xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 91d3e2d42997e57e2abf34b1c24621530819d2eb)
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_flow_driver.h>
28 #include <rte_malloc.h>
29 #include <rte_ip.h>
30 
31 #include <mlx5_glue.h>
32 #include <mlx5_devx_cmds.h>
33 #include <mlx5_prm.h>
34 
35 #include "mlx5_defs.h"
36 #include "mlx5.h"
37 #include "mlx5_flow.h"
38 #include "mlx5_rxtx.h"
39 
40 /* Dev ops structure defined in mlx5.c */
41 extern const struct eth_dev_ops mlx5_dev_ops;
42 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
43 
44 /** Device flow drivers. */
45 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
46 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
47 #endif
48 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
49 
50 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
51 
52 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
53 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
54 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
55 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
56 #endif
57 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
58 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
59 };
60 
61 enum mlx5_expansion {
62 	MLX5_EXPANSION_ROOT,
63 	MLX5_EXPANSION_ROOT_OUTER,
64 	MLX5_EXPANSION_ROOT_ETH_VLAN,
65 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
66 	MLX5_EXPANSION_OUTER_ETH,
67 	MLX5_EXPANSION_OUTER_ETH_VLAN,
68 	MLX5_EXPANSION_OUTER_VLAN,
69 	MLX5_EXPANSION_OUTER_IPV4,
70 	MLX5_EXPANSION_OUTER_IPV4_UDP,
71 	MLX5_EXPANSION_OUTER_IPV4_TCP,
72 	MLX5_EXPANSION_OUTER_IPV6,
73 	MLX5_EXPANSION_OUTER_IPV6_UDP,
74 	MLX5_EXPANSION_OUTER_IPV6_TCP,
75 	MLX5_EXPANSION_VXLAN,
76 	MLX5_EXPANSION_VXLAN_GPE,
77 	MLX5_EXPANSION_GRE,
78 	MLX5_EXPANSION_MPLS,
79 	MLX5_EXPANSION_ETH,
80 	MLX5_EXPANSION_ETH_VLAN,
81 	MLX5_EXPANSION_VLAN,
82 	MLX5_EXPANSION_IPV4,
83 	MLX5_EXPANSION_IPV4_UDP,
84 	MLX5_EXPANSION_IPV4_TCP,
85 	MLX5_EXPANSION_IPV6,
86 	MLX5_EXPANSION_IPV6_UDP,
87 	MLX5_EXPANSION_IPV6_TCP,
88 };
89 
90 /** Supported expansion of items. */
91 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
92 	[MLX5_EXPANSION_ROOT] = {
93 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
94 						 MLX5_EXPANSION_IPV4,
95 						 MLX5_EXPANSION_IPV6),
96 		.type = RTE_FLOW_ITEM_TYPE_END,
97 	},
98 	[MLX5_EXPANSION_ROOT_OUTER] = {
99 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
100 						 MLX5_EXPANSION_OUTER_IPV4,
101 						 MLX5_EXPANSION_OUTER_IPV6),
102 		.type = RTE_FLOW_ITEM_TYPE_END,
103 	},
104 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
105 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
106 		.type = RTE_FLOW_ITEM_TYPE_END,
107 	},
108 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
109 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
110 		.type = RTE_FLOW_ITEM_TYPE_END,
111 	},
112 	[MLX5_EXPANSION_OUTER_ETH] = {
113 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
114 						 MLX5_EXPANSION_OUTER_IPV6,
115 						 MLX5_EXPANSION_MPLS),
116 		.type = RTE_FLOW_ITEM_TYPE_ETH,
117 		.rss_types = 0,
118 	},
119 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
120 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
121 		.type = RTE_FLOW_ITEM_TYPE_ETH,
122 		.rss_types = 0,
123 	},
124 	[MLX5_EXPANSION_OUTER_VLAN] = {
125 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
126 						 MLX5_EXPANSION_OUTER_IPV6),
127 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
128 	},
129 	[MLX5_EXPANSION_OUTER_IPV4] = {
130 		.next = RTE_FLOW_EXPAND_RSS_NEXT
131 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
132 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
133 			 MLX5_EXPANSION_GRE,
134 			 MLX5_EXPANSION_IPV4,
135 			 MLX5_EXPANSION_IPV6),
136 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
137 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
138 			ETH_RSS_NONFRAG_IPV4_OTHER,
139 	},
140 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
141 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
142 						 MLX5_EXPANSION_VXLAN_GPE),
143 		.type = RTE_FLOW_ITEM_TYPE_UDP,
144 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
145 	},
146 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
147 		.type = RTE_FLOW_ITEM_TYPE_TCP,
148 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
149 	},
150 	[MLX5_EXPANSION_OUTER_IPV6] = {
151 		.next = RTE_FLOW_EXPAND_RSS_NEXT
152 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
153 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
154 			 MLX5_EXPANSION_IPV4,
155 			 MLX5_EXPANSION_IPV6),
156 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
157 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
158 			ETH_RSS_NONFRAG_IPV6_OTHER,
159 	},
160 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
161 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
162 						 MLX5_EXPANSION_VXLAN_GPE),
163 		.type = RTE_FLOW_ITEM_TYPE_UDP,
164 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
165 	},
166 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
167 		.type = RTE_FLOW_ITEM_TYPE_TCP,
168 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
169 	},
170 	[MLX5_EXPANSION_VXLAN] = {
171 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
172 						 MLX5_EXPANSION_IPV4,
173 						 MLX5_EXPANSION_IPV6),
174 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
175 	},
176 	[MLX5_EXPANSION_VXLAN_GPE] = {
177 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
178 						 MLX5_EXPANSION_IPV4,
179 						 MLX5_EXPANSION_IPV6),
180 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
181 	},
182 	[MLX5_EXPANSION_GRE] = {
183 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
184 		.type = RTE_FLOW_ITEM_TYPE_GRE,
185 	},
186 	[MLX5_EXPANSION_MPLS] = {
187 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 						 MLX5_EXPANSION_IPV6),
189 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
190 	},
191 	[MLX5_EXPANSION_ETH] = {
192 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
193 						 MLX5_EXPANSION_IPV6),
194 		.type = RTE_FLOW_ITEM_TYPE_ETH,
195 	},
196 	[MLX5_EXPANSION_ETH_VLAN] = {
197 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
198 		.type = RTE_FLOW_ITEM_TYPE_ETH,
199 	},
200 	[MLX5_EXPANSION_VLAN] = {
201 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
202 						 MLX5_EXPANSION_IPV6),
203 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
204 	},
205 	[MLX5_EXPANSION_IPV4] = {
206 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
207 						 MLX5_EXPANSION_IPV4_TCP),
208 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
209 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
210 			ETH_RSS_NONFRAG_IPV4_OTHER,
211 	},
212 	[MLX5_EXPANSION_IPV4_UDP] = {
213 		.type = RTE_FLOW_ITEM_TYPE_UDP,
214 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
215 	},
216 	[MLX5_EXPANSION_IPV4_TCP] = {
217 		.type = RTE_FLOW_ITEM_TYPE_TCP,
218 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
219 	},
220 	[MLX5_EXPANSION_IPV6] = {
221 		.next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
222 						 MLX5_EXPANSION_IPV6_TCP),
223 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
224 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
225 			ETH_RSS_NONFRAG_IPV6_OTHER,
226 	},
227 	[MLX5_EXPANSION_IPV6_UDP] = {
228 		.type = RTE_FLOW_ITEM_TYPE_UDP,
229 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
230 	},
231 	[MLX5_EXPANSION_IPV6_TCP] = {
232 		.type = RTE_FLOW_ITEM_TYPE_TCP,
233 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
234 	},
235 };
236 
237 static const struct rte_flow_ops mlx5_flow_ops = {
238 	.validate = mlx5_flow_validate,
239 	.create = mlx5_flow_create,
240 	.destroy = mlx5_flow_destroy,
241 	.flush = mlx5_flow_flush,
242 	.isolate = mlx5_flow_isolate,
243 	.query = mlx5_flow_query,
244 	.dev_dump = mlx5_flow_dev_dump,
245 };
246 
247 /* Convert FDIR request to Generic flow. */
248 struct mlx5_fdir {
249 	struct rte_flow_attr attr;
250 	struct rte_flow_item items[4];
251 	struct rte_flow_item_eth l2;
252 	struct rte_flow_item_eth l2_mask;
253 	union {
254 		struct rte_flow_item_ipv4 ipv4;
255 		struct rte_flow_item_ipv6 ipv6;
256 	} l3;
257 	union {
258 		struct rte_flow_item_ipv4 ipv4;
259 		struct rte_flow_item_ipv6 ipv6;
260 	} l3_mask;
261 	union {
262 		struct rte_flow_item_udp udp;
263 		struct rte_flow_item_tcp tcp;
264 	} l4;
265 	union {
266 		struct rte_flow_item_udp udp;
267 		struct rte_flow_item_tcp tcp;
268 	} l4_mask;
269 	struct rte_flow_action actions[2];
270 	struct rte_flow_action_queue queue;
271 };
272 
273 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
274 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
275 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
276 };
277 
278 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
279 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
280 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
281 	{ 9, 10, 11 }, { 12, 13, 14 },
282 };
283 
284 /* Tunnel information. */
285 struct mlx5_flow_tunnel_info {
286 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
287 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
288 };
289 
290 static struct mlx5_flow_tunnel_info tunnels_info[] = {
291 	{
292 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
293 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
294 	},
295 	{
296 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
297 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
298 	},
299 	{
300 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
301 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
302 	},
303 	{
304 		.tunnel = MLX5_FLOW_LAYER_GRE,
305 		.ptype = RTE_PTYPE_TUNNEL_GRE,
306 	},
307 	{
308 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
309 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
310 	},
311 	{
312 		.tunnel = MLX5_FLOW_LAYER_MPLS,
313 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
314 	},
315 	{
316 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
317 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
318 	},
319 	{
320 		.tunnel = MLX5_FLOW_LAYER_IPIP,
321 		.ptype = RTE_PTYPE_TUNNEL_IP,
322 	},
323 	{
324 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
325 		.ptype = RTE_PTYPE_TUNNEL_IP,
326 	},
327 	{
328 		.tunnel = MLX5_FLOW_LAYER_GTP,
329 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
330 	},
331 };
332 
333 /**
334  * Translate tag ID to register.
335  *
336  * @param[in] dev
337  *   Pointer to the Ethernet device structure.
338  * @param[in] feature
339  *   The feature that request the register.
340  * @param[in] id
341  *   The request register ID.
342  * @param[out] error
343  *   Error description in case of any.
344  *
345  * @return
346  *   The request register on success, a negative errno
347  *   value otherwise and rte_errno is set.
348  */
349 int
350 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
351 		     enum mlx5_feature_name feature,
352 		     uint32_t id,
353 		     struct rte_flow_error *error)
354 {
355 	struct mlx5_priv *priv = dev->data->dev_private;
356 	struct mlx5_dev_config *config = &priv->config;
357 	enum modify_reg start_reg;
358 	bool skip_mtr_reg = false;
359 
360 	switch (feature) {
361 	case MLX5_HAIRPIN_RX:
362 		return REG_B;
363 	case MLX5_HAIRPIN_TX:
364 		return REG_A;
365 	case MLX5_METADATA_RX:
366 		switch (config->dv_xmeta_en) {
367 		case MLX5_XMETA_MODE_LEGACY:
368 			return REG_B;
369 		case MLX5_XMETA_MODE_META16:
370 			return REG_C_0;
371 		case MLX5_XMETA_MODE_META32:
372 			return REG_C_1;
373 		}
374 		break;
375 	case MLX5_METADATA_TX:
376 		return REG_A;
377 	case MLX5_METADATA_FDB:
378 		switch (config->dv_xmeta_en) {
379 		case MLX5_XMETA_MODE_LEGACY:
380 			return REG_NONE;
381 		case MLX5_XMETA_MODE_META16:
382 			return REG_C_0;
383 		case MLX5_XMETA_MODE_META32:
384 			return REG_C_1;
385 		}
386 		break;
387 	case MLX5_FLOW_MARK:
388 		switch (config->dv_xmeta_en) {
389 		case MLX5_XMETA_MODE_LEGACY:
390 			return REG_NONE;
391 		case MLX5_XMETA_MODE_META16:
392 			return REG_C_1;
393 		case MLX5_XMETA_MODE_META32:
394 			return REG_C_0;
395 		}
396 		break;
397 	case MLX5_MTR_SFX:
398 		/*
399 		 * If meter color and flow match share one register, flow match
400 		 * should use the meter color register for match.
401 		 */
402 		if (priv->mtr_reg_share)
403 			return priv->mtr_color_reg;
404 		else
405 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
406 			       REG_C_3;
407 	case MLX5_MTR_COLOR:
408 		MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
409 		return priv->mtr_color_reg;
410 	case MLX5_COPY_MARK:
411 		/*
412 		 * Metadata COPY_MARK register using is in meter suffix sub
413 		 * flow while with meter. It's safe to share the same register.
414 		 */
415 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
416 	case MLX5_APP_TAG:
417 		/*
418 		 * If meter is enable, it will engage the register for color
419 		 * match and flow match. If meter color match is not using the
420 		 * REG_C_2, need to skip the REG_C_x be used by meter color
421 		 * match.
422 		 * If meter is disable, free to use all available registers.
423 		 */
424 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
425 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
426 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
427 		if (id > (REG_C_7 - start_reg))
428 			return rte_flow_error_set(error, EINVAL,
429 						  RTE_FLOW_ERROR_TYPE_ITEM,
430 						  NULL, "invalid tag id");
431 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
432 			return rte_flow_error_set(error, ENOTSUP,
433 						  RTE_FLOW_ERROR_TYPE_ITEM,
434 						  NULL, "unsupported tag id");
435 		/*
436 		 * This case means meter is using the REG_C_x great than 2.
437 		 * Take care not to conflict with meter color REG_C_x.
438 		 * If the available index REG_C_y >= REG_C_x, skip the
439 		 * color register.
440 		 */
441 		if (skip_mtr_reg && config->flow_mreg_c
442 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
443 			if (config->flow_mreg_c
444 			    [id + 1 + start_reg - REG_C_0] != REG_NONE)
445 				return config->flow_mreg_c
446 					       [id + 1 + start_reg - REG_C_0];
447 			return rte_flow_error_set(error, ENOTSUP,
448 						  RTE_FLOW_ERROR_TYPE_ITEM,
449 						  NULL, "unsupported tag id");
450 		}
451 		return config->flow_mreg_c[id + start_reg - REG_C_0];
452 	}
453 	MLX5_ASSERT(false);
454 	return rte_flow_error_set(error, EINVAL,
455 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
456 				  NULL, "invalid feature name");
457 }
458 
459 /**
460  * Check extensive flow metadata register support.
461  *
462  * @param dev
463  *   Pointer to rte_eth_dev structure.
464  *
465  * @return
466  *   True if device supports extensive flow metadata register, otherwise false.
467  */
468 bool
469 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
470 {
471 	struct mlx5_priv *priv = dev->data->dev_private;
472 	struct mlx5_dev_config *config = &priv->config;
473 
474 	/*
475 	 * Having available reg_c can be regarded inclusively as supporting
476 	 * extensive flow metadata register, which could mean,
477 	 * - metadata register copy action by modify header.
478 	 * - 16 modify header actions is supported.
479 	 * - reg_c's are preserved across different domain (FDB and NIC) on
480 	 *   packet loopback by flow lookup miss.
481 	 */
482 	return config->flow_mreg_c[2] != REG_NONE;
483 }
484 
485 /**
486  * Discover the maximum number of priority available.
487  *
488  * @param[in] dev
489  *   Pointer to the Ethernet device structure.
490  *
491  * @return
492  *   number of supported flow priority on success, a negative errno
493  *   value otherwise and rte_errno is set.
494  */
495 int
496 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
497 {
498 	struct mlx5_priv *priv = dev->data->dev_private;
499 	struct {
500 		struct ibv_flow_attr attr;
501 		struct ibv_flow_spec_eth eth;
502 		struct ibv_flow_spec_action_drop drop;
503 	} flow_attr = {
504 		.attr = {
505 			.num_of_specs = 2,
506 			.port = (uint8_t)priv->ibv_port,
507 		},
508 		.eth = {
509 			.type = IBV_FLOW_SPEC_ETH,
510 			.size = sizeof(struct ibv_flow_spec_eth),
511 		},
512 		.drop = {
513 			.size = sizeof(struct ibv_flow_spec_action_drop),
514 			.type = IBV_FLOW_SPEC_ACTION_DROP,
515 		},
516 	};
517 	struct ibv_flow *flow;
518 	struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
519 	uint16_t vprio[] = { 8, 16 };
520 	int i;
521 	int priority = 0;
522 
523 	if (!drop) {
524 		rte_errno = ENOTSUP;
525 		return -rte_errno;
526 	}
527 	for (i = 0; i != RTE_DIM(vprio); i++) {
528 		flow_attr.attr.priority = vprio[i] - 1;
529 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
530 		if (!flow)
531 			break;
532 		claim_zero(mlx5_glue->destroy_flow(flow));
533 		priority = vprio[i];
534 	}
535 	mlx5_hrxq_drop_release(dev);
536 	switch (priority) {
537 	case 8:
538 		priority = RTE_DIM(priority_map_3);
539 		break;
540 	case 16:
541 		priority = RTE_DIM(priority_map_5);
542 		break;
543 	default:
544 		rte_errno = ENOTSUP;
545 		DRV_LOG(ERR,
546 			"port %u verbs maximum priority: %d expected 8/16",
547 			dev->data->port_id, priority);
548 		return -rte_errno;
549 	}
550 	DRV_LOG(INFO, "port %u flow maximum priority: %d",
551 		dev->data->port_id, priority);
552 	return priority;
553 }
554 
555 /**
556  * Adjust flow priority based on the highest layer and the request priority.
557  *
558  * @param[in] dev
559  *   Pointer to the Ethernet device structure.
560  * @param[in] priority
561  *   The rule base priority.
562  * @param[in] subpriority
563  *   The priority based on the items.
564  *
565  * @return
566  *   The new priority.
567  */
568 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
569 				   uint32_t subpriority)
570 {
571 	uint32_t res = 0;
572 	struct mlx5_priv *priv = dev->data->dev_private;
573 
574 	switch (priv->config.flow_prio) {
575 	case RTE_DIM(priority_map_3):
576 		res = priority_map_3[priority][subpriority];
577 		break;
578 	case RTE_DIM(priority_map_5):
579 		res = priority_map_5[priority][subpriority];
580 		break;
581 	}
582 	return  res;
583 }
584 
585 /**
586  * Verify the @p item specifications (spec, last, mask) are compatible with the
587  * NIC capabilities.
588  *
589  * @param[in] item
590  *   Item specification.
591  * @param[in] mask
592  *   @p item->mask or flow default bit-masks.
593  * @param[in] nic_mask
594  *   Bit-masks covering supported fields by the NIC to compare with user mask.
595  * @param[in] size
596  *   Bit-masks size in bytes.
597  * @param[out] error
598  *   Pointer to error structure.
599  *
600  * @return
601  *   0 on success, a negative errno value otherwise and rte_errno is set.
602  */
603 int
604 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
605 			  const uint8_t *mask,
606 			  const uint8_t *nic_mask,
607 			  unsigned int size,
608 			  struct rte_flow_error *error)
609 {
610 	unsigned int i;
611 
612 	MLX5_ASSERT(nic_mask);
613 	for (i = 0; i < size; ++i)
614 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
615 			return rte_flow_error_set(error, ENOTSUP,
616 						  RTE_FLOW_ERROR_TYPE_ITEM,
617 						  item,
618 						  "mask enables non supported"
619 						  " bits");
620 	if (!item->spec && (item->mask || item->last))
621 		return rte_flow_error_set(error, EINVAL,
622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
623 					  "mask/last without a spec is not"
624 					  " supported");
625 	if (item->spec && item->last) {
626 		uint8_t spec[size];
627 		uint8_t last[size];
628 		unsigned int i;
629 		int ret;
630 
631 		for (i = 0; i < size; ++i) {
632 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
633 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
634 		}
635 		ret = memcmp(spec, last, size);
636 		if (ret != 0)
637 			return rte_flow_error_set(error, EINVAL,
638 						  RTE_FLOW_ERROR_TYPE_ITEM,
639 						  item,
640 						  "range is not valid");
641 	}
642 	return 0;
643 }
644 
645 /**
646  * Adjust the hash fields according to the @p flow information.
647  *
648  * @param[in] dev_flow.
649  *   Pointer to the mlx5_flow.
650  * @param[in] tunnel
651  *   1 when the hash field is for a tunnel item.
652  * @param[in] layer_types
653  *   ETH_RSS_* types.
654  * @param[in] hash_fields
655  *   Item hash fields.
656  *
657  * @return
658  *   The hash fields that should be used.
659  */
660 uint64_t
661 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
662 			    int tunnel __rte_unused, uint64_t layer_types,
663 			    uint64_t hash_fields)
664 {
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 	int rss_request_inner = rss_desc->level >= 2;
667 
668 	/* Check RSS hash level for tunnel. */
669 	if (tunnel && rss_request_inner)
670 		hash_fields |= IBV_RX_HASH_INNER;
671 	else if (tunnel || rss_request_inner)
672 		return 0;
673 #endif
674 	/* Check if requested layer matches RSS hash fields. */
675 	if (!(rss_desc->types & layer_types))
676 		return 0;
677 	return hash_fields;
678 }
679 
680 /**
681  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
682  * if several tunnel rules are used on this queue, the tunnel ptype will be
683  * cleared.
684  *
685  * @param rxq_ctrl
686  *   Rx queue to update.
687  */
688 static void
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
690 {
691 	unsigned int i;
692 	uint32_t tunnel_ptype = 0;
693 
694 	/* Look up for the ptype to use. */
695 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 		if (!rxq_ctrl->flow_tunnels_n[i])
697 			continue;
698 		if (!tunnel_ptype) {
699 			tunnel_ptype = tunnels_info[i].ptype;
700 		} else {
701 			tunnel_ptype = 0;
702 			break;
703 		}
704 	}
705 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
706 }
707 
708 /**
709  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
710  * flow.
711  *
712  * @param[in] dev
713  *   Pointer to the Ethernet device structure.
714  * @param[in] dev_handle
715  *   Pointer to device flow handle structure.
716  */
717 static void
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
719 		       struct mlx5_flow_handle *dev_handle)
720 {
721 	struct mlx5_priv *priv = dev->data->dev_private;
722 	const int mark = dev_handle->mark;
723 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
724 	struct mlx5_hrxq *hrxq;
725 	unsigned int i;
726 
727 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
728 		return;
729 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
730 			      dev_handle->rix_hrxq);
731 	if (!hrxq)
732 		return;
733 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
734 		int idx = hrxq->ind_table->queues[i];
735 		struct mlx5_rxq_ctrl *rxq_ctrl =
736 			container_of((*priv->rxqs)[idx],
737 				     struct mlx5_rxq_ctrl, rxq);
738 
739 		/*
740 		 * To support metadata register copy on Tx loopback,
741 		 * this must be always enabled (metadata may arive
742 		 * from other port - not from local flows only.
743 		 */
744 		if (priv->config.dv_flow_en &&
745 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
746 		    mlx5_flow_ext_mreg_supported(dev)) {
747 			rxq_ctrl->rxq.mark = 1;
748 			rxq_ctrl->flow_mark_n = 1;
749 		} else if (mark) {
750 			rxq_ctrl->rxq.mark = 1;
751 			rxq_ctrl->flow_mark_n++;
752 		}
753 		if (tunnel) {
754 			unsigned int j;
755 
756 			/* Increase the counter matching the flow. */
757 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
758 				if ((tunnels_info[j].tunnel &
759 				     dev_handle->layers) ==
760 				    tunnels_info[j].tunnel) {
761 					rxq_ctrl->flow_tunnels_n[j]++;
762 					break;
763 				}
764 			}
765 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
766 		}
767 	}
768 }
769 
770 /**
771  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
772  *
773  * @param[in] dev
774  *   Pointer to the Ethernet device structure.
775  * @param[in] flow
776  *   Pointer to flow structure.
777  */
778 static void
779 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
780 {
781 	struct mlx5_priv *priv = dev->data->dev_private;
782 	uint32_t handle_idx;
783 	struct mlx5_flow_handle *dev_handle;
784 
785 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
786 		       handle_idx, dev_handle, next)
787 		flow_drv_rxq_flags_set(dev, dev_handle);
788 }
789 
790 /**
791  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
792  * device flow if no other flow uses it with the same kind of request.
793  *
794  * @param dev
795  *   Pointer to Ethernet device.
796  * @param[in] dev_handle
797  *   Pointer to the device flow handle structure.
798  */
799 static void
800 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
801 			struct mlx5_flow_handle *dev_handle)
802 {
803 	struct mlx5_priv *priv = dev->data->dev_private;
804 	const int mark = dev_handle->mark;
805 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
806 	struct mlx5_hrxq *hrxq;
807 	unsigned int i;
808 
809 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
810 		return;
811 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
812 			      dev_handle->rix_hrxq);
813 	if (!hrxq)
814 		return;
815 	MLX5_ASSERT(dev->data->dev_started);
816 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
817 		int idx = hrxq->ind_table->queues[i];
818 		struct mlx5_rxq_ctrl *rxq_ctrl =
819 			container_of((*priv->rxqs)[idx],
820 				     struct mlx5_rxq_ctrl, rxq);
821 
822 		if (priv->config.dv_flow_en &&
823 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
824 		    mlx5_flow_ext_mreg_supported(dev)) {
825 			rxq_ctrl->rxq.mark = 1;
826 			rxq_ctrl->flow_mark_n = 1;
827 		} else if (mark) {
828 			rxq_ctrl->flow_mark_n--;
829 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
830 		}
831 		if (tunnel) {
832 			unsigned int j;
833 
834 			/* Decrease the counter matching the flow. */
835 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
836 				if ((tunnels_info[j].tunnel &
837 				     dev_handle->layers) ==
838 				    tunnels_info[j].tunnel) {
839 					rxq_ctrl->flow_tunnels_n[j]--;
840 					break;
841 				}
842 			}
843 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
844 		}
845 	}
846 }
847 
848 /**
849  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
850  * @p flow if no other flow uses it with the same kind of request.
851  *
852  * @param dev
853  *   Pointer to Ethernet device.
854  * @param[in] flow
855  *   Pointer to the flow.
856  */
857 static void
858 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
859 {
860 	struct mlx5_priv *priv = dev->data->dev_private;
861 	uint32_t handle_idx;
862 	struct mlx5_flow_handle *dev_handle;
863 
864 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
865 		       handle_idx, dev_handle, next)
866 		flow_drv_rxq_flags_trim(dev, dev_handle);
867 }
868 
869 /**
870  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
871  *
872  * @param dev
873  *   Pointer to Ethernet device.
874  */
875 static void
876 flow_rxq_flags_clear(struct rte_eth_dev *dev)
877 {
878 	struct mlx5_priv *priv = dev->data->dev_private;
879 	unsigned int i;
880 
881 	for (i = 0; i != priv->rxqs_n; ++i) {
882 		struct mlx5_rxq_ctrl *rxq_ctrl;
883 		unsigned int j;
884 
885 		if (!(*priv->rxqs)[i])
886 			continue;
887 		rxq_ctrl = container_of((*priv->rxqs)[i],
888 					struct mlx5_rxq_ctrl, rxq);
889 		rxq_ctrl->flow_mark_n = 0;
890 		rxq_ctrl->rxq.mark = 0;
891 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
892 			rxq_ctrl->flow_tunnels_n[j] = 0;
893 		rxq_ctrl->rxq.tunnel = 0;
894 	}
895 }
896 
897 /**
898  * Set the Rx queue dynamic metadata (mask and offset) for a flow
899  *
900  * @param[in] dev
901  *   Pointer to the Ethernet device structure.
902  */
903 void
904 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
905 {
906 	struct mlx5_priv *priv = dev->data->dev_private;
907 	struct mlx5_rxq_data *data;
908 	unsigned int i;
909 
910 	for (i = 0; i != priv->rxqs_n; ++i) {
911 		if (!(*priv->rxqs)[i])
912 			continue;
913 		data = (*priv->rxqs)[i];
914 		if (!rte_flow_dynf_metadata_avail()) {
915 			data->dynf_meta = 0;
916 			data->flow_meta_mask = 0;
917 			data->flow_meta_offset = -1;
918 		} else {
919 			data->dynf_meta = 1;
920 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
921 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
922 		}
923 	}
924 }
925 
926 /*
927  * return a pointer to the desired action in the list of actions.
928  *
929  * @param[in] actions
930  *   The list of actions to search the action in.
931  * @param[in] action
932  *   The action to find.
933  *
934  * @return
935  *   Pointer to the action in the list, if found. NULL otherwise.
936  */
937 const struct rte_flow_action *
938 mlx5_flow_find_action(const struct rte_flow_action *actions,
939 		      enum rte_flow_action_type action)
940 {
941 	if (actions == NULL)
942 		return NULL;
943 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
944 		if (actions->type == action)
945 			return actions;
946 	return NULL;
947 }
948 
949 /*
950  * Validate the flag action.
951  *
952  * @param[in] action_flags
953  *   Bit-fields that holds the actions detected until now.
954  * @param[in] attr
955  *   Attributes of flow that includes this action.
956  * @param[out] error
957  *   Pointer to error structure.
958  *
959  * @return
960  *   0 on success, a negative errno value otherwise and rte_errno is set.
961  */
962 int
963 mlx5_flow_validate_action_flag(uint64_t action_flags,
964 			       const struct rte_flow_attr *attr,
965 			       struct rte_flow_error *error)
966 {
967 	if (action_flags & MLX5_FLOW_ACTION_MARK)
968 		return rte_flow_error_set(error, EINVAL,
969 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
970 					  "can't mark and flag in same flow");
971 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
972 		return rte_flow_error_set(error, EINVAL,
973 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
974 					  "can't have 2 flag"
975 					  " actions in same flow");
976 	if (attr->egress)
977 		return rte_flow_error_set(error, ENOTSUP,
978 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
979 					  "flag action not supported for "
980 					  "egress");
981 	return 0;
982 }
983 
984 /*
985  * Validate the mark action.
986  *
987  * @param[in] action
988  *   Pointer to the queue action.
989  * @param[in] action_flags
990  *   Bit-fields that holds the actions detected until now.
991  * @param[in] attr
992  *   Attributes of flow that includes this action.
993  * @param[out] error
994  *   Pointer to error structure.
995  *
996  * @return
997  *   0 on success, a negative errno value otherwise and rte_errno is set.
998  */
999 int
1000 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1001 			       uint64_t action_flags,
1002 			       const struct rte_flow_attr *attr,
1003 			       struct rte_flow_error *error)
1004 {
1005 	const struct rte_flow_action_mark *mark = action->conf;
1006 
1007 	if (!mark)
1008 		return rte_flow_error_set(error, EINVAL,
1009 					  RTE_FLOW_ERROR_TYPE_ACTION,
1010 					  action,
1011 					  "configuration cannot be null");
1012 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1013 		return rte_flow_error_set(error, EINVAL,
1014 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1015 					  &mark->id,
1016 					  "mark id must in 0 <= id < "
1017 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1018 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1019 		return rte_flow_error_set(error, EINVAL,
1020 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1021 					  "can't flag and mark in same flow");
1022 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1023 		return rte_flow_error_set(error, EINVAL,
1024 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1025 					  "can't have 2 mark actions in same"
1026 					  " flow");
1027 	if (attr->egress)
1028 		return rte_flow_error_set(error, ENOTSUP,
1029 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1030 					  "mark action not supported for "
1031 					  "egress");
1032 	return 0;
1033 }
1034 
1035 /*
1036  * Validate the drop action.
1037  *
1038  * @param[in] action_flags
1039  *   Bit-fields that holds the actions detected until now.
1040  * @param[in] attr
1041  *   Attributes of flow that includes this action.
1042  * @param[out] error
1043  *   Pointer to error structure.
1044  *
1045  * @return
1046  *   0 on success, a negative errno value otherwise and rte_errno is set.
1047  */
1048 int
1049 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1050 			       const struct rte_flow_attr *attr,
1051 			       struct rte_flow_error *error)
1052 {
1053 	if (attr->egress)
1054 		return rte_flow_error_set(error, ENOTSUP,
1055 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1056 					  "drop action not supported for "
1057 					  "egress");
1058 	return 0;
1059 }
1060 
1061 /*
1062  * Validate the queue action.
1063  *
1064  * @param[in] action
1065  *   Pointer to the queue action.
1066  * @param[in] action_flags
1067  *   Bit-fields that holds the actions detected until now.
1068  * @param[in] dev
1069  *   Pointer to the Ethernet device structure.
1070  * @param[in] attr
1071  *   Attributes of flow that includes this action.
1072  * @param[out] error
1073  *   Pointer to error structure.
1074  *
1075  * @return
1076  *   0 on success, a negative errno value otherwise and rte_errno is set.
1077  */
1078 int
1079 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1080 				uint64_t action_flags,
1081 				struct rte_eth_dev *dev,
1082 				const struct rte_flow_attr *attr,
1083 				struct rte_flow_error *error)
1084 {
1085 	struct mlx5_priv *priv = dev->data->dev_private;
1086 	const struct rte_flow_action_queue *queue = action->conf;
1087 
1088 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1089 		return rte_flow_error_set(error, EINVAL,
1090 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1091 					  "can't have 2 fate actions in"
1092 					  " same flow");
1093 	if (!priv->rxqs_n)
1094 		return rte_flow_error_set(error, EINVAL,
1095 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1096 					  NULL, "No Rx queues configured");
1097 	if (queue->index >= priv->rxqs_n)
1098 		return rte_flow_error_set(error, EINVAL,
1099 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1100 					  &queue->index,
1101 					  "queue index out of range");
1102 	if (!(*priv->rxqs)[queue->index])
1103 		return rte_flow_error_set(error, EINVAL,
1104 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1105 					  &queue->index,
1106 					  "queue is not configured");
1107 	if (attr->egress)
1108 		return rte_flow_error_set(error, ENOTSUP,
1109 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1110 					  "queue action not supported for "
1111 					  "egress");
1112 	return 0;
1113 }
1114 
1115 /*
1116  * Validate the rss action.
1117  *
1118  * @param[in] action
1119  *   Pointer to the queue action.
1120  * @param[in] action_flags
1121  *   Bit-fields that holds the actions detected until now.
1122  * @param[in] dev
1123  *   Pointer to the Ethernet device structure.
1124  * @param[in] attr
1125  *   Attributes of flow that includes this action.
1126  * @param[in] item_flags
1127  *   Items that were detected.
1128  * @param[out] error
1129  *   Pointer to error structure.
1130  *
1131  * @return
1132  *   0 on success, a negative errno value otherwise and rte_errno is set.
1133  */
1134 int
1135 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1136 			      uint64_t action_flags,
1137 			      struct rte_eth_dev *dev,
1138 			      const struct rte_flow_attr *attr,
1139 			      uint64_t item_flags,
1140 			      struct rte_flow_error *error)
1141 {
1142 	struct mlx5_priv *priv = dev->data->dev_private;
1143 	const struct rte_flow_action_rss *rss = action->conf;
1144 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1145 	unsigned int i;
1146 
1147 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1148 		return rte_flow_error_set(error, EINVAL,
1149 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1150 					  "can't have 2 fate actions"
1151 					  " in same flow");
1152 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1153 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1154 		return rte_flow_error_set(error, ENOTSUP,
1155 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1156 					  &rss->func,
1157 					  "RSS hash function not supported");
1158 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1159 	if (rss->level > 2)
1160 #else
1161 	if (rss->level > 1)
1162 #endif
1163 		return rte_flow_error_set(error, ENOTSUP,
1164 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1165 					  &rss->level,
1166 					  "tunnel RSS is not supported");
1167 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1168 	if (rss->key_len == 0 && rss->key != NULL)
1169 		return rte_flow_error_set(error, ENOTSUP,
1170 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1171 					  &rss->key_len,
1172 					  "RSS hash key length 0");
1173 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1174 		return rte_flow_error_set(error, ENOTSUP,
1175 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1176 					  &rss->key_len,
1177 					  "RSS hash key too small");
1178 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1179 		return rte_flow_error_set(error, ENOTSUP,
1180 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1181 					  &rss->key_len,
1182 					  "RSS hash key too large");
1183 	if (rss->queue_num > priv->config.ind_table_max_size)
1184 		return rte_flow_error_set(error, ENOTSUP,
1185 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1186 					  &rss->queue_num,
1187 					  "number of queues too large");
1188 	if (rss->types & MLX5_RSS_HF_MASK)
1189 		return rte_flow_error_set(error, ENOTSUP,
1190 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1191 					  &rss->types,
1192 					  "some RSS protocols are not"
1193 					  " supported");
1194 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1195 	    !(rss->types & ETH_RSS_IP))
1196 		return rte_flow_error_set(error, EINVAL,
1197 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1198 					  "L3 partial RSS requested but L3 RSS"
1199 					  " type not specified");
1200 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1201 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1202 		return rte_flow_error_set(error, EINVAL,
1203 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204 					  "L4 partial RSS requested but L4 RSS"
1205 					  " type not specified");
1206 	if (!priv->rxqs_n)
1207 		return rte_flow_error_set(error, EINVAL,
1208 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1209 					  NULL, "No Rx queues configured");
1210 	if (!rss->queue_num)
1211 		return rte_flow_error_set(error, EINVAL,
1212 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1213 					  NULL, "No queues configured");
1214 	for (i = 0; i != rss->queue_num; ++i) {
1215 		if (rss->queue[i] >= priv->rxqs_n)
1216 			return rte_flow_error_set
1217 				(error, EINVAL,
1218 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1219 				 &rss->queue[i], "queue index out of range");
1220 		if (!(*priv->rxqs)[rss->queue[i]])
1221 			return rte_flow_error_set
1222 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1223 				 &rss->queue[i], "queue is not configured");
1224 	}
1225 	if (attr->egress)
1226 		return rte_flow_error_set(error, ENOTSUP,
1227 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1228 					  "rss action not supported for "
1229 					  "egress");
1230 	if (rss->level > 1 &&  !tunnel)
1231 		return rte_flow_error_set(error, EINVAL,
1232 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1233 					  "inner RSS is not supported for "
1234 					  "non-tunnel flows");
1235 	return 0;
1236 }
1237 
1238 /*
1239  * Validate the count action.
1240  *
1241  * @param[in] dev
1242  *   Pointer to the Ethernet device structure.
1243  * @param[in] attr
1244  *   Attributes of flow that includes this action.
1245  * @param[out] error
1246  *   Pointer to error structure.
1247  *
1248  * @return
1249  *   0 on success, a negative errno value otherwise and rte_errno is set.
1250  */
1251 int
1252 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1253 				const struct rte_flow_attr *attr,
1254 				struct rte_flow_error *error)
1255 {
1256 	if (attr->egress)
1257 		return rte_flow_error_set(error, ENOTSUP,
1258 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1259 					  "count action not supported for "
1260 					  "egress");
1261 	return 0;
1262 }
1263 
1264 /**
1265  * Verify the @p attributes will be correctly understood by the NIC and store
1266  * them in the @p flow if everything is correct.
1267  *
1268  * @param[in] dev
1269  *   Pointer to the Ethernet device structure.
1270  * @param[in] attributes
1271  *   Pointer to flow attributes
1272  * @param[out] error
1273  *   Pointer to error structure.
1274  *
1275  * @return
1276  *   0 on success, a negative errno value otherwise and rte_errno is set.
1277  */
1278 int
1279 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1280 			      const struct rte_flow_attr *attributes,
1281 			      struct rte_flow_error *error)
1282 {
1283 	struct mlx5_priv *priv = dev->data->dev_private;
1284 	uint32_t priority_max = priv->config.flow_prio - 1;
1285 
1286 	if (attributes->group)
1287 		return rte_flow_error_set(error, ENOTSUP,
1288 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1289 					  NULL, "groups is not supported");
1290 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1291 	    attributes->priority >= priority_max)
1292 		return rte_flow_error_set(error, ENOTSUP,
1293 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1294 					  NULL, "priority out of range");
1295 	if (attributes->egress)
1296 		return rte_flow_error_set(error, ENOTSUP,
1297 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1298 					  "egress is not supported");
1299 	if (attributes->transfer && !priv->config.dv_esw_en)
1300 		return rte_flow_error_set(error, ENOTSUP,
1301 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1302 					  NULL, "transfer is not supported");
1303 	if (!attributes->ingress)
1304 		return rte_flow_error_set(error, EINVAL,
1305 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1306 					  NULL,
1307 					  "ingress attribute is mandatory");
1308 	return 0;
1309 }
1310 
1311 /**
1312  * Validate ICMP6 item.
1313  *
1314  * @param[in] item
1315  *   Item specification.
1316  * @param[in] item_flags
1317  *   Bit-fields that holds the items detected until now.
1318  * @param[out] error
1319  *   Pointer to error structure.
1320  *
1321  * @return
1322  *   0 on success, a negative errno value otherwise and rte_errno is set.
1323  */
1324 int
1325 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1326 			       uint64_t item_flags,
1327 			       uint8_t target_protocol,
1328 			       struct rte_flow_error *error)
1329 {
1330 	const struct rte_flow_item_icmp6 *mask = item->mask;
1331 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1332 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1333 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1334 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1335 				      MLX5_FLOW_LAYER_OUTER_L4;
1336 	int ret;
1337 
1338 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1339 		return rte_flow_error_set(error, EINVAL,
1340 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1341 					  "protocol filtering not compatible"
1342 					  " with ICMP6 layer");
1343 	if (!(item_flags & l3m))
1344 		return rte_flow_error_set(error, EINVAL,
1345 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1346 					  "IPv6 is mandatory to filter on"
1347 					  " ICMP6");
1348 	if (item_flags & l4m)
1349 		return rte_flow_error_set(error, EINVAL,
1350 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1351 					  "multiple L4 layers not supported");
1352 	if (!mask)
1353 		mask = &rte_flow_item_icmp6_mask;
1354 	ret = mlx5_flow_item_acceptable
1355 		(item, (const uint8_t *)mask,
1356 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1357 		 sizeof(struct rte_flow_item_icmp6), error);
1358 	if (ret < 0)
1359 		return ret;
1360 	return 0;
1361 }
1362 
1363 /**
1364  * Validate ICMP item.
1365  *
1366  * @param[in] item
1367  *   Item specification.
1368  * @param[in] item_flags
1369  *   Bit-fields that holds the items detected until now.
1370  * @param[out] error
1371  *   Pointer to error structure.
1372  *
1373  * @return
1374  *   0 on success, a negative errno value otherwise and rte_errno is set.
1375  */
1376 int
1377 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1378 			     uint64_t item_flags,
1379 			     uint8_t target_protocol,
1380 			     struct rte_flow_error *error)
1381 {
1382 	const struct rte_flow_item_icmp *mask = item->mask;
1383 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1384 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1385 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1386 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1387 				      MLX5_FLOW_LAYER_OUTER_L4;
1388 	int ret;
1389 
1390 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1391 		return rte_flow_error_set(error, EINVAL,
1392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1393 					  "protocol filtering not compatible"
1394 					  " with ICMP layer");
1395 	if (!(item_flags & l3m))
1396 		return rte_flow_error_set(error, EINVAL,
1397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 					  "IPv4 is mandatory to filter"
1399 					  " on ICMP");
1400 	if (item_flags & l4m)
1401 		return rte_flow_error_set(error, EINVAL,
1402 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 					  "multiple L4 layers not supported");
1404 	if (!mask)
1405 		mask = &rte_flow_item_icmp_mask;
1406 	ret = mlx5_flow_item_acceptable
1407 		(item, (const uint8_t *)mask,
1408 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1409 		 sizeof(struct rte_flow_item_icmp), error);
1410 	if (ret < 0)
1411 		return ret;
1412 	return 0;
1413 }
1414 
1415 /**
1416  * Validate Ethernet item.
1417  *
1418  * @param[in] item
1419  *   Item specification.
1420  * @param[in] item_flags
1421  *   Bit-fields that holds the items detected until now.
1422  * @param[out] error
1423  *   Pointer to error structure.
1424  *
1425  * @return
1426  *   0 on success, a negative errno value otherwise and rte_errno is set.
1427  */
1428 int
1429 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1430 			    uint64_t item_flags,
1431 			    struct rte_flow_error *error)
1432 {
1433 	const struct rte_flow_item_eth *mask = item->mask;
1434 	const struct rte_flow_item_eth nic_mask = {
1435 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1436 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1437 		.type = RTE_BE16(0xffff),
1438 	};
1439 	int ret;
1440 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1441 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1442 				       MLX5_FLOW_LAYER_OUTER_L2;
1443 
1444 	if (item_flags & ethm)
1445 		return rte_flow_error_set(error, ENOTSUP,
1446 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1447 					  "multiple L2 layers not supported");
1448 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1449 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1450 		return rte_flow_error_set(error, EINVAL,
1451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1452 					  "L2 layer should not follow "
1453 					  "L3 layers");
1454 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1455 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1456 		return rte_flow_error_set(error, EINVAL,
1457 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1458 					  "L2 layer should not follow VLAN");
1459 	if (!mask)
1460 		mask = &rte_flow_item_eth_mask;
1461 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1462 					(const uint8_t *)&nic_mask,
1463 					sizeof(struct rte_flow_item_eth),
1464 					error);
1465 	return ret;
1466 }
1467 
1468 /**
1469  * Validate VLAN item.
1470  *
1471  * @param[in] item
1472  *   Item specification.
1473  * @param[in] item_flags
1474  *   Bit-fields that holds the items detected until now.
1475  * @param[in] dev
1476  *   Ethernet device flow is being created on.
1477  * @param[out] error
1478  *   Pointer to error structure.
1479  *
1480  * @return
1481  *   0 on success, a negative errno value otherwise and rte_errno is set.
1482  */
1483 int
1484 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1485 			     uint64_t item_flags,
1486 			     struct rte_eth_dev *dev,
1487 			     struct rte_flow_error *error)
1488 {
1489 	const struct rte_flow_item_vlan *spec = item->spec;
1490 	const struct rte_flow_item_vlan *mask = item->mask;
1491 	const struct rte_flow_item_vlan nic_mask = {
1492 		.tci = RTE_BE16(UINT16_MAX),
1493 		.inner_type = RTE_BE16(UINT16_MAX),
1494 	};
1495 	uint16_t vlan_tag = 0;
1496 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1497 	int ret;
1498 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1499 					MLX5_FLOW_LAYER_INNER_L4) :
1500 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1501 					MLX5_FLOW_LAYER_OUTER_L4);
1502 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1503 					MLX5_FLOW_LAYER_OUTER_VLAN;
1504 
1505 	if (item_flags & vlanm)
1506 		return rte_flow_error_set(error, EINVAL,
1507 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1508 					  "multiple VLAN layers not supported");
1509 	else if ((item_flags & l34m) != 0)
1510 		return rte_flow_error_set(error, EINVAL,
1511 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1512 					  "VLAN cannot follow L3/L4 layer");
1513 	if (!mask)
1514 		mask = &rte_flow_item_vlan_mask;
1515 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1516 					(const uint8_t *)&nic_mask,
1517 					sizeof(struct rte_flow_item_vlan),
1518 					error);
1519 	if (ret)
1520 		return ret;
1521 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1522 		struct mlx5_priv *priv = dev->data->dev_private;
1523 
1524 		if (priv->vmwa_context) {
1525 			/*
1526 			 * Non-NULL context means we have a virtual machine
1527 			 * and SR-IOV enabled, we have to create VLAN interface
1528 			 * to make hypervisor to setup E-Switch vport
1529 			 * context correctly. We avoid creating the multiple
1530 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1531 			 */
1532 			return rte_flow_error_set(error, EINVAL,
1533 						  RTE_FLOW_ERROR_TYPE_ITEM,
1534 						  item,
1535 						  "VLAN tag mask is not"
1536 						  " supported in virtual"
1537 						  " environment");
1538 		}
1539 	}
1540 	if (spec) {
1541 		vlan_tag = spec->tci;
1542 		vlan_tag &= mask->tci;
1543 	}
1544 	/*
1545 	 * From verbs perspective an empty VLAN is equivalent
1546 	 * to a packet without VLAN layer.
1547 	 */
1548 	if (!vlan_tag)
1549 		return rte_flow_error_set(error, EINVAL,
1550 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1551 					  item->spec,
1552 					  "VLAN cannot be empty");
1553 	return 0;
1554 }
1555 
1556 /**
1557  * Validate IPV4 item.
1558  *
1559  * @param[in] item
1560  *   Item specification.
1561  * @param[in] item_flags
1562  *   Bit-fields that holds the items detected until now.
1563  * @param[in] acc_mask
1564  *   Acceptable mask, if NULL default internal default mask
1565  *   will be used to check whether item fields are supported.
1566  * @param[out] error
1567  *   Pointer to error structure.
1568  *
1569  * @return
1570  *   0 on success, a negative errno value otherwise and rte_errno is set.
1571  */
1572 int
1573 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1574 			     uint64_t item_flags,
1575 			     uint64_t last_item,
1576 			     uint16_t ether_type,
1577 			     const struct rte_flow_item_ipv4 *acc_mask,
1578 			     struct rte_flow_error *error)
1579 {
1580 	const struct rte_flow_item_ipv4 *mask = item->mask;
1581 	const struct rte_flow_item_ipv4 *spec = item->spec;
1582 	const struct rte_flow_item_ipv4 nic_mask = {
1583 		.hdr = {
1584 			.src_addr = RTE_BE32(0xffffffff),
1585 			.dst_addr = RTE_BE32(0xffffffff),
1586 			.type_of_service = 0xff,
1587 			.next_proto_id = 0xff,
1588 		},
1589 	};
1590 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1591 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1592 				      MLX5_FLOW_LAYER_OUTER_L3;
1593 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1594 				      MLX5_FLOW_LAYER_OUTER_L4;
1595 	int ret;
1596 	uint8_t next_proto = 0xFF;
1597 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1598 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1599 				  MLX5_FLOW_LAYER_INNER_VLAN);
1600 
1601 	if ((last_item & l2_vlan) && ether_type &&
1602 	    ether_type != RTE_ETHER_TYPE_IPV4)
1603 		return rte_flow_error_set(error, EINVAL,
1604 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1605 					  "IPv4 cannot follow L2/VLAN layer "
1606 					  "which ether type is not IPv4");
1607 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1608 		if (mask && spec)
1609 			next_proto = mask->hdr.next_proto_id &
1610 				     spec->hdr.next_proto_id;
1611 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1612 			return rte_flow_error_set(error, EINVAL,
1613 						  RTE_FLOW_ERROR_TYPE_ITEM,
1614 						  item,
1615 						  "multiple tunnel "
1616 						  "not supported");
1617 	}
1618 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1619 		return rte_flow_error_set(error, EINVAL,
1620 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1621 					  "wrong tunnel type - IPv6 specified "
1622 					  "but IPv4 item provided");
1623 	if (item_flags & l3m)
1624 		return rte_flow_error_set(error, ENOTSUP,
1625 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1626 					  "multiple L3 layers not supported");
1627 	else if (item_flags & l4m)
1628 		return rte_flow_error_set(error, EINVAL,
1629 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1630 					  "L3 cannot follow an L4 layer.");
1631 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1632 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1633 		return rte_flow_error_set(error, EINVAL,
1634 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1635 					  "L3 cannot follow an NVGRE layer.");
1636 	if (!mask)
1637 		mask = &rte_flow_item_ipv4_mask;
1638 	else if (mask->hdr.next_proto_id != 0 &&
1639 		 mask->hdr.next_proto_id != 0xff)
1640 		return rte_flow_error_set(error, EINVAL,
1641 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1642 					  "partial mask is not supported"
1643 					  " for protocol");
1644 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1645 					acc_mask ? (const uint8_t *)acc_mask
1646 						 : (const uint8_t *)&nic_mask,
1647 					sizeof(struct rte_flow_item_ipv4),
1648 					error);
1649 	if (ret < 0)
1650 		return ret;
1651 	return 0;
1652 }
1653 
1654 /**
1655  * Validate IPV6 item.
1656  *
1657  * @param[in] item
1658  *   Item specification.
1659  * @param[in] item_flags
1660  *   Bit-fields that holds the items detected until now.
1661  * @param[in] acc_mask
1662  *   Acceptable mask, if NULL default internal default mask
1663  *   will be used to check whether item fields are supported.
1664  * @param[out] error
1665  *   Pointer to error structure.
1666  *
1667  * @return
1668  *   0 on success, a negative errno value otherwise and rte_errno is set.
1669  */
1670 int
1671 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1672 			     uint64_t item_flags,
1673 			     uint64_t last_item,
1674 			     uint16_t ether_type,
1675 			     const struct rte_flow_item_ipv6 *acc_mask,
1676 			     struct rte_flow_error *error)
1677 {
1678 	const struct rte_flow_item_ipv6 *mask = item->mask;
1679 	const struct rte_flow_item_ipv6 *spec = item->spec;
1680 	const struct rte_flow_item_ipv6 nic_mask = {
1681 		.hdr = {
1682 			.src_addr =
1683 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1684 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1685 			.dst_addr =
1686 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1687 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1688 			.vtc_flow = RTE_BE32(0xffffffff),
1689 			.proto = 0xff,
1690 		},
1691 	};
1692 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1693 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1694 				      MLX5_FLOW_LAYER_OUTER_L3;
1695 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1696 				      MLX5_FLOW_LAYER_OUTER_L4;
1697 	int ret;
1698 	uint8_t next_proto = 0xFF;
1699 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1700 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1701 				  MLX5_FLOW_LAYER_INNER_VLAN);
1702 
1703 	if ((last_item & l2_vlan) && ether_type &&
1704 	    ether_type != RTE_ETHER_TYPE_IPV6)
1705 		return rte_flow_error_set(error, EINVAL,
1706 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 					  "IPv6 cannot follow L2/VLAN layer "
1708 					  "which ether type is not IPv6");
1709 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1710 		if (mask && spec)
1711 			next_proto = mask->hdr.proto & spec->hdr.proto;
1712 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1713 			return rte_flow_error_set(error, EINVAL,
1714 						  RTE_FLOW_ERROR_TYPE_ITEM,
1715 						  item,
1716 						  "multiple tunnel "
1717 						  "not supported");
1718 	}
1719 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1720 		return rte_flow_error_set(error, EINVAL,
1721 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1722 					  "wrong tunnel type - IPv4 specified "
1723 					  "but IPv6 item provided");
1724 	if (item_flags & l3m)
1725 		return rte_flow_error_set(error, ENOTSUP,
1726 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1727 					  "multiple L3 layers not supported");
1728 	else if (item_flags & l4m)
1729 		return rte_flow_error_set(error, EINVAL,
1730 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1731 					  "L3 cannot follow an L4 layer.");
1732 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1733 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1734 		return rte_flow_error_set(error, EINVAL,
1735 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1736 					  "L3 cannot follow an NVGRE layer.");
1737 	if (!mask)
1738 		mask = &rte_flow_item_ipv6_mask;
1739 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1740 					acc_mask ? (const uint8_t *)acc_mask
1741 						 : (const uint8_t *)&nic_mask,
1742 					sizeof(struct rte_flow_item_ipv6),
1743 					error);
1744 	if (ret < 0)
1745 		return ret;
1746 	return 0;
1747 }
1748 
1749 /**
1750  * Validate UDP item.
1751  *
1752  * @param[in] item
1753  *   Item specification.
1754  * @param[in] item_flags
1755  *   Bit-fields that holds the items detected until now.
1756  * @param[in] target_protocol
1757  *   The next protocol in the previous item.
1758  * @param[in] flow_mask
1759  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1760  * @param[out] error
1761  *   Pointer to error structure.
1762  *
1763  * @return
1764  *   0 on success, a negative errno value otherwise and rte_errno is set.
1765  */
1766 int
1767 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1768 			    uint64_t item_flags,
1769 			    uint8_t target_protocol,
1770 			    struct rte_flow_error *error)
1771 {
1772 	const struct rte_flow_item_udp *mask = item->mask;
1773 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1774 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1775 				      MLX5_FLOW_LAYER_OUTER_L3;
1776 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1777 				      MLX5_FLOW_LAYER_OUTER_L4;
1778 	int ret;
1779 
1780 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1781 		return rte_flow_error_set(error, EINVAL,
1782 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1783 					  "protocol filtering not compatible"
1784 					  " with UDP layer");
1785 	if (!(item_flags & l3m))
1786 		return rte_flow_error_set(error, EINVAL,
1787 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1788 					  "L3 is mandatory to filter on L4");
1789 	if (item_flags & l4m)
1790 		return rte_flow_error_set(error, EINVAL,
1791 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1792 					  "multiple L4 layers not supported");
1793 	if (!mask)
1794 		mask = &rte_flow_item_udp_mask;
1795 	ret = mlx5_flow_item_acceptable
1796 		(item, (const uint8_t *)mask,
1797 		 (const uint8_t *)&rte_flow_item_udp_mask,
1798 		 sizeof(struct rte_flow_item_udp), error);
1799 	if (ret < 0)
1800 		return ret;
1801 	return 0;
1802 }
1803 
1804 /**
1805  * Validate TCP item.
1806  *
1807  * @param[in] item
1808  *   Item specification.
1809  * @param[in] item_flags
1810  *   Bit-fields that holds the items detected until now.
1811  * @param[in] target_protocol
1812  *   The next protocol in the previous item.
1813  * @param[out] error
1814  *   Pointer to error structure.
1815  *
1816  * @return
1817  *   0 on success, a negative errno value otherwise and rte_errno is set.
1818  */
1819 int
1820 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1821 			    uint64_t item_flags,
1822 			    uint8_t target_protocol,
1823 			    const struct rte_flow_item_tcp *flow_mask,
1824 			    struct rte_flow_error *error)
1825 {
1826 	const struct rte_flow_item_tcp *mask = item->mask;
1827 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1828 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1829 				      MLX5_FLOW_LAYER_OUTER_L3;
1830 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1831 				      MLX5_FLOW_LAYER_OUTER_L4;
1832 	int ret;
1833 
1834 	MLX5_ASSERT(flow_mask);
1835 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1836 		return rte_flow_error_set(error, EINVAL,
1837 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1838 					  "protocol filtering not compatible"
1839 					  " with TCP layer");
1840 	if (!(item_flags & l3m))
1841 		return rte_flow_error_set(error, EINVAL,
1842 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1843 					  "L3 is mandatory to filter on L4");
1844 	if (item_flags & l4m)
1845 		return rte_flow_error_set(error, EINVAL,
1846 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1847 					  "multiple L4 layers not supported");
1848 	if (!mask)
1849 		mask = &rte_flow_item_tcp_mask;
1850 	ret = mlx5_flow_item_acceptable
1851 		(item, (const uint8_t *)mask,
1852 		 (const uint8_t *)flow_mask,
1853 		 sizeof(struct rte_flow_item_tcp), error);
1854 	if (ret < 0)
1855 		return ret;
1856 	return 0;
1857 }
1858 
1859 /**
1860  * Validate VXLAN item.
1861  *
1862  * @param[in] item
1863  *   Item specification.
1864  * @param[in] item_flags
1865  *   Bit-fields that holds the items detected until now.
1866  * @param[in] target_protocol
1867  *   The next protocol in the previous item.
1868  * @param[out] error
1869  *   Pointer to error structure.
1870  *
1871  * @return
1872  *   0 on success, a negative errno value otherwise and rte_errno is set.
1873  */
1874 int
1875 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1876 			      uint64_t item_flags,
1877 			      struct rte_flow_error *error)
1878 {
1879 	const struct rte_flow_item_vxlan *spec = item->spec;
1880 	const struct rte_flow_item_vxlan *mask = item->mask;
1881 	int ret;
1882 	union vni {
1883 		uint32_t vlan_id;
1884 		uint8_t vni[4];
1885 	} id = { .vlan_id = 0, };
1886 
1887 
1888 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1889 		return rte_flow_error_set(error, ENOTSUP,
1890 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1891 					  "multiple tunnel layers not"
1892 					  " supported");
1893 	/*
1894 	 * Verify only UDPv4 is present as defined in
1895 	 * https://tools.ietf.org/html/rfc7348
1896 	 */
1897 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1898 		return rte_flow_error_set(error, EINVAL,
1899 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1900 					  "no outer UDP layer found");
1901 	if (!mask)
1902 		mask = &rte_flow_item_vxlan_mask;
1903 	ret = mlx5_flow_item_acceptable
1904 		(item, (const uint8_t *)mask,
1905 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
1906 		 sizeof(struct rte_flow_item_vxlan),
1907 		 error);
1908 	if (ret < 0)
1909 		return ret;
1910 	if (spec) {
1911 		memcpy(&id.vni[1], spec->vni, 3);
1912 		memcpy(&id.vni[1], mask->vni, 3);
1913 	}
1914 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1915 		return rte_flow_error_set(error, ENOTSUP,
1916 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1917 					  "VXLAN tunnel must be fully defined");
1918 	return 0;
1919 }
1920 
1921 /**
1922  * Validate VXLAN_GPE item.
1923  *
1924  * @param[in] item
1925  *   Item specification.
1926  * @param[in] item_flags
1927  *   Bit-fields that holds the items detected until now.
1928  * @param[in] priv
1929  *   Pointer to the private data structure.
1930  * @param[in] target_protocol
1931  *   The next protocol in the previous item.
1932  * @param[out] error
1933  *   Pointer to error structure.
1934  *
1935  * @return
1936  *   0 on success, a negative errno value otherwise and rte_errno is set.
1937  */
1938 int
1939 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1940 				  uint64_t item_flags,
1941 				  struct rte_eth_dev *dev,
1942 				  struct rte_flow_error *error)
1943 {
1944 	struct mlx5_priv *priv = dev->data->dev_private;
1945 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1946 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1947 	int ret;
1948 	union vni {
1949 		uint32_t vlan_id;
1950 		uint8_t vni[4];
1951 	} id = { .vlan_id = 0, };
1952 
1953 	if (!priv->config.l3_vxlan_en)
1954 		return rte_flow_error_set(error, ENOTSUP,
1955 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1956 					  "L3 VXLAN is not enabled by device"
1957 					  " parameter and/or not configured in"
1958 					  " firmware");
1959 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1960 		return rte_flow_error_set(error, ENOTSUP,
1961 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1962 					  "multiple tunnel layers not"
1963 					  " supported");
1964 	/*
1965 	 * Verify only UDPv4 is present as defined in
1966 	 * https://tools.ietf.org/html/rfc7348
1967 	 */
1968 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1969 		return rte_flow_error_set(error, EINVAL,
1970 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1971 					  "no outer UDP layer found");
1972 	if (!mask)
1973 		mask = &rte_flow_item_vxlan_gpe_mask;
1974 	ret = mlx5_flow_item_acceptable
1975 		(item, (const uint8_t *)mask,
1976 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1977 		 sizeof(struct rte_flow_item_vxlan_gpe),
1978 		 error);
1979 	if (ret < 0)
1980 		return ret;
1981 	if (spec) {
1982 		if (spec->protocol)
1983 			return rte_flow_error_set(error, ENOTSUP,
1984 						  RTE_FLOW_ERROR_TYPE_ITEM,
1985 						  item,
1986 						  "VxLAN-GPE protocol"
1987 						  " not supported");
1988 		memcpy(&id.vni[1], spec->vni, 3);
1989 		memcpy(&id.vni[1], mask->vni, 3);
1990 	}
1991 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1992 		return rte_flow_error_set(error, ENOTSUP,
1993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 					  "VXLAN-GPE tunnel must be fully"
1995 					  " defined");
1996 	return 0;
1997 }
1998 /**
1999  * Validate GRE Key item.
2000  *
2001  * @param[in] item
2002  *   Item specification.
2003  * @param[in] item_flags
2004  *   Bit flags to mark detected items.
2005  * @param[in] gre_item
2006  *   Pointer to gre_item
2007  * @param[out] error
2008  *   Pointer to error structure.
2009  *
2010  * @return
2011  *   0 on success, a negative errno value otherwise and rte_errno is set.
2012  */
2013 int
2014 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2015 				uint64_t item_flags,
2016 				const struct rte_flow_item *gre_item,
2017 				struct rte_flow_error *error)
2018 {
2019 	const rte_be32_t *mask = item->mask;
2020 	int ret = 0;
2021 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2022 	const struct rte_flow_item_gre *gre_spec;
2023 	const struct rte_flow_item_gre *gre_mask;
2024 
2025 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2026 		return rte_flow_error_set(error, ENOTSUP,
2027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 					  "Multiple GRE key not support");
2029 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2030 		return rte_flow_error_set(error, ENOTSUP,
2031 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2032 					  "No preceding GRE header");
2033 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2034 		return rte_flow_error_set(error, ENOTSUP,
2035 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2036 					  "GRE key following a wrong item");
2037 	gre_mask = gre_item->mask;
2038 	if (!gre_mask)
2039 		gre_mask = &rte_flow_item_gre_mask;
2040 	gre_spec = gre_item->spec;
2041 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2042 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2043 		return rte_flow_error_set(error, EINVAL,
2044 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2045 					  "Key bit must be on");
2046 
2047 	if (!mask)
2048 		mask = &gre_key_default_mask;
2049 	ret = mlx5_flow_item_acceptable
2050 		(item, (const uint8_t *)mask,
2051 		 (const uint8_t *)&gre_key_default_mask,
2052 		 sizeof(rte_be32_t), error);
2053 	return ret;
2054 }
2055 
2056 /**
2057  * Validate GRE item.
2058  *
2059  * @param[in] item
2060  *   Item specification.
2061  * @param[in] item_flags
2062  *   Bit flags to mark detected items.
2063  * @param[in] target_protocol
2064  *   The next protocol in the previous item.
2065  * @param[out] error
2066  *   Pointer to error structure.
2067  *
2068  * @return
2069  *   0 on success, a negative errno value otherwise and rte_errno is set.
2070  */
2071 int
2072 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2073 			    uint64_t item_flags,
2074 			    uint8_t target_protocol,
2075 			    struct rte_flow_error *error)
2076 {
2077 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2078 	const struct rte_flow_item_gre *mask = item->mask;
2079 	int ret;
2080 	const struct rte_flow_item_gre nic_mask = {
2081 		.c_rsvd0_ver = RTE_BE16(0xB000),
2082 		.protocol = RTE_BE16(UINT16_MAX),
2083 	};
2084 
2085 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2086 		return rte_flow_error_set(error, EINVAL,
2087 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2088 					  "protocol filtering not compatible"
2089 					  " with this GRE layer");
2090 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2091 		return rte_flow_error_set(error, ENOTSUP,
2092 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2093 					  "multiple tunnel layers not"
2094 					  " supported");
2095 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2096 		return rte_flow_error_set(error, ENOTSUP,
2097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 					  "L3 Layer is missing");
2099 	if (!mask)
2100 		mask = &rte_flow_item_gre_mask;
2101 	ret = mlx5_flow_item_acceptable
2102 		(item, (const uint8_t *)mask,
2103 		 (const uint8_t *)&nic_mask,
2104 		 sizeof(struct rte_flow_item_gre), error);
2105 	if (ret < 0)
2106 		return ret;
2107 #ifndef HAVE_MLX5DV_DR
2108 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2109 	if (spec && (spec->protocol & mask->protocol))
2110 		return rte_flow_error_set(error, ENOTSUP,
2111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2112 					  "without MPLS support the"
2113 					  " specification cannot be used for"
2114 					  " filtering");
2115 #endif
2116 #endif
2117 	return 0;
2118 }
2119 
2120 /**
2121  * Validate Geneve item.
2122  *
2123  * @param[in] item
2124  *   Item specification.
2125  * @param[in] itemFlags
2126  *   Bit-fields that holds the items detected until now.
2127  * @param[in] enPriv
2128  *   Pointer to the private data structure.
2129  * @param[out] error
2130  *   Pointer to error structure.
2131  *
2132  * @return
2133  *   0 on success, a negative errno value otherwise and rte_errno is set.
2134  */
2135 
2136 int
2137 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2138 			       uint64_t item_flags,
2139 			       struct rte_eth_dev *dev,
2140 			       struct rte_flow_error *error)
2141 {
2142 	struct mlx5_priv *priv = dev->data->dev_private;
2143 	const struct rte_flow_item_geneve *spec = item->spec;
2144 	const struct rte_flow_item_geneve *mask = item->mask;
2145 	int ret;
2146 	uint16_t gbhdr;
2147 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2148 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2149 	const struct rte_flow_item_geneve nic_mask = {
2150 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2151 		.vni = "\xff\xff\xff",
2152 		.protocol = RTE_BE16(UINT16_MAX),
2153 	};
2154 
2155 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2156 		return rte_flow_error_set(error, ENOTSUP,
2157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 					  "L3 Geneve is not enabled by device"
2159 					  " parameter and/or not configured in"
2160 					  " firmware");
2161 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2162 		return rte_flow_error_set(error, ENOTSUP,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "multiple tunnel layers not"
2165 					  " supported");
2166 	/*
2167 	 * Verify only UDPv4 is present as defined in
2168 	 * https://tools.ietf.org/html/rfc7348
2169 	 */
2170 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "no outer UDP layer found");
2174 	if (!mask)
2175 		mask = &rte_flow_item_geneve_mask;
2176 	ret = mlx5_flow_item_acceptable
2177 				  (item, (const uint8_t *)mask,
2178 				   (const uint8_t *)&nic_mask,
2179 				   sizeof(struct rte_flow_item_geneve), error);
2180 	if (ret)
2181 		return ret;
2182 	if (spec) {
2183 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2184 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2185 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2186 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2187 			return rte_flow_error_set(error, ENOTSUP,
2188 						  RTE_FLOW_ERROR_TYPE_ITEM,
2189 						  item,
2190 						  "Geneve protocol unsupported"
2191 						  " fields are being used");
2192 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2193 			return rte_flow_error_set
2194 					(error, ENOTSUP,
2195 					 RTE_FLOW_ERROR_TYPE_ITEM,
2196 					 item,
2197 					 "Unsupported Geneve options length");
2198 	}
2199 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2200 		return rte_flow_error_set
2201 				    (error, ENOTSUP,
2202 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2203 				     "Geneve tunnel must be fully defined");
2204 	return 0;
2205 }
2206 
2207 /**
2208  * Validate MPLS item.
2209  *
2210  * @param[in] dev
2211  *   Pointer to the rte_eth_dev structure.
2212  * @param[in] item
2213  *   Item specification.
2214  * @param[in] item_flags
2215  *   Bit-fields that holds the items detected until now.
2216  * @param[in] prev_layer
2217  *   The protocol layer indicated in previous item.
2218  * @param[out] error
2219  *   Pointer to error structure.
2220  *
2221  * @return
2222  *   0 on success, a negative errno value otherwise and rte_errno is set.
2223  */
2224 int
2225 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2226 			     const struct rte_flow_item *item __rte_unused,
2227 			     uint64_t item_flags __rte_unused,
2228 			     uint64_t prev_layer __rte_unused,
2229 			     struct rte_flow_error *error)
2230 {
2231 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2232 	const struct rte_flow_item_mpls *mask = item->mask;
2233 	struct mlx5_priv *priv = dev->data->dev_private;
2234 	int ret;
2235 
2236 	if (!priv->config.mpls_en)
2237 		return rte_flow_error_set(error, ENOTSUP,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "MPLS not supported or"
2240 					  " disabled in firmware"
2241 					  " configuration.");
2242 	/* MPLS over IP, UDP, GRE is allowed */
2243 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2244 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2245 			    MLX5_FLOW_LAYER_GRE)))
2246 		return rte_flow_error_set(error, EINVAL,
2247 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2248 					  "protocol filtering not compatible"
2249 					  " with MPLS layer");
2250 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2251 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2252 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2253 		return rte_flow_error_set(error, ENOTSUP,
2254 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2255 					  "multiple tunnel layers not"
2256 					  " supported");
2257 	if (!mask)
2258 		mask = &rte_flow_item_mpls_mask;
2259 	ret = mlx5_flow_item_acceptable
2260 		(item, (const uint8_t *)mask,
2261 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2262 		 sizeof(struct rte_flow_item_mpls), error);
2263 	if (ret < 0)
2264 		return ret;
2265 	return 0;
2266 #endif
2267 	return rte_flow_error_set(error, ENOTSUP,
2268 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2269 				  "MPLS is not supported by Verbs, please"
2270 				  " update.");
2271 }
2272 
2273 /**
2274  * Validate NVGRE item.
2275  *
2276  * @param[in] item
2277  *   Item specification.
2278  * @param[in] item_flags
2279  *   Bit flags to mark detected items.
2280  * @param[in] target_protocol
2281  *   The next protocol in the previous item.
2282  * @param[out] error
2283  *   Pointer to error structure.
2284  *
2285  * @return
2286  *   0 on success, a negative errno value otherwise and rte_errno is set.
2287  */
2288 int
2289 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2290 			      uint64_t item_flags,
2291 			      uint8_t target_protocol,
2292 			      struct rte_flow_error *error)
2293 {
2294 	const struct rte_flow_item_nvgre *mask = item->mask;
2295 	int ret;
2296 
2297 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2298 		return rte_flow_error_set(error, EINVAL,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "protocol filtering not compatible"
2301 					  " with this GRE layer");
2302 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2303 		return rte_flow_error_set(error, ENOTSUP,
2304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2305 					  "multiple tunnel layers not"
2306 					  " supported");
2307 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2308 		return rte_flow_error_set(error, ENOTSUP,
2309 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2310 					  "L3 Layer is missing");
2311 	if (!mask)
2312 		mask = &rte_flow_item_nvgre_mask;
2313 	ret = mlx5_flow_item_acceptable
2314 		(item, (const uint8_t *)mask,
2315 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2316 		 sizeof(struct rte_flow_item_nvgre), error);
2317 	if (ret < 0)
2318 		return ret;
2319 	return 0;
2320 }
2321 
2322 /* Allocate unique ID for the split Q/RSS subflows. */
2323 static uint32_t
2324 flow_qrss_get_id(struct rte_eth_dev *dev)
2325 {
2326 	struct mlx5_priv *priv = dev->data->dev_private;
2327 	uint32_t qrss_id, ret;
2328 
2329 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2330 	if (ret)
2331 		return 0;
2332 	MLX5_ASSERT(qrss_id);
2333 	return qrss_id;
2334 }
2335 
2336 /* Free unique ID for the split Q/RSS subflows. */
2337 static void
2338 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2339 {
2340 	struct mlx5_priv *priv = dev->data->dev_private;
2341 
2342 	if (qrss_id)
2343 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2344 }
2345 
2346 /**
2347  * Release resource related QUEUE/RSS action split.
2348  *
2349  * @param dev
2350  *   Pointer to Ethernet device.
2351  * @param flow
2352  *   Flow to release id's from.
2353  */
2354 static void
2355 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2356 			     struct rte_flow *flow)
2357 {
2358 	struct mlx5_priv *priv = dev->data->dev_private;
2359 	uint32_t handle_idx;
2360 	struct mlx5_flow_handle *dev_handle;
2361 
2362 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2363 		       handle_idx, dev_handle, next)
2364 		if (dev_handle->split_flow_id)
2365 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2366 }
2367 
2368 static int
2369 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2370 		   const struct rte_flow_attr *attr __rte_unused,
2371 		   const struct rte_flow_item items[] __rte_unused,
2372 		   const struct rte_flow_action actions[] __rte_unused,
2373 		   bool external __rte_unused,
2374 		   int hairpin __rte_unused,
2375 		   struct rte_flow_error *error)
2376 {
2377 	return rte_flow_error_set(error, ENOTSUP,
2378 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2379 }
2380 
2381 static struct mlx5_flow *
2382 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2383 		  const struct rte_flow_attr *attr __rte_unused,
2384 		  const struct rte_flow_item items[] __rte_unused,
2385 		  const struct rte_flow_action actions[] __rte_unused,
2386 		  struct rte_flow_error *error)
2387 {
2388 	rte_flow_error_set(error, ENOTSUP,
2389 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2390 	return NULL;
2391 }
2392 
2393 static int
2394 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2395 		    struct mlx5_flow *dev_flow __rte_unused,
2396 		    const struct rte_flow_attr *attr __rte_unused,
2397 		    const struct rte_flow_item items[] __rte_unused,
2398 		    const struct rte_flow_action actions[] __rte_unused,
2399 		    struct rte_flow_error *error)
2400 {
2401 	return rte_flow_error_set(error, ENOTSUP,
2402 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2403 }
2404 
2405 static int
2406 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2407 		struct rte_flow *flow __rte_unused,
2408 		struct rte_flow_error *error)
2409 {
2410 	return rte_flow_error_set(error, ENOTSUP,
2411 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2412 }
2413 
2414 static void
2415 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2416 		 struct rte_flow *flow __rte_unused)
2417 {
2418 }
2419 
2420 static void
2421 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2422 		  struct rte_flow *flow __rte_unused)
2423 {
2424 }
2425 
2426 static int
2427 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2428 		struct rte_flow *flow __rte_unused,
2429 		const struct rte_flow_action *actions __rte_unused,
2430 		void *data __rte_unused,
2431 		struct rte_flow_error *error)
2432 {
2433 	return rte_flow_error_set(error, ENOTSUP,
2434 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2435 }
2436 
2437 /* Void driver to protect from null pointer reference. */
2438 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2439 	.validate = flow_null_validate,
2440 	.prepare = flow_null_prepare,
2441 	.translate = flow_null_translate,
2442 	.apply = flow_null_apply,
2443 	.remove = flow_null_remove,
2444 	.destroy = flow_null_destroy,
2445 	.query = flow_null_query,
2446 };
2447 
2448 /**
2449  * Select flow driver type according to flow attributes and device
2450  * configuration.
2451  *
2452  * @param[in] dev
2453  *   Pointer to the dev structure.
2454  * @param[in] attr
2455  *   Pointer to the flow attributes.
2456  *
2457  * @return
2458  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2459  */
2460 static enum mlx5_flow_drv_type
2461 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2462 {
2463 	struct mlx5_priv *priv = dev->data->dev_private;
2464 	enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2465 
2466 	if (attr->transfer && priv->config.dv_esw_en)
2467 		type = MLX5_FLOW_TYPE_DV;
2468 	if (!attr->transfer)
2469 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2470 						 MLX5_FLOW_TYPE_VERBS;
2471 	return type;
2472 }
2473 
2474 #define flow_get_drv_ops(type) flow_drv_ops[type]
2475 
2476 /**
2477  * Flow driver validation API. This abstracts calling driver specific functions.
2478  * The type of flow driver is determined according to flow attributes.
2479  *
2480  * @param[in] dev
2481  *   Pointer to the dev structure.
2482  * @param[in] attr
2483  *   Pointer to the flow attributes.
2484  * @param[in] items
2485  *   Pointer to the list of items.
2486  * @param[in] actions
2487  *   Pointer to the list of actions.
2488  * @param[in] external
2489  *   This flow rule is created by request external to PMD.
2490  * @param[in] hairpin
2491  *   Number of hairpin TX actions, 0 means classic flow.
2492  * @param[out] error
2493  *   Pointer to the error structure.
2494  *
2495  * @return
2496  *   0 on success, a negative errno value otherwise and rte_errno is set.
2497  */
2498 static inline int
2499 flow_drv_validate(struct rte_eth_dev *dev,
2500 		  const struct rte_flow_attr *attr,
2501 		  const struct rte_flow_item items[],
2502 		  const struct rte_flow_action actions[],
2503 		  bool external, int hairpin, struct rte_flow_error *error)
2504 {
2505 	const struct mlx5_flow_driver_ops *fops;
2506 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2507 
2508 	fops = flow_get_drv_ops(type);
2509 	return fops->validate(dev, attr, items, actions, external,
2510 			      hairpin, error);
2511 }
2512 
2513 /**
2514  * Flow driver preparation API. This abstracts calling driver specific
2515  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2516  * calculates the size of memory required for device flow, allocates the memory,
2517  * initializes the device flow and returns the pointer.
2518  *
2519  * @note
2520  *   This function initializes device flow structure such as dv or verbs in
2521  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2522  *   rest. For example, adding returning device flow to flow->dev_flow list and
2523  *   setting backward reference to the flow should be done out of this function.
2524  *   layers field is not filled either.
2525  *
2526  * @param[in] dev
2527  *   Pointer to the dev structure.
2528  * @param[in] attr
2529  *   Pointer to the flow attributes.
2530  * @param[in] items
2531  *   Pointer to the list of items.
2532  * @param[in] actions
2533  *   Pointer to the list of actions.
2534  * @param[out] error
2535  *   Pointer to the error structure.
2536  *
2537  * @return
2538  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2539  */
2540 static inline struct mlx5_flow *
2541 flow_drv_prepare(struct rte_eth_dev *dev,
2542 		 const struct rte_flow *flow,
2543 		 const struct rte_flow_attr *attr,
2544 		 const struct rte_flow_item items[],
2545 		 const struct rte_flow_action actions[],
2546 		 struct rte_flow_error *error)
2547 {
2548 	const struct mlx5_flow_driver_ops *fops;
2549 	enum mlx5_flow_drv_type type = flow->drv_type;
2550 
2551 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2552 	fops = flow_get_drv_ops(type);
2553 	return fops->prepare(dev, attr, items, actions, error);
2554 }
2555 
2556 /**
2557  * Flow driver translation API. This abstracts calling driver specific
2558  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2559  * translates a generic flow into a driver flow. flow_drv_prepare() must
2560  * precede.
2561  *
2562  * @note
2563  *   dev_flow->layers could be filled as a result of parsing during translation
2564  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2565  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2566  *   flow->actions could be overwritten even though all the expanded dev_flows
2567  *   have the same actions.
2568  *
2569  * @param[in] dev
2570  *   Pointer to the rte dev structure.
2571  * @param[in, out] dev_flow
2572  *   Pointer to the mlx5 flow.
2573  * @param[in] attr
2574  *   Pointer to the flow attributes.
2575  * @param[in] items
2576  *   Pointer to the list of items.
2577  * @param[in] actions
2578  *   Pointer to the list of actions.
2579  * @param[out] error
2580  *   Pointer to the error structure.
2581  *
2582  * @return
2583  *   0 on success, a negative errno value otherwise and rte_errno is set.
2584  */
2585 static inline int
2586 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2587 		   const struct rte_flow_attr *attr,
2588 		   const struct rte_flow_item items[],
2589 		   const struct rte_flow_action actions[],
2590 		   struct rte_flow_error *error)
2591 {
2592 	const struct mlx5_flow_driver_ops *fops;
2593 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2594 
2595 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2596 	fops = flow_get_drv_ops(type);
2597 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2598 }
2599 
2600 /**
2601  * Flow driver apply API. This abstracts calling driver specific functions.
2602  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2603  * translated driver flows on to device. flow_drv_translate() must precede.
2604  *
2605  * @param[in] dev
2606  *   Pointer to Ethernet device structure.
2607  * @param[in, out] flow
2608  *   Pointer to flow structure.
2609  * @param[out] error
2610  *   Pointer to error structure.
2611  *
2612  * @return
2613  *   0 on success, a negative errno value otherwise and rte_errno is set.
2614  */
2615 static inline int
2616 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2617 	       struct rte_flow_error *error)
2618 {
2619 	const struct mlx5_flow_driver_ops *fops;
2620 	enum mlx5_flow_drv_type type = flow->drv_type;
2621 
2622 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2623 	fops = flow_get_drv_ops(type);
2624 	return fops->apply(dev, flow, error);
2625 }
2626 
2627 /**
2628  * Flow driver remove API. This abstracts calling driver specific functions.
2629  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2630  * on device. All the resources of the flow should be freed by calling
2631  * flow_drv_destroy().
2632  *
2633  * @param[in] dev
2634  *   Pointer to Ethernet device.
2635  * @param[in, out] flow
2636  *   Pointer to flow structure.
2637  */
2638 static inline void
2639 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2640 {
2641 	const struct mlx5_flow_driver_ops *fops;
2642 	enum mlx5_flow_drv_type type = flow->drv_type;
2643 
2644 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2645 	fops = flow_get_drv_ops(type);
2646 	fops->remove(dev, flow);
2647 }
2648 
2649 /**
2650  * Flow driver destroy API. This abstracts calling driver specific functions.
2651  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2652  * on device and releases resources of the flow.
2653  *
2654  * @param[in] dev
2655  *   Pointer to Ethernet device.
2656  * @param[in, out] flow
2657  *   Pointer to flow structure.
2658  */
2659 static inline void
2660 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2661 {
2662 	const struct mlx5_flow_driver_ops *fops;
2663 	enum mlx5_flow_drv_type type = flow->drv_type;
2664 
2665 	flow_mreg_split_qrss_release(dev, flow);
2666 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2667 	fops = flow_get_drv_ops(type);
2668 	fops->destroy(dev, flow);
2669 }
2670 
2671 /**
2672  * Get RSS action from the action list.
2673  *
2674  * @param[in] actions
2675  *   Pointer to the list of actions.
2676  *
2677  * @return
2678  *   Pointer to the RSS action if exist, else return NULL.
2679  */
2680 static const struct rte_flow_action_rss*
2681 flow_get_rss_action(const struct rte_flow_action actions[])
2682 {
2683 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2684 		switch (actions->type) {
2685 		case RTE_FLOW_ACTION_TYPE_RSS:
2686 			return (const struct rte_flow_action_rss *)
2687 			       actions->conf;
2688 		default:
2689 			break;
2690 		}
2691 	}
2692 	return NULL;
2693 }
2694 
2695 static unsigned int
2696 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2697 {
2698 	const struct rte_flow_item *item;
2699 	unsigned int has_vlan = 0;
2700 
2701 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2702 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2703 			has_vlan = 1;
2704 			break;
2705 		}
2706 	}
2707 	if (has_vlan)
2708 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2709 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2710 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2711 			       MLX5_EXPANSION_ROOT_OUTER;
2712 }
2713 
2714 /**
2715  *  Get layer flags from the prefix flow.
2716  *
2717  *  Some flows may be split to several subflows, the prefix subflow gets the
2718  *  match items and the suffix sub flow gets the actions.
2719  *  Some actions need the user defined match item flags to get the detail for
2720  *  the action.
2721  *  This function helps the suffix flow to get the item layer flags from prefix
2722  *  subflow.
2723  *
2724  * @param[in] dev_flow
2725  *   Pointer the created preifx subflow.
2726  *
2727  * @return
2728  *   The layers get from prefix subflow.
2729  */
2730 static inline uint64_t
2731 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2732 {
2733 	uint64_t layers = 0;
2734 
2735 	/*
2736 	 * Layers bits could be localization, but usually the compiler will
2737 	 * help to do the optimization work for source code.
2738 	 * If no decap actions, use the layers directly.
2739 	 */
2740 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2741 		return dev_flow->handle->layers;
2742 	/* Convert L3 layers with decap action. */
2743 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2744 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2745 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2746 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2747 	/* Convert L4 layers with decap action.  */
2748 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2749 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2750 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2751 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2752 	return layers;
2753 }
2754 
2755 /**
2756  * Get metadata split action information.
2757  *
2758  * @param[in] actions
2759  *   Pointer to the list of actions.
2760  * @param[out] qrss
2761  *   Pointer to the return pointer.
2762  * @param[out] qrss_type
2763  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2764  *   if no QUEUE/RSS is found.
2765  * @param[out] encap_idx
2766  *   Pointer to the index of the encap action if exists, otherwise the last
2767  *   action index.
2768  *
2769  * @return
2770  *   Total number of actions.
2771  */
2772 static int
2773 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2774 				       const struct rte_flow_action **qrss,
2775 				       int *encap_idx)
2776 {
2777 	const struct rte_flow_action_raw_encap *raw_encap;
2778 	int actions_n = 0;
2779 	int raw_decap_idx = -1;
2780 
2781 	*encap_idx = -1;
2782 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2783 		switch (actions->type) {
2784 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2785 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2786 			*encap_idx = actions_n;
2787 			break;
2788 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2789 			raw_decap_idx = actions_n;
2790 			break;
2791 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2792 			raw_encap = actions->conf;
2793 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2794 				*encap_idx = raw_decap_idx != -1 ?
2795 						      raw_decap_idx : actions_n;
2796 			break;
2797 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2798 		case RTE_FLOW_ACTION_TYPE_RSS:
2799 			*qrss = actions;
2800 			break;
2801 		default:
2802 			break;
2803 		}
2804 		actions_n++;
2805 	}
2806 	if (*encap_idx == -1)
2807 		*encap_idx = actions_n;
2808 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2809 	return actions_n + 1;
2810 }
2811 
2812 /**
2813  * Check meter action from the action list.
2814  *
2815  * @param[in] actions
2816  *   Pointer to the list of actions.
2817  * @param[out] mtr
2818  *   Pointer to the meter exist flag.
2819  *
2820  * @return
2821  *   Total number of actions.
2822  */
2823 static int
2824 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2825 {
2826 	int actions_n = 0;
2827 
2828 	MLX5_ASSERT(mtr);
2829 	*mtr = 0;
2830 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2831 		switch (actions->type) {
2832 		case RTE_FLOW_ACTION_TYPE_METER:
2833 			*mtr = 1;
2834 			break;
2835 		default:
2836 			break;
2837 		}
2838 		actions_n++;
2839 	}
2840 	/* Count RTE_FLOW_ACTION_TYPE_END. */
2841 	return actions_n + 1;
2842 }
2843 
2844 /**
2845  * Check if the flow should be splited due to hairpin.
2846  * The reason for the split is that in current HW we can't
2847  * support encap on Rx, so if a flow have encap we move it
2848  * to Tx.
2849  *
2850  * @param dev
2851  *   Pointer to Ethernet device.
2852  * @param[in] attr
2853  *   Flow rule attributes.
2854  * @param[in] actions
2855  *   Associated actions (list terminated by the END action).
2856  *
2857  * @return
2858  *   > 0 the number of actions and the flow should be split,
2859  *   0 when no split required.
2860  */
2861 static int
2862 flow_check_hairpin_split(struct rte_eth_dev *dev,
2863 			 const struct rte_flow_attr *attr,
2864 			 const struct rte_flow_action actions[])
2865 {
2866 	int queue_action = 0;
2867 	int action_n = 0;
2868 	int encap = 0;
2869 	const struct rte_flow_action_queue *queue;
2870 	const struct rte_flow_action_rss *rss;
2871 	const struct rte_flow_action_raw_encap *raw_encap;
2872 
2873 	if (!attr->ingress)
2874 		return 0;
2875 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2876 		switch (actions->type) {
2877 		case RTE_FLOW_ACTION_TYPE_QUEUE:
2878 			queue = actions->conf;
2879 			if (queue == NULL)
2880 				return 0;
2881 			if (mlx5_rxq_get_type(dev, queue->index) !=
2882 			    MLX5_RXQ_TYPE_HAIRPIN)
2883 				return 0;
2884 			queue_action = 1;
2885 			action_n++;
2886 			break;
2887 		case RTE_FLOW_ACTION_TYPE_RSS:
2888 			rss = actions->conf;
2889 			if (rss == NULL || rss->queue_num == 0)
2890 				return 0;
2891 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2892 			    MLX5_RXQ_TYPE_HAIRPIN)
2893 				return 0;
2894 			queue_action = 1;
2895 			action_n++;
2896 			break;
2897 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2898 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2899 			encap = 1;
2900 			action_n++;
2901 			break;
2902 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2903 			raw_encap = actions->conf;
2904 			if (raw_encap->size >
2905 			    (sizeof(struct rte_flow_item_eth) +
2906 			     sizeof(struct rte_flow_item_ipv4)))
2907 				encap = 1;
2908 			action_n++;
2909 			break;
2910 		default:
2911 			action_n++;
2912 			break;
2913 		}
2914 	}
2915 	if (encap == 1 && queue_action)
2916 		return action_n;
2917 	return 0;
2918 }
2919 
2920 /* Declare flow create/destroy prototype in advance. */
2921 static uint32_t
2922 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
2923 		 const struct rte_flow_attr *attr,
2924 		 const struct rte_flow_item items[],
2925 		 const struct rte_flow_action actions[],
2926 		 bool external, struct rte_flow_error *error);
2927 
2928 static void
2929 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
2930 		  uint32_t flow_idx);
2931 
2932 /**
2933  * Add a flow of copying flow metadata registers in RX_CP_TBL.
2934  *
2935  * As mark_id is unique, if there's already a registered flow for the mark_id,
2936  * return by increasing the reference counter of the resource. Otherwise, create
2937  * the resource (mcp_res) and flow.
2938  *
2939  * Flow looks like,
2940  *   - If ingress port is ANY and reg_c[1] is mark_id,
2941  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2942  *
2943  * For default flow (zero mark_id), flow is like,
2944  *   - If ingress port is ANY,
2945  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
2946  *
2947  * @param dev
2948  *   Pointer to Ethernet device.
2949  * @param mark_id
2950  *   ID of MARK action, zero means default flow for META.
2951  * @param[out] error
2952  *   Perform verbose error reporting if not NULL.
2953  *
2954  * @return
2955  *   Associated resource on success, NULL otherwise and rte_errno is set.
2956  */
2957 static struct mlx5_flow_mreg_copy_resource *
2958 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2959 			  struct rte_flow_error *error)
2960 {
2961 	struct mlx5_priv *priv = dev->data->dev_private;
2962 	struct rte_flow_attr attr = {
2963 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2964 		.ingress = 1,
2965 	};
2966 	struct mlx5_rte_flow_item_tag tag_spec = {
2967 		.data = mark_id,
2968 	};
2969 	struct rte_flow_item items[] = {
2970 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2971 	};
2972 	struct rte_flow_action_mark ftag = {
2973 		.id = mark_id,
2974 	};
2975 	struct mlx5_flow_action_copy_mreg cp_mreg = {
2976 		.dst = REG_B,
2977 		.src = 0,
2978 	};
2979 	struct rte_flow_action_jump jump = {
2980 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2981 	};
2982 	struct rte_flow_action actions[] = {
2983 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2984 	};
2985 	struct mlx5_flow_mreg_copy_resource *mcp_res;
2986 	uint32_t idx = 0;
2987 	int ret;
2988 
2989 	/* Fill the register fileds in the flow. */
2990 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2991 	if (ret < 0)
2992 		return NULL;
2993 	tag_spec.id = ret;
2994 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2995 	if (ret < 0)
2996 		return NULL;
2997 	cp_mreg.src = ret;
2998 	/* Check if already registered. */
2999 	MLX5_ASSERT(priv->mreg_cp_tbl);
3000 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3001 	if (mcp_res) {
3002 		/* For non-default rule. */
3003 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3004 			mcp_res->refcnt++;
3005 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3006 			    mcp_res->refcnt == 1);
3007 		return mcp_res;
3008 	}
3009 	/* Provide the full width of FLAG specific value. */
3010 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3011 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3012 	/* Build a new flow. */
3013 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3014 		items[0] = (struct rte_flow_item){
3015 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3016 			.spec = &tag_spec,
3017 		};
3018 		items[1] = (struct rte_flow_item){
3019 			.type = RTE_FLOW_ITEM_TYPE_END,
3020 		};
3021 		actions[0] = (struct rte_flow_action){
3022 			.type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3023 			.conf = &ftag,
3024 		};
3025 		actions[1] = (struct rte_flow_action){
3026 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3027 			.conf = &cp_mreg,
3028 		};
3029 		actions[2] = (struct rte_flow_action){
3030 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3031 			.conf = &jump,
3032 		};
3033 		actions[3] = (struct rte_flow_action){
3034 			.type = RTE_FLOW_ACTION_TYPE_END,
3035 		};
3036 	} else {
3037 		/* Default rule, wildcard match. */
3038 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3039 		items[0] = (struct rte_flow_item){
3040 			.type = RTE_FLOW_ITEM_TYPE_END,
3041 		};
3042 		actions[0] = (struct rte_flow_action){
3043 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3044 			.conf = &cp_mreg,
3045 		};
3046 		actions[1] = (struct rte_flow_action){
3047 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3048 			.conf = &jump,
3049 		};
3050 		actions[2] = (struct rte_flow_action){
3051 			.type = RTE_FLOW_ACTION_TYPE_END,
3052 		};
3053 	}
3054 	/* Build a new entry. */
3055 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3056 	if (!mcp_res) {
3057 		rte_errno = ENOMEM;
3058 		return NULL;
3059 	}
3060 	mcp_res->idx = idx;
3061 	/*
3062 	 * The copy Flows are not included in any list. There
3063 	 * ones are referenced from other Flows and can not
3064 	 * be applied, removed, deleted in ardbitrary order
3065 	 * by list traversing.
3066 	 */
3067 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3068 					 actions, false, error);
3069 	if (!mcp_res->rix_flow)
3070 		goto error;
3071 	mcp_res->refcnt++;
3072 	mcp_res->hlist_ent.key = mark_id;
3073 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3074 				&mcp_res->hlist_ent);
3075 	MLX5_ASSERT(!ret);
3076 	if (ret)
3077 		goto error;
3078 	return mcp_res;
3079 error:
3080 	if (mcp_res->rix_flow)
3081 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3082 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3083 	return NULL;
3084 }
3085 
3086 /**
3087  * Release flow in RX_CP_TBL.
3088  *
3089  * @param dev
3090  *   Pointer to Ethernet device.
3091  * @flow
3092  *   Parent flow for wich copying is provided.
3093  */
3094 static void
3095 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3096 			  struct rte_flow *flow)
3097 {
3098 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3099 	struct mlx5_priv *priv = dev->data->dev_private;
3100 
3101 	if (!flow->rix_mreg_copy)
3102 		return;
3103 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3104 				 flow->rix_mreg_copy);
3105 	if (!mcp_res || !priv->mreg_cp_tbl)
3106 		return;
3107 	if (flow->copy_applied) {
3108 		MLX5_ASSERT(mcp_res->appcnt);
3109 		flow->copy_applied = 0;
3110 		--mcp_res->appcnt;
3111 		if (!mcp_res->appcnt) {
3112 			struct rte_flow *mcp_flow = mlx5_ipool_get
3113 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3114 					mcp_res->rix_flow);
3115 
3116 			if (mcp_flow)
3117 				flow_drv_remove(dev, mcp_flow);
3118 		}
3119 	}
3120 	/*
3121 	 * We do not check availability of metadata registers here,
3122 	 * because copy resources are not allocated in this case.
3123 	 */
3124 	if (--mcp_res->refcnt)
3125 		return;
3126 	MLX5_ASSERT(mcp_res->rix_flow);
3127 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3128 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3129 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3130 	flow->rix_mreg_copy = 0;
3131 }
3132 
3133 /**
3134  * Start flow in RX_CP_TBL.
3135  *
3136  * @param dev
3137  *   Pointer to Ethernet device.
3138  * @flow
3139  *   Parent flow for wich copying is provided.
3140  *
3141  * @return
3142  *   0 on success, a negative errno value otherwise and rte_errno is set.
3143  */
3144 static int
3145 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3146 			    struct rte_flow *flow)
3147 {
3148 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3149 	struct mlx5_priv *priv = dev->data->dev_private;
3150 	int ret;
3151 
3152 	if (!flow->rix_mreg_copy || flow->copy_applied)
3153 		return 0;
3154 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3155 				 flow->rix_mreg_copy);
3156 	if (!mcp_res)
3157 		return 0;
3158 	if (!mcp_res->appcnt) {
3159 		struct rte_flow *mcp_flow = mlx5_ipool_get
3160 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3161 				mcp_res->rix_flow);
3162 
3163 		if (mcp_flow) {
3164 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3165 			if (ret)
3166 				return ret;
3167 		}
3168 	}
3169 	++mcp_res->appcnt;
3170 	flow->copy_applied = 1;
3171 	return 0;
3172 }
3173 
3174 /**
3175  * Stop flow in RX_CP_TBL.
3176  *
3177  * @param dev
3178  *   Pointer to Ethernet device.
3179  * @flow
3180  *   Parent flow for wich copying is provided.
3181  */
3182 static void
3183 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3184 			   struct rte_flow *flow)
3185 {
3186 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3187 	struct mlx5_priv *priv = dev->data->dev_private;
3188 
3189 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3190 		return;
3191 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3192 				 flow->rix_mreg_copy);
3193 	if (!mcp_res)
3194 		return;
3195 	MLX5_ASSERT(mcp_res->appcnt);
3196 	--mcp_res->appcnt;
3197 	flow->copy_applied = 0;
3198 	if (!mcp_res->appcnt) {
3199 		struct rte_flow *mcp_flow = mlx5_ipool_get
3200 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3201 				mcp_res->rix_flow);
3202 
3203 		if (mcp_flow)
3204 			flow_drv_remove(dev, mcp_flow);
3205 	}
3206 }
3207 
3208 /**
3209  * Remove the default copy action from RX_CP_TBL.
3210  *
3211  * @param dev
3212  *   Pointer to Ethernet device.
3213  */
3214 static void
3215 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3216 {
3217 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3218 	struct mlx5_priv *priv = dev->data->dev_private;
3219 
3220 	/* Check if default flow is registered. */
3221 	if (!priv->mreg_cp_tbl)
3222 		return;
3223 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3224 					    MLX5_DEFAULT_COPY_ID);
3225 	if (!mcp_res)
3226 		return;
3227 	MLX5_ASSERT(mcp_res->rix_flow);
3228 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3229 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3230 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3231 }
3232 
3233 /**
3234  * Add the default copy action in in RX_CP_TBL.
3235  *
3236  * @param dev
3237  *   Pointer to Ethernet device.
3238  * @param[out] error
3239  *   Perform verbose error reporting if not NULL.
3240  *
3241  * @return
3242  *   0 for success, negative value otherwise and rte_errno is set.
3243  */
3244 static int
3245 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3246 				  struct rte_flow_error *error)
3247 {
3248 	struct mlx5_priv *priv = dev->data->dev_private;
3249 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3250 
3251 	/* Check whether extensive metadata feature is engaged. */
3252 	if (!priv->config.dv_flow_en ||
3253 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3254 	    !mlx5_flow_ext_mreg_supported(dev) ||
3255 	    !priv->sh->dv_regc0_mask)
3256 		return 0;
3257 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3258 	if (!mcp_res)
3259 		return -rte_errno;
3260 	return 0;
3261 }
3262 
3263 /**
3264  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3265  *
3266  * All the flow having Q/RSS action should be split by
3267  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3268  * performs the following,
3269  *   - CQE->flow_tag := reg_c[1] (MARK)
3270  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3271  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3272  * but there should be a flow per each MARK ID set by MARK action.
3273  *
3274  * For the aforementioned reason, if there's a MARK action in flow's action
3275  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3276  * the MARK ID to CQE's flow_tag like,
3277  *   - If reg_c[1] is mark_id,
3278  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3279  *
3280  * For SET_META action which stores value in reg_c[0], as the destination is
3281  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3282  * MARK ID means the default flow. The default flow looks like,
3283  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3284  *
3285  * @param dev
3286  *   Pointer to Ethernet device.
3287  * @param flow
3288  *   Pointer to flow structure.
3289  * @param[in] actions
3290  *   Pointer to the list of actions.
3291  * @param[out] error
3292  *   Perform verbose error reporting if not NULL.
3293  *
3294  * @return
3295  *   0 on success, negative value otherwise and rte_errno is set.
3296  */
3297 static int
3298 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3299 			    struct rte_flow *flow,
3300 			    const struct rte_flow_action *actions,
3301 			    struct rte_flow_error *error)
3302 {
3303 	struct mlx5_priv *priv = dev->data->dev_private;
3304 	struct mlx5_dev_config *config = &priv->config;
3305 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3306 	const struct rte_flow_action_mark *mark;
3307 
3308 	/* Check whether extensive metadata feature is engaged. */
3309 	if (!config->dv_flow_en ||
3310 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3311 	    !mlx5_flow_ext_mreg_supported(dev) ||
3312 	    !priv->sh->dv_regc0_mask)
3313 		return 0;
3314 	/* Find MARK action. */
3315 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3316 		switch (actions->type) {
3317 		case RTE_FLOW_ACTION_TYPE_FLAG:
3318 			mcp_res = flow_mreg_add_copy_action
3319 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3320 			if (!mcp_res)
3321 				return -rte_errno;
3322 			flow->rix_mreg_copy = mcp_res->idx;
3323 			if (dev->data->dev_started) {
3324 				mcp_res->appcnt++;
3325 				flow->copy_applied = 1;
3326 			}
3327 			return 0;
3328 		case RTE_FLOW_ACTION_TYPE_MARK:
3329 			mark = (const struct rte_flow_action_mark *)
3330 				actions->conf;
3331 			mcp_res =
3332 				flow_mreg_add_copy_action(dev, mark->id, error);
3333 			if (!mcp_res)
3334 				return -rte_errno;
3335 			flow->rix_mreg_copy = mcp_res->idx;
3336 			if (dev->data->dev_started) {
3337 				mcp_res->appcnt++;
3338 				flow->copy_applied = 1;
3339 			}
3340 			return 0;
3341 		default:
3342 			break;
3343 		}
3344 	}
3345 	return 0;
3346 }
3347 
3348 #define MLX5_MAX_SPLIT_ACTIONS 24
3349 #define MLX5_MAX_SPLIT_ITEMS 24
3350 
3351 /**
3352  * Split the hairpin flow.
3353  * Since HW can't support encap on Rx we move the encap to Tx.
3354  * If the count action is after the encap then we also
3355  * move the count action. in this case the count will also measure
3356  * the outer bytes.
3357  *
3358  * @param dev
3359  *   Pointer to Ethernet device.
3360  * @param[in] actions
3361  *   Associated actions (list terminated by the END action).
3362  * @param[out] actions_rx
3363  *   Rx flow actions.
3364  * @param[out] actions_tx
3365  *   Tx flow actions..
3366  * @param[out] pattern_tx
3367  *   The pattern items for the Tx flow.
3368  * @param[out] flow_id
3369  *   The flow ID connected to this flow.
3370  *
3371  * @return
3372  *   0 on success.
3373  */
3374 static int
3375 flow_hairpin_split(struct rte_eth_dev *dev,
3376 		   const struct rte_flow_action actions[],
3377 		   struct rte_flow_action actions_rx[],
3378 		   struct rte_flow_action actions_tx[],
3379 		   struct rte_flow_item pattern_tx[],
3380 		   uint32_t *flow_id)
3381 {
3382 	struct mlx5_priv *priv = dev->data->dev_private;
3383 	const struct rte_flow_action_raw_encap *raw_encap;
3384 	const struct rte_flow_action_raw_decap *raw_decap;
3385 	struct mlx5_rte_flow_action_set_tag *set_tag;
3386 	struct rte_flow_action *tag_action;
3387 	struct mlx5_rte_flow_item_tag *tag_item;
3388 	struct rte_flow_item *item;
3389 	char *addr;
3390 	int encap = 0;
3391 
3392 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3393 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3394 		switch (actions->type) {
3395 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3396 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3397 			rte_memcpy(actions_tx, actions,
3398 			       sizeof(struct rte_flow_action));
3399 			actions_tx++;
3400 			break;
3401 		case RTE_FLOW_ACTION_TYPE_COUNT:
3402 			if (encap) {
3403 				rte_memcpy(actions_tx, actions,
3404 					   sizeof(struct rte_flow_action));
3405 				actions_tx++;
3406 			} else {
3407 				rte_memcpy(actions_rx, actions,
3408 					   sizeof(struct rte_flow_action));
3409 				actions_rx++;
3410 			}
3411 			break;
3412 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3413 			raw_encap = actions->conf;
3414 			if (raw_encap->size >
3415 			    (sizeof(struct rte_flow_item_eth) +
3416 			     sizeof(struct rte_flow_item_ipv4))) {
3417 				memcpy(actions_tx, actions,
3418 				       sizeof(struct rte_flow_action));
3419 				actions_tx++;
3420 				encap = 1;
3421 			} else {
3422 				rte_memcpy(actions_rx, actions,
3423 					   sizeof(struct rte_flow_action));
3424 				actions_rx++;
3425 			}
3426 			break;
3427 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3428 			raw_decap = actions->conf;
3429 			if (raw_decap->size <
3430 			    (sizeof(struct rte_flow_item_eth) +
3431 			     sizeof(struct rte_flow_item_ipv4))) {
3432 				memcpy(actions_tx, actions,
3433 				       sizeof(struct rte_flow_action));
3434 				actions_tx++;
3435 			} else {
3436 				rte_memcpy(actions_rx, actions,
3437 					   sizeof(struct rte_flow_action));
3438 				actions_rx++;
3439 			}
3440 			break;
3441 		default:
3442 			rte_memcpy(actions_rx, actions,
3443 				   sizeof(struct rte_flow_action));
3444 			actions_rx++;
3445 			break;
3446 		}
3447 	}
3448 	/* Add set meta action and end action for the Rx flow. */
3449 	tag_action = actions_rx;
3450 	tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3451 	actions_rx++;
3452 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3453 	actions_rx++;
3454 	set_tag = (void *)actions_rx;
3455 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3456 	MLX5_ASSERT(set_tag->id > REG_NONE);
3457 	set_tag->data = *flow_id;
3458 	tag_action->conf = set_tag;
3459 	/* Create Tx item list. */
3460 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3461 	addr = (void *)&pattern_tx[2];
3462 	item = pattern_tx;
3463 	item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3464 	tag_item = (void *)addr;
3465 	tag_item->data = *flow_id;
3466 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3467 	MLX5_ASSERT(set_tag->id > REG_NONE);
3468 	item->spec = tag_item;
3469 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3470 	tag_item = (void *)addr;
3471 	tag_item->data = UINT32_MAX;
3472 	tag_item->id = UINT16_MAX;
3473 	item->mask = tag_item;
3474 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3475 	item->last = NULL;
3476 	item++;
3477 	item->type = RTE_FLOW_ITEM_TYPE_END;
3478 	return 0;
3479 }
3480 
3481 /**
3482  * The last stage of splitting chain, just creates the subflow
3483  * without any modification.
3484  *
3485  * @param[in] dev
3486  *   Pointer to Ethernet device.
3487  * @param[in] flow
3488  *   Parent flow structure pointer.
3489  * @param[in, out] sub_flow
3490  *   Pointer to return the created subflow, may be NULL.
3491  * @param[in] prefix_layers
3492  *   Prefix subflow layers, may be 0.
3493  * @param[in] attr
3494  *   Flow rule attributes.
3495  * @param[in] items
3496  *   Pattern specification (list terminated by the END pattern item).
3497  * @param[in] actions
3498  *   Associated actions (list terminated by the END action).
3499  * @param[in] external
3500  *   This flow rule is created by request external to PMD.
3501  * @param[out] error
3502  *   Perform verbose error reporting if not NULL.
3503  * @return
3504  *   0 on success, negative value otherwise
3505  */
3506 static int
3507 flow_create_split_inner(struct rte_eth_dev *dev,
3508 			struct rte_flow *flow,
3509 			struct mlx5_flow **sub_flow,
3510 			uint64_t prefix_layers,
3511 			const struct rte_flow_attr *attr,
3512 			const struct rte_flow_item items[],
3513 			const struct rte_flow_action actions[],
3514 			bool external, struct rte_flow_error *error)
3515 {
3516 	struct mlx5_flow *dev_flow;
3517 
3518 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3519 	if (!dev_flow)
3520 		return -rte_errno;
3521 	dev_flow->flow = flow;
3522 	dev_flow->external = external;
3523 	/* Subflow object was created, we must include one in the list. */
3524 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3525 		      dev_flow->handle, next);
3526 	/*
3527 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3528 	 * flow may need some user defined item layer flags.
3529 	 */
3530 	if (prefix_layers)
3531 		dev_flow->handle->layers = prefix_layers;
3532 	if (sub_flow)
3533 		*sub_flow = dev_flow;
3534 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3535 }
3536 
3537 /**
3538  * Split the meter flow.
3539  *
3540  * As meter flow will split to three sub flow, other than meter
3541  * action, the other actions make sense to only meter accepts
3542  * the packet. If it need to be dropped, no other additional
3543  * actions should be take.
3544  *
3545  * One kind of special action which decapsulates the L3 tunnel
3546  * header will be in the prefix sub flow, as not to take the
3547  * L3 tunnel header into account.
3548  *
3549  * @param dev
3550  *   Pointer to Ethernet device.
3551  * @param[in] items
3552  *   Pattern specification (list terminated by the END pattern item).
3553  * @param[out] sfx_items
3554  *   Suffix flow match items (list terminated by the END pattern item).
3555  * @param[in] actions
3556  *   Associated actions (list terminated by the END action).
3557  * @param[out] actions_sfx
3558  *   Suffix flow actions.
3559  * @param[out] actions_pre
3560  *   Prefix flow actions.
3561  * @param[out] pattern_sfx
3562  *   The pattern items for the suffix flow.
3563  * @param[out] tag_sfx
3564  *   Pointer to suffix flow tag.
3565  *
3566  * @return
3567  *   0 on success.
3568  */
3569 static int
3570 flow_meter_split_prep(struct rte_eth_dev *dev,
3571 		 const struct rte_flow_item items[],
3572 		 struct rte_flow_item sfx_items[],
3573 		 const struct rte_flow_action actions[],
3574 		 struct rte_flow_action actions_sfx[],
3575 		 struct rte_flow_action actions_pre[])
3576 {
3577 	struct rte_flow_action *tag_action = NULL;
3578 	struct rte_flow_item *tag_item;
3579 	struct mlx5_rte_flow_action_set_tag *set_tag;
3580 	struct rte_flow_error error;
3581 	const struct rte_flow_action_raw_encap *raw_encap;
3582 	const struct rte_flow_action_raw_decap *raw_decap;
3583 	struct mlx5_rte_flow_item_tag *tag_spec;
3584 	struct mlx5_rte_flow_item_tag *tag_mask;
3585 	uint32_t tag_id;
3586 	bool copy_vlan = false;
3587 
3588 	/* Prepare the actions for prefix and suffix flow. */
3589 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3590 		struct rte_flow_action **action_cur = NULL;
3591 
3592 		switch (actions->type) {
3593 		case RTE_FLOW_ACTION_TYPE_METER:
3594 			/* Add the extra tag action first. */
3595 			tag_action = actions_pre;
3596 			tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3597 			actions_pre++;
3598 			action_cur = &actions_pre;
3599 			break;
3600 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3601 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3602 			action_cur = &actions_pre;
3603 			break;
3604 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3605 			raw_encap = actions->conf;
3606 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3607 				action_cur = &actions_pre;
3608 			break;
3609 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3610 			raw_decap = actions->conf;
3611 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3612 				action_cur = &actions_pre;
3613 			break;
3614 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3615 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3616 			copy_vlan = true;
3617 			break;
3618 		default:
3619 			break;
3620 		}
3621 		if (!action_cur)
3622 			action_cur = &actions_sfx;
3623 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3624 		(*action_cur)++;
3625 	}
3626 	/* Add end action to the actions. */
3627 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3628 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3629 	actions_pre++;
3630 	/* Set the tag. */
3631 	set_tag = (void *)actions_pre;
3632 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3633 	/*
3634 	 * Get the id from the qrss_pool to make qrss share the id with meter.
3635 	 */
3636 	tag_id = flow_qrss_get_id(dev);
3637 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3638 	assert(tag_action);
3639 	tag_action->conf = set_tag;
3640 	/* Prepare the suffix subflow items. */
3641 	tag_item = sfx_items++;
3642 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3643 		int item_type = items->type;
3644 
3645 		switch (item_type) {
3646 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
3647 			memcpy(sfx_items, items, sizeof(*sfx_items));
3648 			sfx_items++;
3649 			break;
3650 		case RTE_FLOW_ITEM_TYPE_VLAN:
3651 			if (copy_vlan) {
3652 				memcpy(sfx_items, items, sizeof(*sfx_items));
3653 				/*
3654 				 * Convert to internal match item, it is used
3655 				 * for vlan push and set vid.
3656 				 */
3657 				sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3658 				sfx_items++;
3659 			}
3660 			break;
3661 		default:
3662 			break;
3663 		}
3664 	}
3665 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3666 	sfx_items++;
3667 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3668 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3669 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3670 	tag_mask = tag_spec + 1;
3671 	tag_mask->data = 0xffffff00;
3672 	tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3673 	tag_item->spec = tag_spec;
3674 	tag_item->last = NULL;
3675 	tag_item->mask = tag_mask;
3676 	return tag_id;
3677 }
3678 
3679 /**
3680  * Split action list having QUEUE/RSS for metadata register copy.
3681  *
3682  * Once Q/RSS action is detected in user's action list, the flow action
3683  * should be split in order to copy metadata registers, which will happen in
3684  * RX_CP_TBL like,
3685  *   - CQE->flow_tag := reg_c[1] (MARK)
3686  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3687  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3688  * This is because the last action of each flow must be a terminal action
3689  * (QUEUE, RSS or DROP).
3690  *
3691  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3692  * stored and kept in the mlx5_flow structure per each sub_flow.
3693  *
3694  * The Q/RSS action is replaced with,
3695  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
3696  * And the following JUMP action is added at the end,
3697  *   - JUMP, to RX_CP_TBL.
3698  *
3699  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3700  * flow_create_split_metadata() routine. The flow will look like,
3701  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
3702  *
3703  * @param dev
3704  *   Pointer to Ethernet device.
3705  * @param[out] split_actions
3706  *   Pointer to store split actions to jump to CP_TBL.
3707  * @param[in] actions
3708  *   Pointer to the list of original flow actions.
3709  * @param[in] qrss
3710  *   Pointer to the Q/RSS action.
3711  * @param[in] actions_n
3712  *   Number of original actions.
3713  * @param[out] error
3714  *   Perform verbose error reporting if not NULL.
3715  *
3716  * @return
3717  *   non-zero unique flow_id on success, otherwise 0 and
3718  *   error/rte_error are set.
3719  */
3720 static uint32_t
3721 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3722 			  struct rte_flow_action *split_actions,
3723 			  const struct rte_flow_action *actions,
3724 			  const struct rte_flow_action *qrss,
3725 			  int actions_n, struct rte_flow_error *error)
3726 {
3727 	struct mlx5_rte_flow_action_set_tag *set_tag;
3728 	struct rte_flow_action_jump *jump;
3729 	const int qrss_idx = qrss - actions;
3730 	uint32_t flow_id = 0;
3731 	int ret = 0;
3732 
3733 	/*
3734 	 * Given actions will be split
3735 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3736 	 * - Add jump to mreg CP_TBL.
3737 	 * As a result, there will be one more action.
3738 	 */
3739 	++actions_n;
3740 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3741 	set_tag = (void *)(split_actions + actions_n);
3742 	/*
3743 	 * If tag action is not set to void(it means we are not the meter
3744 	 * suffix flow), add the tag action. Since meter suffix flow already
3745 	 * has the tag added.
3746 	 */
3747 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3748 		/*
3749 		 * Allocate the new subflow ID. This one is unique within
3750 		 * device and not shared with representors. Otherwise,
3751 		 * we would have to resolve multi-thread access synch
3752 		 * issue. Each flow on the shared device is appended
3753 		 * with source vport identifier, so the resulting
3754 		 * flows will be unique in the shared (by master and
3755 		 * representors) domain even if they have coinciding
3756 		 * IDs.
3757 		 */
3758 		flow_id = flow_qrss_get_id(dev);
3759 		if (!flow_id)
3760 			return rte_flow_error_set(error, ENOMEM,
3761 						  RTE_FLOW_ERROR_TYPE_ACTION,
3762 						  NULL, "can't allocate id "
3763 						  "for split Q/RSS subflow");
3764 		/* Internal SET_TAG action to set flow ID. */
3765 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
3766 			.data = flow_id,
3767 		};
3768 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3769 		if (ret < 0)
3770 			return ret;
3771 		set_tag->id = ret;
3772 		/* Construct new actions array. */
3773 		/* Replace QUEUE/RSS action. */
3774 		split_actions[qrss_idx] = (struct rte_flow_action){
3775 			.type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3776 			.conf = set_tag,
3777 		};
3778 	}
3779 	/* JUMP action to jump to mreg copy table (CP_TBL). */
3780 	jump = (void *)(set_tag + 1);
3781 	*jump = (struct rte_flow_action_jump){
3782 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3783 	};
3784 	split_actions[actions_n - 2] = (struct rte_flow_action){
3785 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
3786 		.conf = jump,
3787 	};
3788 	split_actions[actions_n - 1] = (struct rte_flow_action){
3789 		.type = RTE_FLOW_ACTION_TYPE_END,
3790 	};
3791 	return flow_id;
3792 }
3793 
3794 /**
3795  * Extend the given action list for Tx metadata copy.
3796  *
3797  * Copy the given action list to the ext_actions and add flow metadata register
3798  * copy action in order to copy reg_a set by WQE to reg_c[0].
3799  *
3800  * @param[out] ext_actions
3801  *   Pointer to the extended action list.
3802  * @param[in] actions
3803  *   Pointer to the list of actions.
3804  * @param[in] actions_n
3805  *   Number of actions in the list.
3806  * @param[out] error
3807  *   Perform verbose error reporting if not NULL.
3808  * @param[in] encap_idx
3809  *   The encap action inndex.
3810  *
3811  * @return
3812  *   0 on success, negative value otherwise
3813  */
3814 static int
3815 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3816 		       struct rte_flow_action *ext_actions,
3817 		       const struct rte_flow_action *actions,
3818 		       int actions_n, struct rte_flow_error *error,
3819 		       int encap_idx)
3820 {
3821 	struct mlx5_flow_action_copy_mreg *cp_mreg =
3822 		(struct mlx5_flow_action_copy_mreg *)
3823 			(ext_actions + actions_n + 1);
3824 	int ret;
3825 
3826 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3827 	if (ret < 0)
3828 		return ret;
3829 	cp_mreg->dst = ret;
3830 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3831 	if (ret < 0)
3832 		return ret;
3833 	cp_mreg->src = ret;
3834 	if (encap_idx != 0)
3835 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3836 	if (encap_idx == actions_n - 1) {
3837 		ext_actions[actions_n - 1] = (struct rte_flow_action){
3838 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3839 			.conf = cp_mreg,
3840 		};
3841 		ext_actions[actions_n] = (struct rte_flow_action){
3842 			.type = RTE_FLOW_ACTION_TYPE_END,
3843 		};
3844 	} else {
3845 		ext_actions[encap_idx] = (struct rte_flow_action){
3846 			.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3847 			.conf = cp_mreg,
3848 		};
3849 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3850 				sizeof(*ext_actions) * (actions_n - encap_idx));
3851 	}
3852 	return 0;
3853 }
3854 
3855 /**
3856  * The splitting for metadata feature.
3857  *
3858  * - Q/RSS action on NIC Rx should be split in order to pass by
3859  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
3860  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
3861  *
3862  * - All the actions on NIC Tx should have a mreg copy action to
3863  *   copy reg_a from WQE to reg_c[0].
3864  *
3865  * @param dev
3866  *   Pointer to Ethernet device.
3867  * @param[in] flow
3868  *   Parent flow structure pointer.
3869  * @param[in] prefix_layers
3870  *   Prefix flow layer flags.
3871  * @param[in] attr
3872  *   Flow rule attributes.
3873  * @param[in] items
3874  *   Pattern specification (list terminated by the END pattern item).
3875  * @param[in] actions
3876  *   Associated actions (list terminated by the END action).
3877  * @param[in] external
3878  *   This flow rule is created by request external to PMD.
3879  * @param[out] error
3880  *   Perform verbose error reporting if not NULL.
3881  * @return
3882  *   0 on success, negative value otherwise
3883  */
3884 static int
3885 flow_create_split_metadata(struct rte_eth_dev *dev,
3886 			   struct rte_flow *flow,
3887 			   uint64_t prefix_layers,
3888 			   const struct rte_flow_attr *attr,
3889 			   const struct rte_flow_item items[],
3890 			   const struct rte_flow_action actions[],
3891 			   bool external, struct rte_flow_error *error)
3892 {
3893 	struct mlx5_priv *priv = dev->data->dev_private;
3894 	struct mlx5_dev_config *config = &priv->config;
3895 	const struct rte_flow_action *qrss = NULL;
3896 	struct rte_flow_action *ext_actions = NULL;
3897 	struct mlx5_flow *dev_flow = NULL;
3898 	uint32_t qrss_id = 0;
3899 	int mtr_sfx = 0;
3900 	size_t act_size;
3901 	int actions_n;
3902 	int encap_idx;
3903 	int ret;
3904 
3905 	/* Check whether extensive metadata feature is engaged. */
3906 	if (!config->dv_flow_en ||
3907 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3908 	    !mlx5_flow_ext_mreg_supported(dev))
3909 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3910 					       attr, items, actions, external,
3911 					       error);
3912 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3913 							   &encap_idx);
3914 	if (qrss) {
3915 		/* Exclude hairpin flows from splitting. */
3916 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3917 			const struct rte_flow_action_queue *queue;
3918 
3919 			queue = qrss->conf;
3920 			if (mlx5_rxq_get_type(dev, queue->index) ==
3921 			    MLX5_RXQ_TYPE_HAIRPIN)
3922 				qrss = NULL;
3923 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3924 			const struct rte_flow_action_rss *rss;
3925 
3926 			rss = qrss->conf;
3927 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3928 			    MLX5_RXQ_TYPE_HAIRPIN)
3929 				qrss = NULL;
3930 		}
3931 	}
3932 	if (qrss) {
3933 		/* Check if it is in meter suffix table. */
3934 		mtr_sfx = attr->group == (attr->transfer ?
3935 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3936 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3937 		/*
3938 		 * Q/RSS action on NIC Rx should be split in order to pass by
3939 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3940 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3941 		 */
3942 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3943 			   sizeof(struct rte_flow_action_set_tag) +
3944 			   sizeof(struct rte_flow_action_jump);
3945 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3946 		if (!ext_actions)
3947 			return rte_flow_error_set(error, ENOMEM,
3948 						  RTE_FLOW_ERROR_TYPE_ACTION,
3949 						  NULL, "no memory to split "
3950 						  "metadata flow");
3951 		/*
3952 		 * If we are the suffix flow of meter, tag already exist.
3953 		 * Set the tag action to void.
3954 		 */
3955 		if (mtr_sfx)
3956 			ext_actions[qrss - actions].type =
3957 						RTE_FLOW_ACTION_TYPE_VOID;
3958 		else
3959 			ext_actions[qrss - actions].type =
3960 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3961 		/*
3962 		 * Create the new actions list with removed Q/RSS action
3963 		 * and appended set tag and jump to register copy table
3964 		 * (RX_CP_TBL). We should preallocate unique tag ID here
3965 		 * in advance, because it is needed for set tag action.
3966 		 */
3967 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3968 						    qrss, actions_n, error);
3969 		if (!mtr_sfx && !qrss_id) {
3970 			ret = -rte_errno;
3971 			goto exit;
3972 		}
3973 	} else if (attr->egress && !attr->transfer) {
3974 		/*
3975 		 * All the actions on NIC Tx should have a metadata register
3976 		 * copy action to copy reg_a from WQE to reg_c[meta]
3977 		 */
3978 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3979 			   sizeof(struct mlx5_flow_action_copy_mreg);
3980 		ext_actions = rte_zmalloc(__func__, act_size, 0);
3981 		if (!ext_actions)
3982 			return rte_flow_error_set(error, ENOMEM,
3983 						  RTE_FLOW_ERROR_TYPE_ACTION,
3984 						  NULL, "no memory to split "
3985 						  "metadata flow");
3986 		/* Create the action list appended with copy register. */
3987 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3988 					     actions_n, error, encap_idx);
3989 		if (ret < 0)
3990 			goto exit;
3991 	}
3992 	/* Add the unmodified original or prefix subflow. */
3993 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3994 				      items, ext_actions ? ext_actions :
3995 				      actions, external, error);
3996 	if (ret < 0)
3997 		goto exit;
3998 	MLX5_ASSERT(dev_flow);
3999 	if (qrss) {
4000 		const struct rte_flow_attr q_attr = {
4001 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4002 			.ingress = 1,
4003 		};
4004 		/* Internal PMD action to set register. */
4005 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4006 			.data = qrss_id,
4007 			.id = 0,
4008 		};
4009 		struct rte_flow_item q_items[] = {
4010 			{
4011 				.type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4012 				.spec = &q_tag_spec,
4013 				.last = NULL,
4014 				.mask = NULL,
4015 			},
4016 			{
4017 				.type = RTE_FLOW_ITEM_TYPE_END,
4018 			},
4019 		};
4020 		struct rte_flow_action q_actions[] = {
4021 			{
4022 				.type = qrss->type,
4023 				.conf = qrss->conf,
4024 			},
4025 			{
4026 				.type = RTE_FLOW_ACTION_TYPE_END,
4027 			},
4028 		};
4029 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4030 
4031 		/*
4032 		 * Configure the tag item only if there is no meter subflow.
4033 		 * Since tag is already marked in the meter suffix subflow
4034 		 * we can just use the meter suffix items as is.
4035 		 */
4036 		if (qrss_id) {
4037 			/* Not meter subflow. */
4038 			MLX5_ASSERT(!mtr_sfx);
4039 			/*
4040 			 * Put unique id in prefix flow due to it is destroyed
4041 			 * after suffix flow and id will be freed after there
4042 			 * is no actual flows with this id and identifier
4043 			 * reallocation becomes possible (for example, for
4044 			 * other flows in other threads).
4045 			 */
4046 			dev_flow->handle->split_flow_id = qrss_id;
4047 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4048 						   error);
4049 			if (ret < 0)
4050 				goto exit;
4051 			q_tag_spec.id = ret;
4052 		}
4053 		dev_flow = NULL;
4054 		/* Add suffix subflow to execute Q/RSS. */
4055 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4056 					      &q_attr, mtr_sfx ? items :
4057 					      q_items, q_actions,
4058 					      external, error);
4059 		if (ret < 0)
4060 			goto exit;
4061 		/* qrss ID should be freed if failed. */
4062 		qrss_id = 0;
4063 		MLX5_ASSERT(dev_flow);
4064 	}
4065 
4066 exit:
4067 	/*
4068 	 * We do not destroy the partially created sub_flows in case of error.
4069 	 * These ones are included into parent flow list and will be destroyed
4070 	 * by flow_drv_destroy.
4071 	 */
4072 	flow_qrss_free_id(dev, qrss_id);
4073 	rte_free(ext_actions);
4074 	return ret;
4075 }
4076 
4077 /**
4078  * The splitting for meter feature.
4079  *
4080  * - The meter flow will be split to two flows as prefix and
4081  *   suffix flow. The packets make sense only it pass the prefix
4082  *   meter action.
4083  *
4084  * - Reg_C_5 is used for the packet to match betweend prefix and
4085  *   suffix flow.
4086  *
4087  * @param dev
4088  *   Pointer to Ethernet device.
4089  * @param[in] flow
4090  *   Parent flow structure pointer.
4091  * @param[in] attr
4092  *   Flow rule attributes.
4093  * @param[in] items
4094  *   Pattern specification (list terminated by the END pattern item).
4095  * @param[in] actions
4096  *   Associated actions (list terminated by the END action).
4097  * @param[in] external
4098  *   This flow rule is created by request external to PMD.
4099  * @param[out] error
4100  *   Perform verbose error reporting if not NULL.
4101  * @return
4102  *   0 on success, negative value otherwise
4103  */
4104 static int
4105 flow_create_split_meter(struct rte_eth_dev *dev,
4106 			   struct rte_flow *flow,
4107 			   const struct rte_flow_attr *attr,
4108 			   const struct rte_flow_item items[],
4109 			   const struct rte_flow_action actions[],
4110 			   bool external, struct rte_flow_error *error)
4111 {
4112 	struct mlx5_priv *priv = dev->data->dev_private;
4113 	struct rte_flow_action *sfx_actions = NULL;
4114 	struct rte_flow_action *pre_actions = NULL;
4115 	struct rte_flow_item *sfx_items = NULL;
4116 	struct mlx5_flow *dev_flow = NULL;
4117 	struct rte_flow_attr sfx_attr = *attr;
4118 	uint32_t mtr = 0;
4119 	uint32_t mtr_tag_id = 0;
4120 	size_t act_size;
4121 	size_t item_size;
4122 	int actions_n = 0;
4123 	int ret;
4124 
4125 	if (priv->mtr_en)
4126 		actions_n = flow_check_meter_action(actions, &mtr);
4127 	if (mtr) {
4128 		/* The five prefix actions: meter, decap, encap, tag, end. */
4129 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4130 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4131 		/* tag, vlan, port id, end. */
4132 #define METER_SUFFIX_ITEM 4
4133 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4134 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4135 		sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4136 		if (!sfx_actions)
4137 			return rte_flow_error_set(error, ENOMEM,
4138 						  RTE_FLOW_ERROR_TYPE_ACTION,
4139 						  NULL, "no memory to split "
4140 						  "meter flow");
4141 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4142 			     act_size);
4143 		pre_actions = sfx_actions + actions_n;
4144 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4145 						   actions, sfx_actions,
4146 						   pre_actions);
4147 		if (!mtr_tag_id) {
4148 			ret = -rte_errno;
4149 			goto exit;
4150 		}
4151 		/* Add the prefix subflow. */
4152 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4153 					      items, pre_actions, external,
4154 					      error);
4155 		if (ret) {
4156 			ret = -rte_errno;
4157 			goto exit;
4158 		}
4159 		dev_flow->handle->split_flow_id = mtr_tag_id;
4160 		/* Setting the sfx group atrr. */
4161 		sfx_attr.group = sfx_attr.transfer ?
4162 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4163 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4164 	}
4165 	/* Add the prefix subflow. */
4166 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4167 					 flow_get_prefix_layer_flags(dev_flow) :
4168 					 0, &sfx_attr,
4169 					 sfx_items ? sfx_items : items,
4170 					 sfx_actions ? sfx_actions : actions,
4171 					 external, error);
4172 exit:
4173 	if (sfx_actions)
4174 		rte_free(sfx_actions);
4175 	return ret;
4176 }
4177 
4178 /**
4179  * Split the flow to subflow set. The splitters might be linked
4180  * in the chain, like this:
4181  * flow_create_split_outer() calls:
4182  *   flow_create_split_meter() calls:
4183  *     flow_create_split_metadata(meter_subflow_0) calls:
4184  *       flow_create_split_inner(metadata_subflow_0)
4185  *       flow_create_split_inner(metadata_subflow_1)
4186  *       flow_create_split_inner(metadata_subflow_2)
4187  *     flow_create_split_metadata(meter_subflow_1) calls:
4188  *       flow_create_split_inner(metadata_subflow_0)
4189  *       flow_create_split_inner(metadata_subflow_1)
4190  *       flow_create_split_inner(metadata_subflow_2)
4191  *
4192  * This provide flexible way to add new levels of flow splitting.
4193  * The all of successfully created subflows are included to the
4194  * parent flow dev_flow list.
4195  *
4196  * @param dev
4197  *   Pointer to Ethernet device.
4198  * @param[in] flow
4199  *   Parent flow structure pointer.
4200  * @param[in] attr
4201  *   Flow rule attributes.
4202  * @param[in] items
4203  *   Pattern specification (list terminated by the END pattern item).
4204  * @param[in] actions
4205  *   Associated actions (list terminated by the END action).
4206  * @param[in] external
4207  *   This flow rule is created by request external to PMD.
4208  * @param[out] error
4209  *   Perform verbose error reporting if not NULL.
4210  * @return
4211  *   0 on success, negative value otherwise
4212  */
4213 static int
4214 flow_create_split_outer(struct rte_eth_dev *dev,
4215 			struct rte_flow *flow,
4216 			const struct rte_flow_attr *attr,
4217 			const struct rte_flow_item items[],
4218 			const struct rte_flow_action actions[],
4219 			bool external, struct rte_flow_error *error)
4220 {
4221 	int ret;
4222 
4223 	ret = flow_create_split_meter(dev, flow, attr, items,
4224 					 actions, external, error);
4225 	MLX5_ASSERT(ret <= 0);
4226 	return ret;
4227 }
4228 
4229 /**
4230  * Create a flow and add it to @p list.
4231  *
4232  * @param dev
4233  *   Pointer to Ethernet device.
4234  * @param list
4235  *   Pointer to a TAILQ flow list. If this parameter NULL,
4236  *   no list insertion occurred, flow is just created,
4237  *   this is caller's responsibility to track the
4238  *   created flow.
4239  * @param[in] attr
4240  *   Flow rule attributes.
4241  * @param[in] items
4242  *   Pattern specification (list terminated by the END pattern item).
4243  * @param[in] actions
4244  *   Associated actions (list terminated by the END action).
4245  * @param[in] external
4246  *   This flow rule is created by request external to PMD.
4247  * @param[out] error
4248  *   Perform verbose error reporting if not NULL.
4249  *
4250  * @return
4251  *   A flow index on success, 0 otherwise and rte_errno is set.
4252  */
4253 static uint32_t
4254 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4255 		 const struct rte_flow_attr *attr,
4256 		 const struct rte_flow_item items[],
4257 		 const struct rte_flow_action actions[],
4258 		 bool external, struct rte_flow_error *error)
4259 {
4260 	struct mlx5_priv *priv = dev->data->dev_private;
4261 	struct rte_flow *flow = NULL;
4262 	struct mlx5_flow *dev_flow;
4263 	const struct rte_flow_action_rss *rss;
4264 	union {
4265 		struct rte_flow_expand_rss buf;
4266 		uint8_t buffer[2048];
4267 	} expand_buffer;
4268 	union {
4269 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4270 		uint8_t buffer[2048];
4271 	} actions_rx;
4272 	union {
4273 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4274 		uint8_t buffer[2048];
4275 	} actions_hairpin_tx;
4276 	union {
4277 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4278 		uint8_t buffer[2048];
4279 	} items_tx;
4280 	struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4281 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4282 					      priv->rss_desc)[!!priv->flow_idx];
4283 	const struct rte_flow_action *p_actions_rx = actions;
4284 	uint32_t i;
4285 	uint32_t idx = 0;
4286 	int hairpin_flow;
4287 	uint32_t hairpin_id = 0;
4288 	struct rte_flow_attr attr_tx = { .priority = 0 };
4289 	int ret;
4290 
4291 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4292 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4293 				external, hairpin_flow, error);
4294 	if (ret < 0)
4295 		return 0;
4296 	if (hairpin_flow > 0) {
4297 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4298 			rte_errno = EINVAL;
4299 			return 0;
4300 		}
4301 		flow_hairpin_split(dev, actions, actions_rx.actions,
4302 				   actions_hairpin_tx.actions, items_tx.items,
4303 				   &hairpin_id);
4304 		p_actions_rx = actions_rx.actions;
4305 	}
4306 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4307 	if (!flow) {
4308 		rte_errno = ENOMEM;
4309 		goto error_before_flow;
4310 	}
4311 	flow->drv_type = flow_get_drv_type(dev, attr);
4312 	if (hairpin_id != 0)
4313 		flow->hairpin_flow_id = hairpin_id;
4314 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4315 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4316 	memset(rss_desc, 0, sizeof(*rss_desc));
4317 	rss = flow_get_rss_action(p_actions_rx);
4318 	if (rss) {
4319 		/*
4320 		 * The following information is required by
4321 		 * mlx5_flow_hashfields_adjust() in advance.
4322 		 */
4323 		rss_desc->level = rss->level;
4324 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4325 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4326 	}
4327 	flow->dev_handles = 0;
4328 	if (rss && rss->types) {
4329 		unsigned int graph_root;
4330 
4331 		graph_root = find_graph_root(items, rss->level);
4332 		ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4333 					  items, rss->types,
4334 					  mlx5_support_expansion,
4335 					  graph_root);
4336 		MLX5_ASSERT(ret > 0 &&
4337 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4338 	} else {
4339 		buf->entries = 1;
4340 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4341 	}
4342 	/*
4343 	 * Record the start index when there is a nested call. All sub-flows
4344 	 * need to be translated before another calling.
4345 	 * No need to use ping-pong buffer to save memory here.
4346 	 */
4347 	if (priv->flow_idx) {
4348 		MLX5_ASSERT(!priv->flow_nested_idx);
4349 		priv->flow_nested_idx = priv->flow_idx;
4350 	}
4351 	for (i = 0; i < buf->entries; ++i) {
4352 		/*
4353 		 * The splitter may create multiple dev_flows,
4354 		 * depending on configuration. In the simplest
4355 		 * case it just creates unmodified original flow.
4356 		 */
4357 		ret = flow_create_split_outer(dev, flow, attr,
4358 					      buf->entry[i].pattern,
4359 					      p_actions_rx, external,
4360 					      error);
4361 		if (ret < 0)
4362 			goto error;
4363 	}
4364 	/* Create the tx flow. */
4365 	if (hairpin_flow) {
4366 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4367 		attr_tx.ingress = 0;
4368 		attr_tx.egress = 1;
4369 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4370 					    actions_hairpin_tx.actions, error);
4371 		if (!dev_flow)
4372 			goto error;
4373 		dev_flow->flow = flow;
4374 		dev_flow->external = 0;
4375 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4376 			      dev_flow->handle, next);
4377 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4378 					 items_tx.items,
4379 					 actions_hairpin_tx.actions, error);
4380 		if (ret < 0)
4381 			goto error;
4382 	}
4383 	/*
4384 	 * Update the metadata register copy table. If extensive
4385 	 * metadata feature is enabled and registers are supported
4386 	 * we might create the extra rte_flow for each unique
4387 	 * MARK/FLAG action ID.
4388 	 *
4389 	 * The table is updated for ingress Flows only, because
4390 	 * the egress Flows belong to the different device and
4391 	 * copy table should be updated in peer NIC Rx domain.
4392 	 */
4393 	if (attr->ingress &&
4394 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4395 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4396 		if (ret)
4397 			goto error;
4398 	}
4399 	/*
4400 	 * If the flow is external (from application) OR device is started, then
4401 	 * the flow will be applied immediately.
4402 	 */
4403 	if (external || dev->data->dev_started) {
4404 		ret = flow_drv_apply(dev, flow, error);
4405 		if (ret < 0)
4406 			goto error;
4407 	}
4408 	if (list)
4409 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4410 			     flow, next);
4411 	flow_rxq_flags_set(dev, flow);
4412 	/* Nested flow creation index recovery. */
4413 	priv->flow_idx = priv->flow_nested_idx;
4414 	if (priv->flow_nested_idx)
4415 		priv->flow_nested_idx = 0;
4416 	return idx;
4417 error:
4418 	MLX5_ASSERT(flow);
4419 	ret = rte_errno; /* Save rte_errno before cleanup. */
4420 	flow_mreg_del_copy_action(dev, flow);
4421 	flow_drv_destroy(dev, flow);
4422 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4423 	rte_errno = ret; /* Restore rte_errno. */
4424 error_before_flow:
4425 	ret = rte_errno;
4426 	if (hairpin_id)
4427 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4428 				     hairpin_id);
4429 	rte_errno = ret;
4430 	priv->flow_idx = priv->flow_nested_idx;
4431 	if (priv->flow_nested_idx)
4432 		priv->flow_nested_idx = 0;
4433 	return 0;
4434 }
4435 
4436 /**
4437  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4438  * incoming packets to table 1.
4439  *
4440  * Other flow rules, requested for group n, will be created in
4441  * e-switch table n+1.
4442  * Jump action to e-switch group n will be created to group n+1.
4443  *
4444  * Used when working in switchdev mode, to utilise advantages of table 1
4445  * and above.
4446  *
4447  * @param dev
4448  *   Pointer to Ethernet device.
4449  *
4450  * @return
4451  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4452  */
4453 struct rte_flow *
4454 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4455 {
4456 	const struct rte_flow_attr attr = {
4457 		.group = 0,
4458 		.priority = 0,
4459 		.ingress = 1,
4460 		.egress = 0,
4461 		.transfer = 1,
4462 	};
4463 	const struct rte_flow_item pattern = {
4464 		.type = RTE_FLOW_ITEM_TYPE_END,
4465 	};
4466 	struct rte_flow_action_jump jump = {
4467 		.group = 1,
4468 	};
4469 	const struct rte_flow_action actions[] = {
4470 		{
4471 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4472 			.conf = &jump,
4473 		},
4474 		{
4475 			.type = RTE_FLOW_ACTION_TYPE_END,
4476 		},
4477 	};
4478 	struct mlx5_priv *priv = dev->data->dev_private;
4479 	struct rte_flow_error error;
4480 
4481 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4482 						   &attr, &pattern,
4483 						   actions, false, &error);
4484 }
4485 
4486 /**
4487  * Validate a flow supported by the NIC.
4488  *
4489  * @see rte_flow_validate()
4490  * @see rte_flow_ops
4491  */
4492 int
4493 mlx5_flow_validate(struct rte_eth_dev *dev,
4494 		   const struct rte_flow_attr *attr,
4495 		   const struct rte_flow_item items[],
4496 		   const struct rte_flow_action actions[],
4497 		   struct rte_flow_error *error)
4498 {
4499 	int hairpin_flow;
4500 
4501 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4502 	return flow_drv_validate(dev, attr, items, actions,
4503 				true, hairpin_flow, error);
4504 }
4505 
4506 /**
4507  * Create a flow.
4508  *
4509  * @see rte_flow_create()
4510  * @see rte_flow_ops
4511  */
4512 struct rte_flow *
4513 mlx5_flow_create(struct rte_eth_dev *dev,
4514 		 const struct rte_flow_attr *attr,
4515 		 const struct rte_flow_item items[],
4516 		 const struct rte_flow_action actions[],
4517 		 struct rte_flow_error *error)
4518 {
4519 	struct mlx5_priv *priv = dev->data->dev_private;
4520 
4521 	/*
4522 	 * If the device is not started yet, it is not allowed to created a
4523 	 * flow from application. PMD default flows and traffic control flows
4524 	 * are not affected.
4525 	 */
4526 	if (unlikely(!dev->data->dev_started)) {
4527 		DRV_LOG(DEBUG, "port %u is not started when "
4528 			"inserting a flow", dev->data->port_id);
4529 		rte_flow_error_set(error, ENODEV,
4530 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4531 				   NULL,
4532 				   "port not started");
4533 		return NULL;
4534 	}
4535 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4536 				  attr, items, actions, true, error);
4537 }
4538 
4539 /**
4540  * Destroy a flow in a list.
4541  *
4542  * @param dev
4543  *   Pointer to Ethernet device.
4544  * @param list
4545  *   Pointer to the Indexed flow list. If this parameter NULL,
4546  *   there is no flow removal from the list. Be noted that as
4547  *   flow is add to the indexed list, memory of the indexed
4548  *   list points to maybe changed as flow destroyed.
4549  * @param[in] flow_idx
4550  *   Index of flow to destroy.
4551  */
4552 static void
4553 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4554 		  uint32_t flow_idx)
4555 {
4556 	struct mlx5_priv *priv = dev->data->dev_private;
4557 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4558 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4559 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4560 
4561 	if (!flow)
4562 		return;
4563 	/*
4564 	 * Update RX queue flags only if port is started, otherwise it is
4565 	 * already clean.
4566 	 */
4567 	if (dev->data->dev_started)
4568 		flow_rxq_flags_trim(dev, flow);
4569 	if (flow->hairpin_flow_id)
4570 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4571 				     flow->hairpin_flow_id);
4572 	flow_drv_destroy(dev, flow);
4573 	if (list)
4574 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4575 			     flow_idx, flow, next);
4576 	flow_mreg_del_copy_action(dev, flow);
4577 	if (flow->fdir) {
4578 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4579 			if (priv_fdir_flow->rix_flow == flow_idx)
4580 				break;
4581 		}
4582 		if (priv_fdir_flow) {
4583 			LIST_REMOVE(priv_fdir_flow, next);
4584 			rte_free(priv_fdir_flow->fdir);
4585 			rte_free(priv_fdir_flow);
4586 		}
4587 	}
4588 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4589 }
4590 
4591 /**
4592  * Destroy all flows.
4593  *
4594  * @param dev
4595  *   Pointer to Ethernet device.
4596  * @param list
4597  *   Pointer to the Indexed flow list.
4598  * @param active
4599  *   If flushing is called avtively.
4600  */
4601 void
4602 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4603 {
4604 	uint32_t num_flushed = 0;
4605 
4606 	while (*list) {
4607 		flow_list_destroy(dev, list, *list);
4608 		num_flushed++;
4609 	}
4610 	if (active) {
4611 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4612 			dev->data->port_id, num_flushed);
4613 	}
4614 }
4615 
4616 /**
4617  * Remove all flows.
4618  *
4619  * @param dev
4620  *   Pointer to Ethernet device.
4621  * @param list
4622  *   Pointer to the Indexed flow list.
4623  */
4624 void
4625 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
4626 {
4627 	struct mlx5_priv *priv = dev->data->dev_private;
4628 	struct rte_flow *flow = NULL;
4629 	uint32_t idx;
4630 
4631 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4632 		      flow, next) {
4633 		flow_drv_remove(dev, flow);
4634 		flow_mreg_stop_copy_action(dev, flow);
4635 	}
4636 	flow_mreg_del_default_copy_action(dev);
4637 	flow_rxq_flags_clear(dev);
4638 }
4639 
4640 /**
4641  * Add all flows.
4642  *
4643  * @param dev
4644  *   Pointer to Ethernet device.
4645  * @param list
4646  *   Pointer to the Indexed flow list.
4647  *
4648  * @return
4649  *   0 on success, a negative errno value otherwise and rte_errno is set.
4650  */
4651 int
4652 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
4653 {
4654 	struct mlx5_priv *priv = dev->data->dev_private;
4655 	struct rte_flow *flow = NULL;
4656 	struct rte_flow_error error;
4657 	uint32_t idx;
4658 	int ret = 0;
4659 
4660 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4661 	ret = flow_mreg_add_default_copy_action(dev, &error);
4662 	if (ret < 0)
4663 		return -rte_errno;
4664 	/* Apply Flows created by application. */
4665 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
4666 		      flow, next) {
4667 		ret = flow_mreg_start_copy_action(dev, flow);
4668 		if (ret < 0)
4669 			goto error;
4670 		ret = flow_drv_apply(dev, flow, &error);
4671 		if (ret < 0)
4672 			goto error;
4673 		flow_rxq_flags_set(dev, flow);
4674 	}
4675 	return 0;
4676 error:
4677 	ret = rte_errno; /* Save rte_errno before cleanup. */
4678 	mlx5_flow_stop(dev, list);
4679 	rte_errno = ret; /* Restore rte_errno. */
4680 	return -rte_errno;
4681 }
4682 
4683 /**
4684  * Stop all default actions for flows.
4685  *
4686  * @param dev
4687  *   Pointer to Ethernet device.
4688  */
4689 void
4690 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4691 {
4692 	flow_mreg_del_default_copy_action(dev);
4693 	flow_rxq_flags_clear(dev);
4694 }
4695 
4696 /**
4697  * Start all default actions for flows.
4698  *
4699  * @param dev
4700  *   Pointer to Ethernet device.
4701  * @return
4702  *   0 on success, a negative errno value otherwise and rte_errno is set.
4703  */
4704 int
4705 mlx5_flow_start_default(struct rte_eth_dev *dev)
4706 {
4707 	struct rte_flow_error error;
4708 
4709 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4710 	return flow_mreg_add_default_copy_action(dev, &error);
4711 }
4712 
4713 /**
4714  * Allocate intermediate resources for flow creation.
4715  *
4716  * @param dev
4717  *   Pointer to Ethernet device.
4718  */
4719 void
4720 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4721 {
4722 	struct mlx5_priv *priv = dev->data->dev_private;
4723 
4724 	if (!priv->inter_flows) {
4725 		priv->inter_flows = rte_calloc(__func__, 1,
4726 				    MLX5_NUM_MAX_DEV_FLOWS *
4727 				    sizeof(struct mlx5_flow) +
4728 				    (sizeof(struct mlx5_flow_rss_desc) +
4729 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4730 		if (!priv->inter_flows) {
4731 			DRV_LOG(ERR, "can't allocate intermediate memory.");
4732 			return;
4733 		}
4734 	}
4735 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4736 			 [MLX5_NUM_MAX_DEV_FLOWS];
4737 	/* Reset the index. */
4738 	priv->flow_idx = 0;
4739 	priv->flow_nested_idx = 0;
4740 }
4741 
4742 /**
4743  * Free intermediate resources for flows.
4744  *
4745  * @param dev
4746  *   Pointer to Ethernet device.
4747  */
4748 void
4749 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4750 {
4751 	struct mlx5_priv *priv = dev->data->dev_private;
4752 
4753 	rte_free(priv->inter_flows);
4754 	priv->inter_flows = NULL;
4755 }
4756 
4757 /**
4758  * Verify the flow list is empty
4759  *
4760  * @param dev
4761  *  Pointer to Ethernet device.
4762  *
4763  * @return the number of flows not released.
4764  */
4765 int
4766 mlx5_flow_verify(struct rte_eth_dev *dev)
4767 {
4768 	struct mlx5_priv *priv = dev->data->dev_private;
4769 	struct rte_flow *flow;
4770 	uint32_t idx;
4771 	int ret = 0;
4772 
4773 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
4774 		      flow, next) {
4775 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
4776 			dev->data->port_id, (void *)flow);
4777 		++ret;
4778 	}
4779 	return ret;
4780 }
4781 
4782 /**
4783  * Enable default hairpin egress flow.
4784  *
4785  * @param dev
4786  *   Pointer to Ethernet device.
4787  * @param queue
4788  *   The queue index.
4789  *
4790  * @return
4791  *   0 on success, a negative errno value otherwise and rte_errno is set.
4792  */
4793 int
4794 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4795 			    uint32_t queue)
4796 {
4797 	struct mlx5_priv *priv = dev->data->dev_private;
4798 	const struct rte_flow_attr attr = {
4799 		.egress = 1,
4800 		.priority = 0,
4801 	};
4802 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
4803 		.queue = queue,
4804 	};
4805 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
4806 		.queue = UINT32_MAX,
4807 	};
4808 	struct rte_flow_item items[] = {
4809 		{
4810 			.type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4811 			.spec = &queue_spec,
4812 			.last = NULL,
4813 			.mask = &queue_mask,
4814 		},
4815 		{
4816 			.type = RTE_FLOW_ITEM_TYPE_END,
4817 		},
4818 	};
4819 	struct rte_flow_action_jump jump = {
4820 		.group = MLX5_HAIRPIN_TX_TABLE,
4821 	};
4822 	struct rte_flow_action actions[2];
4823 	uint32_t flow_idx;
4824 	struct rte_flow_error error;
4825 
4826 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4827 	actions[0].conf = &jump;
4828 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4829 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4830 				&attr, items, actions, false, &error);
4831 	if (!flow_idx) {
4832 		DRV_LOG(DEBUG,
4833 			"Failed to create ctrl flow: rte_errno(%d),"
4834 			" type(%d), message(%s)",
4835 			rte_errno, error.type,
4836 			error.message ? error.message : " (no stated reason)");
4837 		return -rte_errno;
4838 	}
4839 	return 0;
4840 }
4841 
4842 /**
4843  * Enable a control flow configured from the control plane.
4844  *
4845  * @param dev
4846  *   Pointer to Ethernet device.
4847  * @param eth_spec
4848  *   An Ethernet flow spec to apply.
4849  * @param eth_mask
4850  *   An Ethernet flow mask to apply.
4851  * @param vlan_spec
4852  *   A VLAN flow spec to apply.
4853  * @param vlan_mask
4854  *   A VLAN flow mask to apply.
4855  *
4856  * @return
4857  *   0 on success, a negative errno value otherwise and rte_errno is set.
4858  */
4859 int
4860 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4861 		    struct rte_flow_item_eth *eth_spec,
4862 		    struct rte_flow_item_eth *eth_mask,
4863 		    struct rte_flow_item_vlan *vlan_spec,
4864 		    struct rte_flow_item_vlan *vlan_mask)
4865 {
4866 	struct mlx5_priv *priv = dev->data->dev_private;
4867 	const struct rte_flow_attr attr = {
4868 		.ingress = 1,
4869 		.priority = MLX5_FLOW_PRIO_RSVD,
4870 	};
4871 	struct rte_flow_item items[] = {
4872 		{
4873 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4874 			.spec = eth_spec,
4875 			.last = NULL,
4876 			.mask = eth_mask,
4877 		},
4878 		{
4879 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4880 					      RTE_FLOW_ITEM_TYPE_END,
4881 			.spec = vlan_spec,
4882 			.last = NULL,
4883 			.mask = vlan_mask,
4884 		},
4885 		{
4886 			.type = RTE_FLOW_ITEM_TYPE_END,
4887 		},
4888 	};
4889 	uint16_t queue[priv->reta_idx_n];
4890 	struct rte_flow_action_rss action_rss = {
4891 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4892 		.level = 0,
4893 		.types = priv->rss_conf.rss_hf,
4894 		.key_len = priv->rss_conf.rss_key_len,
4895 		.queue_num = priv->reta_idx_n,
4896 		.key = priv->rss_conf.rss_key,
4897 		.queue = queue,
4898 	};
4899 	struct rte_flow_action actions[] = {
4900 		{
4901 			.type = RTE_FLOW_ACTION_TYPE_RSS,
4902 			.conf = &action_rss,
4903 		},
4904 		{
4905 			.type = RTE_FLOW_ACTION_TYPE_END,
4906 		},
4907 	};
4908 	uint32_t flow_idx;
4909 	struct rte_flow_error error;
4910 	unsigned int i;
4911 
4912 	if (!priv->reta_idx_n || !priv->rxqs_n) {
4913 		return 0;
4914 	}
4915 	for (i = 0; i != priv->reta_idx_n; ++i)
4916 		queue[i] = (*priv->reta_idx)[i];
4917 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
4918 				&attr, items, actions, false, &error);
4919 	if (!flow_idx)
4920 		return -rte_errno;
4921 	return 0;
4922 }
4923 
4924 /**
4925  * Enable a flow control configured from the control plane.
4926  *
4927  * @param dev
4928  *   Pointer to Ethernet device.
4929  * @param eth_spec
4930  *   An Ethernet flow spec to apply.
4931  * @param eth_mask
4932  *   An Ethernet flow mask to apply.
4933  *
4934  * @return
4935  *   0 on success, a negative errno value otherwise and rte_errno is set.
4936  */
4937 int
4938 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4939 	       struct rte_flow_item_eth *eth_spec,
4940 	       struct rte_flow_item_eth *eth_mask)
4941 {
4942 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4943 }
4944 
4945 /**
4946  * Destroy a flow.
4947  *
4948  * @see rte_flow_destroy()
4949  * @see rte_flow_ops
4950  */
4951 int
4952 mlx5_flow_destroy(struct rte_eth_dev *dev,
4953 		  struct rte_flow *flow,
4954 		  struct rte_flow_error *error __rte_unused)
4955 {
4956 	struct mlx5_priv *priv = dev->data->dev_private;
4957 
4958 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
4959 	return 0;
4960 }
4961 
4962 /**
4963  * Destroy all flows.
4964  *
4965  * @see rte_flow_flush()
4966  * @see rte_flow_ops
4967  */
4968 int
4969 mlx5_flow_flush(struct rte_eth_dev *dev,
4970 		struct rte_flow_error *error __rte_unused)
4971 {
4972 	struct mlx5_priv *priv = dev->data->dev_private;
4973 
4974 	mlx5_flow_list_flush(dev, &priv->flows, false);
4975 	return 0;
4976 }
4977 
4978 /**
4979  * Isolated mode.
4980  *
4981  * @see rte_flow_isolate()
4982  * @see rte_flow_ops
4983  */
4984 int
4985 mlx5_flow_isolate(struct rte_eth_dev *dev,
4986 		  int enable,
4987 		  struct rte_flow_error *error)
4988 {
4989 	struct mlx5_priv *priv = dev->data->dev_private;
4990 
4991 	if (dev->data->dev_started) {
4992 		rte_flow_error_set(error, EBUSY,
4993 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4994 				   NULL,
4995 				   "port must be stopped first");
4996 		return -rte_errno;
4997 	}
4998 	priv->isolated = !!enable;
4999 	if (enable)
5000 		dev->dev_ops = &mlx5_dev_ops_isolate;
5001 	else
5002 		dev->dev_ops = &mlx5_dev_ops;
5003 	return 0;
5004 }
5005 
5006 /**
5007  * Query a flow.
5008  *
5009  * @see rte_flow_query()
5010  * @see rte_flow_ops
5011  */
5012 static int
5013 flow_drv_query(struct rte_eth_dev *dev,
5014 	       uint32_t flow_idx,
5015 	       const struct rte_flow_action *actions,
5016 	       void *data,
5017 	       struct rte_flow_error *error)
5018 {
5019 	struct mlx5_priv *priv = dev->data->dev_private;
5020 	const struct mlx5_flow_driver_ops *fops;
5021 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5022 					       [MLX5_IPOOL_RTE_FLOW],
5023 					       flow_idx);
5024 	enum mlx5_flow_drv_type ftype;
5025 
5026 	if (!flow) {
5027 		return rte_flow_error_set(error, ENOENT,
5028 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5029 			  NULL,
5030 			  "invalid flow handle");
5031 	}
5032 	ftype = flow->drv_type;
5033 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5034 	fops = flow_get_drv_ops(ftype);
5035 
5036 	return fops->query(dev, flow, actions, data, error);
5037 }
5038 
5039 /**
5040  * Query a flow.
5041  *
5042  * @see rte_flow_query()
5043  * @see rte_flow_ops
5044  */
5045 int
5046 mlx5_flow_query(struct rte_eth_dev *dev,
5047 		struct rte_flow *flow,
5048 		const struct rte_flow_action *actions,
5049 		void *data,
5050 		struct rte_flow_error *error)
5051 {
5052 	int ret;
5053 
5054 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5055 			     error);
5056 	if (ret < 0)
5057 		return ret;
5058 	return 0;
5059 }
5060 
5061 /**
5062  * Convert a flow director filter to a generic flow.
5063  *
5064  * @param dev
5065  *   Pointer to Ethernet device.
5066  * @param fdir_filter
5067  *   Flow director filter to add.
5068  * @param attributes
5069  *   Generic flow parameters structure.
5070  *
5071  * @return
5072  *   0 on success, a negative errno value otherwise and rte_errno is set.
5073  */
5074 static int
5075 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5076 			 const struct rte_eth_fdir_filter *fdir_filter,
5077 			 struct mlx5_fdir *attributes)
5078 {
5079 	struct mlx5_priv *priv = dev->data->dev_private;
5080 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5081 	const struct rte_eth_fdir_masks *mask =
5082 		&dev->data->dev_conf.fdir_conf.mask;
5083 
5084 	/* Validate queue number. */
5085 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5086 		DRV_LOG(ERR, "port %u invalid queue number %d",
5087 			dev->data->port_id, fdir_filter->action.rx_queue);
5088 		rte_errno = EINVAL;
5089 		return -rte_errno;
5090 	}
5091 	attributes->attr.ingress = 1;
5092 	attributes->items[0] = (struct rte_flow_item) {
5093 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5094 		.spec = &attributes->l2,
5095 		.mask = &attributes->l2_mask,
5096 	};
5097 	switch (fdir_filter->action.behavior) {
5098 	case RTE_ETH_FDIR_ACCEPT:
5099 		attributes->actions[0] = (struct rte_flow_action){
5100 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5101 			.conf = &attributes->queue,
5102 		};
5103 		break;
5104 	case RTE_ETH_FDIR_REJECT:
5105 		attributes->actions[0] = (struct rte_flow_action){
5106 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5107 		};
5108 		break;
5109 	default:
5110 		DRV_LOG(ERR, "port %u invalid behavior %d",
5111 			dev->data->port_id,
5112 			fdir_filter->action.behavior);
5113 		rte_errno = ENOTSUP;
5114 		return -rte_errno;
5115 	}
5116 	attributes->queue.index = fdir_filter->action.rx_queue;
5117 	/* Handle L3. */
5118 	switch (fdir_filter->input.flow_type) {
5119 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5120 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5121 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5122 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5123 			.src_addr = input->flow.ip4_flow.src_ip,
5124 			.dst_addr = input->flow.ip4_flow.dst_ip,
5125 			.time_to_live = input->flow.ip4_flow.ttl,
5126 			.type_of_service = input->flow.ip4_flow.tos,
5127 		};
5128 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5129 			.src_addr = mask->ipv4_mask.src_ip,
5130 			.dst_addr = mask->ipv4_mask.dst_ip,
5131 			.time_to_live = mask->ipv4_mask.ttl,
5132 			.type_of_service = mask->ipv4_mask.tos,
5133 			.next_proto_id = mask->ipv4_mask.proto,
5134 		};
5135 		attributes->items[1] = (struct rte_flow_item){
5136 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5137 			.spec = &attributes->l3,
5138 			.mask = &attributes->l3_mask,
5139 		};
5140 		break;
5141 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5142 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5143 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5144 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5145 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5146 			.proto = input->flow.ipv6_flow.proto,
5147 		};
5148 
5149 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5150 		       input->flow.ipv6_flow.src_ip,
5151 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5152 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5153 		       input->flow.ipv6_flow.dst_ip,
5154 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5155 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5156 		       mask->ipv6_mask.src_ip,
5157 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5158 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5159 		       mask->ipv6_mask.dst_ip,
5160 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5161 		attributes->items[1] = (struct rte_flow_item){
5162 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5163 			.spec = &attributes->l3,
5164 			.mask = &attributes->l3_mask,
5165 		};
5166 		break;
5167 	default:
5168 		DRV_LOG(ERR, "port %u invalid flow type%d",
5169 			dev->data->port_id, fdir_filter->input.flow_type);
5170 		rte_errno = ENOTSUP;
5171 		return -rte_errno;
5172 	}
5173 	/* Handle L4. */
5174 	switch (fdir_filter->input.flow_type) {
5175 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5176 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5177 			.src_port = input->flow.udp4_flow.src_port,
5178 			.dst_port = input->flow.udp4_flow.dst_port,
5179 		};
5180 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5181 			.src_port = mask->src_port_mask,
5182 			.dst_port = mask->dst_port_mask,
5183 		};
5184 		attributes->items[2] = (struct rte_flow_item){
5185 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5186 			.spec = &attributes->l4,
5187 			.mask = &attributes->l4_mask,
5188 		};
5189 		break;
5190 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5191 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5192 			.src_port = input->flow.tcp4_flow.src_port,
5193 			.dst_port = input->flow.tcp4_flow.dst_port,
5194 		};
5195 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5196 			.src_port = mask->src_port_mask,
5197 			.dst_port = mask->dst_port_mask,
5198 		};
5199 		attributes->items[2] = (struct rte_flow_item){
5200 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5201 			.spec = &attributes->l4,
5202 			.mask = &attributes->l4_mask,
5203 		};
5204 		break;
5205 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5206 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5207 			.src_port = input->flow.udp6_flow.src_port,
5208 			.dst_port = input->flow.udp6_flow.dst_port,
5209 		};
5210 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5211 			.src_port = mask->src_port_mask,
5212 			.dst_port = mask->dst_port_mask,
5213 		};
5214 		attributes->items[2] = (struct rte_flow_item){
5215 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5216 			.spec = &attributes->l4,
5217 			.mask = &attributes->l4_mask,
5218 		};
5219 		break;
5220 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5221 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5222 			.src_port = input->flow.tcp6_flow.src_port,
5223 			.dst_port = input->flow.tcp6_flow.dst_port,
5224 		};
5225 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5226 			.src_port = mask->src_port_mask,
5227 			.dst_port = mask->dst_port_mask,
5228 		};
5229 		attributes->items[2] = (struct rte_flow_item){
5230 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5231 			.spec = &attributes->l4,
5232 			.mask = &attributes->l4_mask,
5233 		};
5234 		break;
5235 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5236 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5237 		break;
5238 	default:
5239 		DRV_LOG(ERR, "port %u invalid flow type%d",
5240 			dev->data->port_id, fdir_filter->input.flow_type);
5241 		rte_errno = ENOTSUP;
5242 		return -rte_errno;
5243 	}
5244 	return 0;
5245 }
5246 
5247 #define FLOW_FDIR_CMP(f1, f2, fld) \
5248 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5249 
5250 /**
5251  * Compare two FDIR flows. If items and actions are identical, the two flows are
5252  * regarded as same.
5253  *
5254  * @param dev
5255  *   Pointer to Ethernet device.
5256  * @param f1
5257  *   FDIR flow to compare.
5258  * @param f2
5259  *   FDIR flow to compare.
5260  *
5261  * @return
5262  *   Zero on match, 1 otherwise.
5263  */
5264 static int
5265 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5266 {
5267 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5268 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5269 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5270 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5271 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5272 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5273 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5274 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5275 		return 1;
5276 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5277 	    FLOW_FDIR_CMP(f1, f2, queue))
5278 		return 1;
5279 	return 0;
5280 }
5281 
5282 /**
5283  * Search device flow list to find out a matched FDIR flow.
5284  *
5285  * @param dev
5286  *   Pointer to Ethernet device.
5287  * @param fdir_flow
5288  *   FDIR flow to lookup.
5289  *
5290  * @return
5291  *   Index of flow if found, 0 otherwise.
5292  */
5293 static uint32_t
5294 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5295 {
5296 	struct mlx5_priv *priv = dev->data->dev_private;
5297 	uint32_t flow_idx = 0;
5298 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5299 
5300 	MLX5_ASSERT(fdir_flow);
5301 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5302 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5303 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5304 				dev->data->port_id, flow_idx);
5305 			flow_idx = priv_fdir_flow->rix_flow;
5306 			break;
5307 		}
5308 	}
5309 	return flow_idx;
5310 }
5311 
5312 /**
5313  * Add new flow director filter and store it in list.
5314  *
5315  * @param dev
5316  *   Pointer to Ethernet device.
5317  * @param fdir_filter
5318  *   Flow director filter to add.
5319  *
5320  * @return
5321  *   0 on success, a negative errno value otherwise and rte_errno is set.
5322  */
5323 static int
5324 flow_fdir_filter_add(struct rte_eth_dev *dev,
5325 		     const struct rte_eth_fdir_filter *fdir_filter)
5326 {
5327 	struct mlx5_priv *priv = dev->data->dev_private;
5328 	struct mlx5_fdir *fdir_flow;
5329 	struct rte_flow *flow;
5330 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5331 	uint32_t flow_idx;
5332 	int ret;
5333 
5334 	fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5335 	if (!fdir_flow) {
5336 		rte_errno = ENOMEM;
5337 		return -rte_errno;
5338 	}
5339 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5340 	if (ret)
5341 		goto error;
5342 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5343 	if (flow_idx) {
5344 		rte_errno = EEXIST;
5345 		goto error;
5346 	}
5347 	priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5348 				     0);
5349 	if (!priv_fdir_flow) {
5350 		rte_errno = ENOMEM;
5351 		goto error;
5352 	}
5353 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5354 				    fdir_flow->items, fdir_flow->actions, true,
5355 				    NULL);
5356 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5357 	if (!flow)
5358 		goto error;
5359 	flow->fdir = 1;
5360 	priv_fdir_flow->fdir = fdir_flow;
5361 	priv_fdir_flow->rix_flow = flow_idx;
5362 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5363 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5364 		dev->data->port_id, (void *)flow);
5365 	return 0;
5366 error:
5367 	rte_free(priv_fdir_flow);
5368 	rte_free(fdir_flow);
5369 	return -rte_errno;
5370 }
5371 
5372 /**
5373  * Delete specific filter.
5374  *
5375  * @param dev
5376  *   Pointer to Ethernet device.
5377  * @param fdir_filter
5378  *   Filter to be deleted.
5379  *
5380  * @return
5381  *   0 on success, a negative errno value otherwise and rte_errno is set.
5382  */
5383 static int
5384 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5385 			const struct rte_eth_fdir_filter *fdir_filter)
5386 {
5387 	struct mlx5_priv *priv = dev->data->dev_private;
5388 	uint32_t flow_idx;
5389 	struct mlx5_fdir fdir_flow = {
5390 		.attr.group = 0,
5391 	};
5392 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5393 	int ret;
5394 
5395 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5396 	if (ret)
5397 		return -rte_errno;
5398 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5399 		/* Find the fdir in priv list */
5400 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5401 			break;
5402 	}
5403 	if (!priv_fdir_flow)
5404 		return 0;
5405 	LIST_REMOVE(priv_fdir_flow, next);
5406 	flow_idx = priv_fdir_flow->rix_flow;
5407 	flow_list_destroy(dev, &priv->flows, flow_idx);
5408 	rte_free(priv_fdir_flow->fdir);
5409 	rte_free(priv_fdir_flow);
5410 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5411 		dev->data->port_id, flow_idx);
5412 	return 0;
5413 }
5414 
5415 /**
5416  * Update queue for specific filter.
5417  *
5418  * @param dev
5419  *   Pointer to Ethernet device.
5420  * @param fdir_filter
5421  *   Filter to be updated.
5422  *
5423  * @return
5424  *   0 on success, a negative errno value otherwise and rte_errno is set.
5425  */
5426 static int
5427 flow_fdir_filter_update(struct rte_eth_dev *dev,
5428 			const struct rte_eth_fdir_filter *fdir_filter)
5429 {
5430 	int ret;
5431 
5432 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5433 	if (ret)
5434 		return ret;
5435 	return flow_fdir_filter_add(dev, fdir_filter);
5436 }
5437 
5438 /**
5439  * Flush all filters.
5440  *
5441  * @param dev
5442  *   Pointer to Ethernet device.
5443  */
5444 static void
5445 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5446 {
5447 	struct mlx5_priv *priv = dev->data->dev_private;
5448 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5449 
5450 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5451 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5452 		LIST_REMOVE(priv_fdir_flow, next);
5453 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5454 		rte_free(priv_fdir_flow->fdir);
5455 		rte_free(priv_fdir_flow);
5456 	}
5457 }
5458 
5459 /**
5460  * Get flow director information.
5461  *
5462  * @param dev
5463  *   Pointer to Ethernet device.
5464  * @param[out] fdir_info
5465  *   Resulting flow director information.
5466  */
5467 static void
5468 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5469 {
5470 	struct rte_eth_fdir_masks *mask =
5471 		&dev->data->dev_conf.fdir_conf.mask;
5472 
5473 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5474 	fdir_info->guarant_spc = 0;
5475 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5476 	fdir_info->max_flexpayload = 0;
5477 	fdir_info->flow_types_mask[0] = 0;
5478 	fdir_info->flex_payload_unit = 0;
5479 	fdir_info->max_flex_payload_segment_num = 0;
5480 	fdir_info->flex_payload_limit = 0;
5481 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5482 }
5483 
5484 /**
5485  * Deal with flow director operations.
5486  *
5487  * @param dev
5488  *   Pointer to Ethernet device.
5489  * @param filter_op
5490  *   Operation to perform.
5491  * @param arg
5492  *   Pointer to operation-specific structure.
5493  *
5494  * @return
5495  *   0 on success, a negative errno value otherwise and rte_errno is set.
5496  */
5497 static int
5498 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5499 		    void *arg)
5500 {
5501 	enum rte_fdir_mode fdir_mode =
5502 		dev->data->dev_conf.fdir_conf.mode;
5503 
5504 	if (filter_op == RTE_ETH_FILTER_NOP)
5505 		return 0;
5506 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5507 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5508 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5509 			dev->data->port_id, fdir_mode);
5510 		rte_errno = EINVAL;
5511 		return -rte_errno;
5512 	}
5513 	switch (filter_op) {
5514 	case RTE_ETH_FILTER_ADD:
5515 		return flow_fdir_filter_add(dev, arg);
5516 	case RTE_ETH_FILTER_UPDATE:
5517 		return flow_fdir_filter_update(dev, arg);
5518 	case RTE_ETH_FILTER_DELETE:
5519 		return flow_fdir_filter_delete(dev, arg);
5520 	case RTE_ETH_FILTER_FLUSH:
5521 		flow_fdir_filter_flush(dev);
5522 		break;
5523 	case RTE_ETH_FILTER_INFO:
5524 		flow_fdir_info_get(dev, arg);
5525 		break;
5526 	default:
5527 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5528 			dev->data->port_id, filter_op);
5529 		rte_errno = EINVAL;
5530 		return -rte_errno;
5531 	}
5532 	return 0;
5533 }
5534 
5535 /**
5536  * Manage filter operations.
5537  *
5538  * @param dev
5539  *   Pointer to Ethernet device structure.
5540  * @param filter_type
5541  *   Filter type.
5542  * @param filter_op
5543  *   Operation to perform.
5544  * @param arg
5545  *   Pointer to operation-specific structure.
5546  *
5547  * @return
5548  *   0 on success, a negative errno value otherwise and rte_errno is set.
5549  */
5550 int
5551 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5552 		     enum rte_filter_type filter_type,
5553 		     enum rte_filter_op filter_op,
5554 		     void *arg)
5555 {
5556 	switch (filter_type) {
5557 	case RTE_ETH_FILTER_GENERIC:
5558 		if (filter_op != RTE_ETH_FILTER_GET) {
5559 			rte_errno = EINVAL;
5560 			return -rte_errno;
5561 		}
5562 		*(const void **)arg = &mlx5_flow_ops;
5563 		return 0;
5564 	case RTE_ETH_FILTER_FDIR:
5565 		return flow_fdir_ctrl_func(dev, filter_op, arg);
5566 	default:
5567 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
5568 			dev->data->port_id, filter_type);
5569 		rte_errno = ENOTSUP;
5570 		return -rte_errno;
5571 	}
5572 	return 0;
5573 }
5574 
5575 /**
5576  * Create the needed meter and suffix tables.
5577  *
5578  * @param[in] dev
5579  *   Pointer to Ethernet device.
5580  * @param[in] fm
5581  *   Pointer to the flow meter.
5582  *
5583  * @return
5584  *   Pointer to table set on success, NULL otherwise.
5585  */
5586 struct mlx5_meter_domains_infos *
5587 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5588 			  const struct mlx5_flow_meter *fm)
5589 {
5590 	const struct mlx5_flow_driver_ops *fops;
5591 
5592 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5593 	return fops->create_mtr_tbls(dev, fm);
5594 }
5595 
5596 /**
5597  * Destroy the meter table set.
5598  *
5599  * @param[in] dev
5600  *   Pointer to Ethernet device.
5601  * @param[in] tbl
5602  *   Pointer to the meter table set.
5603  *
5604  * @return
5605  *   0 on success.
5606  */
5607 int
5608 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5609 			   struct mlx5_meter_domains_infos *tbls)
5610 {
5611 	const struct mlx5_flow_driver_ops *fops;
5612 
5613 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5614 	return fops->destroy_mtr_tbls(dev, tbls);
5615 }
5616 
5617 /**
5618  * Create policer rules.
5619  *
5620  * @param[in] dev
5621  *   Pointer to Ethernet device.
5622  * @param[in] fm
5623  *   Pointer to flow meter structure.
5624  * @param[in] attr
5625  *   Pointer to flow attributes.
5626  *
5627  * @return
5628  *   0 on success, -1 otherwise.
5629  */
5630 int
5631 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5632 			       struct mlx5_flow_meter *fm,
5633 			       const struct rte_flow_attr *attr)
5634 {
5635 	const struct mlx5_flow_driver_ops *fops;
5636 
5637 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5638 	return fops->create_policer_rules(dev, fm, attr);
5639 }
5640 
5641 /**
5642  * Destroy policer rules.
5643  *
5644  * @param[in] fm
5645  *   Pointer to flow meter structure.
5646  * @param[in] attr
5647  *   Pointer to flow attributes.
5648  *
5649  * @return
5650  *   0 on success, -1 otherwise.
5651  */
5652 int
5653 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5654 				struct mlx5_flow_meter *fm,
5655 				const struct rte_flow_attr *attr)
5656 {
5657 	const struct mlx5_flow_driver_ops *fops;
5658 
5659 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5660 	return fops->destroy_policer_rules(dev, fm, attr);
5661 }
5662 
5663 /**
5664  * Allocate a counter.
5665  *
5666  * @param[in] dev
5667  *   Pointer to Ethernet device structure.
5668  *
5669  * @return
5670  *   Index to allocated counter  on success, 0 otherwise.
5671  */
5672 uint32_t
5673 mlx5_counter_alloc(struct rte_eth_dev *dev)
5674 {
5675 	const struct mlx5_flow_driver_ops *fops;
5676 	struct rte_flow_attr attr = { .transfer = 0 };
5677 
5678 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5679 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5680 		return fops->counter_alloc(dev);
5681 	}
5682 	DRV_LOG(ERR,
5683 		"port %u counter allocate is not supported.",
5684 		 dev->data->port_id);
5685 	return 0;
5686 }
5687 
5688 /**
5689  * Free a counter.
5690  *
5691  * @param[in] dev
5692  *   Pointer to Ethernet device structure.
5693  * @param[in] cnt
5694  *   Index to counter to be free.
5695  */
5696 void
5697 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5698 {
5699 	const struct mlx5_flow_driver_ops *fops;
5700 	struct rte_flow_attr attr = { .transfer = 0 };
5701 
5702 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5703 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5704 		fops->counter_free(dev, cnt);
5705 		return;
5706 	}
5707 	DRV_LOG(ERR,
5708 		"port %u counter free is not supported.",
5709 		 dev->data->port_id);
5710 }
5711 
5712 /**
5713  * Query counter statistics.
5714  *
5715  * @param[in] dev
5716  *   Pointer to Ethernet device structure.
5717  * @param[in] cnt
5718  *   Index to counter to query.
5719  * @param[in] clear
5720  *   Set to clear counter statistics.
5721  * @param[out] pkts
5722  *   The counter hits packets number to save.
5723  * @param[out] bytes
5724  *   The counter hits bytes number to save.
5725  *
5726  * @return
5727  *   0 on success, a negative errno value otherwise.
5728  */
5729 int
5730 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5731 		   bool clear, uint64_t *pkts, uint64_t *bytes)
5732 {
5733 	const struct mlx5_flow_driver_ops *fops;
5734 	struct rte_flow_attr attr = { .transfer = 0 };
5735 
5736 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5737 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5738 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
5739 	}
5740 	DRV_LOG(ERR,
5741 		"port %u counter query is not supported.",
5742 		 dev->data->port_id);
5743 	return -ENOTSUP;
5744 }
5745 
5746 #define MLX5_POOL_QUERY_FREQ_US 1000000
5747 
5748 /**
5749  * Set the periodic procedure for triggering asynchronous batch queries for all
5750  * the counter pools.
5751  *
5752  * @param[in] sh
5753  *   Pointer to mlx5_ibv_shared object.
5754  */
5755 void
5756 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5757 {
5758 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5759 	uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5760 	uint32_t us;
5761 
5762 	cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5763 	pools_n += rte_atomic16_read(&cont->n_valid);
5764 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5765 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5766 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5767 		sh->cmng.query_thread_on = 0;
5768 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
5769 	} else {
5770 		sh->cmng.query_thread_on = 1;
5771 	}
5772 }
5773 
5774 /**
5775  * The periodic procedure for triggering asynchronous batch queries for all the
5776  * counter pools. This function is probably called by the host thread.
5777  *
5778  * @param[in] arg
5779  *   The parameter for the alarm process.
5780  */
5781 void
5782 mlx5_flow_query_alarm(void *arg)
5783 {
5784 	struct mlx5_ibv_shared *sh = arg;
5785 	struct mlx5_devx_obj *dcs;
5786 	uint16_t offset;
5787 	int ret;
5788 	uint8_t batch = sh->cmng.batch;
5789 	uint16_t pool_index = sh->cmng.pool_index;
5790 	struct mlx5_pools_container *cont;
5791 	struct mlx5_pools_container *mcont;
5792 	struct mlx5_flow_counter_pool *pool;
5793 
5794 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5795 		goto set_alarm;
5796 next_container:
5797 	cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5798 	mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5799 	/* Check if resize was done and need to flip a container. */
5800 	if (cont != mcont) {
5801 		if (cont->pools) {
5802 			/* Clean the old container. */
5803 			rte_free(cont->pools);
5804 			memset(cont, 0, sizeof(*cont));
5805 		}
5806 		rte_cio_wmb();
5807 		 /* Flip the host container. */
5808 		sh->cmng.mhi[batch] ^= (uint8_t)2;
5809 		cont = mcont;
5810 	}
5811 	if (!cont->pools) {
5812 		/* 2 empty containers case is unexpected. */
5813 		if (unlikely(batch != sh->cmng.batch))
5814 			goto set_alarm;
5815 		batch ^= 0x1;
5816 		pool_index = 0;
5817 		goto next_container;
5818 	}
5819 	pool = cont->pools[pool_index];
5820 	if (pool->raw_hw)
5821 		/* There is a pool query in progress. */
5822 		goto set_alarm;
5823 	pool->raw_hw =
5824 		LIST_FIRST(&sh->cmng.free_stat_raws);
5825 	if (!pool->raw_hw)
5826 		/* No free counter statistics raw memory. */
5827 		goto set_alarm;
5828 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5829 							      (&pool->a64_dcs);
5830 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5831 	/*
5832 	 * Identify the counters released between query trigger and query
5833 	 * handle more effiecntly. The counter released in this gap period
5834 	 * should wait for a new round of query as the new arrived packets
5835 	 * will not be taken into account.
5836 	 */
5837 	rte_atomic64_add(&pool->start_query_gen, 1);
5838 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5839 					       offset, NULL, NULL,
5840 					       pool->raw_hw->mem_mng->dm->id,
5841 					       (void *)(uintptr_t)
5842 					       (pool->raw_hw->data + offset),
5843 					       sh->devx_comp,
5844 					       (uint64_t)(uintptr_t)pool);
5845 	if (ret) {
5846 		rte_atomic64_sub(&pool->start_query_gen, 1);
5847 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5848 			" %d", pool->min_dcs->id);
5849 		pool->raw_hw = NULL;
5850 		goto set_alarm;
5851 	}
5852 	pool->raw_hw->min_dcs_id = dcs->id;
5853 	LIST_REMOVE(pool->raw_hw, next);
5854 	sh->cmng.pending_queries++;
5855 	pool_index++;
5856 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5857 		batch ^= 0x1;
5858 		pool_index = 0;
5859 	}
5860 set_alarm:
5861 	sh->cmng.batch = batch;
5862 	sh->cmng.pool_index = pool_index;
5863 	mlx5_set_query_alarm(sh);
5864 }
5865 
5866 /**
5867  * Handler for the HW respond about ready values from an asynchronous batch
5868  * query. This function is probably called by the host thread.
5869  *
5870  * @param[in] sh
5871  *   The pointer to the shared IB device context.
5872  * @param[in] async_id
5873  *   The Devx async ID.
5874  * @param[in] status
5875  *   The status of the completion.
5876  */
5877 void
5878 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5879 				  uint64_t async_id, int status)
5880 {
5881 	struct mlx5_flow_counter_pool *pool =
5882 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5883 	struct mlx5_counter_stats_raw *raw_to_free;
5884 
5885 	if (unlikely(status)) {
5886 		rte_atomic64_sub(&pool->start_query_gen, 1);
5887 		raw_to_free = pool->raw_hw;
5888 	} else {
5889 		raw_to_free = pool->raw;
5890 		rte_spinlock_lock(&pool->sl);
5891 		pool->raw = pool->raw_hw;
5892 		rte_spinlock_unlock(&pool->sl);
5893 		MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5894 			    rte_atomic64_read(&pool->start_query_gen));
5895 		rte_atomic64_set(&pool->end_query_gen,
5896 				 rte_atomic64_read(&pool->start_query_gen));
5897 		/* Be sure the new raw counters data is updated in memory. */
5898 		rte_cio_wmb();
5899 	}
5900 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5901 	pool->raw_hw = NULL;
5902 	sh->cmng.pending_queries--;
5903 }
5904 
5905 /**
5906  * Translate the rte_flow group index to HW table value.
5907  *
5908  * @param[in] attributes
5909  *   Pointer to flow attributes
5910  * @param[in] external
5911  *   Value is part of flow rule created by request external to PMD.
5912  * @param[in] group
5913  *   rte_flow group index value.
5914  * @param[out] fdb_def_rule
5915  *   Whether fdb jump to table 1 is configured.
5916  * @param[out] table
5917  *   HW table value.
5918  * @param[out] error
5919  *   Pointer to error structure.
5920  *
5921  * @return
5922  *   0 on success, a negative errno value otherwise and rte_errno is set.
5923  */
5924 int
5925 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5926 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
5927 			 struct rte_flow_error *error)
5928 {
5929 	if (attributes->transfer && external && fdb_def_rule) {
5930 		if (group == UINT32_MAX)
5931 			return rte_flow_error_set
5932 						(error, EINVAL,
5933 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5934 						 NULL,
5935 						 "group index not supported");
5936 		*table = group + 1;
5937 	} else {
5938 		*table = group;
5939 	}
5940 	return 0;
5941 }
5942 
5943 /**
5944  * Discover availability of metadata reg_c's.
5945  *
5946  * Iteratively use test flows to check availability.
5947  *
5948  * @param[in] dev
5949  *   Pointer to the Ethernet device structure.
5950  *
5951  * @return
5952  *   0 on success, a negative errno value otherwise and rte_errno is set.
5953  */
5954 int
5955 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5956 {
5957 	struct mlx5_priv *priv = dev->data->dev_private;
5958 	struct mlx5_dev_config *config = &priv->config;
5959 	enum modify_reg idx;
5960 	int n = 0;
5961 
5962 	/* reg_c[0] and reg_c[1] are reserved. */
5963 	config->flow_mreg_c[n++] = REG_C_0;
5964 	config->flow_mreg_c[n++] = REG_C_1;
5965 	/* Discover availability of other reg_c's. */
5966 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5967 		struct rte_flow_attr attr = {
5968 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5969 			.priority = MLX5_FLOW_PRIO_RSVD,
5970 			.ingress = 1,
5971 		};
5972 		struct rte_flow_item items[] = {
5973 			[0] = {
5974 				.type = RTE_FLOW_ITEM_TYPE_END,
5975 			},
5976 		};
5977 		struct rte_flow_action actions[] = {
5978 			[0] = {
5979 				.type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5980 				.conf = &(struct mlx5_flow_action_copy_mreg){
5981 					.src = REG_C_1,
5982 					.dst = idx,
5983 				},
5984 			},
5985 			[1] = {
5986 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
5987 				.conf = &(struct rte_flow_action_jump){
5988 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5989 				},
5990 			},
5991 			[2] = {
5992 				.type = RTE_FLOW_ACTION_TYPE_END,
5993 			},
5994 		};
5995 		uint32_t flow_idx;
5996 		struct rte_flow *flow;
5997 		struct rte_flow_error error;
5998 
5999 		if (!config->dv_flow_en)
6000 			break;
6001 		/* Create internal flow, validation skips copy action. */
6002 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6003 					    actions, false, &error);
6004 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6005 				      flow_idx);
6006 		if (!flow)
6007 			continue;
6008 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6009 			config->flow_mreg_c[n++] = idx;
6010 		flow_list_destroy(dev, NULL, flow_idx);
6011 	}
6012 	for (; n < MLX5_MREG_C_NUM; ++n)
6013 		config->flow_mreg_c[n] = REG_NONE;
6014 	return 0;
6015 }
6016 
6017 /**
6018  * Dump flow raw hw data to file
6019  *
6020  * @param[in] dev
6021  *    The pointer to Ethernet device.
6022  * @param[in] file
6023  *   A pointer to a file for output.
6024  * @param[out] error
6025  *   Perform verbose error reporting if not NULL. PMDs initialize this
6026  *   structure in case of error only.
6027  * @return
6028  *   0 on success, a nagative value otherwise.
6029  */
6030 int
6031 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6032 		   FILE *file,
6033 		   struct rte_flow_error *error __rte_unused)
6034 {
6035 	struct mlx5_priv *priv = dev->data->dev_private;
6036 	struct mlx5_ibv_shared *sh = priv->sh;
6037 
6038 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6039 				       sh->tx_domain, file);
6040 }
6041