-fields that holds the items detected until now.
1477  * @param[in] dev
1478  *   Ethernet device flow is being created on.
1479  * @param[out] error
1480  *   Pointer to error structure.
1481  *
1482  * @return
1483  *   0 on success, a negative errno value otherwise and rte_errno is set.
1484  */
1485 int
1486 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1487 			     uint64_t item_flags,
1488 			     struct rte_eth_dev *dev,
1489 			     struct rte_flow_error *error)
1490 {
1491 	const struct rte_flow_item_vlan *spec = item->spec;
1492 	const struct rte_flow_item_vlan *mask = item->mask;
1493 	const struct rte_flow_item_vlan nic_mask = {
1494 		.tci = RTE_BE16(UINT16_MAX),
1495 		.inner_type = RTE_BE16(UINT16_MAX),
1496 	};
1497 	uint16_t vlan_tag = 0;
1498 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1499 	int ret;
1500 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1501 					MLX5_FLOW_LAYER_INNER_L4) :
1502 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1503 					MLX5_FLOW_LAYER_OUTER_L4);
1504 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1505 					MLX5_FLOW_LAYER_OUTER_VLAN;
1506 
1507 	if (item_flags & vlanm)
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1510 					  "multiple VLAN layers not supported");
1511 	else if ((item_flags & l34m) != 0)
1512 		return rte_flow_error_set(error, EINVAL,
1513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1514 					  "VLAN cannot follow L3/L4 layer");
1515 	if (!mask)
1516 		mask = &rte_flow_item_vlan_mask;
1517 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1518 					(const uint8_t *)&nic_mask,
1519 					sizeof(struct rte_flow_item_vlan),
1520 					error);
1521 	if (ret)
1522 		return ret;
1523 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1524 		struct mlx5_priv *priv = dev->data->dev_private;
1525 
1526 		if (priv->vmwa_context) {
1527 			/*
1528 			 * Non-NULL context means we have a virtual machine
1529 			 * and SR-IOV enabled, we have to create VLAN interface
1530 			 * to make hypervisor to setup E-Switch vport
1531 			 * context correctly. We avoid creating the multiple
1532 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1533 			 */
1534 			return rte_flow_error_set(error, EINVAL,
1535 						  RTE_FLOW_ERROR_TYPE_ITEM,
1536 						  item,
1537 						  "VLAN tag mask is not"
1538 						  " supported in virtual"
1539 						  " environment");
1540 		}
1541 	}
1542 	if (spec) {
1543 		vlan_tag = spec->tci;
1544 		vlan_tag &= mask->tci;
1545 	}
1546 	/*
1547 	 * From verbs perspective an empty VLAN is equivalent
1548 	 * to a packet without VLAN layer.
1549 	 */
1550 	if (!vlan_tag)
1551 		return rte_flow_error_set(error, EINVAL,
1552 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1553 					  item->spec,
1554 					  "VLAN cannot be empty");
1555 	return 0;
1556 }
1557 
1558 /**
1559  * Validate IPV4 item.
1560  *
1561  * @param[in] item
1562  *   Item specification.
1563  * @param[in] item_flags
1564  *   Bit-fields that holds the items detected until now.
1565  * @param[in] acc_mask
1566  *   Acceptable mask, if NULL default internal default mask
1567  *   will be used to check whether item fields are supported.
1568  * @param[out] error
1569  *   Pointer to error structure.
1570  *
1571  * @return
1572  *   0 on success, a negative errno value otherwise and rte_errno is set.
1573  */
1574 int
1575 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1576 			     uint64_t item_flags,
1577 			     uint64_t last_item,
1578 			     uint16_t ether_type,
1579 			     const struct rte_flow_item_ipv4 *acc_mask,
1580 			     struct rte_flow_error *error)
1581 {
1582 	const struct rte_flow_item_ipv4 *mask = item->mask;
1583 	const struct rte_flow_item_ipv4 *spec = item->spec;
1584 	const struct rte_flow_item_ipv4 nic_mask = {
1585 		.hdr = {
1586 			.src_addr = RTE_BE32(0xffffffff),
1587 			.dst_addr = RTE_BE32(0xffffffff),
1588 			.type_of_service = 0xff,
1589 			.next_proto_id = 0xff,
1590 		},
1591 	};
1592 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1593 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1594 				      MLX5_FLOW_LAYER_OUTER_L3;
1595 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1596 				      MLX5_FLOW_LAYER_OUTER_L4;
1597 	int ret;
1598 	uint8_t next_proto = 0xFF;
1599 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1600 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1601 				  MLX5_FLOW_LAYER_INNER_VLAN);
1602 
1603 	if ((last_item & l2_vlan) && ether_type &&
1604 	    ether_type != RTE_ETHER_TYPE_IPV4)
1605 		return rte_flow_error_set(error, EINVAL,
1606 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1607 					  "IPv4 cannot follow L2/VLAN layer "
1608 					  "which ether type is not IPv4");
1609 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1610 		if (mask && spec)
1611 			next_proto = mask->hdr.next_proto_id &
1612 				     spec->hdr.next_proto_id;
1613 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1614 			return rte_flow_error_set(error, EINVAL,
1615 						  RTE_FLOW_ERROR_TYPE_ITEM,
1616 						  item,
1617 						  "multiple tunnel "
1618 						  "not supported");
1619 	}
1620 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1621 		return rte_flow_error_set(error, EINVAL,
1622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1623 					  "wrong tunnel type - IPv6 specified "
1624 					  "but IPv4 item provided");
1625 	if (item_flags & l3m)
1626 		return rte_flow_error_set(error, ENOTSUP,
1627 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1628 					  "multiple L3 layers not supported");
1629 	else if (item_flags & l4m)
1630 		return rte_flow_error_set(error, EINVAL,
1631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1632 					  "L3 cannot follow an L4 layer.");
1633 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1634 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1635 		return rte_flow_error_set(error, EINVAL,
1636 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1637 					  "L3 cannot follow an NVGRE layer.");
1638 	if (!mask)
1639 		mask = &rte_flow_item_ipv4_mask;
1640 	else if (mask->hdr.next_proto_id != 0 &&
1641 		 mask->hdr.next_proto_id != 0xff)
1642 		return rte_flow_error_set(error, EINVAL,
1643 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1644 					  "partial mask is not supported"
1645 					  " for protocol");
1646 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1647 					acc_mask ? (const uint8_t *)acc_mask
1648 						 : (const uint8_t *)&nic_mask,
1649 					sizeof(struct rte_flow_item_ipv4),
1650 					error);
1651 	if (ret < 0)
1652 		return ret;
1653 	return 0;
1654 }
1655 
1656 /**
1657  * Validate IPV6 item.
1658  *
1659  * @param[in] item
1660  *   Item specification.
1661  * @param[in] item_flags
1662  *   Bit-fields that holds the items detected until now.
1663  * @param[in] acc_mask
1664  *   Acceptable mask, if NULL default internal default mask
1665  *   will be used to check whether item fields are supported.
1666  * @param[out] error
1667  *   Pointer to error structure.
1668  *
1669  * @return
1670  *   0 on success, a negative errno value otherwise and rte_errno is set.
1671  */
1672 int
1673 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1674 			     uint64_t item_flags,
1675 			     uint64_t last_item,
1676 			     uint16_t ether_type,
1677 			     const struct rte_flow_item_ipv6 *acc_mask,
1678 			     struct rte_flow_error *error)
1679 {
1680 	const struct rte_flow_item_ipv6 *mask = item->mask;
1681 	const struct rte_flow_item_ipv6 *spec = item->spec;
1682 	const struct rte_flow_item_ipv6 nic_mask = {
1683 		.hdr = {
1684 			.src_addr =
1685 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1686 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1687 			.dst_addr =
1688 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1689 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1690 			.vtc_flow = RTE_BE32(0xffffffff),
1691 			.proto = 0xff,
1692 		},
1693 	};
1694 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1695 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1696 				      MLX5_FLOW_LAYER_OUTER_L3;
1697 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1698 				      MLX5_FLOW_LAYER_OUTER_L4;
1699 	int ret;
1700 	uint8_t next_proto = 0xFF;
1701 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1702 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1703 				  MLX5_FLOW_LAYER_INNER_VLAN);
1704 
1705 	if ((last_item & l2_vlan) && ether_type &&
1706 	    ether_type != RTE_ETHER_TYPE_IPV6)
1707 		return rte_flow_error_set(error, EINVAL,
1708 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1709 					  "IPv6 cannot follow L2/VLAN layer "
1710 					  "which ether type is not IPv6");
1711 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1712 		if (mask && spec)
1713 			next_proto = mask->hdr.proto & spec->hdr.proto;
1714 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1715 			return rte_flow_error_set(error, EINVAL,
1716 						  RTE_FLOW_ERROR_TYPE_ITEM,
1717 						  item,
1718 						  "multiple tunnel "
1719 						  "not supported");
1720 	}
1721 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1722 		return rte_flow_error_set(error, EINVAL,
1723 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1724 					  "wrong tunnel type - IPv4 specified "
1725 					  "but IPv6 item provided");
1726 	if (item_flags & l3m)
1727 		return rte_flow_error_set(error, ENOTSUP,
1728 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1729 					  "multiple L3 layers not supported");
1730 	else if (item_flags & l4m)
1731 		return rte_flow_error_set(error, EINVAL,
1732 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1733 					  "L3 cannot follow an L4 layer.");
1734 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1735 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1736 		return rte_flow_error_set(error, EINVAL,
1737 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1738 					  "L3 cannot follow an NVGRE layer.");
1739 	if (!mask)
1740 		mask = &rte_flow_item_ipv6_mask;
1741 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1742 					acc_mask ? (const uint8_t *)acc_mask
1743 						 : (const uint8_t *)&nic_mask,
1744 					sizeof(struct rte_flow_item_ipv6),
1745 					error);
1746 	if (ret < 0)
1747 		return ret;
1748 	return 0;
1749 }
1750 
1751 /**
1752  * Validate UDP item.
1753  *
1754  * @param[in] item
1755  *   Item specification.
1756  * @param[in] item_flags
1757  *   Bit-fields that holds the items detected until now.
1758  * @param[in] target_protocol
1759  *   The next protocol in the previous item.
1760  * @param[in] flow_mask
1761  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1762  * @param[out] error
1763  *   Pointer to error structure.
1764  *
1765  * @return
1766  *   0 on success, a negative errno value otherwise and rte_errno is set.
1767  */
1768 int
1769 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1770 			    uint64_t item_flags,
1771 			    uint8_t target_protocol,
1772 			    struct rte_flow_error *error)
1773 {
1774 	const struct rte_flow_item_udp *mask = item->mask;
1775 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1776 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1777 				      MLX5_FLOW_LAYER_OUTER_L3;
1778 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1779 				      MLX5_FLOW_LAYER_OUTER_L4;
1780 	int ret;
1781 
1782 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1783 		return rte_flow_error_set(error, EINVAL,
1784 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1785 					  "protocol filtering not compatible"
1786 					  " with UDP layer");
1787 	if (!(item_flags & l3m))
1788 		return rte_flow_error_set(error, EINVAL,
1789 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1790 					  "L3 is mandatory to filter on L4");
1791 	if (item_flags & l4m)
1792 		return rte_flow_error_set(error, EINVAL,
1793 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1794 					  "multiple L4 layers not supported");
1795 	if (!mask)
1796 		mask = &rte_flow_item_udp_mask;
1797 	ret = mlx5_flow_item_acceptable
1798 		(item, (const uint8_t *)mask,
1799 		 (const uint8_t *)&rte_flow_item_udp_mask,
1800 		 sizeof(struct rte_flow_item_udp), error);
1801 	if (ret < 0)
1802 		return ret;
1803 	return 0;
1804 }
1805 
1806 /**
1807  * Validate TCP item.
1808  *
1809  * @param[in] item
1810  *   Item specification.
1811  * @param[in] item_flags
1812  *   Bit-fields that holds the items detected until now.
1813  * @param[in] target_protocol
1814  *   The next protocol in the previous item.
1815  * @param[out] error
1816  *   Pointer to error structure.
1817  *
1818  * @return
1819  *   0 on success, a negative errno value otherwise and rte_errno is set.
1820  */
1821 int
1822 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1823 			    uint64_t item_flags,
1824 			    uint8_t target_protocol,
1825 			    const struct rte_flow_item_tcp *flow_mask,
1826 			    struct rte_flow_error *error)
1827 {
1828 	const struct rte_flow_item_tcp *mask = item->mask;
1829 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1830 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1831 				      MLX5_FLOW_LAYER_OUTER_L3;
1832 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1833 				      MLX5_FLOW_LAYER_OUTER_L4;
1834 	int ret;
1835 
1836 	MLX5_ASSERT(flow_mask);
1837 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1838 		return rte_flow_error_set(error, EINVAL,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "protocol filtering not compatible"
1841 					  " with TCP layer");
1842 	if (!(item_flags & l3m))
1843 		return rte_flow_error_set(error, EINVAL,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "L3 is mandatory to filter on L4");
1846 	if (item_flags & l4m)
1847 		return rte_flow_error_set(error, EINVAL,
1848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1849 					  "multiple L4 layers not supported");
1850 	if (!mask)
1851 		mask = &rte_flow_item_tcp_mask;
1852 	ret = mlx5_flow_item_acceptable
1853 		(item, (const uint8_t *)mask,
1854 		 (const uint8_t *)flow_mask,
1855 		 sizeof(struct rte_flow_item_tcp), error);
1856 	if (ret < 0)
1857 		return ret;
1858 	return 0;
1859 }
1860 
1861 /**
1862  * Validate VXLAN item.
1863  *
1864  * @param[in] item
1865  *   Item specification.
1866  * @param[in] item_flags
1867  *   Bit-fields that holds the items detected until now.
1868  * @param[in] target_protocol
1869  *   The next protocol in the previous item.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 int
1877 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1878 			      uint64_t item_flags,
1879 			      struct rte_flow_error *error)
1880 {
1881 	const struct rte_flow_item_vxlan *spec = item->spec;
1882 	const struct rte_flow_item_vxlan *mask = item->mask;
1883 	int ret;
1884 	union vni {
1885 		uint32_t vlan_id;
1886 		uint8_t vni[4];
1887 	} id = { .vlan_id = 0, };
1888 
1889 
1890 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1891 		return rte_flow_error_set(error, ENOTSUP,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "multiple tunnel layers not"
1894 					  " supported");
1895 	/*
1896 	 * Verify only UDPv4 is present as defined in
1897 	 * https://tools.ietf.org/html/rfc7348
1898 	 */
1899 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1900 		return rte_flow_error_set(error, EINVAL,
1901 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1902 					  "no outer UDP layer found");
1903 	if (!mask)
1904 		mask = &rte_flow_item_vxlan_mask;
1905 	ret = mlx5_flow_item_acceptable
1906 		(item, (const uint8_t *)mask,
1907 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1908 		 sizeof(struct rte_flow_item_vxlan),
1909 		 error);
1910 	if (ret < 0)
1911 		return ret;
1912 	if (spec) {
1913 		memcpy(&id.vni[1], spec->vni, 3);
1914 		memcpy(&id.vni[1], mask->vni, 3);
1915 	}
1916 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1917 		return rte_flow_error_set(error, ENOTSUP,
1918 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1919 					  "VXLAN tunnel must be fully defined");
1920 	return 0;
1921 }
1922 
1923 /**
1924  * Validate VXLAN_GPE item.
1925  *
1926  * @param[in] item
1927  *   Item specification.
1928  * @param[in] item_flags
1929  *   Bit-fields that holds the items detected until now.
1930  * @param[in] priv
1931  *   Pointer to the private data structure.
1932  * @param[in] target_protocol
1933  *   The next protocol in the previous item.
1934  * @param[out] error
1935  *   Pointer to error structure.
1936  *
1937  * @return
1938  *   0 on success, a negative errno value otherwise and rte_errno is set.
1939  */
1940 int
1941 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1942 				  uint64_t item_flags,
1943 				  struct rte_eth_dev *dev,
1944 				  struct rte_flow_error *error)
1945 {
1946 	struct mlx5_priv *priv = dev->data->dev_private;
1947 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1948 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1949 	int ret;
1950 	union vni {
1951 		uint32_t vlan_id;
1952 		uint8_t vni[4];
1953 	} id = { .vlan_id = 0, };
1954 
1955 	if (!priv->config.l3_vxlan_en)
1956 		return rte_flow_error_set(error, ENOTSUP,
1957 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1958 					  "L3 VXLAN is not enabled by device"
1959 					  " parameter and/or not configured in"
1960 					  " firmware");
1961 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1962 		return rte_flow_error_set(error, ENOTSUP,
1963 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1964 					  "multiple tunnel layers not"
1965 					  " supported");
1966 	/*
1967 	 * Verify only UDPv4 is present as defined in
1968 	 * https://tools.ietf.org/html/rfc7348
1969 	 */
1970 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1971 		return rte_flow_error_set(error, EINVAL,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "no outer UDP layer found");
1974 	if (!mask)
1975 		mask = &rte_flow_item_vxlan_gpe_mask;
1976 	ret = mlx5_flow_item_acceptable
1977 		(item, (const uint8_t *)mask,
1978 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1979 		 sizeof(struct rte_flow_item_vxlan_gpe),
1980 		 error);
1981 	if (ret < 0)
1982 		return ret;
1983 	if (spec) {
1984 		if (spec->protocol)
1985 			return rte_flow_error_set(error, ENOTSUP,
1986 						  RTE_FLOW_ERROR_TYPE_ITEM,
1987 						  item,
1988 						  "VxLAN-GPE protocol"
1989 						  " not supported");
1990 		memcpy(&id.vni[1], spec->vni, 3);
1991 		memcpy(&id.vni[1], mask->vni, 3);
1992 	}
1993 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1994 		return rte_flow_error_set(error, ENOTSUP,
1995 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1996 					  "VXLAN-GPE tunnel must be fully"
1997 					  " defined");
1998 	return 0;
1999 }
2000 /**
2001  * Validate GRE Key item.
2002  *
2003  * @param[in] item
2004  *   Item specification.
2005  * @param[in] item_flags
2006  *   Bit flags to mark detected items.
2007  * @param[in] gre_item
2008  *   Pointer to gre_item
2009  * @param[out] error
2010  *   Pointer to error structure.
2011  *
2012  * @return
2013  *   0 on success, a negative errno value otherwise and rte_errno is set.
2014  */
2015 int
2016 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2017 				uint64_t item_flags,
2018 				const struct rte_flow_item *gre_item,
2019 				struct rte_flow_error *error)
2020 {
2021 	const rte_be32_t *mask = item->mask;
2022 	int ret = 0;
2023 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2024 	const struct rte_flow_item_gre *gre_spec;
2025 	const struct rte_flow_item_gre *gre_mask;
2026 
2027 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2028 		return rte_flow_error_set(error, ENOTSUP,
2029 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2030 					  "Multiple GRE key not support");
2031 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2032 		return rte_flow_error_set(error, ENOTSUP,
2033 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2034 					  "No preceding GRE header");
2035 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2036 		return rte_flow_error_set(error, ENOTSUP,
2037 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2038 					  "GRE key following a wrong item");
2039 	gre_mask = gre_item->mask;
2040 	if (!gre_mask)
2041 		gre_mask = &rte_flow_item_gre_mask;
2042 	gre_spec = gre_item->spec;
2043 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2044 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "Key bit must be on");
2048 
2049 	if (!mask)
2050 		mask = &gre_key_default_mask;
2051 	ret = mlx5_flow_item_acceptable
2052 		(item, (const uint8_t *)mask,
2053 		 (const uint8_t *)&gre_key_default_mask,
2054 		 sizeof(rte_be32_t), error);
2055 	return ret;
2056 }
2057 
2058 /**
2059  * Validate GRE item.
2060  *
2061  * @param[in] item
2062  *   Item specification.
2063  * @param[in] item_flags
2064  *   Bit flags to mark detected items.
2065  * @param[in] target_protocol
2066  *   The next protocol in the previous item.
2067  * @param[out] error
2068  *   Pointer to error structure.
2069  *
2070  * @return
2071  *   0 on success, a negative errno value otherwise and rte_errno is set.
2072  */
2073 int
2074 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2075 			    uint64_t item_flags,
2076 			    uint8_t target_protocol,
2077 			    struct rte_flow_error *error)
2078 {
2079 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2080 	const struct rte_flow_item_gre *mask = item->mask;
2081 	int ret;
2082 	const struct rte_flow_item_gre nic_mask = {
2083 		.c_rsvd0_ver = RTE_BE16(0xB000),
2084 		.protocol = RTE_BE16(UINT16_MAX),
2085 	};
2086 
2087 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2088 		return rte_flow_error_set(error, EINVAL,
2089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2090 					  "protocol filtering not compatible"
2091 					  " with this GRE layer");
2092 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2093 		return rte_flow_error_set(error, ENOTSUP,
2094 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 					  "multiple tunnel layers not"
2096 					  " supported");
2097 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2098 		return rte_flow_error_set(error, ENOTSUP,
2099 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2100 					  "L3 Layer is missing");
2101 	if (!mask)
2102 		mask = &rte_flow_item_gre_mask;
2103 	ret = mlx5_flow_item_acceptable
2104 		(item, (const uint8_t *)mask,
2105 		 (const uint8_t *)&nic_mask,
2106 		 sizeof(struct rte_flow_item_gre), error);
2107 	if (ret < 0)
2108 		return ret;
2109 #ifndef HAVE_MLX5DV_DR
2110 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2111 	if (spec && (spec->protocol & mask->protocol))
2112 		return rte_flow_error_set(error, ENOTSUP,
2113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 					  "without MPLS support the"
2115 					  " specification cannot be used for"
2116 					  " filtering");
2117 #endif
2118 #endif
2119 	return 0;
2120 }
2121 
2122 /**
2123  * Validate Geneve item.
2124  *
2125  * @param[in] item
2126  *   Item specification.
2127  * @param[in] itemFlags
2128  *   Bit-fields that holds the items detected until now.
2129  * @param[in] enPriv
2130  *   Pointer to the private data structure.
2131  * @param[out] error
2132  *   Pointer to error structure.
2133  *
2134  * @return
2135  *   0 on success, a negative errno value otherwise and rte_errno is set.
2136  */
2137 
2138 int
2139 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2140 			       uint64_t item_flags,
2141 			       struct rte_eth_dev *dev,
2142 			       struct rte_flow_error *error)
2143 {
2144 	struct mlx5_priv *priv = dev->data->dev_private;
2145 	const struct rte_flow_item_geneve *spec = item->spec;
2146 	const struct rte_flow_item_geneve *mask = item->mask;
2147 	int ret;
2148 	uint16_t gbhdr;
2149 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2150 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2151 	const struct rte_flow_item_geneve nic_mask = {
2152 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2153 		.vni = "\xff\xff\xff",
2154 		.protocol = RTE_BE16(UINT16_MAX),
2155 	};
2156 
2157 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2158 		return rte_flow_error_set(error, ENOTSUP,
2159 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2160 					  "L3 Geneve is not enabled by device"
2161 					  " parameter and/or not configured in"
2162 					  " firmware");
2163 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2164 		return rte_flow_error_set(error, ENOTSUP,
2165 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2166 					  "multiple tunnel layers not"
2167 					  " supported");
2168 	/*
2169 	 * Verify only UDPv4 is present as defined in
2170 	 * https://tools.ietf.org/html/rfc7348
2171 	 */
2172 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2173 		return rte_flow_error_set(error, EINVAL,
2174 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2175 					  "no outer UDP layer found");
2176 	if (!mask)
2177 		mask = &rte_flow_item_geneve_mask;
2178 	ret = mlx5_flow_item_acceptable
2179 				  (item, (const uint8_t *)mask,
2180 				   (const uint8_t *)&nic_mask,
2181 				   sizeof(struct rte_flow_item_geneve), error);
2182 	if (ret)
2183 		return ret;
2184 	if (spec) {
2185 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2186 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2187 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2188 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2189 			return rte_flow_error_set(error, ENOTSUP,
2190 						  RTE_FLOW_ERROR_TYPE_ITEM,
2191 						  item,
2192 						  "Geneve protocol unsupported"
2193 						  " fields are being used");
2194 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2195 			return rte_flow_error_set
2196 					(error, ENOTSUP,
2197 					 RTE_FLOW_ERROR_TYPE_ITEM,
2198 					 item,
2199 					 "Unsupported Geneve options length");
2200 	}
2201 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2202 		return rte_flow_error_set
2203 				    (error, ENOTSUP,
2204 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2205 				     "Geneve tunnel must be fully defined");
2206 	return 0;
2207 }
2208 
2209 /**
2210  * Validate MPLS item.
2211  *
2212  * @param[in] dev
2213  *   Pointer to the rte_eth_dev structure.
2214  * @param[in] item
2215  *   Item specification.
2216  * @param[in] item_flags
2217  *   Bit-fields that holds the items detected until now.
2218  * @param[in] prev_layer
2219  *   The protocol layer indicated in previous item.
2220  * @param[out] error
2221  *   Pointer to error structure.
2222  *
2223  * @return
2224  *   0 on success, a negative errno value otherwise and rte_errno is set.
2225  */
2226 int
2227 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2228 			     const struct rte_flow_item *item __rte_unused,
2229 			     uint64_t item_flags __rte_unused,
2230 			     uint64_t prev_layer __rte_unused,
2231 			     struct rte_flow_error *error)
2232 {
2233 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2234 	const struct rte_flow_item_mpls *mask = item->mask;
2235 	struct mlx5_priv *priv = dev->data->dev_private;
2236 	int ret;
2237 
2238 	if (!priv->config.mpls_en)
2239 		return rte_flow_error_set(error, ENOTSUP,
2240 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2241 					  "MPLS not supported or"
2242 					  " disabled in firmware"
2243 					  " configuration.");
2244 	/* MPLS over IP, UDP, GRE is allowed */
2245 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2246 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2247 			    MLX5_FLOW_LAYER_GRE)))
2248 		return rte_flow_error_set(error, EINVAL,
2249 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2250 					  "protocol filtering not compatible"
2251 					  " with MPLS layer");
2252 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2253 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2254 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2255 		return rte_flow_error_set(error, ENOTSUP,
2256 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2257 					  "multiple tunnel layers not"
2258 					  " supported");
2259 	if (!mask)
2260 		mask = &rte_flow_item_mpls_mask;
2261 	ret = mlx5_flow_item_acceptable
2262 		(item, (const uint8_t *)mask,
2263 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2264 		 sizeof(struct rte_flow_item_mpls), error);
2265 	if (ret < 0)
2266 		return ret;
2267 	return 0;
2268 #else
2269 	return rte_flow_error_set(error, ENOTSUP,
2270 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 				  "MPLS is not supported by Verbs, please"
2272 				  " update.");
2273 #endif
2274 }
2275 
2276 /**
2277  * Validate NVGRE item.
2278  *
2279  * @param[in] item
2280  *   Item specification.
2281  * @param[in] item_flags
2282  *   Bit flags to mark detected items.
2283  * @param[in] target_protocol
2284  *   The next protocol in the previous item.
2285  * @param[out] error
2286  *   Pointer to error structure.
2287  *
2288  * @return
2289  *   0 on success, a negative errno value otherwise and rte_errno is set.
2290  */
2291 int
2292 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2293 			      uint64_t item_flags,
2294 			      uint8_t target_protocol,
2295 			      struct rte_flow_error *error)
2296 {
2297 	const struct rte_flow_item_nvgre *mask = item->mask;
2298 	int ret;
2299 
2300 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2301 		return rte_flow_error_set(error, EINVAL,
2302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 					  "protocol filtering not compatible"
2304 					  " with this GRE layer");
2305 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2306 		return rte_flow_error_set(error, ENOTSUP,
2307 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2308 					  "multiple tunnel layers not"
2309 					  " supported");
2310 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2311 		return rte_flow_error_set(error, ENOTSUP,
2312 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2313 					  "L3 Layer is missing");
2314 	if (!mask)
2315 		mask = &rte_flow_item_nvgre_mask;
2316 	ret = mlx5_flow_item_acceptable
2317 		(item, (const uint8_t *)mask,
2318 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2319 		 sizeof(struct rte_flow_item_nvgre), error);
2320 	if (ret < 0)
2321 		return ret;
2322 	return 0;
2323 }
2324 
2325 /* Allocate unique ID for the split Q/RSS subflows. */
2326 static uint32_t
2327 flow_qrss_get_id(struct rte_eth_dev *dev)
2328 {
2329 	struct mlx5_priv *priv = dev->data->dev_private;
2330 	uint32_t qrss_id, ret;
2331 
2332 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2333 	if (ret)
2334 		return 0;
2335 	MLX5_ASSERT(qrss_id);
2336 	return qrss_id;
2337 }
2338 
2339 /* Free unique ID for the split Q/RSS subflows. */
2340 static void
2341 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2342 {
2343 	struct mlx5_priv *priv = dev->data->dev_private;
2344 
2345 	if (qrss_id)
2346 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2347 }
2348 
2349 /**
2350  * Release resource related QUEUE/RSS action split.
2351  *
2352  * @param dev
2353  *   Pointer to Ethernet device.
2354  * @param flow
2355  *   Flow to release id's from.
2356  */
2357 static void
2358 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2359 			     struct rte_flow *flow)
2360 {
2361 	struct mlx5_priv *priv = dev->data->dev_private;
2362 	uint32_t handle_idx;
2363 	struct mlx5_flow_handle *dev_handle;
2364 
2365 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2366 		       handle_idx, dev_handle, next)
2367 		if (dev_handle->split_flow_id)
2368 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2369 }
2370 
2371 static int
2372 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2373 		   const struct rte_flow_attr *attr __rte_unused,
2374 		   const struct rte_flow_item items[] __rte_unused,
2375 		   const struct rte_flow_action actions[] __rte_unused,
2376 		   bool external __rte_unused,
2377 		   int hairpin __rte_unused,
2378 		   struct rte_flow_error *error)
2379 {
2380 	return rte_flow_error_set(error, ENOTSUP,
2381 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2382 }
2383 
2384 static struct mlx5_flow *
2385 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2386 		  const struct rte_flow_attr *attr __rte_unused,
2387 		  const struct rte_flow_item items[] __rte_unused,
2388 		  const struct rte_flow_action actions[] __rte_unused,
2389 		  struct rte_flow_error *error)
2390 {
2391 	rte_flow_error_set(error, ENOTSUP,
2392 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2393 	return NULL;
2394 }
2395 
2396 static int
2397 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2398 		    struct mlx5_flow *dev_flow __rte_unused,
2399 		    const struct rte_flow_attr *attr __rte_unused,
2400 		    const struct rte_flow_item items[] __rte_unused,
2401 		    const struct rte_flow_action actions[] __rte_unused,
2402 		    struct rte_flow_error *error)
2403 {
2404 	return rte_flow_error_set(error, ENOTSUP,
2405 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2406 }
2407 
2408 static int
2409 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2410 		struct rte_flow *flow __rte_unused,
2411 		struct rte_flow_error *error)
2412 {
2413 	return rte_flow_error_set(error, ENOTSUP,
2414 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2415 }
2416 
2417 static void
2418 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2419 		 struct rte_flow *flow __rte_unused)
2420 {
2421 }
2422 
2423 static void
2424 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2425 		  struct rte_flow *flow __rte_unused)
2426 {
2427 }
2428 
2429 static int
2430 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2431 		struct rte_flow *flow __rte_unused,
2432 		const struct rte_flow_action *actions __rte_unused,
2433 		void *data __rte_unused,
2434 		struct rte_flow_error *error)
2435 {
2436 	return rte_flow_error_set(error, ENOTSUP,
2437 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2438 }
2439 
2440 /* Void driver to protect from null pointer reference. */
2441 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2442 	.validate = flow_null_validate,
2443 	.prepare = flow_null_prepare,
2444 	.translate = flow_null_translate,
2445 	.apply = flow_null_apply,
2446 	.remove = flow_null_remove,
2447 	.destroy = flow_null_destroy,
2448 	.query = flow_null_query,
2449 };
2450 
2451 /**
2452  * Select flow driver type according to flow attributes and device
2453  * configuration.
2454  *
2455  * @param[in] dev
2456  *   Pointer to the dev structure.
2457  * @param[in] attr
2458  *   Pointer to the flow attributes.
2459  *
2460  * @return
2461  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2462  */
2463 static enum mlx5_flow_drv_type
2464 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2465 {
2466 	struct mlx5_priv *priv = dev->data->dev_private;
2467 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2468 
2469 	if (attr->transfer && priv->config.dv_esw_en)
2470 		type = MLX5_FLOW_TYPE_DV;
2471 	if (!attr->transfer)
2472 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2473 						 MLX5_FLOW_TYPE_VERBS;
2474 	return type;
2475 }
2476 
2477 #define flow_get_drv_ops(type) flow_drv_ops[type]
2478 
2479 /**
2480  * Flow driver validation API. This abstracts calling driver specific functions.
2481  * The type of flow driver is determined according to flow attributes.
2482  *
2483  * @param[in] dev
2484  *   Pointer to the dev structure.
2485  * @param[in] attr
2486  *   Pointer to the flow attributes.
2487  * @param[in] items
2488  *   Pointer to the list of items.
2489  * @param[in] actions
2490  *   Pointer to the list of actions.
2491  * @param[in] external
2492  *   This flow rule is created by request external to PMD.
2493  * @param[in] hairpin
2494  *   Number of hairpin TX actions, 0 means classic flow.
2495  * @param[out] error
2496  *   Pointer to the error structure.
2497  *
2498  * @return
2499  *   0 on success, a negative errno value otherwise and rte_errno is set.
2500  */
2501 static inline int
2502 flow_drv_validate(struct rte_eth_dev *dev,
2503 		  const struct rte_flow_attr *attr,
2504 		  const struct rte_flow_item items[],
2505 		  const struct rte_flow_action actions[],
2506 		  bool external, int hairpin, struct rte_flow_error *error)
2507 {
2508 	const struct mlx5_flow_driver_ops *fops;
2509 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2510 
2511 	fops = flow_get_drv_ops(type);
2512 	return fops->validate(dev, attr, items, actions, external,
2513 			      hairpin, error);
2514 }
2515 
2516 /**
2517  * Flow driver preparation API. This abstracts calling driver specific
2518  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2519  * calculates the size of memory required for device flow, allocates the memory,
2520  * initializes the device flow and returns the pointer.
2521  *
2522  * @note
2523  *   This function initializes device flow structure such as dv or verbs in
2524  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2525  *   rest. For example, adding returning device flow to flow->dev_flow list and
2526  *   setting backward reference to the flow should be done out of this function.
2527  *   layers field is not filled either.
2528  *
2529  * @param[in] dev
2530  *   Pointer to the dev structure.
2531  * @param[in] attr
2532  *   Pointer to the flow attributes.
2533  * @param[in] items
2534  *   Pointer to the list of items.
2535  * @param[in] actions
2536  *   Pointer to the list of actions.
2537  * @param[in] flow_idx
2538  *   This memory pool index to the flow.
2539  * @param[out] error
2540  *   Pointer to the error structure.
2541  *
2542  * @return
2543  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2544  */
2545 static inline struct mlx5_flow *
2546 flow_drv_prepare(struct rte_eth_dev *dev,
2547 		 const struct rte_flow *flow,
2548 		 const struct rte_flow_attr *attr,
2549 		 const struct rte_flow_item items[],
2550 		 const struct rte_flow_action actions[],
2551 		 uint32_t flow_idx,
2552 		 struct rte_flow_error *error)
2553 {
2554 	const struct mlx5_flow_driver_ops *fops;
2555 	enum mlx5_flow_drv_type type = flow->drv_type;
2556 	struct mlx5_flow *mlx5_flow = NULL;
2557 
2558 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2559 	fops = flow_get_drv_ops(type);
2560 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2561 	if (mlx5_flow)
2562 		mlx5_flow->flow_idx = flow_idx;
2563 	return mlx5_flow;
2564 }
2565 
2566 /**
2567  * Flow driver translation API. This abstracts calling driver specific
2568  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2569  * translates a generic flow into a driver flow. flow_drv_prepare() must
2570  * precede.
2571  *
2572  * @note
2573  *   dev_flow->layers could be filled as a result of parsing during translation
2574  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2575  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2576  *   flow->actions could be overwritten even though all the expanded dev_flows
2577  *   have the same actions.
2578  *
2579  * @param[in] dev
2580  *   Pointer to the rte dev structure.
2581  * @param[in, out] dev_flow
2582  *   Pointer to the mlx5 flow.
2583  * @param[in] attr
2584  *   Pointer to the flow attributes.
2585  * @param[in] items
2586  *   Pointer to the list of items.
2587  * @param[in] actions
2588  *   Pointer to the list of actions.
2589  * @param[out] error
2590  *   Pointer to the error structure.
2591  *
2592  * @return
2593  *   0 on success, a negative errno value otherwise and rte_errno is set.
2594  */
2595 static inline int
2596 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2597 		   const struct rte_flow_attr *attr,
2598 		   const struct rte_flow_item items[],
2599 		   const struct rte_flow_action actions[],
2600 		   struct rte_flow_error *error)
2601 {
2602 	const struct mlx5_flow_driver_ops *fops;
2603 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2604 
2605 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2606 	fops = flow_get_drv_ops(type);
2607 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2608 }
2609 
2610 /**
2611  * Flow driver apply API. This abstracts calling driver specific functions.
2612  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2613  * translated driver flows on to device. flow_drv_translate() must precede.
2614  *
2615  * @param[in] dev
2616  *   Pointer to Ethernet device structure.
2617  * @param[in, out] flow
2618  *   Pointer to flow structure.
2619  * @param[out] error
2620  *   Pointer to error structure.
2621  *
2622  * @return
2623  *   0 on success, a negative errno value otherwise and rte_errno is set.
2624  */
2625 static inline int
2626 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2627 	       struct rte_flow_error *error)
2628 {
2629 	const struct mlx5_flow_driver_ops *fops;
2630 	enum mlx5_flow_drv_type type = flow->drv_type;
2631 
2632 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2633 	fops = flow_get_drv_ops(type);
2634 	return fops->apply(dev, flow, error);
2635 }
2636 
2637 /**
2638  * Flow driver remove API. This abstracts calling driver specific functions.
2639  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2640  * on device. All the resources of the flow should be freed by calling
2641  * flow_drv_destroy().
2642  *
2643  * @param[in] dev
2644  *   Pointer to Ethernet device.
2645  * @param[in, out] flow
2646  *   Pointer to flow structure.
2647  */
2648 static inline void
2649 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2650 {
2651 	const struct mlx5_flow_driver_ops *fops;
2652 	enum mlx5_flow_drv_type type = flow->drv_type;
2653 
2654 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2655 	fops = flow_get_drv_ops(type);
2656 	fops->remove(dev, flow);
2657 }
2658 
2659 /**
2660  * Flow driver destroy API. This abstracts calling driver specific functions.
2661  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2662  * on device and releases resources of the flow.
2663  *
2664  * @param[in] dev
2665  *   Pointer to Ethernet device.
2666  * @param[in, out] flow
2667  *   Pointer to flow structure.
2668  */
2669 static inline void
2670 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2671 {
2672 	const struct mlx5_flow_driver_ops *fops;
2673 	enum mlx5_flow_drv_type type = flow->drv_type;
2674 
2675 	flow_mreg_split_qrss_release(dev, flow);
2676 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2677 	fops = flow_get_drv_ops(type);
2678 	fops->destroy(dev, flow);
2679 }
2680 
2681 /**
2682  * Get RSS action from the action list.
2683  *
2684  * @param[in] actions
2685  *   Pointer to the list of actions.
2686  *
2687  * @return
2688  *   Pointer to the RSS action if exist, else return NULL.
2689  */
2690 static const struct rte_flow_action_rss*
2691 flow_get_rss_action(const struct rte_flow_action actions[])
2692 {
2693 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2694 		switch (actions->type) {
2695 		case RTE_FLOW_ACTION_TYPE_RSS:
2696 			return (const struct rte_flow_action_rss *)
2697 			       actions->conf;
2698 		default:
2699 			break;
2700 		}
2701 	}
2702 	return NULL;
2703 }
2704 
2705 static unsigned int
2706 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2707 {
2708 	const struct rte_flow_item *item;
2709 	unsigned int has_vlan = 0;
2710 
2711 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2712 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2713 			has_vlan = 1;
2714 			break;
2715 		}
2716 	}
2717 	if (has_vlan)
2718 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2719 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2720 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2721 			       MLX5_EXPANSION_ROOT_OUTER;
2722 }
2723 
2724 /**
2725  *  Get layer flags from the prefix flow.
2726  *
2727  *  Some flows may be split to several subflows, the prefix subflow gets the
2728  *  match items and the suffix sub flow gets the actions.
2729  *  Some actions need the user defined match item flags to get the detail for
2730  *  the action.
2731  *  This function helps the suffix flow to get the item layer flags from prefix
2732  *  subflow.
2733  *
2734  * @param[in] dev_flow
2735  *   Pointer the created preifx subflow.
2736  *
2737  * @return
2738  *   The layers get from prefix subflow.
2739  */
2740 static inline uint64_t
2741 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2742 {
2743 	uint64_t layers = 0;
2744 
2745 	/*
2746 	 * Layers bits could be localization, but usually the compiler will
2747 	 * help to do the optimization work for source code.
2748 	 * If no decap actions, use the layers directly.
2749 	 */
2750 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2751 		return dev_flow->handle->layers;
2752 	/* Convert L3 layers with decap action. */
2753 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2754 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2755 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2756 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2757 	/* Convert L4 layers with decap action.  */
2758 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2759 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2760 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2761 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2762 	return layers;
2763 }
2764 
2765 /**
2766  * Get metadata split action information.
2767  *
2768  * @param[in] actions
2769  *   Pointer to the list of actions.
2770  * @param[out] qrss
2771  *   Pointer to the return pointer.
2772  * @param[out] qrss_type
2773  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2774  *   if no QUEUE/RSS is found.
2775  * @param[out] encap_idx
2776  *   Pointer to the index of the encap action if exists, otherwise the last
2777  *   action index.
2778  *
2779  * @return
2780  *   Total number of actions.
2781  */
2782 static int
2783 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2784 				       const struct rte_flow_action **qrss,
2785 				       int *encap_idx)
2786 {
2787 	const struct rte_flow_action_raw_encap *raw_encap;
2788 	int actions_n = 0;
2789 	int raw_decap_idx = -1;
2790 
2791 	*encap_idx = -1;
2792 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2793 		switch (actions->type) {
2794 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2795 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2796 			*encap_idx = actions_n;
2797 			break;
2798 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2799 			raw_decap_idx = actions_n;
2800 			break;
2801 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2802 			raw_encap = actions->conf;
2803 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2804 				*encap_idx = raw_decap_idx != -1 ?
2805 						      raw_decap_idx : actions_n;
2806 			break;
2807 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2808 		case RTE_FLOW_ACTION_TYPE_RSS:
2809 			*qrss = actions;
2810 			break;
2811 		default:
2812 			break;
2813 		}
2814 		actions_n++;
2815 	}
2816 	if (*encap_idx == -1)
2817 		*encap_idx = actions_n;
2818 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2819 	return actions_n + 1;
2820 }
2821 
2822 /**
2823  * Check meter action from the action list.
2824  *
2825  * @param[in] actions
2826  *   Pointer to the list of actions.
2827  * @param[out] mtr
2828  *   Pointer to the meter exist flag.
2829  *
2830  * @return
2831  *   Total number of actions.
2832  */
2833 static int
2834 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2835 {
2836 	int actions_n = 0;
2837 
2838 	MLX5_ASSERT(mtr);
2839 	*mtr = 0;
2840 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2841 		switch (actions->type) {
2842 		case RTE_FLOW_ACTION_TYPE_METER:
2843 			*mtr = 1;
2844 			break;
2845 		default:
2846 			break;
2847 		}
2848 		actions_n++;
2849 	}
2850 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2851 	return actions_n + 1;
2852 }
2853 
2854 /**
2855  * Check if the flow should be splited due to hairpin.
2856  * The reason for the split is that in current HW we can't
2857  * support encap on Rx, so if a flow have encap we move it
2858  * to Tx.
2859  *
2860  * @param dev
2861  *   Pointer to Ethernet device.
2862  * @param[in] attr
2863  *   Flow rule attributes.
2864  * @param[in] actions
2865  *   Associated actions (list terminated by the END action).
2866  *
2867  * @return
2868  *   > 0 the number of actions and the flow should be split,
2869  *   0 when no split required.
2870  */
2871 static int
2872 flow_check_hairpin_split(struct rte_eth_dev *dev,
2873 			 const struct rte_flow_attr *attr,
2874 			 const struct rte_flow_action actions[])
2875 {
2876 	int queue_action = 0;
2877 	int action_n = 0;
2878 	int encap = 0;
2879 	const struct rte_flow_action_queue *queue;
2880 	const struct rte_flow_action_rss *rss;
2881 	const struct rte_flow_action_raw_encap *raw_encap;
2882 
2883 	if (!attr->ingress)
2884 		return 0;
2885 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2886 		switch (actions->type) {
2887 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2888 			queue = actions->conf;
2889 			if (queue == NULL)
2890 				return 0;
2891 			if (mlx5_rxq_get_type(dev, queue->index) !=
2892 			    MLX5_RXQ_TYPE_HAIRPIN)
2893 				return 0;
2894 			queue_action = 1;
2895 			action_n++;
2896 			break;
2897 		case RTE_FLOW_ACTION_TYPE_RSS:
2898 			rss = actions->conf;
2899 			if (rss == NULL || rss->queue_num == 0)
2900 				return 0;
2901 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2902 			    MLX5_RXQ_TYPE_HAIRPIN)
2903 				return 0;
2904 			queue_action = 1;
2905 			action_n++;
2906 			break;
2907 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2908 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2909 			encap = 1;
2910 			action_n++;
2911 			break;
2912 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2913 			raw_encap = actions->conf;
2914 			if (raw_encap->size >
2915 			    (sizeof(struct rte_flow_item_eth) +
2916 			     sizeof(struct rte_flow_item_ipv4)))
2917 				encap = 1;
2918 			action_n++;
2919 			break;
2920 		default:
2921 			action_n++;
2922 			break;
2923 		}
2924 	}
2925 	if (encap == 1 && queue_action)
2926 		return action_n;
2927 	return 0;
2928 }
2929 
2930 /* Declare flow create/destroy prototype in advance. */
2931 static uint32_t
2932 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2933 		 const struct rte_flow_attr *attr,
2934 		 const struct rte_flow_item items[],
2935 		 const struct rte_flow_action actions[],
2936 		 bool external, struct rte_flow_error *error);
2937 
2938 static void
2939 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2940 		  uint32_t flow_idx);
2941 
2942 /**
2943  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2944  *
2945  * As mark_id is unique, if there's already a registered flow for the mark_id,
2946  * return by increasing the reference counter of the resource. Otherwise, create
2947  * the resource (mcp_res) and flow.
2948  *
2949  * Flow looks like,
2950  *   - If ingress port is ANY and reg_c[1] is mark_id,
2951  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2952  *
2953  * For default flow (zero mark_id), flow is like,
2954  *   - If ingress port is ANY,
2955  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2956  *
2957  * @param dev
2958  *   Pointer to Ethernet device.
2959  * @param mark_id
2960  *   ID of MARK action, zero means default flow for META.
2961  * @param[out] error
2962  *   Perform verbose error reporting if not NULL.
2963  *
2964  * @return
2965  *   Associated resource on success, NULL otherwise and rte_errno is set.
2966  */
2967 static struct mlx5_flow_mreg_copy_resource *
2968 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2969 			  struct rte_flow_error *error)
2970 {
2971 	struct mlx5_priv *priv = dev->data->dev_private;
2972 	struct rte_flow_attr attr = {
2973 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2974 		.ingress = 1,
2975 	};
2976 	struct mlx5_rte_flow_item_tag tag_spec = {
2977 		.data = mark_id,
2978 	};
2979 	struct rte_flow_item items[] = {
2980 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2981 	};
2982 	struct rte_flow_action_mark ftag = {
2983 		.id = mark_id,
2984 	};
2985 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2986 		.dst = REG_B,
2987 		.src = 0,
2988 	};
2989 	struct rte_flow_action_jump jump = {
2990 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2991 	};
2992 	struct rte_flow_action actions[] = {
2993 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2994 	};
2995 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2996 	uint32_t idx = 0;
2997 	int ret;
2998 
2999 	/* Fill the register fileds in the flow. */
3000 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3001 	if (ret < 0)
3002 		return NULL;
3003 	tag_spec.id = ret;
3004 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3005 	if (ret < 0)
3006 		return NULL;
3007 	cp_mreg.src = ret;
3008 	/* Check if already registered. */
3009 	MLX5_ASSERT(priv->mreg_cp_tbl);
3010 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3011 	if (mcp_res) {
3012 		/* For non-default rule. */
3013 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3014 			mcp_res->refcnt++;
3015 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3016 			    mcp_res->refcnt == 1);
3017 		return mcp_res;
3018 	}
3019 	/* Provide the full width of FLAG specific value. */
3020 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3021 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3022 	/* Build a new flow. */
3023 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3024 		items[0] = (struct rte_flow_item){
3025 			.type = (enum rte_flow_item_type)
3026 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3027 			.spec = &tag_spec,
3028 		};
3029 		items[1] = (struct rte_flow_item){
3030 			.type = RTE_FLOW_ITEM_TYPE_END,
3031 		};
3032 		actions[0] = (struct rte_flow_action){
3033 			.type = (enum rte_flow_action_type)
3034 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3035 			.conf = &ftag,
3036 		};
3037 		actions[1] = (struct rte_flow_action){
3038 			.type = (enum rte_flow_action_type)
3039 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3040 			.conf = &cp_mreg,
3041 		};
3042 		actions[2] = (struct rte_flow_action){
3043 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3044 			.conf = &jump,
3045 		};
3046 		actions[3] = (struct rte_flow_action){
3047 			.type = RTE_FLOW_ACTION_TYPE_END,
3048 		};
3049 	} else {
3050 		/* Default rule, wildcard match. */
3051 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3052 		items[0] = (struct rte_flow_item){
3053 			.type = RTE_FLOW_ITEM_TYPE_END,
3054 		};
3055 		actions[0] = (struct rte_flow_action){
3056 			.type = (enum rte_flow_action_type)
3057 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3058 			.conf = &cp_mreg,
3059 		};
3060 		actions[1] = (struct rte_flow_action){
3061 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3062 			.conf = &jump,
3063 		};
3064 		actions[2] = (struct rte_flow_action){
3065 			.type = RTE_FLOW_ACTION_TYPE_END,
3066 		};
3067 	}
3068 	/* Build a new entry. */
3069 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3070 	if (!mcp_res) {
3071 		rte_errno = ENOMEM;
3072 		return NULL;
3073 	}
3074 	mcp_res->idx = idx;
3075 	/*
3076 	 * The copy Flows are not included in any list. There
3077 	 * ones are referenced from other Flows and can not
3078 	 * be applied, removed, deleted in ardbitrary order
3079 	 * by list traversing.
3080 	 */
3081 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3082 					 actions, false, error);
3083 	if (!mcp_res->rix_flow)
3084 		goto error;
3085 	mcp_res->refcnt++;
3086 	mcp_res->hlist_ent.key = mark_id;
3087 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3088 				&mcp_res->hlist_ent);
3089 	MLX5_ASSERT(!ret);
3090 	if (ret)
3091 		goto error;
3092 	return mcp_res;
3093 error:
3094 	if (mcp_res->rix_flow)
3095 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3096 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3097 	return NULL;
3098 }
3099 
3100 /**
3101  * Release flow in RX_CP_TBL.
3102  *
3103  * @param dev
3104  *   Pointer to Ethernet device.
3105  * @flow
3106  *   Parent flow for wich copying is provided.
3107  */
3108 static void
3109 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3110 			  struct rte_flow *flow)
3111 {
3112 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3113 	struct mlx5_priv *priv = dev->data->dev_private;
3114 
3115 	if (!flow->rix_mreg_copy)
3116 		return;
3117 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3118 				 flow->rix_mreg_copy);
3119 	if (!mcp_res || !priv->mreg_cp_tbl)
3120 		return;
3121 	if (flow->copy_applied) {
3122 		MLX5_ASSERT(mcp_res->appcnt);
3123 		flow->copy_applied = 0;
3124 		--mcp_res->appcnt;
3125 		if (!mcp_res->appcnt) {
3126 			struct rte_flow *mcp_flow = mlx5_ipool_get
3127 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3128 					mcp_res->rix_flow);
3129 
3130 			if (mcp_flow)
3131 				flow_drv_remove(dev, mcp_flow);
3132 		}
3133 	}
3134 	/*
3135 	 * We do not check availability of metadata registers here,
3136 	 * because copy resources are not allocated in this case.
3137 	 */
3138 	if (--mcp_res->refcnt)
3139 		return;
3140 	MLX5_ASSERT(mcp_res->rix_flow);
3141 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3142 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3143 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3144 	flow->rix_mreg_copy = 0;
3145 }
3146 
3147 /**
3148  * Start flow in RX_CP_TBL.
3149  *
3150  * @param dev
3151  *   Pointer to Ethernet device.
3152  * @flow
3153  *   Parent flow for wich copying is provided.
3154  *
3155  * @return
3156  *   0 on success, a negative errno value otherwise and rte_errno is set.
3157  */
3158 static int
3159 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3160 			    struct rte_flow *flow)
3161 {
3162 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3163 	struct mlx5_priv *priv = dev->data->dev_private;
3164 	int ret;
3165 
3166 	if (!flow->rix_mreg_copy || flow->copy_applied)
3167 		return 0;
3168 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3169 				 flow->rix_mreg_copy);
3170 	if (!mcp_res)
3171 		return 0;
3172 	if (!mcp_res->appcnt) {
3173 		struct rte_flow *mcp_flow = mlx5_ipool_get
3174 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3175 				mcp_res->rix_flow);
3176 
3177 		if (mcp_flow) {
3178 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3179 			if (ret)
3180 				return ret;
3181 		}
3182 	}
3183 	++mcp_res->appcnt;
3184 	flow->copy_applied = 1;
3185 	return 0;
3186 }
3187 
3188 /**
3189  * Stop flow in RX_CP_TBL.
3190  *
3191  * @param dev
3192  *   Pointer to Ethernet device.
3193  * @flow
3194  *   Parent flow for wich copying is provided.
3195  */
3196 static void
3197 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3198 			   struct rte_flow *flow)
3199 {
3200 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3201 	struct mlx5_priv *priv = dev->data->dev_private;
3202 
3203 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3204 		return;
3205 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3206 				 flow->rix_mreg_copy);
3207 	if (!mcp_res)
3208 		return;
3209 	MLX5_ASSERT(mcp_res->appcnt);
3210 	--mcp_res->appcnt;
3211 	flow->copy_applied = 0;
3212 	if (!mcp_res->appcnt) {
3213 		struct rte_flow *mcp_flow = mlx5_ipool_get
3214 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3215 				mcp_res->rix_flow);
3216 
3217 		if (mcp_flow)
3218 			flow_drv_remove(dev, mcp_flow);
3219 	}
3220 }
3221 
3222 /**
3223  * Remove the default copy action from RX_CP_TBL.
3224  *
3225  * @param dev
3226  *   Pointer to Ethernet device.
3227  */
3228 static void
3229 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3230 {
3231 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3232 	struct mlx5_priv *priv = dev->data->dev_private;
3233 
3234 	/* Check if default flow is registered. */
3235 	if (!priv->mreg_cp_tbl)
3236 		return;
3237 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3238 					    MLX5_DEFAULT_COPY_ID);
3239 	if (!mcp_res)
3240 		return;
3241 	MLX5_ASSERT(mcp_res->rix_flow);
3242 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3243 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3244 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3245 }
3246 
3247 /**
3248  * Add the default copy action in in RX_CP_TBL.
3249  *
3250  * @param dev
3251  *   Pointer to Ethernet device.
3252  * @param[out] error
3253  *   Perform verbose error reporting if not NULL.
3254  *
3255  * @return
3256  *   0 for success, negative value otherwise and rte_errno is set.
3257  */
3258 static int
3259 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3260 				  struct rte_flow_error *error)
3261 {
3262 	struct mlx5_priv *priv = dev->data->dev_private;
3263 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3264 
3265 	/* Check whether extensive metadata feature is engaged. */
3266 	if (!priv->config.dv_flow_en ||
3267 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3268 	    !mlx5_flow_ext_mreg_supported(dev) ||
3269 	    !priv->sh->dv_regc0_mask)
3270 		return 0;
3271 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3272 	if (!mcp_res)
3273 		return -rte_errno;
3274 	return 0;
3275 }
3276 
3277 /**
3278  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3279  *
3280  * All the flow having Q/RSS action should be split by
3281  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3282  * performs the following,
3283  *   - CQE->flow_tag := reg_c[1] (MARK)
3284  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3285  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3286  * but there should be a flow per each MARK ID set by MARK action.
3287  *
3288  * For the aforementioned reason, if there's a MARK action in flow's action
3289  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3290  * the MARK ID to CQE's flow_tag like,
3291  *   - If reg_c[1] is mark_id,
3292  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3293  *
3294  * For SET_META action which stores value in reg_c[0], as the destination is
3295  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3296  * MARK ID means the default flow. The default flow looks like,
3297  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3298  *
3299  * @param dev
3300  *   Pointer to Ethernet device.
3301  * @param flow
3302  *   Pointer to flow structure.
3303  * @param[in] actions
3304  *   Pointer to the list of actions.
3305  * @param[out] error
3306  *   Perform verbose error reporting if not NULL.
3307  *
3308  * @return
3309  *   0 on success, negative value otherwise and rte_errno is set.
3310  */
3311 static int
3312 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3313 			    struct rte_flow *flow,
3314 			    const struct rte_flow_action *actions,
3315 			    struct rte_flow_error *error)
3316 {
3317 	struct mlx5_priv *priv = dev->data->dev_private;
3318 	struct mlx5_dev_config *config = &priv->config;
3319 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3320 	const struct rte_flow_action_mark *mark;
3321 
3322 	/* Check whether extensive metadata feature is engaged. */
3323 	if (!config->dv_flow_en ||
3324 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3325 	    !mlx5_flow_ext_mreg_supported(dev) ||
3326 	    !priv->sh->dv_regc0_mask)
3327 		return 0;
3328 	/* Find MARK action. */
3329 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3330 		switch (actions->type) {
3331 		case RTE_FLOW_ACTION_TYPE_FLAG:
3332 			mcp_res = flow_mreg_add_copy_action
3333 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3334 			if (!mcp_res)
3335 				return -rte_errno;
3336 			flow->rix_mreg_copy = mcp_res->idx;
3337 			if (dev->data->dev_started) {
3338 				mcp_res->appcnt++;
3339 				flow->copy_applied = 1;
3340 			}
3341 			return 0;
3342 		case RTE_FLOW_ACTION_TYPE_MARK:
3343 			mark = (const struct rte_flow_action_mark *)
3344 				actions->conf;
3345 			mcp_res =
3346 				flow_mreg_add_copy_action(dev, mark->id, error);
3347 			if (!mcp_res)
3348 				return -rte_errno;
3349 			flow->rix_mreg_copy = mcp_res->idx;
3350 			if (dev->data->dev_started) {
3351 				mcp_res->appcnt++;
3352 				flow->copy_applied = 1;
3353 			}
3354 			return 0;
3355 		default:
3356 			break;
3357 		}
3358 	}
3359 	return 0;
3360 }
3361 
3362 #define MLX5_MAX_SPLIT_ACTIONS 24
3363 #define MLX5_MAX_SPLIT_ITEMS 24
3364 
3365 /**
3366  * Split the hairpin flow.
3367  * Since HW can't support encap on Rx we move the encap to Tx.
3368  * If the count action is after the encap then we also
3369  * move the count action. in this case the count will also measure
3370  * the outer bytes.
3371  *
3372  * @param dev
3373  *   Pointer to Ethernet device.
3374  * @param[in] actions
3375  *   Associated actions (list terminated by the END action).
3376  * @param[out] actions_rx
3377  *   Rx flow actions.
3378  * @param[out] actions_tx
3379  *   Tx flow actions..
3380  * @param[out] pattern_tx
3381  *   The pattern items for the Tx flow.
3382  * @param[out] flow_id
3383  *   The flow ID connected to this flow.
3384  *
3385  * @return
3386  *   0 on success.
3387  */
3388 static int
3389 flow_hairpin_split(struct rte_eth_dev *dev,
3390 		   const struct rte_flow_action actions[],
3391 		   struct rte_flow_action actions_rx[],
3392 		   struct rte_flow_action actions_tx[],
3393 		   struct rte_flow_item pattern_tx[],
3394 		   uint32_t *flow_id)
3395 {
3396 	struct mlx5_priv *priv = dev->data->dev_private;
3397 	const struct rte_flow_action_raw_encap *raw_encap;
3398 	const struct rte_flow_action_raw_decap *raw_decap;
3399 	struct mlx5_rte_flow_action_set_tag *set_tag;
3400 	struct rte_flow_action *tag_action;
3401 	struct mlx5_rte_flow_item_tag *tag_item;
3402 	struct rte_flow_item *item;
3403 	char *addr;
3404 	int encap = 0;
3405 
3406 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3407 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3408 		switch (actions->type) {
3409 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3410 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3411 			rte_memcpy(actions_tx, actions,
3412 			       sizeof(struct rte_flow_action));
3413 			actions_tx++;
3414 			break;
3415 		case RTE_FLOW_ACTION_TYPE_COUNT:
3416 			if (encap) {
3417 				rte_memcpy(actions_tx, actions,
3418 					   sizeof(struct rte_flow_action));
3419 				actions_tx++;
3420 			} else {
3421 				rte_memcpy(actions_rx, actions,
3422 					   sizeof(struct rte_flow_action));
3423 				actions_rx++;
3424 			}
3425 			break;
3426 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3427 			raw_encap = actions->conf;
3428 			if (raw_encap->size >
3429 			    (sizeof(struct rte_flow_item_eth) +
3430 			     sizeof(struct rte_flow_item_ipv4))) {
3431 				memcpy(actions_tx, actions,
3432 				       sizeof(struct rte_flow_action));
3433 				actions_tx++;
3434 				encap = 1;
3435 			} else {
3436 				rte_memcpy(actions_rx, actions,
3437 					   sizeof(struct rte_flow_action));
3438 				actions_rx++;
3439 			}
3440 			break;
3441 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3442 			raw_decap = actions->conf;
3443 			if (raw_decap->size <
3444 			    (sizeof(struct rte_flow_item_eth) +
3445 			     sizeof(struct rte_flow_item_ipv4))) {
3446 				memcpy(actions_tx, actions,
3447 				       sizeof(struct rte_flow_action));
3448 				actions_tx++;
3449 			} else {
3450 				rte_memcpy(actions_rx, actions,
3451 					   sizeof(struct rte_flow_action));
3452 				actions_rx++;
3453 			}
3454 			break;
3455 		default:
3456 			rte_memcpy(actions_rx, actions,
3457 				   sizeof(struct rte_flow_action));
3458 			actions_rx++;
3459 			break;
3460 		}
3461 	}
3462 	/* Add set meta action and end action for the Rx flow. */
3463 	tag_action = actions_rx;
3464 	tag_action->type = (enum rte_flow_action_type)
3465 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3466 	actions_rx++;
3467 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3468 	actions_rx++;
3469 	set_tag = (void *)actions_rx;
3470 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3471 	MLX5_ASSERT(set_tag->id > REG_NONE);
3472 	set_tag->data = *flow_id;
3473 	tag_action->conf = set_tag;
3474 	/* Create Tx item list. */
3475 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3476 	addr = (void *)&pattern_tx[2];
3477 	item = pattern_tx;
3478 	item->type = (enum rte_flow_item_type)
3479 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3480 	tag_item = (void *)addr;
3481 	tag_item->data = *flow_id;
3482 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3483 	MLX5_ASSERT(set_tag->id > REG_NONE);
3484 	item->spec = tag_item;
3485 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3486 	tag_item = (void *)addr;
3487 	tag_item->data = UINT32_MAX;
3488 	tag_item->id = UINT16_MAX;
3489 	item->mask = tag_item;
3490 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3491 	item->last = NULL;
3492 	item++;
3493 	item->type = RTE_FLOW_ITEM_TYPE_END;
3494 	return 0;
3495 }
3496 
3497 /**
3498  * The last stage of splitting chain, just creates the subflow
3499  * without any modification.
3500  *
3501  * @param[in] dev
3502  *   Pointer to Ethernet device.
3503  * @param[in] flow
3504  *   Parent flow structure pointer.
3505  * @param[in, out] sub_flow
3506  *   Pointer to return the created subflow, may be NULL.
3507  * @param[in] prefix_layers
3508  *   Prefix subflow layers, may be 0.
3509  * @param[in] attr
3510  *   Flow rule attributes.
3511  * @param[in] items
3512  *   Pattern specification (list terminated by the END pattern item).
3513  * @param[in] actions
3514  *   Associated actions (list terminated by the END action).
3515  * @param[in] external
3516  *   This flow rule is created by request external to PMD.
3517  * @param[in] flow_idx
3518  *   This memory pool index to the flow.
3519  * @param[out] error
3520  *   Perform verbose error reporting if not NULL.
3521  * @return
3522  *   0 on success, negative value otherwise
3523  */
3524 static int
3525 flow_create_split_inner(struct rte_eth_dev *dev,
3526 			struct rte_flow *flow,
3527 			struct mlx5_flow **sub_flow,
3528 			uint64_t prefix_layers,
3529 			const struct rte_flow_attr *attr,
3530 			const struct rte_flow_item items[],
3531 			const struct rte_flow_action actions[],
3532 			bool external, uint32_t flow_idx,
3533 			struct rte_flow_error *error)
3534 {
3535 	struct mlx5_flow *dev_flow;
3536 
3537 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3538 		flow_idx, error);
3539 	if (!dev_flow)
3540 		return -rte_errno;
3541 	dev_flow->flow = flow;
3542 	dev_flow->external = external;
3543 	/* Subflow object was created, we must include one in the list. */
3544 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3545 		      dev_flow->handle, next);
3546 	/*
3547 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3548 	 * flow may need some user defined item layer flags.
3549 	 */
3550 	if (prefix_layers)
3551 		dev_flow->handle->layers = prefix_layers;
3552 	if (sub_flow)
3553 		*sub_flow = dev_flow;
3554 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3555 }
3556 
3557 /**
3558  * Split the meter flow.
3559  *
3560  * As meter flow will split to three sub flow, other than meter
3561  * action, the other actions make sense to only meter accepts
3562  * the packet. If it need to be dropped, no other additional
3563  * actions should be take.
3564  *
3565  * One kind of special action which decapsulates the L3 tunnel
3566  * header will be in the prefix sub flow, as not to take the
3567  * L3 tunnel header into account.
3568  *
3569  * @param dev
3570  *   Pointer to Ethernet device.
3571  * @param[in] items
3572  *   Pattern specification (list terminated by the END pattern item).
3573  * @param[out] sfx_items
3574  *   Suffix flow match items (list terminated by the END pattern item).
3575  * @param[in] actions
3576  *   Associated actions (list terminated by the END action).
3577  * @param[out] actions_sfx
3578  *   Suffix flow actions.
3579  * @param[out] actions_pre
3580  *   Prefix flow actions.
3581  * @param[out] pattern_sfx
3582  *   The pattern items for the suffix flow.
3583  * @param[out] tag_sfx
3584  *   Pointer to suffix flow tag.
3585  *
3586  * @return
3587  *   0 on success.
3588  */
3589 static int
3590 flow_meter_split_prep(struct rte_eth_dev *dev,
3591 		 const struct rte_flow_item items[],
3592 		 struct rte_flow_item sfx_items[],
3593 		 const struct rte_flow_action actions[],
3594 		 struct rte_flow_action actions_sfx[],
3595 		 struct rte_flow_action actions_pre[])
3596 {
3597 	struct rte_flow_action *tag_action = NULL;
3598 	struct rte_flow_item *tag_item;
3599 	struct mlx5_rte_flow_action_set_tag *set_tag;
3600 	struct rte_flow_error error;
3601 	const struct rte_flow_action_raw_encap *raw_encap;
3602 	const struct rte_flow_action_raw_decap *raw_decap;
3603 	struct mlx5_rte_flow_item_tag *tag_spec;
3604 	struct mlx5_rte_flow_item_tag *tag_mask;
3605 	uint32_t tag_id;
3606 	bool copy_vlan = false;
3607 
3608 	/* Prepare the actions for prefix and suffix flow. */
3609 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3610 		struct rte_flow_action **action_cur = NULL;
3611 
3612 		switch (actions->type) {
3613 		case RTE_FLOW_ACTION_TYPE_METER:
3614 			/* Add the extra tag action first. */
3615 			tag_action = actions_pre;
3616 			tag_action->type = (enum rte_flow_action_type)
3617 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3618 			actions_pre++;
3619 			action_cur = &actions_pre;
3620 			break;
3621 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3622 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3623 			action_cur = &actions_pre;
3624 			break;
3625 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3626 			raw_encap = actions->conf;
3627 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3628 				action_cur = &actions_pre;
3629 			break;
3630 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3631 			raw_decap = actions->conf;
3632 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3633 				action_cur = &actions_pre;
3634 			break;
3635 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3636 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3637 			copy_vlan = true;
3638 			break;
3639 		default:
3640 			break;
3641 		}
3642 		if (!action_cur)
3643 			action_cur = &actions_sfx;
3644 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3645 		(*action_cur)++;
3646 	}
3647 	/* Add end action to the actions. */
3648 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3649 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3650 	actions_pre++;
3651 	/* Set the tag. */
3652 	set_tag = (void *)actions_pre;
3653 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3654 	/*
3655 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3656 	 */
3657 	tag_id = flow_qrss_get_id(dev);
3658 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3659 	assert(tag_action);
3660 	tag_action->conf = set_tag;
3661 	/* Prepare the suffix subflow items. */
3662 	tag_item = sfx_items++;
3663 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3664 		int item_type = items->type;
3665 
3666 		switch (item_type) {
3667 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3668 			memcpy(sfx_items, items, sizeof(*sfx_items));
3669 			sfx_items++;
3670 			break;
3671 		case RTE_FLOW_ITEM_TYPE_VLAN:
3672 			if (copy_vlan) {
3673 				memcpy(sfx_items, items, sizeof(*sfx_items));
3674 				/*
3675 				 * Convert to internal match item, it is used
3676 				 * for vlan push and set vid.
3677 				 */
3678 				sfx_items->type = (enum rte_flow_item_type)
3679 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3680 				sfx_items++;
3681 			}
3682 			break;
3683 		default:
3684 			break;
3685 		}
3686 	}
3687 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3688 	sfx_items++;
3689 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3690 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3691 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3692 	tag_mask = tag_spec + 1;
3693 	tag_mask->data = 0xffffff00;
3694 	tag_item->type = (enum rte_flow_item_type)
3695 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3696 	tag_item->spec = tag_spec;
3697 	tag_item->last = NULL;
3698 	tag_item->mask = tag_mask;
3699 	return tag_id;
3700 }
3701 
3702 /**
3703  * Split action list having QUEUE/RSS for metadata register copy.
3704  *
3705  * Once Q/RSS action is detected in user's action list, the flow action
3706  * should be split in order to copy metadata registers, which will happen in
3707  * RX_CP_TBL like,
3708  *   - CQE->flow_tag := reg_c[1] (MARK)
3709  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3710  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3711  * This is because the last action of each flow must be a terminal action
3712  * (QUEUE, RSS or DROP).
3713  *
3714  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3715  * stored and kept in the mlx5_flow structure per each sub_flow.
3716  *
3717  * The Q/RSS action is replaced with,
3718  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3719  * And the following JUMP action is added at the end,
3720  *   - JUMP, to RX_CP_TBL.
3721  *
3722  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3723  * flow_create_split_metadata() routine. The flow will look like,
3724  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3725  *
3726  * @param dev
3727  *   Pointer to Ethernet device.
3728  * @param[out] split_actions
3729  *   Pointer to store split actions to jump to CP_TBL.
3730  * @param[in] actions
3731  *   Pointer to the list of original flow actions.
3732  * @param[in] qrss
3733  *   Pointer to the Q/RSS action.
3734  * @param[in] actions_n
3735  *   Number of original actions.
3736  * @param[out] error
3737  *   Perform verbose error reporting if not NULL.
3738  *
3739  * @return
3740  *   non-zero unique flow_id on success, otherwise 0 and
3741  *   error/rte_error are set.
3742  */
3743 static uint32_t
3744 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3745 			  struct rte_flow_action *split_actions,
3746 			  const struct rte_flow_action *actions,
3747 			  const struct rte_flow_action *qrss,
3748 			  int actions_n, struct rte_flow_error *error)
3749 {
3750 	struct mlx5_rte_flow_action_set_tag *set_tag;
3751 	struct rte_flow_action_jump *jump;
3752 	const int qrss_idx = qrss - actions;
3753 	uint32_t flow_id = 0;
3754 	int ret = 0;
3755 
3756 	/*
3757 	 * Given actions will be split
3758 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3759 	 * - Add jump to mreg CP_TBL.
3760 	 * As a result, there will be one more action.
3761 	 */
3762 	++actions_n;
3763 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3764 	set_tag = (void *)(split_actions + actions_n);
3765 	/*
3766 	 * If tag action is not set to void(it means we are not the meter
3767 	 * suffix flow), add the tag action. Since meter suffix flow already
3768 	 * has the tag added.
3769 	 */
3770 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3771 		/*
3772 		 * Allocate the new subflow ID. This one is unique within
3773 		 * device and not shared with representors. Otherwise,
3774 		 * we would have to resolve multi-thread access synch
3775 		 * issue. Each flow on the shared device is appended
3776 		 * with source vport identifier, so the resulting
3777 		 * flows will be unique in the shared (by master and
3778 		 * representors) domain even if they have coinciding
3779 		 * IDs.
3780 		 */
3781 		flow_id = flow_qrss_get_id(dev);
3782 		if (!flow_id)
3783 			return rte_flow_error_set(error, ENOMEM,
3784 						  RTE_FLOW_ERROR_TYPE_ACTION,
3785 						  NULL, "can't allocate id "
3786 						  "for split Q/RSS subflow");
3787 		/* Internal SET_TAG action to set flow ID. */
3788 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3789 			.data = flow_id,
3790 		};
3791 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3792 		if (ret < 0)
3793 			return ret;
3794 		set_tag->id = ret;
3795 		/* Construct new actions array. */
3796 		/* Replace QUEUE/RSS action. */
3797 		split_actions[qrss_idx] = (struct rte_flow_action){
3798 			.type = (enum rte_flow_action_type)
3799 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3800 			.conf = set_tag,
3801 		};
3802 	}
3803 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3804 	jump = (void *)(set_tag + 1);
3805 	*jump = (struct rte_flow_action_jump){
3806 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3807 	};
3808 	split_actions[actions_n - 2] = (struct rte_flow_action){
3809 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3810 		.conf = jump,
3811 	};
3812 	split_actions[actions_n - 1] = (struct rte_flow_action){
3813 		.type = RTE_FLOW_ACTION_TYPE_END,
3814 	};
3815 	return flow_id;
3816 }
3817 
3818 /**
3819  * Extend the given action list for Tx metadata copy.
3820  *
3821  * Copy the given action list to the ext_actions and add flow metadata register
3822  * copy action in order to copy reg_a set by WQE to reg_c[0].
3823  *
3824  * @param[out] ext_actions
3825  *   Pointer to the extended action list.
3826  * @param[in] actions
3827  *   Pointer to the list of actions.
3828  * @param[in] actions_n
3829  *   Number of actions in the list.
3830  * @param[out] error
3831  *   Perform verbose error reporting if not NULL.
3832  * @param[in] encap_idx
3833  *   The encap action inndex.
3834  *
3835  * @return
3836  *   0 on success, negative value otherwise
3837  */
3838 static int
3839 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3840 		       struct rte_flow_action *ext_actions,
3841 		       const struct rte_flow_action *actions,
3842 		       int actions_n, struct rte_flow_error *error,
3843 		       int encap_idx)
3844 {
3845 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3846 		(struct mlx5_flow_action_copy_mreg *)
3847 			(ext_actions + actions_n + 1);
3848 	int ret;
3849 
3850 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3851 	if (ret < 0)
3852 		return ret;
3853 	cp_mreg->dst = ret;
3854 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3855 	if (ret < 0)
3856 		return ret;
3857 	cp_mreg->src = ret;
3858 	if (encap_idx != 0)
3859 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3860 	if (encap_idx == actions_n - 1) {
3861 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3862 			.type = (enum rte_flow_action_type)
3863 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3864 			.conf = cp_mreg,
3865 		};
3866 		ext_actions[actions_n] = (struct rte_flow_action){
3867 			.type = RTE_FLOW_ACTION_TYPE_END,
3868 		};
3869 	} else {
3870 		ext_actions[encap_idx] = (struct rte_flow_action){
3871 			.type = (enum rte_flow_action_type)
3872 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3873 			.conf = cp_mreg,
3874 		};
3875 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3876 				sizeof(*ext_actions) * (actions_n - encap_idx));
3877 	}
3878 	return 0;
3879 }
3880 
3881 /**
3882  * The splitting for metadata feature.
3883  *
3884  * - Q/RSS action on NIC Rx should be split in order to pass by
3885  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3886  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3887  *
3888  * - All the actions on NIC Tx should have a mreg copy action to
3889  *   copy reg_a from WQE to reg_c[0].
3890  *
3891  * @param dev
3892  *   Pointer to Ethernet device.
3893  * @param[in] flow
3894  *   Parent flow structure pointer.
3895  * @param[in] prefix_layers
3896  *   Prefix flow layer flags.
3897  * @param[in] attr
3898  *   Flow rule attributes.
3899  * @param[in] items
3900  *   Pattern specification (list terminated by the END pattern item).
3901  * @param[in] actions
3902  *   Associated actions (list terminated by the END action).
3903  * @param[in] external
3904  *   This flow rule is created by request external to PMD.
3905  * @param[in] flow_idx
3906  *   This memory pool index to the flow.
3907  * @param[out] error
3908  *   Perform verbose error reporting if not NULL.
3909  * @return
3910  *   0 on success, negative value otherwise
3911  */
3912 static int
3913 flow_create_split_metadata(struct rte_eth_dev *dev,
3914 			   struct rte_flow *flow,
3915 			   uint64_t prefix_layers,
3916 			   const struct rte_flow_attr *attr,
3917 			   const struct rte_flow_item items[],
3918 			   const struct rte_flow_action actions[],
3919 			   bool external, uint32_t flow_idx,
3920 			   struct rte_flow_error *error)
3921 {
3922 	struct mlx5_priv *priv = dev->data->dev_private;
3923 	struct mlx5_dev_config *config = &priv->config;
3924 	const struct rte_flow_action *qrss = NULL;
3925 	struct rte_flow_action *ext_actions = NULL;
3926 	struct mlx5_flow *dev_flow = NULL;
3927 	uint32_t qrss_id = 0;
3928 	int mtr_sfx = 0;
3929 	size_t act_size;
3930 	int actions_n;
3931 	int encap_idx;
3932 	int ret;
3933 
3934 	/* Check whether extensive metadata feature is engaged. */
3935 	if (!config->dv_flow_en ||
3936 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3937 	    !mlx5_flow_ext_mreg_supported(dev))
3938 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3939 					       attr, items, actions, external,
3940 					       flow_idx, error);
3941 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3942 							   &encap_idx);
3943 	if (qrss) {
3944 		/* Exclude hairpin flows from splitting. */
3945 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3946 			const struct rte_flow_action_queue *queue;
3947 
3948 			queue = qrss->conf;
3949 			if (mlx5_rxq_get_type(dev, queue->index) ==
3950 			    MLX5_RXQ_TYPE_HAIRPIN)
3951 				qrss = NULL;
3952 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3953 			const struct rte_flow_action_rss *rss;
3954 
3955 			rss = qrss->conf;
3956 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3957 			    MLX5_RXQ_TYPE_HAIRPIN)
3958 				qrss = NULL;
3959 		}
3960 	}
3961 	if (qrss) {
3962 		/* Check if it is in meter suffix table. */
3963 		mtr_sfx = attr->group == (attr->transfer ?
3964 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3965 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3966 		/*
3967 		 * Q/RSS action on NIC Rx should be split in order to pass by
3968 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3969 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3970 		 */
3971 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3972 			   sizeof(struct rte_flow_action_set_tag) +
3973 			   sizeof(struct rte_flow_action_jump);
3974 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3975 		if (!ext_actions)
3976 			return rte_flow_error_set(error, ENOMEM,
3977 						  RTE_FLOW_ERROR_TYPE_ACTION,
3978 						  NULL, "no memory to split "
3979 						  "metadata flow");
3980 		/*
3981 		 * If we are the suffix flow of meter, tag already exist.
3982 		 * Set the tag action to void.
3983 		 */
3984 		if (mtr_sfx)
3985 			ext_actions[qrss - actions].type =
3986 						RTE_FLOW_ACTION_TYPE_VOID;
3987 		else
3988 			ext_actions[qrss - actions].type =
3989 						(enum rte_flow_action_type)
3990 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3991 		/*
3992 		 * Create the new actions list with removed Q/RSS action
3993 		 * and appended set tag and jump to register copy table
3994 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3995 		 * in advance, because it is needed for set tag action.
3996 		 */
3997 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3998 						    qrss, actions_n, error);
3999 		if (!mtr_sfx && !qrss_id) {
4000 			ret = -rte_errno;
4001 			goto exit;
4002 		}
4003 	} else if (attr->egress && !attr->transfer) {
4004 		/*
4005 		 * All the actions on NIC Tx should have a metadata register
4006 		 * copy action to copy reg_a from WQE to reg_c[meta]
4007 		 */
4008 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4009 			   sizeof(struct mlx5_flow_action_copy_mreg);
4010 		ext_actions = rte_zmalloc(__func__, act_size, 0);
4011 		if (!ext_actions)
4012 			return rte_flow_error_set(error, ENOMEM,
4013 						  RTE_FLOW_ERROR_TYPE_ACTION,
4014 						  NULL, "no memory to split "
4015 						  "metadata flow");
4016 		/* Create the action list appended with copy register. */
4017 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4018 					     actions_n, error, encap_idx);
4019 		if (ret < 0)
4020 			goto exit;
4021 	}
4022 	/* Add the unmodified original or prefix subflow. */
4023 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4024 				      items, ext_actions ? ext_actions :
4025 				      actions, external, flow_idx, error);
4026 	if (ret < 0)
4027 		goto exit;
4028 	MLX5_ASSERT(dev_flow);
4029 	if (qrss) {
4030 		const struct rte_flow_attr q_attr = {
4031 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4032 			.ingress = 1,
4033 		};
4034 		/* Internal PMD action to set register. */
4035 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4036 			.data = qrss_id,
4037 			.id = 0,
4038 		};
4039 		struct rte_flow_item q_items[] = {
4040 			{
4041 				.type = (enum rte_flow_item_type)
4042 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4043 				.spec = &q_tag_spec,
4044 				.last = NULL,
4045 				.mask = NULL,
4046 			},
4047 			{
4048 				.type = RTE_FLOW_ITEM_TYPE_END,
4049 			},
4050 		};
4051 		struct rte_flow_action q_actions[] = {
4052 			{
4053 				.type = qrss->type,
4054 				.conf = qrss->conf,
4055 			},
4056 			{
4057 				.type = RTE_FLOW_ACTION_TYPE_END,
4058 			},
4059 		};
4060 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4061 
4062 		/*
4063 		 * Configure the tag item only if there is no meter subflow.
4064 		 * Since tag is already marked in the meter suffix subflow
4065 		 * we can just use the meter suffix items as is.
4066 		 */
4067 		if (qrss_id) {
4068 			/* Not meter subflow. */
4069 			MLX5_ASSERT(!mtr_sfx);
4070 			/*
4071 			 * Put unique id in prefix flow due to it is destroyed
4072 			 * after suffix flow and id will be freed after there
4073 			 * is no actual flows with this id and identifier
4074 			 * reallocation becomes possible (for example, for
4075 			 * other flows in other threads).
4076 			 */
4077 			dev_flow->handle->split_flow_id = qrss_id;
4078 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4079 						   error);
4080 			if (ret < 0)
4081 				goto exit;
4082 			q_tag_spec.id = ret;
4083 		}
4084 		dev_flow = NULL;
4085 		/* Add suffix subflow to execute Q/RSS. */
4086 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4087 					      &q_attr, mtr_sfx ? items :
4088 					      q_items, q_actions,
4089 					      external, flow_idx, error);
4090 		if (ret < 0)
4091 			goto exit;
4092 		/* qrss ID should be freed if failed. */
4093 		qrss_id = 0;
4094 		MLX5_ASSERT(dev_flow);
4095 	}
4096 
4097 exit:
4098 	/*
4099 	 * We do not destroy the partially created sub_flows in case of error.
4100 	 * These ones are included into parent flow list and will be destroyed
4101 	 * by flow_drv_destroy.
4102 	 */
4103 	flow_qrss_free_id(dev, qrss_id);
4104 	rte_free(ext_actions);
4105 	return ret;
4106 }
4107 
4108 /**
4109  * The splitting for meter feature.
4110  *
4111  * - The meter flow will be split to two flows as prefix and
4112  *   suffix flow. The packets make sense only it pass the prefix
4113  *   meter action.
4114  *
4115  * - Reg_C_5 is used for the packet to match betweend prefix and
4116  *   suffix flow.
4117  *
4118  * @param dev
4119  *   Pointer to Ethernet device.
4120  * @param[in] flow
4121  *   Parent flow structure pointer.
4122  * @param[in] attr
4123  *   Flow rule attributes.
4124  * @param[in] items
4125  *   Pattern specification (list terminated by the END pattern item).
4126  * @param[in] actions
4127  *   Associated actions (list terminated by the END action).
4128  * @param[in] external
4129  *   This flow rule is created by request external to PMD.
4130  * @param[in] flow_idx
4131  *   This memory pool index to the flow.
4132  * @param[out] error
4133  *   Perform verbose error reporting if not NULL.
4134  * @return
4135  *   0 on success, negative value otherwise
4136  */
4137 static int
4138 flow_create_split_meter(struct rte_eth_dev *dev,
4139 			   struct rte_flow *flow,
4140 			   const struct rte_flow_attr *attr,
4141 			   const struct rte_flow_item items[],
4142 			   const struct rte_flow_action actions[],
4143 			   bool external, uint32_t flow_idx,
4144 			   struct rte_flow_error *error)
4145 {
4146 	struct mlx5_priv *priv = dev->data->dev_private;
4147 	struct rte_flow_action *sfx_actions = NULL;
4148 	struct rte_flow_action *pre_actions = NULL;
4149 	struct rte_flow_item *sfx_items = NULL;
4150 	struct mlx5_flow *dev_flow = NULL;
4151 	struct rte_flow_attr sfx_attr = *attr;
4152 	uint32_t mtr = 0;
4153 	uint32_t mtr_tag_id = 0;
4154 	size_t act_size;
4155 	size_t item_size;
4156 	int actions_n = 0;
4157 	int ret;
4158 
4159 	if (priv->mtr_en)
4160 		actions_n = flow_check_meter_action(actions, &mtr);
4161 	if (mtr) {
4162 		/* The five prefix actions: meter, decap, encap, tag, end. */
4163 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4164 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4165 		/* tag, vlan, port id, end. */
4166 #define METER_SUFFIX_ITEM 4
4167 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4168 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4169 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4170 		if (!sfx_actions)
4171 			return rte_flow_error_set(error, ENOMEM,
4172 						  RTE_FLOW_ERROR_TYPE_ACTION,
4173 						  NULL, "no memory to split "
4174 						  "meter flow");
4175 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4176 			     act_size);
4177 		pre_actions = sfx_actions + actions_n;
4178 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4179 						   actions, sfx_actions,
4180 						   pre_actions);
4181 		if (!mtr_tag_id) {
4182 			ret = -rte_errno;
4183 			goto exit;
4184 		}
4185 		/* Add the prefix subflow. */
4186 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4187 					      items, pre_actions, external,
4188 					      flow_idx, error);
4189 		if (ret) {
4190 			ret = -rte_errno;
4191 			goto exit;
4192 		}
4193 		dev_flow->handle->split_flow_id = mtr_tag_id;
4194 		/* Setting the sfx group atrr. */
4195 		sfx_attr.group = sfx_attr.transfer ?
4196 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4197 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4198 	}
4199 	/* Add the prefix subflow. */
4200 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4201 					 flow_get_prefix_layer_flags(dev_flow) :
4202 					 0, &sfx_attr,
4203 					 sfx_items ? sfx_items : items,
4204 					 sfx_actions ? sfx_actions : actions,
4205 					 external, flow_idx, error);
4206 exit:
4207 	if (sfx_actions)
4208 		rte_free(sfx_actions);
4209 	return ret;
4210 }
4211 
4212 /**
4213  * Split the flow to subflow set. The splitters might be linked
4214  * in the chain, like this:
4215  * flow_create_split_outer() calls:
4216  *   flow_create_split_meter() calls:
4217  *     flow_create_split_metadata(meter_subflow_0) calls:
4218  *       flow_create_split_inner(metadata_subflow_0)
4219  *       flow_create_split_inner(metadata_subflow_1)
4220  *       flow_create_split_inner(metadata_subflow_2)
4221  *     flow_create_split_metadata(meter_subflow_1) calls:
4222  *       flow_create_split_inner(metadata_subflow_0)
4223  *       flow_create_split_inner(metadata_subflow_1)
4224  *       flow_create_split_inner(metadata_subflow_2)
4225  *
4226  * This provide flexible way to add new levels of flow splitting.
4227  * The all of successfully created subflows are included to the
4228  * parent flow dev_flow list.
4229  *
4230  * @param dev
4231  *   Pointer to Ethernet device.
4232  * @param[in] flow
4233  *   Parent flow structure pointer.
4234  * @param[in] attr
4235  *   Flow rule attributes.
4236  * @param[in] items
4237  *   Pattern specification (list terminated by the END pattern item).
4238  * @param[in] actions
4239  *   Associated actions (list terminated by the END action).
4240  * @param[in] external
4241  *   This flow rule is created by request external to PMD.
4242  * @param[in] flow_idx
4243  *   This memory pool index to the flow.
4244  * @param[out] error
4245  *   Perform verbose error reporting if not NULL.
4246  * @return
4247  *   0 on success, negative value otherwise
4248  */
4249 static int
4250 flow_create_split_outer(struct rte_eth_dev *dev,
4251 			struct rte_flow *flow,
4252 			const struct rte_flow_attr *attr,
4253 			const struct rte_flow_item items[],
4254 			const struct rte_flow_action actions[],
4255 			bool external, uint32_t flow_idx,
4256 			struct rte_flow_error *error)
4257 {
4258 	int ret;
4259 
4260 	ret = flow_create_split_meter(dev, flow, attr, items,
4261 					 actions, external, flow_idx, error);
4262 	MLX5_ASSERT(ret <= 0);
4263 	return ret;
4264 }
4265 
4266 /**
4267  * Create a flow and add it to @p list.
4268  *
4269  * @param dev
4270  *   Pointer to Ethernet device.
4271  * @param list
4272  *   Pointer to a TAILQ flow list. If this parameter NULL,
4273  *   no list insertion occurred, flow is just created,
4274  *   this is caller's responsibility to track the
4275  *   created flow.
4276  * @param[in] attr
4277  *   Flow rule attributes.
4278  * @param[in] items
4279  *   Pattern specification (list terminated by the END pattern item).
4280  * @param[in] actions
4281  *   Associated actions (list terminated by the END action).
4282  * @param[in] external
4283  *   This flow rule is created by request external to PMD.
4284  * @param[out] error
4285  *   Perform verbose error reporting if not NULL.
4286  *
4287  * @return
4288  *   A flow index on success, 0 otherwise and rte_errno is set.
4289  */
4290 static uint32_t
4291 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4292 		 const struct rte_flow_attr *attr,
4293 		 const struct rte_flow_item items[],
4294 		 const struct rte_flow_action actions[],
4295 		 bool external, struct rte_flow_error *error)
4296 {
4297 	struct mlx5_priv *priv = dev->data->dev_private;
4298 	struct rte_flow *flow = NULL;
4299 	struct mlx5_flow *dev_flow;
4300 	const struct rte_flow_action_rss *rss;
4301 	union {
4302 		struct rte_flow_expand_rss buf;
4303 		uint8_t buffer[2048];
4304 	} expand_buffer;
4305 	union {
4306 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4307 		uint8_t buffer[2048];
4308 	} actions_rx;
4309 	union {
4310 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4311 		uint8_t buffer[2048];
4312 	} actions_hairpin_tx;
4313 	union {
4314 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4315 		uint8_t buffer[2048];
4316 	} items_tx;
4317 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4318 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4319 					      priv->rss_desc)[!!priv->flow_idx];
4320 	const struct rte_flow_action *p_actions_rx = actions;
4321 	uint32_t i;
4322 	uint32_t idx = 0;
4323 	int hairpin_flow;
4324 	uint32_t hairpin_id = 0;
4325 	struct rte_flow_attr attr_tx = { .priority = 0 };
4326 	int ret;
4327 
4328 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4329 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4330 				external, hairpin_flow, error);
4331 	if (ret < 0)
4332 		return 0;
4333 	if (hairpin_flow > 0) {
4334 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4335 			rte_errno = EINVAL;
4336 			return 0;
4337 		}
4338 		flow_hairpin_split(dev, actions, actions_rx.actions,
4339 				   actions_hairpin_tx.actions, items_tx.items,
4340 				   &hairpin_id);
4341 		p_actions_rx = actions_rx.actions;
4342 	}
4343 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4344 	if (!flow) {
4345 		rte_errno = ENOMEM;
4346 		goto error_before_flow;
4347 	}
4348 	flow->drv_type = flow_get_drv_type(dev, attr);
4349 	if (hairpin_id != 0)
4350 		flow->hairpin_flow_id = hairpin_id;
4351 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4352 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4353 	memset(rss_desc, 0, sizeof(*rss_desc));
4354 	rss = flow_get_rss_action(p_actions_rx);
4355 	if (rss) {
4356 		/*
4357 		 * The following information is required by
4358 		 * mlx5_flow_hashfields_adjust() in advance.
4359 		 */
4360 		rss_desc->level = rss->level;
4361 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4362 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4363 	}
4364 	flow->dev_handles = 0;
4365 	if (rss && rss->types) {
4366 		unsigned int graph_root;
4367 
4368 		graph_root = find_graph_root(items, rss->level);
4369 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4370 					  items, rss->types,
4371 					  mlx5_support_expansion,
4372 					  graph_root);
4373 		MLX5_ASSERT(ret > 0 &&
4374 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4375 	} else {
4376 		buf->entries = 1;
4377 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4378 	}
4379 	/*
4380 	 * Record the start index when there is a nested call. All sub-flows
4381 	 * need to be translated before another calling.
4382 	 * No need to use ping-pong buffer to save memory here.
4383 	 */
4384 	if (priv->flow_idx) {
4385 		MLX5_ASSERT(!priv->flow_nested_idx);
4386 		priv->flow_nested_idx = priv->flow_idx;
4387 	}
4388 	for (i = 0; i < buf->entries; ++i) {
4389 		/*
4390 		 * The splitter may create multiple dev_flows,
4391 		 * depending on configuration. In the simplest
4392 		 * case it just creates unmodified original flow.
4393 		 */
4394 		ret = flow_create_split_outer(dev, flow, attr,
4395 					      buf->entry[i].pattern,
4396 					      p_actions_rx, external, idx,
4397 					      error);
4398 		if (ret < 0)
4399 			goto error;
4400 	}
4401 	/* Create the tx flow. */
4402 	if (hairpin_flow) {
4403 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4404 		attr_tx.ingress = 0;
4405 		attr_tx.egress = 1;
4406 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4407 					 actions_hairpin_tx.actions,
4408 					 idx, error);
4409 		if (!dev_flow)
4410 			goto error;
4411 		dev_flow->flow = flow;
4412 		dev_flow->external = 0;
4413 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4414 			      dev_flow->handle, next);
4415 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4416 					 items_tx.items,
4417 					 actions_hairpin_tx.actions, error);
4418 		if (ret < 0)
4419 			goto error;
4420 	}
4421 	/*
4422 	 * Update the metadata register copy table. If extensive
4423 	 * metadata feature is enabled and registers are supported
4424 	 * we might create the extra rte_flow for each unique
4425 	 * MARK/FLAG action ID.
4426 	 *
4427 	 * The table is updated for ingress Flows only, because
4428 	 * the egress Flows belong to the different device and
4429 	 * copy table should be updated in peer NIC Rx domain.
4430 	 */
4431 	if (attr->ingress &&
4432 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4433 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4434 		if (ret)
4435 			goto error;
4436 	}
4437 	/*
4438 	 * If the flow is external (from application) OR device is started, then
4439 	 * the flow will be applied immediately.
4440 	 */
4441 	if (external || dev->data->dev_started) {
4442 		ret = flow_drv_apply(dev, flow, error);
4443 		if (ret < 0)
4444 			goto error;
4445 	}
4446 	if (list)
4447 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4448 			     flow, next);
4449 	flow_rxq_flags_set(dev, flow);
4450 	/* Nested flow creation index recovery. */
4451 	priv->flow_idx = priv->flow_nested_idx;
4452 	if (priv->flow_nested_idx)
4453 		priv->flow_nested_idx = 0;
4454 	return idx;
4455 error:
4456 	MLX5_ASSERT(flow);
4457 	ret = rte_errno; /* Save rte_errno before cleanup. */
4458 	flow_mreg_del_copy_action(dev, flow);
4459 	flow_drv_destroy(dev, flow);
4460 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4461 	rte_errno = ret; /* Restore rte_errno. */
4462 error_before_flow:
4463 	ret = rte_errno;
4464 	if (hairpin_id)
4465 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4466 				     hairpin_id);
4467 	rte_errno = ret;
4468 	priv->flow_idx = priv->flow_nested_idx;
4469 	if (priv->flow_nested_idx)
4470 		priv->flow_nested_idx = 0;
4471 	return 0;
4472 }
4473 
4474 /**
4475  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4476  * incoming packets to table 1.
4477  *
4478  * Other flow rules, requested for group n, will be created in
4479  * e-switch table n+1.
4480  * Jump action to e-switch group n will be created to group n+1.
4481  *
4482  * Used when working in switchdev mode, to utilise advantages of table 1
4483  * and above.
4484  *
4485  * @param dev
4486  *   Pointer to Ethernet device.
4487  *
4488  * @return
4489  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4490  */
4491 struct rte_flow *
4492 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4493 {
4494 	const struct rte_flow_attr attr = {
4495 		.group = 0,
4496 		.priority = 0,
4497 		.ingress = 1,
4498 		.egress = 0,
4499 		.transfer = 1,
4500 	};
4501 	const struct rte_flow_item pattern = {
4502 		.type = RTE_FLOW_ITEM_TYPE_END,
4503 	};
4504 	struct rte_flow_action_jump jump = {
4505 		.group = 1,
4506 	};
4507 	const struct rte_flow_action actions[] = {
4508 		{
4509 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4510 			.conf = &jump,
4511 		},
4512 		{
4513 			.type = RTE_FLOW_ACTION_TYPE_END,
4514 		},
4515 	};
4516 	struct mlx5_priv *priv = dev->data->dev_private;
4517 	struct rte_flow_error error;
4518 
4519 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4520 						   &attr, &pattern,
4521 						   actions, false, &error);
4522 }
4523 
4524 /**
4525  * Validate a flow supported by the NIC.
4526  *
4527  * @see rte_flow_validate()
4528  * @see rte_flow_ops
4529  */
4530 int
4531 mlx5_flow_validate(struct rte_eth_dev *dev,
4532 		   const struct rte_flow_attr *attr,
4533 		   const struct rte_flow_item items[],
4534 		   const struct rte_flow_action actions[],
4535 		   struct rte_flow_error *error)
4536 {
4537 	int hairpin_flow;
4538 
4539 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4540 	return flow_drv_validate(dev, attr, items, actions,
4541 				true, hairpin_flow, error);
4542 }
4543 
4544 /**
4545  * Create a flow.
4546  *
4547  * @see rte_flow_create()
4548  * @see rte_flow_ops
4549  */
4550 struct rte_flow *
4551 mlx5_flow_create(struct rte_eth_dev *dev,
4552 		 const struct rte_flow_attr *attr,
4553 		 const struct rte_flow_item items[],
4554 		 const struct rte_flow_action actions[],
4555 		 struct rte_flow_error *error)
4556 {
4557 	struct mlx5_priv *priv = dev->data->dev_private;
4558 
4559 	/*
4560 	 * If the device is not started yet, it is not allowed to created a
4561 	 * flow from application. PMD default flows and traffic control flows
4562 	 * are not affected.
4563 	 */
4564 	if (unlikely(!dev->data->dev_started)) {
4565 		DRV_LOG(DEBUG, "port %u is not started when "
4566 			"inserting a flow", dev->data->port_id);
4567 		rte_flow_error_set(error, ENODEV,
4568 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4569 				   NULL,
4570 				   "port not started");
4571 		return NULL;
4572 	}
4573 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4574 				  attr, items, actions, true, error);
4575 }
4576 
4577 /**
4578  * Destroy a flow in a list.
4579  *
4580  * @param dev
4581  *   Pointer to Ethernet device.
4582  * @param list
4583  *   Pointer to the Indexed flow list. If this parameter NULL,
4584  *   there is no flow removal from the list. Be noted that as
4585  *   flow is add to the indexed list, memory of the indexed
4586  *   list points to maybe changed as flow destroyed.
4587  * @param[in] flow_idx
4588  *   Index of flow to destroy.
4589  */
4590 static void
4591 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4592 		  uint32_t flow_idx)
4593 {
4594 	struct mlx5_priv *priv = dev->data->dev_private;
4595 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4596 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4597 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4598 
4599 	if (!flow)
4600 		return;
4601 	/*
4602 	 * Update RX queue flags only if port is started, otherwise it is
4603 	 * already clean.
4604 	 */
4605 	if (dev->data->dev_started)
4606 		flow_rxq_flags_trim(dev, flow);
4607 	if (flow->hairpin_flow_id)
4608 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4609 				     flow->hairpin_flow_id);
4610 	flow_drv_destroy(dev, flow);
4611 	if (list)
4612 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4613 			     flow_idx, flow, next);
4614 	flow_mreg_del_copy_action(dev, flow);
4615 	if (flow->fdir) {
4616 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4617 			if (priv_fdir_flow->rix_flow == flow_idx)
4618 				break;
4619 		}
4620 		if (priv_fdir_flow) {
4621 			LIST_REMOVE(priv_fdir_flow, next);
4622 			rte_free(priv_fdir_flow->fdir);
4623 			rte_free(priv_fdir_flow);
4624 		}
4625 	}
4626 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4627 }
4628 
4629 /**
4630  * Destroy all flows.
4631  *
4632  * @param dev
4633  *   Pointer to Ethernet device.
4634  * @param list
4635  *   Pointer to the Indexed flow list.
4636  * @param active
4637  *   If flushing is called avtively.
4638  */
4639 void
4640 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4641 {
4642 	uint32_t num_flushed = 0;
4643 
4644 	while (*list) {
4645 		flow_list_destroy(dev, list, *list);
4646 		num_flushed++;
4647 	}
4648 	if (active) {
4649 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4650 			dev->data->port_id, num_flushed);
4651 	}
4652 }
4653 
4654 /**
4655  * Remove all flows.
4656  *
4657  * @param dev
4658  *   Pointer to Ethernet device.
4659  * @param list
4660  *   Pointer to the Indexed flow list.
4661  */
4662 void
4663 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4664 {
4665 	struct mlx5_priv *priv = dev->data->dev_private;
4666 	struct rte_flow *flow = NULL;
4667 	uint32_t idx;
4668 
4669 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4670 		      flow, next) {
4671 		flow_drv_remove(dev, flow);
4672 		flow_mreg_stop_copy_action(dev, flow);
4673 	}
4674 	flow_mreg_del_default_copy_action(dev);
4675 	flow_rxq_flags_clear(dev);
4676 }
4677 
4678 /**
4679  * Add all flows.
4680  *
4681  * @param dev
4682  *   Pointer to Ethernet device.
4683  * @param list
4684  *   Pointer to the Indexed flow list.
4685  *
4686  * @return
4687  *   0 on success, a negative errno value otherwise and rte_errno is set.
4688  */
4689 int
4690 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4691 {
4692 	struct mlx5_priv *priv = dev->data->dev_private;
4693 	struct rte_flow *flow = NULL;
4694 	struct rte_flow_error error;
4695 	uint32_t idx;
4696 	int ret = 0;
4697 
4698 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4699 	ret = flow_mreg_add_default_copy_action(dev, &error);
4700 	if (ret < 0)
4701 		return -rte_errno;
4702 	/* Apply Flows created by application. */
4703 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4704 		      flow, next) {
4705 		ret = flow_mreg_start_copy_action(dev, flow);
4706 		if (ret < 0)
4707 			goto error;
4708 		ret = flow_drv_apply(dev, flow, &error);
4709 		if (ret < 0)
4710 			goto error;
4711 		flow_rxq_flags_set(dev, flow);
4712 	}
4713 	return 0;
4714 error:
4715 	ret = rte_errno; /* Save rte_errno before cleanup. */
4716 	mlx5_flow_stop(dev, list);
4717 	rte_errno = ret; /* Restore rte_errno. */
4718 	return -rte_errno;
4719 }
4720 
4721 /**
4722  * Stop all default actions for flows.
4723  *
4724  * @param dev
4725  *   Pointer to Ethernet device.
4726  */
4727 void
4728 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4729 {
4730 	flow_mreg_del_default_copy_action(dev);
4731 	flow_rxq_flags_clear(dev);
4732 }
4733 
4734 /**
4735  * Start all default actions for flows.
4736  *
4737  * @param dev
4738  *   Pointer to Ethernet device.
4739  * @return
4740  *   0 on success, a negative errno value otherwise and rte_errno is set.
4741  */
4742 int
4743 mlx5_flow_start_default(struct rte_eth_dev *dev)
4744 {
4745 	struct rte_flow_error error;
4746 
4747 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4748 	return flow_mreg_add_default_copy_action(dev, &error);
4749 }
4750 
4751 /**
4752  * Allocate intermediate resources for flow creation.
4753  *
4754  * @param dev
4755  *   Pointer to Ethernet device.
4756  */
4757 void
4758 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4759 {
4760 	struct mlx5_priv *priv = dev->data->dev_private;
4761 
4762 	if (!priv->inter_flows) {
4763 		priv->inter_flows = rte_calloc(__func__, 1,
4764 				    MLX5_NUM_MAX_DEV_FLOWS *
4765 				    sizeof(struct mlx5_flow) +
4766 				    (sizeof(struct mlx5_flow_rss_desc) +
4767 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4768 		if (!priv->inter_flows) {
4769 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4770 			return;
4771 		}
4772 	}
4773 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4774 			 [MLX5_NUM_MAX_DEV_FLOWS];
4775 	/* Reset the index. */
4776 	priv->flow_idx = 0;
4777 	priv->flow_nested_idx = 0;
4778 }
4779 
4780 /**
4781  * Free intermediate resources for flows.
4782  *
4783  * @param dev
4784  *   Pointer to Ethernet device.
4785  */
4786 void
4787 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4788 {
4789 	struct mlx5_priv *priv = dev->data->dev_private;
4790 
4791 	rte_free(priv->inter_flows);
4792 	priv->inter_flows = NULL;
4793 }
4794 
4795 /**
4796  * Verify the flow list is empty
4797  *
4798  * @param dev
4799  *  Pointer to Ethernet device.
4800  *
4801  * @return the number of flows not released.
4802  */
4803 int
4804 mlx5_flow_verify(struct rte_eth_dev *dev)
4805 {
4806 	struct mlx5_priv *priv = dev->data->dev_private;
4807 	struct rte_flow *flow;
4808 	uint32_t idx;
4809 	int ret = 0;
4810 
4811 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4812 		      flow, next) {
4813 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4814 			dev->data->port_id, (void *)flow);
4815 		++ret;
4816 	}
4817 	return ret;
4818 }
4819 
4820 /**
4821  * Enable default hairpin egress flow.
4822  *
4823  * @param dev
4824  *   Pointer to Ethernet device.
4825  * @param queue
4826  *   The queue index.
4827  *
4828  * @return
4829  *   0 on success, a negative errno value otherwise and rte_errno is set.
4830  */
4831 int
4832 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4833 			    uint32_t queue)
4834 {
4835 	struct mlx5_priv *priv = dev->data->dev_private;
4836 	const struct rte_flow_attr attr = {
4837 		.egress = 1,
4838 		.priority = 0,
4839 	};
4840 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4841 		.queue = queue,
4842 	};
4843 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4844 		.queue = UINT32_MAX,
4845 	};
4846 	struct rte_flow_item items[] = {
4847 		{
4848 			.type = (enum rte_flow_item_type)
4849 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4850 			.spec = &queue_spec,
4851 			.last = NULL,
4852 			.mask = &queue_mask,
4853 		},
4854 		{
4855 			.type = RTE_FLOW_ITEM_TYPE_END,
4856 		},
4857 	};
4858 	struct rte_flow_action_jump jump = {
4859 		.group = MLX5_HAIRPIN_TX_TABLE,
4860 	};
4861 	struct rte_flow_action actions[2];
4862 	uint32_t flow_idx;
4863 	struct rte_flow_error error;
4864 
4865 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4866 	actions[0].conf = &jump;
4867 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4868 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4869 				&attr, items, actions, false, &error);
4870 	if (!flow_idx) {
4871 		DRV_LOG(DEBUG,
4872 			"Failed to create ctrl flow: rte_errno(%d),"
4873 			" type(%d), message(%s)",
4874 			rte_errno, error.type,
4875 			error.message ? error.message : " (no stated reason)");
4876 		return -rte_errno;
4877 	}
4878 	return 0;
4879 }
4880 
4881 /**
4882  * Enable a control flow configured from the control plane.
4883  *
4884  * @param dev
4885  *   Pointer to Ethernet device.
4886  * @param eth_spec
4887  *   An Ethernet flow spec to apply.
4888  * @param eth_mask
4889  *   An Ethernet flow mask to apply.
4890  * @param vlan_spec
4891  *   A VLAN flow spec to apply.
4892  * @param vlan_mask
4893  *   A VLAN flow mask to apply.
4894  *
4895  * @return
4896  *   0 on success, a negative errno value otherwise and rte_errno is set.
4897  */
4898 int
4899 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4900 		    struct rte_flow_item_eth *eth_spec,
4901 		    struct rte_flow_item_eth *eth_mask,
4902 		    struct rte_flow_item_vlan *vlan_spec,
4903 		    struct rte_flow_item_vlan *vlan_mask)
4904 {
4905 	struct mlx5_priv *priv = dev->data->dev_private;
4906 	const struct rte_flow_attr attr = {
4907 		.ingress = 1,
4908 		.priority = MLX5_FLOW_PRIO_RSVD,
4909 	};
4910 	struct rte_flow_item items[] = {
4911 		{
4912 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4913 			.spec = eth_spec,
4914 			.last = NULL,
4915 			.mask = eth_mask,
4916 		},
4917 		{
4918 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4919 					      RTE_FLOW_ITEM_TYPE_END,
4920 			.spec = vlan_spec,
4921 			.last = NULL,
4922 			.mask = vlan_mask,
4923 		},
4924 		{
4925 			.type = RTE_FLOW_ITEM_TYPE_END,
4926 		},
4927 	};
4928 	uint16_t queue[priv->reta_idx_n];
4929 	struct rte_flow_action_rss action_rss = {
4930 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4931 		.level = 0,
4932 		.types = priv->rss_conf.rss_hf,
4933 		.key_len = priv->rss_conf.rss_key_len,
4934 		.queue_num = priv->reta_idx_n,
4935 		.key = priv->rss_conf.rss_key,
4936 		.queue = queue,
4937 	};
4938 	struct rte_flow_action actions[] = {
4939 		{
4940 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4941 			.conf = &action_rss,
4942 		},
4943 		{
4944 			.type = RTE_FLOW_ACTION_TYPE_END,
4945 		},
4946 	};
4947 	uint32_t flow_idx;
4948 	struct rte_flow_error error;
4949 	unsigned int i;
4950 
4951 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4952 		return 0;
4953 	}
4954 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4955 		action_rss.types = 0;
4956 	for (i = 0; i != priv->reta_idx_n; ++i)
4957 		queue[i] = (*priv->reta_idx)[i];
4958 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4959 				&attr, items, actions, false, &error);
4960 	if (!flow_idx)
4961 		return -rte_errno;
4962 	return 0;
4963 }
4964 
4965 /**
4966  * Enable a flow control configured from the control plane.
4967  *
4968  * @param dev
4969  *   Pointer to Ethernet device.
4970  * @param eth_spec
4971  *   An Ethernet flow spec to apply.
4972  * @param eth_mask
4973  *   An Ethernet flow mask to apply.
4974  *
4975  * @return
4976  *   0 on success, a negative errno value otherwise and rte_errno is set.
4977  */
4978 int
4979 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4980 	       struct rte_flow_item_eth *eth_spec,
4981 	       struct rte_flow_item_eth *eth_mask)
4982 {
4983 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4984 }
4985 
4986 /**
4987  * Destroy a flow.
4988  *
4989  * @see rte_flow_destroy()
4990  * @see rte_flow_ops
4991  */
4992 int
4993 mlx5_flow_destroy(struct rte_eth_dev *dev,
4994 		  struct rte_flow *flow,
4995 		  struct rte_flow_error *error __rte_unused)
4996 {
4997 	struct mlx5_priv *priv = dev->data->dev_private;
4998 
4999 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5000 	return 0;
5001 }
5002 
5003 /**
5004  * Destroy all flows.
5005  *
5006  * @see rte_flow_flush()
5007  * @see rte_flow_ops
5008  */
5009 int
5010 mlx5_flow_flush(struct rte_eth_dev *dev,
5011 		struct rte_flow_error *error __rte_unused)
5012 {
5013 	struct mlx5_priv *priv = dev->data->dev_private;
5014 
5015 	mlx5_flow_list_flush(dev, &priv->flows, false);
5016 	return 0;
5017 }
5018 
5019 /**
5020  * Isolated mode.
5021  *
5022  * @see rte_flow_isolate()
5023  * @see rte_flow_ops
5024  */
5025 int
5026 mlx5_flow_isolate(struct rte_eth_dev *dev,
5027 		  int enable,
5028 		  struct rte_flow_error *error)
5029 {
5030 	struct mlx5_priv *priv = dev->data->dev_private;
5031 
5032 	if (dev->data->dev_started) {
5033 		rte_flow_error_set(error, EBUSY,
5034 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5035 				   NULL,
5036 				   "port must be stopped first");
5037 		return -rte_errno;
5038 	}
5039 	priv->isolated = !!enable;
5040 	if (enable)
5041 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
5042 	else
5043 		dev->dev_ops = &mlx5_os_dev_ops;
5044 	return 0;
5045 }
5046 
5047 /**
5048  * Query a flow.
5049  *
5050  * @see rte_flow_query()
5051  * @see rte_flow_ops
5052  */
5053 static int
5054 flow_drv_query(struct rte_eth_dev *dev,
5055 	       uint32_t flow_idx,
5056 	       const struct rte_flow_action *actions,
5057 	       void *data,
5058 	       struct rte_flow_error *error)
5059 {
5060 	struct mlx5_priv *priv = dev->data->dev_private;
5061 	const struct mlx5_flow_driver_ops *fops;
5062 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5063 					       [MLX5_IPOOL_RTE_FLOW],
5064 					       flow_idx);
5065 	enum mlx5_flow_drv_type ftype;
5066 
5067 	if (!flow) {
5068 		return rte_flow_error_set(error, ENOENT,
5069 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5070 			  NULL,
5071 			  "invalid flow handle");
5072 	}
5073 	ftype = flow->drv_type;
5074 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5075 	fops = flow_get_drv_ops(ftype);
5076 
5077 	return fops->query(dev, flow, actions, data, error);
5078 }
5079 
5080 /**
5081  * Query a flow.
5082  *
5083  * @see rte_flow_query()
5084  * @see rte_flow_ops
5085  */
5086 int
5087 mlx5_flow_query(struct rte_eth_dev *dev,
5088 		struct rte_flow *flow,
5089 		const struct rte_flow_action *actions,
5090 		void *data,
5091 		struct rte_flow_error *error)
5092 {
5093 	int ret;
5094 
5095 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5096 			     error);
5097 	if (ret < 0)
5098 		return ret;
5099 	return 0;
5100 }
5101 
5102 /**
5103  * Convert a flow director filter to a generic flow.
5104  *
5105  * @param dev
5106  *   Pointer to Ethernet device.
5107  * @param fdir_filter
5108  *   Flow director filter to add.
5109  * @param attributes
5110  *   Generic flow parameters structure.
5111  *
5112  * @return
5113  *   0 on success, a negative errno value otherwise and rte_errno is set.
5114  */
5115 static int
5116 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5117 			 const struct rte_eth_fdir_filter *fdir_filter,
5118 			 struct mlx5_fdir *attributes)
5119 {
5120 	struct mlx5_priv *priv = dev->data->dev_private;
5121 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5122 	const struct rte_eth_fdir_masks *mask =
5123 		&dev->data->dev_conf.fdir_conf.mask;
5124 
5125 	/* Validate queue number. */
5126 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5127 		DRV_LOG(ERR, "port %u invalid queue number %d",
5128 			dev->data->port_id, fdir_filter->action.rx_queue);
5129 		rte_errno = EINVAL;
5130 		return -rte_errno;
5131 	}
5132 	attributes->attr.ingress = 1;
5133 	attributes->items[0] = (struct rte_flow_item) {
5134 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5135 		.spec = &attributes->l2,
5136 		.mask = &attributes->l2_mask,
5137 	};
5138 	switch (fdir_filter->action.behavior) {
5139 	case RTE_ETH_FDIR_ACCEPT:
5140 		attributes->actions[0] = (struct rte_flow_action){
5141 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5142 			.conf = &attributes->queue,
5143 		};
5144 		break;
5145 	case RTE_ETH_FDIR_REJECT:
5146 		attributes->actions[0] = (struct rte_flow_action){
5147 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5148 		};
5149 		break;
5150 	default:
5151 		DRV_LOG(ERR, "port %u invalid behavior %d",
5152 			dev->data->port_id,
5153 			fdir_filter->action.behavior);
5154 		rte_errno = ENOTSUP;
5155 		return -rte_errno;
5156 	}
5157 	attributes->queue.index = fdir_filter->action.rx_queue;
5158 	/* Handle L3. */
5159 	switch (fdir_filter->input.flow_type) {
5160 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5161 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5162 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5163 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5164 			.src_addr = input->flow.ip4_flow.src_ip,
5165 			.dst_addr = input->flow.ip4_flow.dst_ip,
5166 			.time_to_live = input->flow.ip4_flow.ttl,
5167 			.type_of_service = input->flow.ip4_flow.tos,
5168 		};
5169 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5170 			.src_addr = mask->ipv4_mask.src_ip,
5171 			.dst_addr = mask->ipv4_mask.dst_ip,
5172 			.time_to_live = mask->ipv4_mask.ttl,
5173 			.type_of_service = mask->ipv4_mask.tos,
5174 			.next_proto_id = mask->ipv4_mask.proto,
5175 		};
5176 		attributes->items[1] = (struct rte_flow_item){
5177 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5178 			.spec = &attributes->l3,
5179 			.mask = &attributes->l3_mask,
5180 		};
5181 		break;
5182 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5183 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5184 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5185 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5186 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5187 			.proto = input->flow.ipv6_flow.proto,
5188 		};
5189 
5190 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5191 		       input->flow.ipv6_flow.src_ip,
5192 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5193 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5194 		       input->flow.ipv6_flow.dst_ip,
5195 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5196 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5197 		       mask->ipv6_mask.src_ip,
5198 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5199 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5200 		       mask->ipv6_mask.dst_ip,
5201 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5202 		attributes->items[1] = (struct rte_flow_item){
5203 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5204 			.spec = &attributes->l3,
5205 			.mask = &attributes->l3_mask,
5206 		};
5207 		break;
5208 	default:
5209 		DRV_LOG(ERR, "port %u invalid flow type%d",
5210 			dev->data->port_id, fdir_filter->input.flow_type);
5211 		rte_errno = ENOTSUP;
5212 		return -rte_errno;
5213 	}
5214 	/* Handle L4. */
5215 	switch (fdir_filter->input.flow_type) {
5216 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5217 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5218 			.src_port = input->flow.udp4_flow.src_port,
5219 			.dst_port = input->flow.udp4_flow.dst_port,
5220 		};
5221 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5222 			.src_port = mask->src_port_mask,
5223 			.dst_port = mask->dst_port_mask,
5224 		};
5225 		attributes->items[2] = (struct rte_flow_item){
5226 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5227 			.spec = &attributes->l4,
5228 			.mask = &attributes->l4_mask,
5229 		};
5230 		break;
5231 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5232 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5233 			.src_port = input->flow.tcp4_flow.src_port,
5234 			.dst_port = input->flow.tcp4_flow.dst_port,
5235 		};
5236 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5237 			.src_port = mask->src_port_mask,
5238 			.dst_port = mask->dst_port_mask,
5239 		};
5240 		attributes->items[2] = (struct rte_flow_item){
5241 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5242 			.spec = &attributes->l4,
5243 			.mask = &attributes->l4_mask,
5244 		};
5245 		break;
5246 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5247 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5248 			.src_port = input->flow.udp6_flow.src_port,
5249 			.dst_port = input->flow.udp6_flow.dst_port,
5250 		};
5251 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5252 			.src_port = mask->src_port_mask,
5253 			.dst_port = mask->dst_port_mask,
5254 		};
5255 		attributes->items[2] = (struct rte_flow_item){
5256 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5257 			.spec = &attributes->l4,
5258 			.mask = &attributes->l4_mask,
5259 		};
5260 		break;
5261 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5262 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5263 			.src_port = input->flow.tcp6_flow.src_port,
5264 			.dst_port = input->flow.tcp6_flow.dst_port,
5265 		};
5266 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5267 			.src_port = mask->src_port_mask,
5268 			.dst_port = mask->dst_port_mask,
5269 		};
5270 		attributes->items[2] = (struct rte_flow_item){
5271 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5272 			.spec = &attributes->l4,
5273 			.mask = &attributes->l4_mask,
5274 		};
5275 		break;
5276 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5277 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5278 		break;
5279 	default:
5280 		DRV_LOG(ERR, "port %u invalid flow type%d",
5281 			dev->data->port_id, fdir_filter->input.flow_type);
5282 		rte_errno = ENOTSUP;
5283 		return -rte_errno;
5284 	}
5285 	return 0;
5286 }
5287 
5288 #define FLOW_FDIR_CMP(f1, f2, fld) \
5289 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5290 
5291 /**
5292  * Compare two FDIR flows. If items and actions are identical, the two flows are
5293  * regarded as same.
5294  *
5295  * @param dev
5296  *   Pointer to Ethernet device.
5297  * @param f1
5298  *   FDIR flow to compare.
5299  * @param f2
5300  *   FDIR flow to compare.
5301  *
5302  * @return
5303  *   Zero on match, 1 otherwise.
5304  */
5305 static int
5306 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5307 {
5308 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5309 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5310 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5311 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5312 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5313 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5314 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5315 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5316 		return 1;
5317 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5318 	    FLOW_FDIR_CMP(f1, f2, queue))
5319 		return 1;
5320 	return 0;
5321 }
5322 
5323 /**
5324  * Search device flow list to find out a matched FDIR flow.
5325  *
5326  * @param dev
5327  *   Pointer to Ethernet device.
5328  * @param fdir_flow
5329  *   FDIR flow to lookup.
5330  *
5331  * @return
5332  *   Index of flow if found, 0 otherwise.
5333  */
5334 static uint32_t
5335 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5336 {
5337 	struct mlx5_priv *priv = dev->data->dev_private;
5338 	uint32_t flow_idx = 0;
5339 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5340 
5341 	MLX5_ASSERT(fdir_flow);
5342 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5343 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5344 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5345 				dev->data->port_id, flow_idx);
5346 			flow_idx = priv_fdir_flow->rix_flow;
5347 			break;
5348 		}
5349 	}
5350 	return flow_idx;
5351 }
5352 
5353 /**
5354  * Add new flow director filter and store it in list.
5355  *
5356  * @param dev
5357  *   Pointer to Ethernet device.
5358  * @param fdir_filter
5359  *   Flow director filter to add.
5360  *
5361  * @return
5362  *   0 on success, a negative errno value otherwise and rte_errno is set.
5363  */
5364 static int
5365 flow_fdir_filter_add(struct rte_eth_dev *dev,
5366 		     const struct rte_eth_fdir_filter *fdir_filter)
5367 {
5368 	struct mlx5_priv *priv = dev->data->dev_private;
5369 	struct mlx5_fdir *fdir_flow;
5370 	struct rte_flow *flow;
5371 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5372 	uint32_t flow_idx;
5373 	int ret;
5374 
5375 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5376 	if (!fdir_flow) {
5377 		rte_errno = ENOMEM;
5378 		return -rte_errno;
5379 	}
5380 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5381 	if (ret)
5382 		goto error;
5383 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5384 	if (flow_idx) {
5385 		rte_errno = EEXIST;
5386 		goto error;
5387 	}
5388 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5389 				     0);
5390 	if (!priv_fdir_flow) {
5391 		rte_errno = ENOMEM;
5392 		goto error;
5393 	}
5394 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5395 				    fdir_flow->items, fdir_flow->actions, true,
5396 				    NULL);
5397 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5398 	if (!flow)
5399 		goto error;
5400 	flow->fdir = 1;
5401 	priv_fdir_flow->fdir = fdir_flow;
5402 	priv_fdir_flow->rix_flow = flow_idx;
5403 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5404 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5405 		dev->data->port_id, (void *)flow);
5406 	return 0;
5407 error:
5408 	rte_free(priv_fdir_flow);
5409 	rte_free(fdir_flow);
5410 	return -rte_errno;
5411 }
5412 
5413 /**
5414  * Delete specific filter.
5415  *
5416  * @param dev
5417  *   Pointer to Ethernet device.
5418  * @param fdir_filter
5419  *   Filter to be deleted.
5420  *
5421  * @return
5422  *   0 on success, a negative errno value otherwise and rte_errno is set.
5423  */
5424 static int
5425 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5426 			const struct rte_eth_fdir_filter *fdir_filter)
5427 {
5428 	struct mlx5_priv *priv = dev->data->dev_private;
5429 	uint32_t flow_idx;
5430 	struct mlx5_fdir fdir_flow = {
5431 		.attr.group = 0,
5432 	};
5433 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5434 	int ret;
5435 
5436 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5437 	if (ret)
5438 		return -rte_errno;
5439 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5440 		/* Find the fdir in priv list */
5441 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5442 			break;
5443 	}
5444 	if (!priv_fdir_flow)
5445 		return 0;
5446 	LIST_REMOVE(priv_fdir_flow, next);
5447 	flow_idx = priv_fdir_flow->rix_flow;
5448 	flow_list_destroy(dev, &priv->flows, flow_idx);
5449 	rte_free(priv_fdir_flow->fdir);
5450 	rte_free(priv_fdir_flow);
5451 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5452 		dev->data->port_id, flow_idx);
5453 	return 0;
5454 }
5455 
5456 /**
5457  * Update queue for specific filter.
5458  *
5459  * @param dev
5460  *   Pointer to Ethernet device.
5461  * @param fdir_filter
5462  *   Filter to be updated.
5463  *
5464  * @return
5465  *   0 on success, a negative errno value otherwise and rte_errno is set.
5466  */
5467 static int
5468 flow_fdir_filter_update(struct rte_eth_dev *dev,
5469 			const struct rte_eth_fdir_filter *fdir_filter)
5470 {
5471 	int ret;
5472 
5473 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5474 	if (ret)
5475 		return ret;
5476 	return flow_fdir_filter_add(dev, fdir_filter);
5477 }
5478 
5479 /**
5480  * Flush all filters.
5481  *
5482  * @param dev
5483  *   Pointer to Ethernet device.
5484  */
5485 static void
5486 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5487 {
5488 	struct mlx5_priv *priv = dev->data->dev_private;
5489 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5490 
5491 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5492 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5493 		LIST_REMOVE(priv_fdir_flow, next);
5494 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5495 		rte_free(priv_fdir_flow->fdir);
5496 		rte_free(priv_fdir_flow);
5497 	}
5498 }
5499 
5500 /**
5501  * Get flow director information.
5502  *
5503  * @param dev
5504  *   Pointer to Ethernet device.
5505  * @param[out] fdir_info
5506  *   Resulting flow director information.
5507  */
5508 static void
5509 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5510 {
5511 	struct rte_eth_fdir_masks *mask =
5512 		&dev->data->dev_conf.fdir_conf.mask;
5513 
5514 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5515 	fdir_info->guarant_spc = 0;
5516 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5517 	fdir_info->max_flexpayload = 0;
5518 	fdir_info->flow_types_mask[0] = 0;
5519 	fdir_info->flex_payload_unit = 0;
5520 	fdir_info->max_flex_payload_segment_num = 0;
5521 	fdir_info->flex_payload_limit = 0;
5522 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5523 }
5524 
5525 /**
5526  * Deal with flow director operations.
5527  *
5528  * @param dev
5529  *   Pointer to Ethernet device.
5530  * @param filter_op
5531  *   Operation to perform.
5532  * @param arg
5533  *   Pointer to operation-specific structure.
5534  *
5535  * @return
5536  *   0 on success, a negative errno value otherwise and rte_errno is set.
5537  */
5538 static int
5539 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5540 		    void *arg)
5541 {
5542 	enum rte_fdir_mode fdir_mode =
5543 		dev->data->dev_conf.fdir_conf.mode;
5544 
5545 	if (filter_op == RTE_ETH_FILTER_NOP)
5546 		return 0;
5547 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5548 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5549 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5550 			dev->data->port_id, fdir_mode);
5551 		rte_errno = EINVAL;
5552 		return -rte_errno;
5553 	}
5554 	switch (filter_op) {
5555 	case RTE_ETH_FILTER_ADD:
5556 		return flow_fdir_filter_add(dev, arg);
5557 	case RTE_ETH_FILTER_UPDATE:
5558 		return flow_fdir_filter_update(dev, arg);
5559 	case RTE_ETH_FILTER_DELETE:
5560 		return flow_fdir_filter_delete(dev, arg);
5561 	case RTE_ETH_FILTER_FLUSH:
5562 		flow_fdir_filter_flush(dev);
5563 		break;
5564 	case RTE_ETH_FILTER_INFO:
5565 		flow_fdir_info_get(dev, arg);
5566 		break;
5567 	default:
5568 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5569 			dev->data->port_id, filter_op);
5570 		rte_errno = EINVAL;
5571 		return -rte_errno;
5572 	}
5573 	return 0;
5574 }
5575 
5576 /**
5577  * Manage filter operations.
5578  *
5579  * @param dev
5580  *   Pointer to Ethernet device structure.
5581  * @param filter_type
5582  *   Filter type.
5583  * @param filter_op
5584  *   Operation to perform.
5585  * @param arg
5586  *   Pointer to operation-specific structure.
5587  *
5588  * @return
5589  *   0 on success, a negative errno value otherwise and rte_errno is set.
5590  */
5591 int
5592 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5593 		     enum rte_filter_type filter_type,
5594 		     enum rte_filter_op filter_op,
5595 		     void *arg)
5596 {
5597 	switch (filter_type) {
5598 	case RTE_ETH_FILTER_GENERIC:
5599 		if (filter_op != RTE_ETH_FILTER_GET) {
5600 			rte_errno = EINVAL;
5601 			return -rte_errno;
5602 		}
5603 		*(const void **)arg = &mlx5_flow_ops;
5604 		return 0;
5605 	case RTE_ETH_FILTER_FDIR:
5606 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5607 	default:
5608 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5609 			dev->data->port_id, filter_type);
5610 		rte_errno = ENOTSUP;
5611 		return -rte_errno;
5612 	}
5613 	return 0;
5614 }
5615 
5616 /**
5617  * Create the needed meter and suffix tables.
5618  *
5619  * @param[in] dev
5620  *   Pointer to Ethernet device.
5621  * @param[in] fm
5622  *   Pointer to the flow meter.
5623  *
5624  * @return
5625  *   Pointer to table set on success, NULL otherwise.
5626  */
5627 struct mlx5_meter_domains_infos *
5628 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5629 			  const struct mlx5_flow_meter *fm)
5630 {
5631 	const struct mlx5_flow_driver_ops *fops;
5632 
5633 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5634 	return fops->create_mtr_tbls(dev, fm);
5635 }
5636 
5637 /**
5638  * Destroy the meter table set.
5639  *
5640  * @param[in] dev
5641  *   Pointer to Ethernet device.
5642  * @param[in] tbl
5643  *   Pointer to the meter table set.
5644  *
5645  * @return
5646  *   0 on success.
5647  */
5648 int
5649 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5650 			   struct mlx5_meter_domains_infos *tbls)
5651 {
5652 	const struct mlx5_flow_driver_ops *fops;
5653 
5654 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5655 	return fops->destroy_mtr_tbls(dev, tbls);
5656 }
5657 
5658 /**
5659  * Create policer rules.
5660  *
5661  * @param[in] dev
5662  *   Pointer to Ethernet device.
5663  * @param[in] fm
5664  *   Pointer to flow meter structure.
5665  * @param[in] attr
5666  *   Pointer to flow attributes.
5667  *
5668  * @return
5669  *   0 on success, -1 otherwise.
5670  */
5671 int
5672 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5673 			       struct mlx5_flow_meter *fm,
5674 			       const struct rte_flow_attr *attr)
5675 {
5676 	const struct mlx5_flow_driver_ops *fops;
5677 
5678 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5679 	return fops->create_policer_rules(dev, fm, attr);
5680 }
5681 
5682 /**
5683  * Destroy policer rules.
5684  *
5685  * @param[in] fm
5686  *   Pointer to flow meter structure.
5687  * @param[in] attr
5688  *   Pointer to flow attributes.
5689  *
5690  * @return
5691  *   0 on success, -1 otherwise.
5692  */
5693 int
5694 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5695 				struct mlx5_flow_meter *fm,
5696 				const struct rte_flow_attr *attr)
5697 {
5698 	const struct mlx5_flow_driver_ops *fops;
5699 
5700 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5701 	return fops->destroy_policer_rules(dev, fm, attr);
5702 }
5703 
5704 /**
5705  * Allocate a counter.
5706  *
5707  * @param[in] dev
5708  *   Pointer to Ethernet device structure.
5709  *
5710  * @return
5711  *   Index to allocated counter  on success, 0 otherwise.
5712  */
5713 uint32_t
5714 mlx5_counter_alloc(struct rte_eth_dev *dev)
5715 {
5716 	const struct mlx5_flow_driver_ops *fops;
5717 	struct rte_flow_attr attr = { .transfer = 0 };
5718 
5719 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5720 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5721 		return fops->counter_alloc(dev);
5722 	}
5723 	DRV_LOG(ERR,
5724 		"port %u counter allocate is not supported.",
5725 		 dev->data->port_id);
5726 	return 0;
5727 }
5728 
5729 /**
5730  * Free a counter.
5731  *
5732  * @param[in] dev
5733  *   Pointer to Ethernet device structure.
5734  * @param[in] cnt
5735  *   Index to counter to be free.
5736  */
5737 void
5738 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5739 {
5740 	const struct mlx5_flow_driver_ops *fops;
5741 	struct rte_flow_attr attr = { .transfer = 0 };
5742 
5743 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5744 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5745 		fops->counter_free(dev, cnt);
5746 		return;
5747 	}
5748 	DRV_LOG(ERR,
5749 		"port %u counter free is not supported.",
5750 		 dev->data->port_id);
5751 }
5752 
5753 /**
5754  * Query counter statistics.
5755  *
5756  * @param[in] dev
5757  *   Pointer to Ethernet device structure.
5758  * @param[in] cnt
5759  *   Index to counter to query.
5760  * @param[in] clear
5761  *   Set to clear counter statistics.
5762  * @param[out] pkts
5763  *   The counter hits packets number to save.
5764  * @param[out] bytes
5765  *   The counter hits bytes number to save.
5766  *
5767  * @return
5768  *   0 on success, a negative errno value otherwise.
5769  */
5770 int
5771 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5772 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5773 {
5774 	const struct mlx5_flow_driver_ops *fops;
5775 	struct rte_flow_attr attr = { .transfer = 0 };
5776 
5777 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5778 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5779 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5780 	}
5781 	DRV_LOG(ERR,
5782 		"port %u counter query is not supported.",
5783 		 dev->data->port_id);
5784 	return -ENOTSUP;
5785 }
5786 
5787 #define MLX5_POOL_QUERY_FREQ_US 1000000
5788 
5789 /**
5790  * Get number of all validate pools.
5791  *
5792  * @param[in] sh
5793  *   Pointer to mlx5_dev_ctx_shared object.
5794  *
5795  * @return
5796  *   The number of all validate pools.
5797  */
5798 static uint32_t
5799 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
5800 {
5801 	int i;
5802 	uint32_t pools_n = 0;
5803 
5804 	for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
5805 		pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
5806 	return pools_n;
5807 }
5808 
5809 /**
5810  * Set the periodic procedure for triggering asynchronous batch queries for all
5811  * the counter pools.
5812  *
5813  * @param[in] sh
5814  *   Pointer to mlx5_dev_ctx_shared object.
5815  */
5816 void
5817 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
5818 {
5819 	uint32_t pools_n, us;
5820 
5821 	pools_n = mlx5_get_all_valid_pool_count(sh);
5822 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5823 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5824 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5825 		sh->cmng.query_thread_on = 0;
5826 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5827 	} else {
5828 		sh->cmng.query_thread_on = 1;
5829 	}
5830 }
5831 
5832 /**
5833  * The periodic procedure for triggering asynchronous batch queries for all the
5834  * counter pools. This function is probably called by the host thread.
5835  *
5836  * @param[in] arg
5837  *   The parameter for the alarm process.
5838  */
5839 void
5840 mlx5_flow_query_alarm(void *arg)
5841 {
5842 	struct mlx5_dev_ctx_shared *sh = arg;
5843 	struct mlx5_devx_obj *dcs;
5844 	uint16_t offset;
5845 	int ret;
5846 	uint8_t batch = sh->cmng.batch;
5847 	uint8_t age = sh->cmng.age;
5848 	uint16_t pool_index = sh->cmng.pool_index;
5849 	struct mlx5_pools_container *cont;
5850 	struct mlx5_flow_counter_pool *pool;
5851 	int cont_loop = MLX5_CCONT_TYPE_MAX;
5852 
5853 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5854 		goto set_alarm;
5855 next_container:
5856 	cont = MLX5_CNT_CONTAINER(sh, batch, age);
5857 	rte_spinlock_lock(&cont->resize_sl);
5858 	if (!cont->pools) {
5859 		rte_spinlock_unlock(&cont->resize_sl);
5860 		/* Check if all the containers are empty. */
5861 		if (unlikely(--cont_loop == 0))
5862 			goto set_alarm;
5863 		batch ^= 0x1;
5864 		pool_index = 0;
5865 		if (batch == 0 && pool_index == 0) {
5866 			age ^= 0x1;
5867 			sh->cmng.batch = batch;
5868 			sh->cmng.age = age;
5869 		}
5870 		goto next_container;
5871 	}
5872 	pool = cont->pools[pool_index];
5873 	rte_spinlock_unlock(&cont->resize_sl);
5874 	if (pool->raw_hw)
5875 		/* There is a pool query in progress. */
5876 		goto set_alarm;
5877 	pool->raw_hw =
5878 		LIST_FIRST(&sh->cmng.free_stat_raws);
5879 	if (!pool->raw_hw)
5880 		/* No free counter statistics raw memory. */
5881 		goto set_alarm;
5882 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5883 							      (&pool->a64_dcs);
5884 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5885 	/*
5886 	 * Identify the counters released between query trigger and query
5887 	 * handle more effiecntly. The counter released in this gap period
5888 	 * should wait for a new round of query as the new arrived packets
5889 	 * will not be taken into account.
5890 	 */
5891 	rte_atomic64_add(&pool->start_query_gen, 1);
5892 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5893 					       offset, NULL, NULL,
5894 					       pool->raw_hw->mem_mng->dm->id,
5895 					       (void *)(uintptr_t)
5896 					       (pool->raw_hw->data + offset),
5897 					       sh->devx_comp,
5898 					       (uint64_t)(uintptr_t)pool);
5899 	if (ret) {
5900 		rte_atomic64_sub(&pool->start_query_gen, 1);
5901 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5902 			" %d", pool->min_dcs->id);
5903 		pool->raw_hw = NULL;
5904 		goto set_alarm;
5905 	}
5906 	pool->raw_hw->min_dcs_id = dcs->id;
5907 	LIST_REMOVE(pool->raw_hw, next);
5908 	sh->cmng.pending_queries++;
5909 	pool_index++;
5910 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5911 		batch ^= 0x1;
5912 		pool_index = 0;
5913 		if (batch == 0 && pool_index == 0)
5914 			age ^= 0x1;
5915 	}
5916 set_alarm:
5917 	sh->cmng.batch = batch;
5918 	sh->cmng.pool_index = pool_index;
5919 	sh->cmng.age = age;
5920 	mlx5_set_query_alarm(sh);
5921 }
5922 
5923 /**
5924  * Check and callback event for new aged flow in the counter pool
5925  *
5926  * @param[in] sh
5927  *   Pointer to mlx5_dev_ctx_shared object.
5928  * @param[in] pool
5929  *   Pointer to Current counter pool.
5930  */
5931 static void
5932 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
5933 		   struct mlx5_flow_counter_pool *pool)
5934 {
5935 	struct mlx5_priv *priv;
5936 	struct mlx5_flow_counter *cnt;
5937 	struct mlx5_age_info *age_info;
5938 	struct mlx5_age_param *age_param;
5939 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
5940 	struct mlx5_counter_stats_raw *prev = pool->raw;
5941 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
5942 	uint32_t i;
5943 
5944 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
5945 		cnt = MLX5_POOL_GET_CNT(pool, i);
5946 		age_param = MLX5_CNT_TO_AGE(cnt);
5947 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
5948 			continue;
5949 		if (cur->data[i].hits != prev->data[i].hits) {
5950 			age_param->expire = curr + age_param->timeout;
5951 			continue;
5952 		}
5953 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
5954 			continue;
5955 		/**
5956 		 * Hold the lock first, or if between the
5957 		 * state AGE_TMOUT and tailq operation the
5958 		 * release happened, the release procedure
5959 		 * may delete a non-existent tailq node.
5960 		 */
5961 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
5962 		age_info = GET_PORT_AGE_INFO(priv);
5963 		rte_spinlock_lock(&age_info->aged_sl);
5964 		/* If the cpmset fails, release happens. */
5965 		if (rte_atomic16_cmpset((volatile uint16_t *)
5966 					&age_param->state,
5967 					AGE_CANDIDATE,
5968 					AGE_TMOUT) ==
5969 					AGE_CANDIDATE) {
5970 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
5971 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
5972 		}
5973 		rte_spinlock_unlock(&age_info->aged_sl);
5974 	}
5975 	for (i = 0; i < sh->max_port; i++) {
5976 		age_info = &sh->port[i].age_info;
5977 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
5978 			continue;
5979 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
5980 			_rte_eth_dev_callback_process
5981 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
5982 				RTE_ETH_EVENT_FLOW_AGED, NULL);
5983 		age_info->flags = 0;
5984 	}
5985 }
5986 
5987 /**
5988  * Handler for the HW respond about ready values from an asynchronous batch
5989  * query. This function is probably called by the host thread.
5990  *
5991  * @param[in] sh
5992  *   The pointer to the shared device context.
5993  * @param[in] async_id
5994  *   The Devx async ID.
5995  * @param[in] status
5996  *   The status of the completion.
5997  */
5998 void
5999 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6000 				  uint64_t async_id, int status)
6001 {
6002 	struct mlx5_flow_counter_pool *pool =
6003 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6004 	struct mlx5_counter_stats_raw *raw_to_free;
6005 
6006 	if (unlikely(status)) {
6007 		rte_atomic64_sub(&pool->start_query_gen, 1);
6008 		raw_to_free = pool->raw_hw;
6009 	} else {
6010 		raw_to_free = pool->raw;
6011 		if (IS_AGE_POOL(pool))
6012 			mlx5_flow_aging_check(sh, pool);
6013 		rte_spinlock_lock(&pool->sl);
6014 		pool->raw = pool->raw_hw;
6015 		rte_spinlock_unlock(&pool->sl);
6016 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
6017 			    rte_atomic64_read(&pool->start_query_gen));
6018 		rte_atomic64_set(&pool->end_query_gen,
6019 				 rte_atomic64_read(&pool->start_query_gen));
6020 		/* Be sure the new raw counters data is updated in memory. */
6021 		rte_cio_wmb();
6022 	}
6023 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6024 	pool->raw_hw = NULL;
6025 	sh->cmng.pending_queries--;
6026 }
6027 
6028 /**
6029  * Translate the rte_flow group index to HW table value.
6030  *
6031  * @param[in] attributes
6032  *   Pointer to flow attributes
6033  * @param[in] external
6034  *   Value is part of flow rule created by request external to PMD.
6035  * @param[in] group
6036  *   rte_flow group index value.
6037  * @param[out] fdb_def_rule
6038  *   Whether fdb jump to table 1 is configured.
6039  * @param[out] table
6040  *   HW table value.
6041  * @param[out] error
6042  *   Pointer to error structure.
6043  *
6044  * @return
6045  *   0 on success, a negative errno value otherwise and rte_errno is set.
6046  */
6047 int
6048 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6049 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6050 			 struct rte_flow_error *error)
6051 {
6052 	if (attributes->transfer && external && fdb_def_rule) {
6053 		if (group == UINT32_MAX)
6054 			return rte_flow_error_set
6055 						(error, EINVAL,
6056 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6057 						 NULL,
6058 						 "group index not supported");
6059 		*table = group + 1;
6060 	} else {
6061 		*table = group;
6062 	}
6063 	return 0;
6064 }
6065 
6066 /**
6067  * Discover availability of metadata reg_c's.
6068  *
6069  * Iteratively use test flows to check availability.
6070  *
6071  * @param[in] dev
6072  *   Pointer to the Ethernet device structure.
6073  *
6074  * @return
6075  *   0 on success, a negative errno value otherwise and rte_errno is set.
6076  */
6077 int
6078 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6079 {
6080 	struct mlx5_priv *priv = dev->data->dev_private;
6081 	struct mlx5_dev_config *config = &priv->config;
6082 	enum modify_reg idx;
6083 	int n = 0;
6084 
6085 	/* reg_c[0] and reg_c[1] are reserved. */
6086 	config->flow_mreg_c[n++] = REG_C_0;
6087 	config->flow_mreg_c[n++] = REG_C_1;
6088 	/* Discover availability of other reg_c's. */
6089 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6090 		struct rte_flow_attr attr = {
6091 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6092 			.priority = MLX5_FLOW_PRIO_RSVD,
6093 			.ingress = 1,
6094 		};
6095 		struct rte_flow_item items[] = {
6096 			[0] = {
6097 				.type = RTE_FLOW_ITEM_TYPE_END,
6098 			},
6099 		};
6100 		struct rte_flow_action actions[] = {
6101 			[0] = {
6102 				.type = (enum rte_flow_action_type)
6103 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6104 				.conf = &(struct mlx5_flow_action_copy_mreg){
6105 					.src = REG_C_1,
6106 					.dst = idx,
6107 				},
6108 			},
6109 			[1] = {
6110 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6111 				.conf = &(struct rte_flow_action_jump){
6112 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6113 				},
6114 			},
6115 			[2] = {
6116 				.type = RTE_FLOW_ACTION_TYPE_END,
6117 			},
6118 		};
6119 		uint32_t flow_idx;
6120 		struct rte_flow *flow;
6121 		struct rte_flow_error error;
6122 
6123 		if (!config->dv_flow_en)
6124 			break;
6125 		/* Create internal flow, validation skips copy action. */
6126 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6127 					    actions, false, &error);
6128 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6129 				      flow_idx);
6130 		if (!flow)
6131 			continue;
6132 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6133 			config->flow_mreg_c[n++] = idx;
6134 		flow_list_destroy(dev, NULL, flow_idx);
6135 	}
6136 	for (; n < MLX5_MREG_C_NUM; ++n)
6137 		config->flow_mreg_c[n] = REG_NONE;
6138 	return 0;
6139 }
6140 
6141 /**
6142  * Dump flow raw hw data to file
6143  *
6144  * @param[in] dev
6145  *    The pointer to Ethernet device.
6146  * @param[in] file
6147  *   A pointer to a file for output.
6148  * @param[out] error
6149  *   Perform verbose error reporting if not NULL. PMDs initialize this
6150  *   structure in case of error only.
6151  * @return
6152  *   0 on success, a nagative value otherwise.
6153  */
6154 int
6155 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6156 		   FILE *file,
6157 		   struct rte_flow_error *error __rte_unused)
6158 {
6159 	struct mlx5_priv *priv = dev->data->dev_private;
6160 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6161 
6162 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6163 				       sh->tx_domain, file);
6164 }
6165 
6166 /**
6167  * Get aged-out flows.
6168  *
6169  * @param[in] dev
6170  *   Pointer to the Ethernet device structure.
6171  * @param[in] context
6172  *   The address of an array of pointers to the aged-out flows contexts.
6173  * @param[in] nb_countexts
6174  *   The length of context array pointers.
6175  * @param[out] error
6176  *   Perform verbose error reporting if not NULL. Initialized in case of
6177  *   error only.
6178  *
6179  * @return
6180  *   how many contexts get in success, otherwise negative errno value.
6181  *   if nb_contexts is 0, return the amount of all aged contexts.
6182  *   if nb_contexts is not 0 , return the amount of aged flows reported
6183  *   in the context array.
6184  */
6185 int
6186 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6187 			uint32_t nb_contexts, struct rte_flow_error *error)
6188 {
6189 	const struct mlx5_flow_driver_ops *fops;
6190 	struct rte_flow_attr attr = { .transfer = 0 };
6191 
6192 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6193 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6194 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6195 						    error);
6196 	}
6197 	DRV_LOG(ERR,
6198 		"port %u get aged flows is not supported.",
6199 		 dev->data->port_id);
6200 	return -ENOTSUP;
6201 }
6202