xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b4edeaf3efd51e677e5c10cc30a07c250a739316)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rx.h"
32 #include "mlx5_tx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 struct tunnel_default_miss_ctx {
37 	uint16_t *queue;
38 	__extension__
39 	union {
40 		struct rte_flow_action_rss action_rss;
41 		struct rte_flow_action_queue miss_queue;
42 		struct rte_flow_action_jump miss_jump;
43 		uint8_t raw[0];
44 	};
45 };
46 
47 static int
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 			     struct rte_flow *flow,
50 			     const struct rte_flow_attr *attr,
51 			     const struct rte_flow_action *app_actions,
52 			     uint32_t flow_idx,
53 			     const struct mlx5_flow_tunnel *tunnel,
54 			     struct tunnel_default_miss_ctx *ctx,
55 			     struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
58 static void
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
60 static uint32_t
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 				const struct mlx5_flow_tunnel *tunnel,
63 				uint32_t group, uint32_t *table,
64 				struct rte_flow_error *error);
65 
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
68 
69 
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
72 
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
74 
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
79 #endif
80 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
82 };
83 
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
86 	(const int []){ \
87 		__VA_ARGS__, 0, \
88 	}
89 
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 	const int *const next;
93 	/**<
94 	 * List of next node indexes. Index 0 is interpreted as a terminator.
95 	 */
96 	const enum rte_flow_item_type type;
97 	/**< Pattern item type of current node. */
98 	uint64_t rss_types;
99 	/**<
100 	 * RSS types bit-field associated with this node
101 	 * (see ETH_RSS_* definitions).
102 	 */
103 	uint8_t optional;
104 	/**< optional expand field. Default 0 to expand, 1 not go deeper. */
105 };
106 
107 /** Object returned by mlx5_flow_expand_rss(). */
108 struct mlx5_flow_expand_rss {
109 	uint32_t entries;
110 	/**< Number of entries @p patterns and @p priorities. */
111 	struct {
112 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
113 		uint32_t priority; /**< Priority offset for each expansion. */
114 	} entry[];
115 };
116 
117 static void
118 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
119 
120 static bool
121 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
122 {
123 	switch (item->type) {
124 	case RTE_FLOW_ITEM_TYPE_ETH:
125 	case RTE_FLOW_ITEM_TYPE_VLAN:
126 	case RTE_FLOW_ITEM_TYPE_IPV4:
127 	case RTE_FLOW_ITEM_TYPE_IPV6:
128 	case RTE_FLOW_ITEM_TYPE_UDP:
129 	case RTE_FLOW_ITEM_TYPE_TCP:
130 	case RTE_FLOW_ITEM_TYPE_VXLAN:
131 	case RTE_FLOW_ITEM_TYPE_NVGRE:
132 	case RTE_FLOW_ITEM_TYPE_GRE:
133 	case RTE_FLOW_ITEM_TYPE_GENEVE:
134 	case RTE_FLOW_ITEM_TYPE_MPLS:
135 		return true;
136 	default:
137 		break;
138 	}
139 	return false;
140 }
141 
142 static enum rte_flow_item_type
143 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
144 {
145 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
146 	uint16_t ether_type = 0;
147 	uint16_t ether_type_m;
148 	uint8_t ip_next_proto = 0;
149 	uint8_t ip_next_proto_m;
150 
151 	if (item == NULL || item->spec == NULL)
152 		return ret;
153 	switch (item->type) {
154 	case RTE_FLOW_ITEM_TYPE_ETH:
155 		if (item->mask)
156 			ether_type_m = ((const struct rte_flow_item_eth *)
157 						(item->mask))->type;
158 		else
159 			ether_type_m = rte_flow_item_eth_mask.type;
160 		if (ether_type_m != RTE_BE16(0xFFFF))
161 			break;
162 		ether_type = ((const struct rte_flow_item_eth *)
163 				(item->spec))->type;
164 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
165 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
166 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
167 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
168 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
169 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
170 		else
171 			ret = RTE_FLOW_ITEM_TYPE_END;
172 		break;
173 	case RTE_FLOW_ITEM_TYPE_VLAN:
174 		if (item->mask)
175 			ether_type_m = ((const struct rte_flow_item_vlan *)
176 						(item->mask))->inner_type;
177 		else
178 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
179 		if (ether_type_m != RTE_BE16(0xFFFF))
180 			break;
181 		ether_type = ((const struct rte_flow_item_vlan *)
182 				(item->spec))->inner_type;
183 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
184 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
185 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
186 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
187 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
188 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
189 		else
190 			ret = RTE_FLOW_ITEM_TYPE_END;
191 		break;
192 	case RTE_FLOW_ITEM_TYPE_IPV4:
193 		if (item->mask)
194 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
195 					(item->mask))->hdr.next_proto_id;
196 		else
197 			ip_next_proto_m =
198 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
199 		if (ip_next_proto_m != 0xFF)
200 			break;
201 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
202 				(item->spec))->hdr.next_proto_id;
203 		if (ip_next_proto == IPPROTO_UDP)
204 			ret = RTE_FLOW_ITEM_TYPE_UDP;
205 		else if (ip_next_proto == IPPROTO_TCP)
206 			ret = RTE_FLOW_ITEM_TYPE_TCP;
207 		else if (ip_next_proto == IPPROTO_IP)
208 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
209 		else if (ip_next_proto == IPPROTO_IPV6)
210 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
211 		else
212 			ret = RTE_FLOW_ITEM_TYPE_END;
213 		break;
214 	case RTE_FLOW_ITEM_TYPE_IPV6:
215 		if (item->mask)
216 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
217 						(item->mask))->hdr.proto;
218 		else
219 			ip_next_proto_m =
220 				rte_flow_item_ipv6_mask.hdr.proto;
221 		if (ip_next_proto_m != 0xFF)
222 			break;
223 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
224 				(item->spec))->hdr.proto;
225 		if (ip_next_proto == IPPROTO_UDP)
226 			ret = RTE_FLOW_ITEM_TYPE_UDP;
227 		else if (ip_next_proto == IPPROTO_TCP)
228 			ret = RTE_FLOW_ITEM_TYPE_TCP;
229 		else if (ip_next_proto == IPPROTO_IP)
230 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
231 		else if (ip_next_proto == IPPROTO_IPV6)
232 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
233 		else
234 			ret = RTE_FLOW_ITEM_TYPE_END;
235 		break;
236 	default:
237 		ret = RTE_FLOW_ITEM_TYPE_VOID;
238 		break;
239 	}
240 	return ret;
241 }
242 
243 #define MLX5_RSS_EXP_ELT_N 16
244 
245 /**
246  * Expand RSS flows into several possible flows according to the RSS hash
247  * fields requested and the driver capabilities.
248  *
249  * @param[out] buf
250  *   Buffer to store the result expansion.
251  * @param[in] size
252  *   Buffer size in bytes. If 0, @p buf can be NULL.
253  * @param[in] pattern
254  *   User flow pattern.
255  * @param[in] types
256  *   RSS types to expand (see ETH_RSS_* definitions).
257  * @param[in] graph
258  *   Input graph to expand @p pattern according to @p types.
259  * @param[in] graph_root_index
260  *   Index of root node in @p graph, typically 0.
261  *
262  * @return
263  *   A positive value representing the size of @p buf in bytes regardless of
264  *   @p size on success, a negative errno value otherwise and rte_errno is
265  *   set, the following errors are defined:
266  *
267  *   -E2BIG: graph-depth @p graph is too deep.
268  *   -EINVAL: @p size has not enough space for expanded pattern.
269  */
270 static int
271 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
272 		     const struct rte_flow_item *pattern, uint64_t types,
273 		     const struct mlx5_flow_expand_node graph[],
274 		     int graph_root_index)
275 {
276 	const struct rte_flow_item *item;
277 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
278 	const int *next_node;
279 	const int *stack[MLX5_RSS_EXP_ELT_N];
280 	int stack_pos = 0;
281 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
282 	unsigned int i;
283 	size_t lsize;
284 	size_t user_pattern_size = 0;
285 	void *addr = NULL;
286 	const struct mlx5_flow_expand_node *next = NULL;
287 	struct rte_flow_item missed_item;
288 	int missed = 0;
289 	int elt = 0;
290 	const struct rte_flow_item *last_item = NULL;
291 
292 	memset(&missed_item, 0, sizeof(missed_item));
293 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
294 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
295 	if (lsize > size)
296 		return -EINVAL;
297 	buf->entry[0].priority = 0;
298 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
299 	buf->entries = 0;
300 	addr = buf->entry[0].pattern;
301 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
302 		if (!mlx5_flow_is_rss_expandable_item(item)) {
303 			user_pattern_size += sizeof(*item);
304 			continue;
305 		}
306 		last_item = item;
307 		for (i = 0; node->next && node->next[i]; ++i) {
308 			next = &graph[node->next[i]];
309 			if (next->type == item->type)
310 				break;
311 		}
312 		if (next)
313 			node = next;
314 		user_pattern_size += sizeof(*item);
315 	}
316 	user_pattern_size += sizeof(*item); /* Handle END item. */
317 	lsize += user_pattern_size;
318 	if (lsize > size)
319 		return -EINVAL;
320 	/* Copy the user pattern in the first entry of the buffer. */
321 	rte_memcpy(addr, pattern, user_pattern_size);
322 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
323 	buf->entries = 1;
324 	/* Start expanding. */
325 	memset(flow_items, 0, sizeof(flow_items));
326 	user_pattern_size -= sizeof(*item);
327 	/*
328 	 * Check if the last valid item has spec set, need complete pattern,
329 	 * and the pattern can be used for expansion.
330 	 */
331 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
332 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
333 		/* Item type END indicates expansion is not required. */
334 		return lsize;
335 	}
336 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
337 		next = NULL;
338 		missed = 1;
339 		for (i = 0; node->next && node->next[i]; ++i) {
340 			next = &graph[node->next[i]];
341 			if (next->type == missed_item.type) {
342 				flow_items[0].type = missed_item.type;
343 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
344 				break;
345 			}
346 			next = NULL;
347 		}
348 	}
349 	if (next && missed) {
350 		elt = 2; /* missed item + item end. */
351 		node = next;
352 		lsize += elt * sizeof(*item) + user_pattern_size;
353 		if (lsize > size)
354 			return -EINVAL;
355 		if (node->rss_types & types) {
356 			buf->entry[buf->entries].priority = 1;
357 			buf->entry[buf->entries].pattern = addr;
358 			buf->entries++;
359 			rte_memcpy(addr, buf->entry[0].pattern,
360 				   user_pattern_size);
361 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
362 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
363 			addr = (void *)(((uintptr_t)addr) +
364 					elt * sizeof(*item));
365 		}
366 	}
367 	memset(flow_items, 0, sizeof(flow_items));
368 	next_node = node->next;
369 	stack[stack_pos] = next_node;
370 	node = next_node ? &graph[*next_node] : NULL;
371 	while (node) {
372 		flow_items[stack_pos].type = node->type;
373 		if (node->rss_types & types) {
374 			size_t n;
375 			/*
376 			 * compute the number of items to copy from the
377 			 * expansion and copy it.
378 			 * When the stack_pos is 0, there are 1 element in it,
379 			 * plus the addition END item.
380 			 */
381 			elt = stack_pos + 2;
382 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
383 			lsize += elt * sizeof(*item) + user_pattern_size;
384 			if (lsize > size)
385 				return -EINVAL;
386 			n = elt * sizeof(*item);
387 			buf->entry[buf->entries].priority =
388 				stack_pos + 1 + missed;
389 			buf->entry[buf->entries].pattern = addr;
390 			buf->entries++;
391 			rte_memcpy(addr, buf->entry[0].pattern,
392 				   user_pattern_size);
393 			addr = (void *)(((uintptr_t)addr) +
394 					user_pattern_size);
395 			rte_memcpy(addr, &missed_item,
396 				   missed * sizeof(*item));
397 			addr = (void *)(((uintptr_t)addr) +
398 				missed * sizeof(*item));
399 			rte_memcpy(addr, flow_items, n);
400 			addr = (void *)(((uintptr_t)addr) + n);
401 		}
402 		/* Go deeper. */
403 		if (!node->optional && node->next) {
404 			next_node = node->next;
405 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
406 				rte_errno = E2BIG;
407 				return -rte_errno;
408 			}
409 			stack[stack_pos] = next_node;
410 		} else if (*(next_node + 1)) {
411 			/* Follow up with the next possibility. */
412 			++next_node;
413 		} else {
414 			/* Move to the next path. */
415 			if (stack_pos)
416 				next_node = stack[--stack_pos];
417 			next_node++;
418 			stack[stack_pos] = next_node;
419 		}
420 		node = *next_node ? &graph[*next_node] : NULL;
421 	};
422 	return lsize;
423 }
424 
425 enum mlx5_expansion {
426 	MLX5_EXPANSION_ROOT,
427 	MLX5_EXPANSION_ROOT_OUTER,
428 	MLX5_EXPANSION_ROOT_ETH_VLAN,
429 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
430 	MLX5_EXPANSION_OUTER_ETH,
431 	MLX5_EXPANSION_OUTER_ETH_VLAN,
432 	MLX5_EXPANSION_OUTER_VLAN,
433 	MLX5_EXPANSION_OUTER_IPV4,
434 	MLX5_EXPANSION_OUTER_IPV4_UDP,
435 	MLX5_EXPANSION_OUTER_IPV4_TCP,
436 	MLX5_EXPANSION_OUTER_IPV6,
437 	MLX5_EXPANSION_OUTER_IPV6_UDP,
438 	MLX5_EXPANSION_OUTER_IPV6_TCP,
439 	MLX5_EXPANSION_VXLAN,
440 	MLX5_EXPANSION_VXLAN_GPE,
441 	MLX5_EXPANSION_GRE,
442 	MLX5_EXPANSION_NVGRE,
443 	MLX5_EXPANSION_GRE_KEY,
444 	MLX5_EXPANSION_MPLS,
445 	MLX5_EXPANSION_ETH,
446 	MLX5_EXPANSION_ETH_VLAN,
447 	MLX5_EXPANSION_VLAN,
448 	MLX5_EXPANSION_IPV4,
449 	MLX5_EXPANSION_IPV4_UDP,
450 	MLX5_EXPANSION_IPV4_TCP,
451 	MLX5_EXPANSION_IPV6,
452 	MLX5_EXPANSION_IPV6_UDP,
453 	MLX5_EXPANSION_IPV6_TCP,
454 };
455 
456 /** Supported expansion of items. */
457 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
458 	[MLX5_EXPANSION_ROOT] = {
459 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
460 						  MLX5_EXPANSION_IPV4,
461 						  MLX5_EXPANSION_IPV6),
462 		.type = RTE_FLOW_ITEM_TYPE_END,
463 	},
464 	[MLX5_EXPANSION_ROOT_OUTER] = {
465 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
466 						  MLX5_EXPANSION_OUTER_IPV4,
467 						  MLX5_EXPANSION_OUTER_IPV6),
468 		.type = RTE_FLOW_ITEM_TYPE_END,
469 	},
470 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
471 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
472 		.type = RTE_FLOW_ITEM_TYPE_END,
473 	},
474 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
475 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
476 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
477 		.type = RTE_FLOW_ITEM_TYPE_END,
478 	},
479 	[MLX5_EXPANSION_OUTER_ETH] = {
480 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
481 						  MLX5_EXPANSION_OUTER_IPV6),
482 		.type = RTE_FLOW_ITEM_TYPE_ETH,
483 		.rss_types = 0,
484 	},
485 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
486 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
487 		.type = RTE_FLOW_ITEM_TYPE_ETH,
488 		.rss_types = 0,
489 	},
490 	[MLX5_EXPANSION_OUTER_VLAN] = {
491 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
492 						  MLX5_EXPANSION_OUTER_IPV6),
493 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
494 	},
495 	[MLX5_EXPANSION_OUTER_IPV4] = {
496 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
497 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
498 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
499 			 MLX5_EXPANSION_GRE,
500 			 MLX5_EXPANSION_NVGRE,
501 			 MLX5_EXPANSION_IPV4,
502 			 MLX5_EXPANSION_IPV6),
503 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
504 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
505 			ETH_RSS_NONFRAG_IPV4_OTHER,
506 	},
507 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
508 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 						  MLX5_EXPANSION_VXLAN_GPE,
510 						  MLX5_EXPANSION_MPLS),
511 		.type = RTE_FLOW_ITEM_TYPE_UDP,
512 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
513 	},
514 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
515 		.type = RTE_FLOW_ITEM_TYPE_TCP,
516 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
517 	},
518 	[MLX5_EXPANSION_OUTER_IPV6] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
520 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
521 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
522 			 MLX5_EXPANSION_IPV4,
523 			 MLX5_EXPANSION_IPV6,
524 			 MLX5_EXPANSION_GRE,
525 			 MLX5_EXPANSION_NVGRE),
526 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
527 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
528 			ETH_RSS_NONFRAG_IPV6_OTHER,
529 	},
530 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
531 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
532 						  MLX5_EXPANSION_VXLAN_GPE,
533 						  MLX5_EXPANSION_MPLS),
534 		.type = RTE_FLOW_ITEM_TYPE_UDP,
535 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
536 	},
537 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
538 		.type = RTE_FLOW_ITEM_TYPE_TCP,
539 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
540 	},
541 	[MLX5_EXPANSION_VXLAN] = {
542 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
543 						  MLX5_EXPANSION_IPV4,
544 						  MLX5_EXPANSION_IPV6),
545 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
546 	},
547 	[MLX5_EXPANSION_VXLAN_GPE] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
549 						  MLX5_EXPANSION_IPV4,
550 						  MLX5_EXPANSION_IPV6),
551 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
552 	},
553 	[MLX5_EXPANSION_GRE] = {
554 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
555 						  MLX5_EXPANSION_IPV6,
556 						  MLX5_EXPANSION_GRE_KEY,
557 						  MLX5_EXPANSION_MPLS),
558 		.type = RTE_FLOW_ITEM_TYPE_GRE,
559 	},
560 	[MLX5_EXPANSION_GRE_KEY] = {
561 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
562 						  MLX5_EXPANSION_IPV6,
563 						  MLX5_EXPANSION_MPLS),
564 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
565 		.optional = 1,
566 	},
567 	[MLX5_EXPANSION_NVGRE] = {
568 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
569 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
570 	},
571 	[MLX5_EXPANSION_MPLS] = {
572 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
573 						  MLX5_EXPANSION_IPV6,
574 						  MLX5_EXPANSION_ETH),
575 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
576 	},
577 	[MLX5_EXPANSION_ETH] = {
578 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
579 						  MLX5_EXPANSION_IPV6),
580 		.type = RTE_FLOW_ITEM_TYPE_ETH,
581 	},
582 	[MLX5_EXPANSION_ETH_VLAN] = {
583 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
584 		.type = RTE_FLOW_ITEM_TYPE_ETH,
585 	},
586 	[MLX5_EXPANSION_VLAN] = {
587 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
588 						  MLX5_EXPANSION_IPV6),
589 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
590 	},
591 	[MLX5_EXPANSION_IPV4] = {
592 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
593 						  MLX5_EXPANSION_IPV4_TCP),
594 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
595 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
596 			ETH_RSS_NONFRAG_IPV4_OTHER,
597 	},
598 	[MLX5_EXPANSION_IPV4_UDP] = {
599 		.type = RTE_FLOW_ITEM_TYPE_UDP,
600 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
601 	},
602 	[MLX5_EXPANSION_IPV4_TCP] = {
603 		.type = RTE_FLOW_ITEM_TYPE_TCP,
604 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
605 	},
606 	[MLX5_EXPANSION_IPV6] = {
607 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
608 						  MLX5_EXPANSION_IPV6_TCP),
609 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
610 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
611 			ETH_RSS_NONFRAG_IPV6_OTHER,
612 	},
613 	[MLX5_EXPANSION_IPV6_UDP] = {
614 		.type = RTE_FLOW_ITEM_TYPE_UDP,
615 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
616 	},
617 	[MLX5_EXPANSION_IPV6_TCP] = {
618 		.type = RTE_FLOW_ITEM_TYPE_TCP,
619 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
620 	},
621 };
622 
623 static struct rte_flow_action_handle *
624 mlx5_action_handle_create(struct rte_eth_dev *dev,
625 			  const struct rte_flow_indir_action_conf *conf,
626 			  const struct rte_flow_action *action,
627 			  struct rte_flow_error *error);
628 static int mlx5_action_handle_destroy
629 				(struct rte_eth_dev *dev,
630 				 struct rte_flow_action_handle *handle,
631 				 struct rte_flow_error *error);
632 static int mlx5_action_handle_update
633 				(struct rte_eth_dev *dev,
634 				 struct rte_flow_action_handle *handle,
635 				 const void *update,
636 				 struct rte_flow_error *error);
637 static int mlx5_action_handle_query
638 				(struct rte_eth_dev *dev,
639 				 const struct rte_flow_action_handle *handle,
640 				 void *data,
641 				 struct rte_flow_error *error);
642 static int
643 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
644 		    struct rte_flow_tunnel *app_tunnel,
645 		    struct rte_flow_action **actions,
646 		    uint32_t *num_of_actions,
647 		    struct rte_flow_error *error);
648 static int
649 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
650 		       struct rte_flow_tunnel *app_tunnel,
651 		       struct rte_flow_item **items,
652 		       uint32_t *num_of_items,
653 		       struct rte_flow_error *error);
654 static int
655 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
656 			      struct rte_flow_item *pmd_items,
657 			      uint32_t num_items, struct rte_flow_error *err);
658 static int
659 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
660 				struct rte_flow_action *pmd_actions,
661 				uint32_t num_actions,
662 				struct rte_flow_error *err);
663 static int
664 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
665 				  struct rte_mbuf *m,
666 				  struct rte_flow_restore_info *info,
667 				  struct rte_flow_error *err);
668 
669 static const struct rte_flow_ops mlx5_flow_ops = {
670 	.validate = mlx5_flow_validate,
671 	.create = mlx5_flow_create,
672 	.destroy = mlx5_flow_destroy,
673 	.flush = mlx5_flow_flush,
674 	.isolate = mlx5_flow_isolate,
675 	.query = mlx5_flow_query,
676 	.dev_dump = mlx5_flow_dev_dump,
677 	.get_aged_flows = mlx5_flow_get_aged_flows,
678 	.action_handle_create = mlx5_action_handle_create,
679 	.action_handle_destroy = mlx5_action_handle_destroy,
680 	.action_handle_update = mlx5_action_handle_update,
681 	.action_handle_query = mlx5_action_handle_query,
682 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
683 	.tunnel_match = mlx5_flow_tunnel_match,
684 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
685 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
686 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
687 };
688 
689 /* Tunnel information. */
690 struct mlx5_flow_tunnel_info {
691 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
692 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
693 };
694 
695 static struct mlx5_flow_tunnel_info tunnels_info[] = {
696 	{
697 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
698 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
699 	},
700 	{
701 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
702 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
703 	},
704 	{
705 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
706 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
707 	},
708 	{
709 		.tunnel = MLX5_FLOW_LAYER_GRE,
710 		.ptype = RTE_PTYPE_TUNNEL_GRE,
711 	},
712 	{
713 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
714 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
715 	},
716 	{
717 		.tunnel = MLX5_FLOW_LAYER_MPLS,
718 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
719 	},
720 	{
721 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
722 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
723 	},
724 	{
725 		.tunnel = MLX5_FLOW_LAYER_IPIP,
726 		.ptype = RTE_PTYPE_TUNNEL_IP,
727 	},
728 	{
729 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
730 		.ptype = RTE_PTYPE_TUNNEL_IP,
731 	},
732 	{
733 		.tunnel = MLX5_FLOW_LAYER_GTP,
734 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
735 	},
736 };
737 
738 
739 
740 /**
741  * Translate tag ID to register.
742  *
743  * @param[in] dev
744  *   Pointer to the Ethernet device structure.
745  * @param[in] feature
746  *   The feature that request the register.
747  * @param[in] id
748  *   The request register ID.
749  * @param[out] error
750  *   Error description in case of any.
751  *
752  * @return
753  *   The request register on success, a negative errno
754  *   value otherwise and rte_errno is set.
755  */
756 int
757 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
758 		     enum mlx5_feature_name feature,
759 		     uint32_t id,
760 		     struct rte_flow_error *error)
761 {
762 	struct mlx5_priv *priv = dev->data->dev_private;
763 	struct mlx5_dev_config *config = &priv->config;
764 	enum modify_reg start_reg;
765 	bool skip_mtr_reg = false;
766 
767 	switch (feature) {
768 	case MLX5_HAIRPIN_RX:
769 		return REG_B;
770 	case MLX5_HAIRPIN_TX:
771 		return REG_A;
772 	case MLX5_METADATA_RX:
773 		switch (config->dv_xmeta_en) {
774 		case MLX5_XMETA_MODE_LEGACY:
775 			return REG_B;
776 		case MLX5_XMETA_MODE_META16:
777 			return REG_C_0;
778 		case MLX5_XMETA_MODE_META32:
779 			return REG_C_1;
780 		}
781 		break;
782 	case MLX5_METADATA_TX:
783 		return REG_A;
784 	case MLX5_METADATA_FDB:
785 		switch (config->dv_xmeta_en) {
786 		case MLX5_XMETA_MODE_LEGACY:
787 			return REG_NON;
788 		case MLX5_XMETA_MODE_META16:
789 			return REG_C_0;
790 		case MLX5_XMETA_MODE_META32:
791 			return REG_C_1;
792 		}
793 		break;
794 	case MLX5_FLOW_MARK:
795 		switch (config->dv_xmeta_en) {
796 		case MLX5_XMETA_MODE_LEGACY:
797 			return REG_NON;
798 		case MLX5_XMETA_MODE_META16:
799 			return REG_C_1;
800 		case MLX5_XMETA_MODE_META32:
801 			return REG_C_0;
802 		}
803 		break;
804 	case MLX5_MTR_ID:
805 		/*
806 		 * If meter color and meter id share one register, flow match
807 		 * should use the meter color register for match.
808 		 */
809 		if (priv->mtr_reg_share)
810 			return priv->mtr_color_reg;
811 		else
812 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
813 			       REG_C_3;
814 	case MLX5_MTR_COLOR:
815 	case MLX5_ASO_FLOW_HIT:
816 	case MLX5_ASO_CONNTRACK:
817 		/* All features use the same REG_C. */
818 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
819 		return priv->mtr_color_reg;
820 	case MLX5_COPY_MARK:
821 		/*
822 		 * Metadata COPY_MARK register using is in meter suffix sub
823 		 * flow while with meter. It's safe to share the same register.
824 		 */
825 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
826 	case MLX5_APP_TAG:
827 		/*
828 		 * If meter is enable, it will engage the register for color
829 		 * match and flow match. If meter color match is not using the
830 		 * REG_C_2, need to skip the REG_C_x be used by meter color
831 		 * match.
832 		 * If meter is disable, free to use all available registers.
833 		 */
834 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
835 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
836 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
837 		if (id > (uint32_t)(REG_C_7 - start_reg))
838 			return rte_flow_error_set(error, EINVAL,
839 						  RTE_FLOW_ERROR_TYPE_ITEM,
840 						  NULL, "invalid tag id");
841 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
842 			return rte_flow_error_set(error, ENOTSUP,
843 						  RTE_FLOW_ERROR_TYPE_ITEM,
844 						  NULL, "unsupported tag id");
845 		/*
846 		 * This case means meter is using the REG_C_x great than 2.
847 		 * Take care not to conflict with meter color REG_C_x.
848 		 * If the available index REG_C_y >= REG_C_x, skip the
849 		 * color register.
850 		 */
851 		if (skip_mtr_reg && config->flow_mreg_c
852 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
853 			if (id >= (uint32_t)(REG_C_7 - start_reg))
854 				return rte_flow_error_set(error, EINVAL,
855 						       RTE_FLOW_ERROR_TYPE_ITEM,
856 							NULL, "invalid tag id");
857 			if (config->flow_mreg_c
858 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
859 				return config->flow_mreg_c
860 					       [id + 1 + start_reg - REG_C_0];
861 			return rte_flow_error_set(error, ENOTSUP,
862 						  RTE_FLOW_ERROR_TYPE_ITEM,
863 						  NULL, "unsupported tag id");
864 		}
865 		return config->flow_mreg_c[id + start_reg - REG_C_0];
866 	}
867 	MLX5_ASSERT(false);
868 	return rte_flow_error_set(error, EINVAL,
869 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
870 				  NULL, "invalid feature name");
871 }
872 
873 /**
874  * Check extensive flow metadata register support.
875  *
876  * @param dev
877  *   Pointer to rte_eth_dev structure.
878  *
879  * @return
880  *   True if device supports extensive flow metadata register, otherwise false.
881  */
882 bool
883 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
884 {
885 	struct mlx5_priv *priv = dev->data->dev_private;
886 	struct mlx5_dev_config *config = &priv->config;
887 
888 	/*
889 	 * Having available reg_c can be regarded inclusively as supporting
890 	 * extensive flow metadata register, which could mean,
891 	 * - metadata register copy action by modify header.
892 	 * - 16 modify header actions is supported.
893 	 * - reg_c's are preserved across different domain (FDB and NIC) on
894 	 *   packet loopback by flow lookup miss.
895 	 */
896 	return config->flow_mreg_c[2] != REG_NON;
897 }
898 
899 /**
900  * Get the lowest priority.
901  *
902  * @param[in] dev
903  *   Pointer to the Ethernet device structure.
904  * @param[in] attributes
905  *   Pointer to device flow rule attributes.
906  *
907  * @return
908  *   The value of lowest priority of flow.
909  */
910 uint32_t
911 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
912 			  const struct rte_flow_attr *attr)
913 {
914 	struct mlx5_priv *priv = dev->data->dev_private;
915 
916 	if (!attr->group && !attr->transfer)
917 		return priv->config.flow_prio - 2;
918 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
919 }
920 
921 /**
922  * Calculate matcher priority of the flow.
923  *
924  * @param[in] dev
925  *   Pointer to the Ethernet device structure.
926  * @param[in] attr
927  *   Pointer to device flow rule attributes.
928  * @param[in] subpriority
929  *   The priority based on the items.
930  * @return
931  *   The matcher priority of the flow.
932  */
933 uint16_t
934 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
935 			  const struct rte_flow_attr *attr,
936 			  uint32_t subpriority)
937 {
938 	uint16_t priority = (uint16_t)attr->priority;
939 	struct mlx5_priv *priv = dev->data->dev_private;
940 
941 	if (!attr->group && !attr->transfer) {
942 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
943 			priority = priv->config.flow_prio - 1;
944 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
945 	}
946 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
947 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
948 	return priority * 3 + subpriority;
949 }
950 
951 /**
952  * Verify the @p item specifications (spec, last, mask) are compatible with the
953  * NIC capabilities.
954  *
955  * @param[in] item
956  *   Item specification.
957  * @param[in] mask
958  *   @p item->mask or flow default bit-masks.
959  * @param[in] nic_mask
960  *   Bit-masks covering supported fields by the NIC to compare with user mask.
961  * @param[in] size
962  *   Bit-masks size in bytes.
963  * @param[in] range_accepted
964  *   True if range of values is accepted for specific fields, false otherwise.
965  * @param[out] error
966  *   Pointer to error structure.
967  *
968  * @return
969  *   0 on success, a negative errno value otherwise and rte_errno is set.
970  */
971 int
972 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
973 			  const uint8_t *mask,
974 			  const uint8_t *nic_mask,
975 			  unsigned int size,
976 			  bool range_accepted,
977 			  struct rte_flow_error *error)
978 {
979 	unsigned int i;
980 
981 	MLX5_ASSERT(nic_mask);
982 	for (i = 0; i < size; ++i)
983 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
984 			return rte_flow_error_set(error, ENOTSUP,
985 						  RTE_FLOW_ERROR_TYPE_ITEM,
986 						  item,
987 						  "mask enables non supported"
988 						  " bits");
989 	if (!item->spec && (item->mask || item->last))
990 		return rte_flow_error_set(error, EINVAL,
991 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
992 					  "mask/last without a spec is not"
993 					  " supported");
994 	if (item->spec && item->last && !range_accepted) {
995 		uint8_t spec[size];
996 		uint8_t last[size];
997 		unsigned int i;
998 		int ret;
999 
1000 		for (i = 0; i < size; ++i) {
1001 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1002 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1003 		}
1004 		ret = memcmp(spec, last, size);
1005 		if (ret != 0)
1006 			return rte_flow_error_set(error, EINVAL,
1007 						  RTE_FLOW_ERROR_TYPE_ITEM,
1008 						  item,
1009 						  "range is not valid");
1010 	}
1011 	return 0;
1012 }
1013 
1014 /**
1015  * Adjust the hash fields according to the @p flow information.
1016  *
1017  * @param[in] dev_flow.
1018  *   Pointer to the mlx5_flow.
1019  * @param[in] tunnel
1020  *   1 when the hash field is for a tunnel item.
1021  * @param[in] layer_types
1022  *   ETH_RSS_* types.
1023  * @param[in] hash_fields
1024  *   Item hash fields.
1025  *
1026  * @return
1027  *   The hash fields that should be used.
1028  */
1029 uint64_t
1030 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1031 			    int tunnel __rte_unused, uint64_t layer_types,
1032 			    uint64_t hash_fields)
1033 {
1034 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1035 	int rss_request_inner = rss_desc->level >= 2;
1036 
1037 	/* Check RSS hash level for tunnel. */
1038 	if (tunnel && rss_request_inner)
1039 		hash_fields |= IBV_RX_HASH_INNER;
1040 	else if (tunnel || rss_request_inner)
1041 		return 0;
1042 #endif
1043 	/* Check if requested layer matches RSS hash fields. */
1044 	if (!(rss_desc->types & layer_types))
1045 		return 0;
1046 	return hash_fields;
1047 }
1048 
1049 /**
1050  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1051  * if several tunnel rules are used on this queue, the tunnel ptype will be
1052  * cleared.
1053  *
1054  * @param rxq_ctrl
1055  *   Rx queue to update.
1056  */
1057 static void
1058 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1059 {
1060 	unsigned int i;
1061 	uint32_t tunnel_ptype = 0;
1062 
1063 	/* Look up for the ptype to use. */
1064 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1065 		if (!rxq_ctrl->flow_tunnels_n[i])
1066 			continue;
1067 		if (!tunnel_ptype) {
1068 			tunnel_ptype = tunnels_info[i].ptype;
1069 		} else {
1070 			tunnel_ptype = 0;
1071 			break;
1072 		}
1073 	}
1074 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1075 }
1076 
1077 /**
1078  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1079  * flow.
1080  *
1081  * @param[in] dev
1082  *   Pointer to the Ethernet device structure.
1083  * @param[in] dev_handle
1084  *   Pointer to device flow handle structure.
1085  */
1086 void
1087 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1088 		       struct mlx5_flow_handle *dev_handle)
1089 {
1090 	struct mlx5_priv *priv = dev->data->dev_private;
1091 	const int mark = dev_handle->mark;
1092 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1093 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1094 	unsigned int i;
1095 
1096 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1097 		struct mlx5_hrxq *hrxq;
1098 
1099 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1100 			      dev_handle->rix_hrxq);
1101 		if (hrxq)
1102 			ind_tbl = hrxq->ind_table;
1103 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1104 		struct mlx5_shared_action_rss *shared_rss;
1105 
1106 		shared_rss = mlx5_ipool_get
1107 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1108 			 dev_handle->rix_srss);
1109 		if (shared_rss)
1110 			ind_tbl = shared_rss->ind_tbl;
1111 	}
1112 	if (!ind_tbl)
1113 		return;
1114 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1115 		int idx = ind_tbl->queues[i];
1116 		struct mlx5_rxq_ctrl *rxq_ctrl =
1117 			container_of((*priv->rxqs)[idx],
1118 				     struct mlx5_rxq_ctrl, rxq);
1119 
1120 		/*
1121 		 * To support metadata register copy on Tx loopback,
1122 		 * this must be always enabled (metadata may arive
1123 		 * from other port - not from local flows only.
1124 		 */
1125 		if (priv->config.dv_flow_en &&
1126 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1127 		    mlx5_flow_ext_mreg_supported(dev)) {
1128 			rxq_ctrl->rxq.mark = 1;
1129 			rxq_ctrl->flow_mark_n = 1;
1130 		} else if (mark) {
1131 			rxq_ctrl->rxq.mark = 1;
1132 			rxq_ctrl->flow_mark_n++;
1133 		}
1134 		if (tunnel) {
1135 			unsigned int j;
1136 
1137 			/* Increase the counter matching the flow. */
1138 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1139 				if ((tunnels_info[j].tunnel &
1140 				     dev_handle->layers) ==
1141 				    tunnels_info[j].tunnel) {
1142 					rxq_ctrl->flow_tunnels_n[j]++;
1143 					break;
1144 				}
1145 			}
1146 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1147 		}
1148 	}
1149 }
1150 
1151 /**
1152  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1153  *
1154  * @param[in] dev
1155  *   Pointer to the Ethernet device structure.
1156  * @param[in] flow
1157  *   Pointer to flow structure.
1158  */
1159 static void
1160 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1161 {
1162 	struct mlx5_priv *priv = dev->data->dev_private;
1163 	uint32_t handle_idx;
1164 	struct mlx5_flow_handle *dev_handle;
1165 
1166 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1167 		       handle_idx, dev_handle, next)
1168 		flow_drv_rxq_flags_set(dev, dev_handle);
1169 }
1170 
1171 /**
1172  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1173  * device flow if no other flow uses it with the same kind of request.
1174  *
1175  * @param dev
1176  *   Pointer to Ethernet device.
1177  * @param[in] dev_handle
1178  *   Pointer to the device flow handle structure.
1179  */
1180 static void
1181 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1182 			struct mlx5_flow_handle *dev_handle)
1183 {
1184 	struct mlx5_priv *priv = dev->data->dev_private;
1185 	const int mark = dev_handle->mark;
1186 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1187 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1188 	unsigned int i;
1189 
1190 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1191 		struct mlx5_hrxq *hrxq;
1192 
1193 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1194 			      dev_handle->rix_hrxq);
1195 		if (hrxq)
1196 			ind_tbl = hrxq->ind_table;
1197 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1198 		struct mlx5_shared_action_rss *shared_rss;
1199 
1200 		shared_rss = mlx5_ipool_get
1201 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1202 			 dev_handle->rix_srss);
1203 		if (shared_rss)
1204 			ind_tbl = shared_rss->ind_tbl;
1205 	}
1206 	if (!ind_tbl)
1207 		return;
1208 	MLX5_ASSERT(dev->data->dev_started);
1209 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1210 		int idx = ind_tbl->queues[i];
1211 		struct mlx5_rxq_ctrl *rxq_ctrl =
1212 			container_of((*priv->rxqs)[idx],
1213 				     struct mlx5_rxq_ctrl, rxq);
1214 
1215 		if (priv->config.dv_flow_en &&
1216 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1217 		    mlx5_flow_ext_mreg_supported(dev)) {
1218 			rxq_ctrl->rxq.mark = 1;
1219 			rxq_ctrl->flow_mark_n = 1;
1220 		} else if (mark) {
1221 			rxq_ctrl->flow_mark_n--;
1222 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1223 		}
1224 		if (tunnel) {
1225 			unsigned int j;
1226 
1227 			/* Decrease the counter matching the flow. */
1228 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1229 				if ((tunnels_info[j].tunnel &
1230 				     dev_handle->layers) ==
1231 				    tunnels_info[j].tunnel) {
1232 					rxq_ctrl->flow_tunnels_n[j]--;
1233 					break;
1234 				}
1235 			}
1236 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1237 		}
1238 	}
1239 }
1240 
1241 /**
1242  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1243  * @p flow if no other flow uses it with the same kind of request.
1244  *
1245  * @param dev
1246  *   Pointer to Ethernet device.
1247  * @param[in] flow
1248  *   Pointer to the flow.
1249  */
1250 static void
1251 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1252 {
1253 	struct mlx5_priv *priv = dev->data->dev_private;
1254 	uint32_t handle_idx;
1255 	struct mlx5_flow_handle *dev_handle;
1256 
1257 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1258 		       handle_idx, dev_handle, next)
1259 		flow_drv_rxq_flags_trim(dev, dev_handle);
1260 }
1261 
1262 /**
1263  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1264  *
1265  * @param dev
1266  *   Pointer to Ethernet device.
1267  */
1268 static void
1269 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1270 {
1271 	struct mlx5_priv *priv = dev->data->dev_private;
1272 	unsigned int i;
1273 
1274 	for (i = 0; i != priv->rxqs_n; ++i) {
1275 		struct mlx5_rxq_ctrl *rxq_ctrl;
1276 		unsigned int j;
1277 
1278 		if (!(*priv->rxqs)[i])
1279 			continue;
1280 		rxq_ctrl = container_of((*priv->rxqs)[i],
1281 					struct mlx5_rxq_ctrl, rxq);
1282 		rxq_ctrl->flow_mark_n = 0;
1283 		rxq_ctrl->rxq.mark = 0;
1284 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1285 			rxq_ctrl->flow_tunnels_n[j] = 0;
1286 		rxq_ctrl->rxq.tunnel = 0;
1287 	}
1288 }
1289 
1290 /**
1291  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1292  *
1293  * @param[in] dev
1294  *   Pointer to the Ethernet device structure.
1295  */
1296 void
1297 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1298 {
1299 	struct mlx5_priv *priv = dev->data->dev_private;
1300 	struct mlx5_rxq_data *data;
1301 	unsigned int i;
1302 
1303 	for (i = 0; i != priv->rxqs_n; ++i) {
1304 		if (!(*priv->rxqs)[i])
1305 			continue;
1306 		data = (*priv->rxqs)[i];
1307 		if (!rte_flow_dynf_metadata_avail()) {
1308 			data->dynf_meta = 0;
1309 			data->flow_meta_mask = 0;
1310 			data->flow_meta_offset = -1;
1311 			data->flow_meta_port_mask = 0;
1312 		} else {
1313 			data->dynf_meta = 1;
1314 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1315 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1316 			data->flow_meta_port_mask = (uint32_t)~0;
1317 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1318 				data->flow_meta_port_mask >>= 16;
1319 		}
1320 	}
1321 }
1322 
1323 /*
1324  * return a pointer to the desired action in the list of actions.
1325  *
1326  * @param[in] actions
1327  *   The list of actions to search the action in.
1328  * @param[in] action
1329  *   The action to find.
1330  *
1331  * @return
1332  *   Pointer to the action in the list, if found. NULL otherwise.
1333  */
1334 const struct rte_flow_action *
1335 mlx5_flow_find_action(const struct rte_flow_action *actions,
1336 		      enum rte_flow_action_type action)
1337 {
1338 	if (actions == NULL)
1339 		return NULL;
1340 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1341 		if (actions->type == action)
1342 			return actions;
1343 	return NULL;
1344 }
1345 
1346 /*
1347  * Validate the flag action.
1348  *
1349  * @param[in] action_flags
1350  *   Bit-fields that holds the actions detected until now.
1351  * @param[in] attr
1352  *   Attributes of flow that includes this action.
1353  * @param[out] error
1354  *   Pointer to error structure.
1355  *
1356  * @return
1357  *   0 on success, a negative errno value otherwise and rte_errno is set.
1358  */
1359 int
1360 mlx5_flow_validate_action_flag(uint64_t action_flags,
1361 			       const struct rte_flow_attr *attr,
1362 			       struct rte_flow_error *error)
1363 {
1364 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1365 		return rte_flow_error_set(error, EINVAL,
1366 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1367 					  "can't mark and flag in same flow");
1368 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1369 		return rte_flow_error_set(error, EINVAL,
1370 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1371 					  "can't have 2 flag"
1372 					  " actions in same flow");
1373 	if (attr->egress)
1374 		return rte_flow_error_set(error, ENOTSUP,
1375 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1376 					  "flag action not supported for "
1377 					  "egress");
1378 	return 0;
1379 }
1380 
1381 /*
1382  * Validate the mark action.
1383  *
1384  * @param[in] action
1385  *   Pointer to the queue action.
1386  * @param[in] action_flags
1387  *   Bit-fields that holds the actions detected until now.
1388  * @param[in] attr
1389  *   Attributes of flow that includes this action.
1390  * @param[out] error
1391  *   Pointer to error structure.
1392  *
1393  * @return
1394  *   0 on success, a negative errno value otherwise and rte_errno is set.
1395  */
1396 int
1397 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1398 			       uint64_t action_flags,
1399 			       const struct rte_flow_attr *attr,
1400 			       struct rte_flow_error *error)
1401 {
1402 	const struct rte_flow_action_mark *mark = action->conf;
1403 
1404 	if (!mark)
1405 		return rte_flow_error_set(error, EINVAL,
1406 					  RTE_FLOW_ERROR_TYPE_ACTION,
1407 					  action,
1408 					  "configuration cannot be null");
1409 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1410 		return rte_flow_error_set(error, EINVAL,
1411 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1412 					  &mark->id,
1413 					  "mark id must in 0 <= id < "
1414 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1415 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1416 		return rte_flow_error_set(error, EINVAL,
1417 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1418 					  "can't flag and mark in same flow");
1419 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1420 		return rte_flow_error_set(error, EINVAL,
1421 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1422 					  "can't have 2 mark actions in same"
1423 					  " flow");
1424 	if (attr->egress)
1425 		return rte_flow_error_set(error, ENOTSUP,
1426 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1427 					  "mark action not supported for "
1428 					  "egress");
1429 	return 0;
1430 }
1431 
1432 /*
1433  * Validate the drop action.
1434  *
1435  * @param[in] action_flags
1436  *   Bit-fields that holds the actions detected until now.
1437  * @param[in] attr
1438  *   Attributes of flow that includes this action.
1439  * @param[out] error
1440  *   Pointer to error structure.
1441  *
1442  * @return
1443  *   0 on success, a negative errno value otherwise and rte_errno is set.
1444  */
1445 int
1446 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1447 			       const struct rte_flow_attr *attr,
1448 			       struct rte_flow_error *error)
1449 {
1450 	if (attr->egress)
1451 		return rte_flow_error_set(error, ENOTSUP,
1452 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1453 					  "drop action not supported for "
1454 					  "egress");
1455 	return 0;
1456 }
1457 
1458 /*
1459  * Validate the queue action.
1460  *
1461  * @param[in] action
1462  *   Pointer to the queue action.
1463  * @param[in] action_flags
1464  *   Bit-fields that holds the actions detected until now.
1465  * @param[in] dev
1466  *   Pointer to the Ethernet device structure.
1467  * @param[in] attr
1468  *   Attributes of flow that includes this action.
1469  * @param[out] error
1470  *   Pointer to error structure.
1471  *
1472  * @return
1473  *   0 on success, a negative errno value otherwise and rte_errno is set.
1474  */
1475 int
1476 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1477 				uint64_t action_flags,
1478 				struct rte_eth_dev *dev,
1479 				const struct rte_flow_attr *attr,
1480 				struct rte_flow_error *error)
1481 {
1482 	struct mlx5_priv *priv = dev->data->dev_private;
1483 	const struct rte_flow_action_queue *queue = action->conf;
1484 
1485 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1486 		return rte_flow_error_set(error, EINVAL,
1487 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1488 					  "can't have 2 fate actions in"
1489 					  " same flow");
1490 	if (!priv->rxqs_n)
1491 		return rte_flow_error_set(error, EINVAL,
1492 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1493 					  NULL, "No Rx queues configured");
1494 	if (queue->index >= priv->rxqs_n)
1495 		return rte_flow_error_set(error, EINVAL,
1496 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1497 					  &queue->index,
1498 					  "queue index out of range");
1499 	if (!(*priv->rxqs)[queue->index])
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1502 					  &queue->index,
1503 					  "queue is not configured");
1504 	if (attr->egress)
1505 		return rte_flow_error_set(error, ENOTSUP,
1506 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1507 					  "queue action not supported for "
1508 					  "egress");
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Validate the rss action.
1514  *
1515  * @param[in] dev
1516  *   Pointer to the Ethernet device structure.
1517  * @param[in] action
1518  *   Pointer to the queue action.
1519  * @param[out] error
1520  *   Pointer to error structure.
1521  *
1522  * @return
1523  *   0 on success, a negative errno value otherwise and rte_errno is set.
1524  */
1525 int
1526 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1527 			 const struct rte_flow_action *action,
1528 			 struct rte_flow_error *error)
1529 {
1530 	struct mlx5_priv *priv = dev->data->dev_private;
1531 	const struct rte_flow_action_rss *rss = action->conf;
1532 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1533 	unsigned int i;
1534 
1535 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1536 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1537 		return rte_flow_error_set(error, ENOTSUP,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 					  &rss->func,
1540 					  "RSS hash function not supported");
1541 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1542 	if (rss->level > 2)
1543 #else
1544 	if (rss->level > 1)
1545 #endif
1546 		return rte_flow_error_set(error, ENOTSUP,
1547 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1548 					  &rss->level,
1549 					  "tunnel RSS is not supported");
1550 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1551 	if (rss->key_len == 0 && rss->key != NULL)
1552 		return rte_flow_error_set(error, ENOTSUP,
1553 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1554 					  &rss->key_len,
1555 					  "RSS hash key length 0");
1556 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1559 					  &rss->key_len,
1560 					  "RSS hash key too small");
1561 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1562 		return rte_flow_error_set(error, ENOTSUP,
1563 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1564 					  &rss->key_len,
1565 					  "RSS hash key too large");
1566 	if (rss->queue_num > priv->config.ind_table_max_size)
1567 		return rte_flow_error_set(error, ENOTSUP,
1568 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 					  &rss->queue_num,
1570 					  "number of queues too large");
1571 	if (rss->types & MLX5_RSS_HF_MASK)
1572 		return rte_flow_error_set(error, ENOTSUP,
1573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1574 					  &rss->types,
1575 					  "some RSS protocols are not"
1576 					  " supported");
1577 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1578 	    !(rss->types & ETH_RSS_IP))
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1581 					  "L3 partial RSS requested but L3 RSS"
1582 					  " type not specified");
1583 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1584 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1585 		return rte_flow_error_set(error, EINVAL,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1587 					  "L4 partial RSS requested but L4 RSS"
1588 					  " type not specified");
1589 	if (!priv->rxqs_n)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 					  NULL, "No Rx queues configured");
1593 	if (!rss->queue_num)
1594 		return rte_flow_error_set(error, EINVAL,
1595 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1596 					  NULL, "No queues configured");
1597 	for (i = 0; i != rss->queue_num; ++i) {
1598 		struct mlx5_rxq_ctrl *rxq_ctrl;
1599 
1600 		if (rss->queue[i] >= priv->rxqs_n)
1601 			return rte_flow_error_set
1602 				(error, EINVAL,
1603 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1604 				 &rss->queue[i], "queue index out of range");
1605 		if (!(*priv->rxqs)[rss->queue[i]])
1606 			return rte_flow_error_set
1607 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1608 				 &rss->queue[i], "queue is not configured");
1609 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1610 					struct mlx5_rxq_ctrl, rxq);
1611 		if (i == 0)
1612 			rxq_type = rxq_ctrl->type;
1613 		if (rxq_type != rxq_ctrl->type)
1614 			return rte_flow_error_set
1615 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1616 				 &rss->queue[i],
1617 				 "combining hairpin and regular RSS queues is not supported");
1618 	}
1619 	return 0;
1620 }
1621 
1622 /*
1623  * Validate the rss action.
1624  *
1625  * @param[in] action
1626  *   Pointer to the queue action.
1627  * @param[in] action_flags
1628  *   Bit-fields that holds the actions detected until now.
1629  * @param[in] dev
1630  *   Pointer to the Ethernet device structure.
1631  * @param[in] attr
1632  *   Attributes of flow that includes this action.
1633  * @param[in] item_flags
1634  *   Items that were detected.
1635  * @param[out] error
1636  *   Pointer to error structure.
1637  *
1638  * @return
1639  *   0 on success, a negative errno value otherwise and rte_errno is set.
1640  */
1641 int
1642 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1643 			      uint64_t action_flags,
1644 			      struct rte_eth_dev *dev,
1645 			      const struct rte_flow_attr *attr,
1646 			      uint64_t item_flags,
1647 			      struct rte_flow_error *error)
1648 {
1649 	const struct rte_flow_action_rss *rss = action->conf;
1650 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1651 	int ret;
1652 
1653 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1654 		return rte_flow_error_set(error, EINVAL,
1655 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1656 					  "can't have 2 fate actions"
1657 					  " in same flow");
1658 	ret = mlx5_validate_action_rss(dev, action, error);
1659 	if (ret)
1660 		return ret;
1661 	if (attr->egress)
1662 		return rte_flow_error_set(error, ENOTSUP,
1663 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1664 					  "rss action not supported for "
1665 					  "egress");
1666 	if (rss->level > 1 && !tunnel)
1667 		return rte_flow_error_set(error, EINVAL,
1668 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1669 					  "inner RSS is not supported for "
1670 					  "non-tunnel flows");
1671 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1672 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1673 		return rte_flow_error_set(error, EINVAL,
1674 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1675 					  "RSS on eCPRI is not supported now");
1676 	}
1677 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1678 	    !(item_flags &
1679 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1680 	    rss->level > 1)
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1683 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1684 	return 0;
1685 }
1686 
1687 /*
1688  * Validate the default miss action.
1689  *
1690  * @param[in] action_flags
1691  *   Bit-fields that holds the actions detected until now.
1692  * @param[out] error
1693  *   Pointer to error structure.
1694  *
1695  * @return
1696  *   0 on success, a negative errno value otherwise and rte_errno is set.
1697  */
1698 int
1699 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1700 				const struct rte_flow_attr *attr,
1701 				struct rte_flow_error *error)
1702 {
1703 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1704 		return rte_flow_error_set(error, EINVAL,
1705 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1706 					  "can't have 2 fate actions in"
1707 					  " same flow");
1708 	if (attr->egress)
1709 		return rte_flow_error_set(error, ENOTSUP,
1710 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1711 					  "default miss action not supported "
1712 					  "for egress");
1713 	if (attr->group)
1714 		return rte_flow_error_set(error, ENOTSUP,
1715 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1716 					  "only group 0 is supported");
1717 	if (attr->transfer)
1718 		return rte_flow_error_set(error, ENOTSUP,
1719 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1720 					  NULL, "transfer is not supported");
1721 	return 0;
1722 }
1723 
1724 /*
1725  * Validate the count action.
1726  *
1727  * @param[in] dev
1728  *   Pointer to the Ethernet device structure.
1729  * @param[in] attr
1730  *   Attributes of flow that includes this action.
1731  * @param[out] error
1732  *   Pointer to error structure.
1733  *
1734  * @return
1735  *   0 on success, a negative errno value otherwise and rte_errno is set.
1736  */
1737 int
1738 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1739 				const struct rte_flow_attr *attr,
1740 				struct rte_flow_error *error)
1741 {
1742 	if (attr->egress)
1743 		return rte_flow_error_set(error, ENOTSUP,
1744 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1745 					  "count action not supported for "
1746 					  "egress");
1747 	return 0;
1748 }
1749 
1750 /*
1751  * Validate the ASO CT action.
1752  *
1753  * @param[in] dev
1754  *   Pointer to the Ethernet device structure.
1755  * @param[in] conntrack
1756  *   Pointer to the CT action profile.
1757  * @param[out] error
1758  *   Pointer to error structure.
1759  *
1760  * @return
1761  *   0 on success, a negative errno value otherwise and rte_errno is set.
1762  */
1763 int
1764 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1765 			const struct rte_flow_action_conntrack *conntrack,
1766 			struct rte_flow_error *error)
1767 {
1768 	RTE_SET_USED(dev);
1769 
1770 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1771 		return rte_flow_error_set(error, EINVAL,
1772 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1773 					  "Invalid CT state");
1774 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1775 		return rte_flow_error_set(error, EINVAL,
1776 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1777 					  "Invalid last TCP packet flag");
1778 	return 0;
1779 }
1780 
1781 /**
1782  * Verify the @p attributes will be correctly understood by the NIC and store
1783  * them in the @p flow if everything is correct.
1784  *
1785  * @param[in] dev
1786  *   Pointer to the Ethernet device structure.
1787  * @param[in] attributes
1788  *   Pointer to flow attributes
1789  * @param[out] error
1790  *   Pointer to error structure.
1791  *
1792  * @return
1793  *   0 on success, a negative errno value otherwise and rte_errno is set.
1794  */
1795 int
1796 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1797 			      const struct rte_flow_attr *attributes,
1798 			      struct rte_flow_error *error)
1799 {
1800 	struct mlx5_priv *priv = dev->data->dev_private;
1801 	uint32_t priority_max = priv->config.flow_prio - 1;
1802 
1803 	if (attributes->group)
1804 		return rte_flow_error_set(error, ENOTSUP,
1805 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1806 					  NULL, "groups is not supported");
1807 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1808 	    attributes->priority >= priority_max)
1809 		return rte_flow_error_set(error, ENOTSUP,
1810 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1811 					  NULL, "priority out of range");
1812 	if (attributes->egress)
1813 		return rte_flow_error_set(error, ENOTSUP,
1814 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1815 					  "egress is not supported");
1816 	if (attributes->transfer && !priv->config.dv_esw_en)
1817 		return rte_flow_error_set(error, ENOTSUP,
1818 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1819 					  NULL, "transfer is not supported");
1820 	if (!attributes->ingress)
1821 		return rte_flow_error_set(error, EINVAL,
1822 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1823 					  NULL,
1824 					  "ingress attribute is mandatory");
1825 	return 0;
1826 }
1827 
1828 /**
1829  * Validate ICMP6 item.
1830  *
1831  * @param[in] item
1832  *   Item specification.
1833  * @param[in] item_flags
1834  *   Bit-fields that holds the items detected until now.
1835  * @param[in] ext_vlan_sup
1836  *   Whether extended VLAN features are supported or not.
1837  * @param[out] error
1838  *   Pointer to error structure.
1839  *
1840  * @return
1841  *   0 on success, a negative errno value otherwise and rte_errno is set.
1842  */
1843 int
1844 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1845 			       uint64_t item_flags,
1846 			       uint8_t target_protocol,
1847 			       struct rte_flow_error *error)
1848 {
1849 	const struct rte_flow_item_icmp6 *mask = item->mask;
1850 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1851 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1852 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1853 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1854 				      MLX5_FLOW_LAYER_OUTER_L4;
1855 	int ret;
1856 
1857 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1858 		return rte_flow_error_set(error, EINVAL,
1859 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1860 					  "protocol filtering not compatible"
1861 					  " with ICMP6 layer");
1862 	if (!(item_flags & l3m))
1863 		return rte_flow_error_set(error, EINVAL,
1864 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1865 					  "IPv6 is mandatory to filter on"
1866 					  " ICMP6");
1867 	if (item_flags & l4m)
1868 		return rte_flow_error_set(error, EINVAL,
1869 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1870 					  "multiple L4 layers not supported");
1871 	if (!mask)
1872 		mask = &rte_flow_item_icmp6_mask;
1873 	ret = mlx5_flow_item_acceptable
1874 		(item, (const uint8_t *)mask,
1875 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1876 		 sizeof(struct rte_flow_item_icmp6),
1877 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1878 	if (ret < 0)
1879 		return ret;
1880 	return 0;
1881 }
1882 
1883 /**
1884  * Validate ICMP item.
1885  *
1886  * @param[in] item
1887  *   Item specification.
1888  * @param[in] item_flags
1889  *   Bit-fields that holds the items detected until now.
1890  * @param[out] error
1891  *   Pointer to error structure.
1892  *
1893  * @return
1894  *   0 on success, a negative errno value otherwise and rte_errno is set.
1895  */
1896 int
1897 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1898 			     uint64_t item_flags,
1899 			     uint8_t target_protocol,
1900 			     struct rte_flow_error *error)
1901 {
1902 	const struct rte_flow_item_icmp *mask = item->mask;
1903 	const struct rte_flow_item_icmp nic_mask = {
1904 		.hdr.icmp_type = 0xff,
1905 		.hdr.icmp_code = 0xff,
1906 		.hdr.icmp_ident = RTE_BE16(0xffff),
1907 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1908 	};
1909 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1910 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1911 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1912 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1913 				      MLX5_FLOW_LAYER_OUTER_L4;
1914 	int ret;
1915 
1916 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1917 		return rte_flow_error_set(error, EINVAL,
1918 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1919 					  "protocol filtering not compatible"
1920 					  " with ICMP layer");
1921 	if (!(item_flags & l3m))
1922 		return rte_flow_error_set(error, EINVAL,
1923 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1924 					  "IPv4 is mandatory to filter"
1925 					  " on ICMP");
1926 	if (item_flags & l4m)
1927 		return rte_flow_error_set(error, EINVAL,
1928 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1929 					  "multiple L4 layers not supported");
1930 	if (!mask)
1931 		mask = &nic_mask;
1932 	ret = mlx5_flow_item_acceptable
1933 		(item, (const uint8_t *)mask,
1934 		 (const uint8_t *)&nic_mask,
1935 		 sizeof(struct rte_flow_item_icmp),
1936 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1937 	if (ret < 0)
1938 		return ret;
1939 	return 0;
1940 }
1941 
1942 /**
1943  * Validate Ethernet item.
1944  *
1945  * @param[in] item
1946  *   Item specification.
1947  * @param[in] item_flags
1948  *   Bit-fields that holds the items detected until now.
1949  * @param[out] error
1950  *   Pointer to error structure.
1951  *
1952  * @return
1953  *   0 on success, a negative errno value otherwise and rte_errno is set.
1954  */
1955 int
1956 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1957 			    uint64_t item_flags, bool ext_vlan_sup,
1958 			    struct rte_flow_error *error)
1959 {
1960 	const struct rte_flow_item_eth *mask = item->mask;
1961 	const struct rte_flow_item_eth nic_mask = {
1962 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1963 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1964 		.type = RTE_BE16(0xffff),
1965 		.has_vlan = ext_vlan_sup ? 1 : 0,
1966 	};
1967 	int ret;
1968 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1969 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1970 				       MLX5_FLOW_LAYER_OUTER_L2;
1971 
1972 	if (item_flags & ethm)
1973 		return rte_flow_error_set(error, ENOTSUP,
1974 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1975 					  "multiple L2 layers not supported");
1976 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1977 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1978 		return rte_flow_error_set(error, EINVAL,
1979 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1980 					  "L2 layer should not follow "
1981 					  "L3 layers");
1982 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1983 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1984 		return rte_flow_error_set(error, EINVAL,
1985 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1986 					  "L2 layer should not follow VLAN");
1987 	if (!mask)
1988 		mask = &rte_flow_item_eth_mask;
1989 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1990 					(const uint8_t *)&nic_mask,
1991 					sizeof(struct rte_flow_item_eth),
1992 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1993 	return ret;
1994 }
1995 
1996 /**
1997  * Validate VLAN item.
1998  *
1999  * @param[in] item
2000  *   Item specification.
2001  * @param[in] item_flags
2002  *   Bit-fields that holds the items detected until now.
2003  * @param[in] dev
2004  *   Ethernet device flow is being created on.
2005  * @param[out] error
2006  *   Pointer to error structure.
2007  *
2008  * @return
2009  *   0 on success, a negative errno value otherwise and rte_errno is set.
2010  */
2011 int
2012 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2013 			     uint64_t item_flags,
2014 			     struct rte_eth_dev *dev,
2015 			     struct rte_flow_error *error)
2016 {
2017 	const struct rte_flow_item_vlan *spec = item->spec;
2018 	const struct rte_flow_item_vlan *mask = item->mask;
2019 	const struct rte_flow_item_vlan nic_mask = {
2020 		.tci = RTE_BE16(UINT16_MAX),
2021 		.inner_type = RTE_BE16(UINT16_MAX),
2022 	};
2023 	uint16_t vlan_tag = 0;
2024 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2025 	int ret;
2026 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2027 					MLX5_FLOW_LAYER_INNER_L4) :
2028 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2029 					MLX5_FLOW_LAYER_OUTER_L4);
2030 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2031 					MLX5_FLOW_LAYER_OUTER_VLAN;
2032 
2033 	if (item_flags & vlanm)
2034 		return rte_flow_error_set(error, EINVAL,
2035 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2036 					  "multiple VLAN layers not supported");
2037 	else if ((item_flags & l34m) != 0)
2038 		return rte_flow_error_set(error, EINVAL,
2039 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2040 					  "VLAN cannot follow L3/L4 layer");
2041 	if (!mask)
2042 		mask = &rte_flow_item_vlan_mask;
2043 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2044 					(const uint8_t *)&nic_mask,
2045 					sizeof(struct rte_flow_item_vlan),
2046 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2047 	if (ret)
2048 		return ret;
2049 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2050 		struct mlx5_priv *priv = dev->data->dev_private;
2051 
2052 		if (priv->vmwa_context) {
2053 			/*
2054 			 * Non-NULL context means we have a virtual machine
2055 			 * and SR-IOV enabled, we have to create VLAN interface
2056 			 * to make hypervisor to setup E-Switch vport
2057 			 * context correctly. We avoid creating the multiple
2058 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2059 			 */
2060 			return rte_flow_error_set(error, EINVAL,
2061 						  RTE_FLOW_ERROR_TYPE_ITEM,
2062 						  item,
2063 						  "VLAN tag mask is not"
2064 						  " supported in virtual"
2065 						  " environment");
2066 		}
2067 	}
2068 	if (spec) {
2069 		vlan_tag = spec->tci;
2070 		vlan_tag &= mask->tci;
2071 	}
2072 	/*
2073 	 * From verbs perspective an empty VLAN is equivalent
2074 	 * to a packet without VLAN layer.
2075 	 */
2076 	if (!vlan_tag)
2077 		return rte_flow_error_set(error, EINVAL,
2078 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2079 					  item->spec,
2080 					  "VLAN cannot be empty");
2081 	return 0;
2082 }
2083 
2084 /**
2085  * Validate IPV4 item.
2086  *
2087  * @param[in] item
2088  *   Item specification.
2089  * @param[in] item_flags
2090  *   Bit-fields that holds the items detected until now.
2091  * @param[in] last_item
2092  *   Previous validated item in the pattern items.
2093  * @param[in] ether_type
2094  *   Type in the ethernet layer header (including dot1q).
2095  * @param[in] acc_mask
2096  *   Acceptable mask, if NULL default internal default mask
2097  *   will be used to check whether item fields are supported.
2098  * @param[in] range_accepted
2099  *   True if range of values is accepted for specific fields, false otherwise.
2100  * @param[out] error
2101  *   Pointer to error structure.
2102  *
2103  * @return
2104  *   0 on success, a negative errno value otherwise and rte_errno is set.
2105  */
2106 int
2107 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2108 			     uint64_t item_flags,
2109 			     uint64_t last_item,
2110 			     uint16_t ether_type,
2111 			     const struct rte_flow_item_ipv4 *acc_mask,
2112 			     bool range_accepted,
2113 			     struct rte_flow_error *error)
2114 {
2115 	const struct rte_flow_item_ipv4 *mask = item->mask;
2116 	const struct rte_flow_item_ipv4 *spec = item->spec;
2117 	const struct rte_flow_item_ipv4 nic_mask = {
2118 		.hdr = {
2119 			.src_addr = RTE_BE32(0xffffffff),
2120 			.dst_addr = RTE_BE32(0xffffffff),
2121 			.type_of_service = 0xff,
2122 			.next_proto_id = 0xff,
2123 		},
2124 	};
2125 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2126 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2127 				      MLX5_FLOW_LAYER_OUTER_L3;
2128 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2129 				      MLX5_FLOW_LAYER_OUTER_L4;
2130 	int ret;
2131 	uint8_t next_proto = 0xFF;
2132 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2133 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2134 				  MLX5_FLOW_LAYER_INNER_VLAN);
2135 
2136 	if ((last_item & l2_vlan) && ether_type &&
2137 	    ether_type != RTE_ETHER_TYPE_IPV4)
2138 		return rte_flow_error_set(error, EINVAL,
2139 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2140 					  "IPv4 cannot follow L2/VLAN layer "
2141 					  "which ether type is not IPv4");
2142 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2143 		if (mask && spec)
2144 			next_proto = mask->hdr.next_proto_id &
2145 				     spec->hdr.next_proto_id;
2146 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2147 			return rte_flow_error_set(error, EINVAL,
2148 						  RTE_FLOW_ERROR_TYPE_ITEM,
2149 						  item,
2150 						  "multiple tunnel "
2151 						  "not supported");
2152 	}
2153 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2154 		return rte_flow_error_set(error, EINVAL,
2155 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2156 					  "wrong tunnel type - IPv6 specified "
2157 					  "but IPv4 item provided");
2158 	if (item_flags & l3m)
2159 		return rte_flow_error_set(error, ENOTSUP,
2160 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2161 					  "multiple L3 layers not supported");
2162 	else if (item_flags & l4m)
2163 		return rte_flow_error_set(error, EINVAL,
2164 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2165 					  "L3 cannot follow an L4 layer.");
2166 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2167 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2168 		return rte_flow_error_set(error, EINVAL,
2169 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2170 					  "L3 cannot follow an NVGRE layer.");
2171 	if (!mask)
2172 		mask = &rte_flow_item_ipv4_mask;
2173 	else if (mask->hdr.next_proto_id != 0 &&
2174 		 mask->hdr.next_proto_id != 0xff)
2175 		return rte_flow_error_set(error, EINVAL,
2176 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2177 					  "partial mask is not supported"
2178 					  " for protocol");
2179 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2180 					acc_mask ? (const uint8_t *)acc_mask
2181 						 : (const uint8_t *)&nic_mask,
2182 					sizeof(struct rte_flow_item_ipv4),
2183 					range_accepted, error);
2184 	if (ret < 0)
2185 		return ret;
2186 	return 0;
2187 }
2188 
2189 /**
2190  * Validate IPV6 item.
2191  *
2192  * @param[in] item
2193  *   Item specification.
2194  * @param[in] item_flags
2195  *   Bit-fields that holds the items detected until now.
2196  * @param[in] last_item
2197  *   Previous validated item in the pattern items.
2198  * @param[in] ether_type
2199  *   Type in the ethernet layer header (including dot1q).
2200  * @param[in] acc_mask
2201  *   Acceptable mask, if NULL default internal default mask
2202  *   will be used to check whether item fields are supported.
2203  * @param[out] error
2204  *   Pointer to error structure.
2205  *
2206  * @return
2207  *   0 on success, a negative errno value otherwise and rte_errno is set.
2208  */
2209 int
2210 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2211 			     uint64_t item_flags,
2212 			     uint64_t last_item,
2213 			     uint16_t ether_type,
2214 			     const struct rte_flow_item_ipv6 *acc_mask,
2215 			     struct rte_flow_error *error)
2216 {
2217 	const struct rte_flow_item_ipv6 *mask = item->mask;
2218 	const struct rte_flow_item_ipv6 *spec = item->spec;
2219 	const struct rte_flow_item_ipv6 nic_mask = {
2220 		.hdr = {
2221 			.src_addr =
2222 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2223 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2224 			.dst_addr =
2225 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2226 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2227 			.vtc_flow = RTE_BE32(0xffffffff),
2228 			.proto = 0xff,
2229 		},
2230 	};
2231 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2232 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2233 				      MLX5_FLOW_LAYER_OUTER_L3;
2234 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2235 				      MLX5_FLOW_LAYER_OUTER_L4;
2236 	int ret;
2237 	uint8_t next_proto = 0xFF;
2238 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2239 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2240 				  MLX5_FLOW_LAYER_INNER_VLAN);
2241 
2242 	if ((last_item & l2_vlan) && ether_type &&
2243 	    ether_type != RTE_ETHER_TYPE_IPV6)
2244 		return rte_flow_error_set(error, EINVAL,
2245 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2246 					  "IPv6 cannot follow L2/VLAN layer "
2247 					  "which ether type is not IPv6");
2248 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2249 		next_proto = spec->hdr.proto;
2250 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2251 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2252 			return rte_flow_error_set(error, EINVAL,
2253 						  RTE_FLOW_ERROR_TYPE_ITEM,
2254 						  item,
2255 						  "multiple tunnel "
2256 						  "not supported");
2257 	}
2258 	if (next_proto == IPPROTO_HOPOPTS  ||
2259 	    next_proto == IPPROTO_ROUTING  ||
2260 	    next_proto == IPPROTO_FRAGMENT ||
2261 	    next_proto == IPPROTO_ESP	   ||
2262 	    next_proto == IPPROTO_AH	   ||
2263 	    next_proto == IPPROTO_DSTOPTS)
2264 		return rte_flow_error_set(error, EINVAL,
2265 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2266 					  "IPv6 proto (next header) should "
2267 					  "not be set as extension header");
2268 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2269 		return rte_flow_error_set(error, EINVAL,
2270 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 					  "wrong tunnel type - IPv4 specified "
2272 					  "but IPv6 item provided");
2273 	if (item_flags & l3m)
2274 		return rte_flow_error_set(error, ENOTSUP,
2275 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 					  "multiple L3 layers not supported");
2277 	else if (item_flags & l4m)
2278 		return rte_flow_error_set(error, EINVAL,
2279 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2280 					  "L3 cannot follow an L4 layer.");
2281 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2282 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2283 		return rte_flow_error_set(error, EINVAL,
2284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 					  "L3 cannot follow an NVGRE layer.");
2286 	if (!mask)
2287 		mask = &rte_flow_item_ipv6_mask;
2288 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2289 					acc_mask ? (const uint8_t *)acc_mask
2290 						 : (const uint8_t *)&nic_mask,
2291 					sizeof(struct rte_flow_item_ipv6),
2292 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2293 	if (ret < 0)
2294 		return ret;
2295 	return 0;
2296 }
2297 
2298 /**
2299  * Validate UDP item.
2300  *
2301  * @param[in] item
2302  *   Item specification.
2303  * @param[in] item_flags
2304  *   Bit-fields that holds the items detected until now.
2305  * @param[in] target_protocol
2306  *   The next protocol in the previous item.
2307  * @param[in] flow_mask
2308  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2309  * @param[out] error
2310  *   Pointer to error structure.
2311  *
2312  * @return
2313  *   0 on success, a negative errno value otherwise and rte_errno is set.
2314  */
2315 int
2316 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2317 			    uint64_t item_flags,
2318 			    uint8_t target_protocol,
2319 			    struct rte_flow_error *error)
2320 {
2321 	const struct rte_flow_item_udp *mask = item->mask;
2322 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2323 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2324 				      MLX5_FLOW_LAYER_OUTER_L3;
2325 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2326 				      MLX5_FLOW_LAYER_OUTER_L4;
2327 	int ret;
2328 
2329 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2330 		return rte_flow_error_set(error, EINVAL,
2331 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2332 					  "protocol filtering not compatible"
2333 					  " with UDP layer");
2334 	if (!(item_flags & l3m))
2335 		return rte_flow_error_set(error, EINVAL,
2336 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2337 					  "L3 is mandatory to filter on L4");
2338 	if (item_flags & l4m)
2339 		return rte_flow_error_set(error, EINVAL,
2340 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2341 					  "multiple L4 layers not supported");
2342 	if (!mask)
2343 		mask = &rte_flow_item_udp_mask;
2344 	ret = mlx5_flow_item_acceptable
2345 		(item, (const uint8_t *)mask,
2346 		 (const uint8_t *)&rte_flow_item_udp_mask,
2347 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2348 		 error);
2349 	if (ret < 0)
2350 		return ret;
2351 	return 0;
2352 }
2353 
2354 /**
2355  * Validate TCP item.
2356  *
2357  * @param[in] item
2358  *   Item specification.
2359  * @param[in] item_flags
2360  *   Bit-fields that holds the items detected until now.
2361  * @param[in] target_protocol
2362  *   The next protocol in the previous item.
2363  * @param[out] error
2364  *   Pointer to error structure.
2365  *
2366  * @return
2367  *   0 on success, a negative errno value otherwise and rte_errno is set.
2368  */
2369 int
2370 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2371 			    uint64_t item_flags,
2372 			    uint8_t target_protocol,
2373 			    const struct rte_flow_item_tcp *flow_mask,
2374 			    struct rte_flow_error *error)
2375 {
2376 	const struct rte_flow_item_tcp *mask = item->mask;
2377 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2378 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2379 				      MLX5_FLOW_LAYER_OUTER_L3;
2380 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2381 				      MLX5_FLOW_LAYER_OUTER_L4;
2382 	int ret;
2383 
2384 	MLX5_ASSERT(flow_mask);
2385 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2386 		return rte_flow_error_set(error, EINVAL,
2387 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2388 					  "protocol filtering not compatible"
2389 					  " with TCP layer");
2390 	if (!(item_flags & l3m))
2391 		return rte_flow_error_set(error, EINVAL,
2392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2393 					  "L3 is mandatory to filter on L4");
2394 	if (item_flags & l4m)
2395 		return rte_flow_error_set(error, EINVAL,
2396 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2397 					  "multiple L4 layers not supported");
2398 	if (!mask)
2399 		mask = &rte_flow_item_tcp_mask;
2400 	ret = mlx5_flow_item_acceptable
2401 		(item, (const uint8_t *)mask,
2402 		 (const uint8_t *)flow_mask,
2403 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2404 		 error);
2405 	if (ret < 0)
2406 		return ret;
2407 	return 0;
2408 }
2409 
2410 /**
2411  * Validate VXLAN item.
2412  *
2413  * @param[in] dev
2414  *   Pointer to the Ethernet device structure.
2415  * @param[in] item
2416  *   Item specification.
2417  * @param[in] item_flags
2418  *   Bit-fields that holds the items detected until now.
2419  * @param[in] attr
2420  *   Flow rule attributes.
2421  * @param[out] error
2422  *   Pointer to error structure.
2423  *
2424  * @return
2425  *   0 on success, a negative errno value otherwise and rte_errno is set.
2426  */
2427 int
2428 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2429 			      const struct rte_flow_item *item,
2430 			      uint64_t item_flags,
2431 			      const struct rte_flow_attr *attr,
2432 			      struct rte_flow_error *error)
2433 {
2434 	const struct rte_flow_item_vxlan *spec = item->spec;
2435 	const struct rte_flow_item_vxlan *mask = item->mask;
2436 	int ret;
2437 	struct mlx5_priv *priv = dev->data->dev_private;
2438 	union vni {
2439 		uint32_t vlan_id;
2440 		uint8_t vni[4];
2441 	} id = { .vlan_id = 0, };
2442 	const struct rte_flow_item_vxlan nic_mask = {
2443 		.vni = "\xff\xff\xff",
2444 		.rsvd1 = 0xff,
2445 	};
2446 	const struct rte_flow_item_vxlan *valid_mask;
2447 
2448 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2449 		return rte_flow_error_set(error, ENOTSUP,
2450 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2451 					  "multiple tunnel layers not"
2452 					  " supported");
2453 	valid_mask = &rte_flow_item_vxlan_mask;
2454 	/*
2455 	 * Verify only UDPv4 is present as defined in
2456 	 * https://tools.ietf.org/html/rfc7348
2457 	 */
2458 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2459 		return rte_flow_error_set(error, EINVAL,
2460 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 					  "no outer UDP layer found");
2462 	if (!mask)
2463 		mask = &rte_flow_item_vxlan_mask;
2464 	/* FDB domain & NIC domain non-zero group */
2465 	if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2466 		valid_mask = &nic_mask;
2467 	/* Group zero in NIC domain */
2468 	if (!attr->group && !attr->transfer && priv->sh->tunnel_header_0_1)
2469 		valid_mask = &nic_mask;
2470 	ret = mlx5_flow_item_acceptable
2471 		(item, (const uint8_t *)mask,
2472 		 (const uint8_t *)valid_mask,
2473 		 sizeof(struct rte_flow_item_vxlan),
2474 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2475 	if (ret < 0)
2476 		return ret;
2477 	if (spec) {
2478 		memcpy(&id.vni[1], spec->vni, 3);
2479 		memcpy(&id.vni[1], mask->vni, 3);
2480 	}
2481 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2482 		return rte_flow_error_set(error, ENOTSUP,
2483 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2484 					  "VXLAN tunnel must be fully defined");
2485 	return 0;
2486 }
2487 
2488 /**
2489  * Validate VXLAN_GPE item.
2490  *
2491  * @param[in] item
2492  *   Item specification.
2493  * @param[in] item_flags
2494  *   Bit-fields that holds the items detected until now.
2495  * @param[in] priv
2496  *   Pointer to the private data structure.
2497  * @param[in] target_protocol
2498  *   The next protocol in the previous item.
2499  * @param[out] error
2500  *   Pointer to error structure.
2501  *
2502  * @return
2503  *   0 on success, a negative errno value otherwise and rte_errno is set.
2504  */
2505 int
2506 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2507 				  uint64_t item_flags,
2508 				  struct rte_eth_dev *dev,
2509 				  struct rte_flow_error *error)
2510 {
2511 	struct mlx5_priv *priv = dev->data->dev_private;
2512 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2513 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2514 	int ret;
2515 	union vni {
2516 		uint32_t vlan_id;
2517 		uint8_t vni[4];
2518 	} id = { .vlan_id = 0, };
2519 
2520 	if (!priv->config.l3_vxlan_en)
2521 		return rte_flow_error_set(error, ENOTSUP,
2522 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2523 					  "L3 VXLAN is not enabled by device"
2524 					  " parameter and/or not configured in"
2525 					  " firmware");
2526 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2527 		return rte_flow_error_set(error, ENOTSUP,
2528 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2529 					  "multiple tunnel layers not"
2530 					  " supported");
2531 	/*
2532 	 * Verify only UDPv4 is present as defined in
2533 	 * https://tools.ietf.org/html/rfc7348
2534 	 */
2535 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2536 		return rte_flow_error_set(error, EINVAL,
2537 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2538 					  "no outer UDP layer found");
2539 	if (!mask)
2540 		mask = &rte_flow_item_vxlan_gpe_mask;
2541 	ret = mlx5_flow_item_acceptable
2542 		(item, (const uint8_t *)mask,
2543 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2544 		 sizeof(struct rte_flow_item_vxlan_gpe),
2545 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2546 	if (ret < 0)
2547 		return ret;
2548 	if (spec) {
2549 		if (spec->protocol)
2550 			return rte_flow_error_set(error, ENOTSUP,
2551 						  RTE_FLOW_ERROR_TYPE_ITEM,
2552 						  item,
2553 						  "VxLAN-GPE protocol"
2554 						  " not supported");
2555 		memcpy(&id.vni[1], spec->vni, 3);
2556 		memcpy(&id.vni[1], mask->vni, 3);
2557 	}
2558 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2559 		return rte_flow_error_set(error, ENOTSUP,
2560 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2561 					  "VXLAN-GPE tunnel must be fully"
2562 					  " defined");
2563 	return 0;
2564 }
2565 /**
2566  * Validate GRE Key item.
2567  *
2568  * @param[in] item
2569  *   Item specification.
2570  * @param[in] item_flags
2571  *   Bit flags to mark detected items.
2572  * @param[in] gre_item
2573  *   Pointer to gre_item
2574  * @param[out] error
2575  *   Pointer to error structure.
2576  *
2577  * @return
2578  *   0 on success, a negative errno value otherwise and rte_errno is set.
2579  */
2580 int
2581 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2582 				uint64_t item_flags,
2583 				const struct rte_flow_item *gre_item,
2584 				struct rte_flow_error *error)
2585 {
2586 	const rte_be32_t *mask = item->mask;
2587 	int ret = 0;
2588 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2589 	const struct rte_flow_item_gre *gre_spec;
2590 	const struct rte_flow_item_gre *gre_mask;
2591 
2592 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2593 		return rte_flow_error_set(error, ENOTSUP,
2594 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2595 					  "Multiple GRE key not support");
2596 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2597 		return rte_flow_error_set(error, ENOTSUP,
2598 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2599 					  "No preceding GRE header");
2600 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2601 		return rte_flow_error_set(error, ENOTSUP,
2602 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2603 					  "GRE key following a wrong item");
2604 	gre_mask = gre_item->mask;
2605 	if (!gre_mask)
2606 		gre_mask = &rte_flow_item_gre_mask;
2607 	gre_spec = gre_item->spec;
2608 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2609 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2610 		return rte_flow_error_set(error, EINVAL,
2611 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2612 					  "Key bit must be on");
2613 
2614 	if (!mask)
2615 		mask = &gre_key_default_mask;
2616 	ret = mlx5_flow_item_acceptable
2617 		(item, (const uint8_t *)mask,
2618 		 (const uint8_t *)&gre_key_default_mask,
2619 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2620 	return ret;
2621 }
2622 
2623 /**
2624  * Validate GRE item.
2625  *
2626  * @param[in] item
2627  *   Item specification.
2628  * @param[in] item_flags
2629  *   Bit flags to mark detected items.
2630  * @param[in] target_protocol
2631  *   The next protocol in the previous item.
2632  * @param[out] error
2633  *   Pointer to error structure.
2634  *
2635  * @return
2636  *   0 on success, a negative errno value otherwise and rte_errno is set.
2637  */
2638 int
2639 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2640 			    uint64_t item_flags,
2641 			    uint8_t target_protocol,
2642 			    struct rte_flow_error *error)
2643 {
2644 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2645 	const struct rte_flow_item_gre *mask = item->mask;
2646 	int ret;
2647 	const struct rte_flow_item_gre nic_mask = {
2648 		.c_rsvd0_ver = RTE_BE16(0xB000),
2649 		.protocol = RTE_BE16(UINT16_MAX),
2650 	};
2651 
2652 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2653 		return rte_flow_error_set(error, EINVAL,
2654 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2655 					  "protocol filtering not compatible"
2656 					  " with this GRE layer");
2657 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2658 		return rte_flow_error_set(error, ENOTSUP,
2659 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2660 					  "multiple tunnel layers not"
2661 					  " supported");
2662 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2663 		return rte_flow_error_set(error, ENOTSUP,
2664 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2665 					  "L3 Layer is missing");
2666 	if (!mask)
2667 		mask = &rte_flow_item_gre_mask;
2668 	ret = mlx5_flow_item_acceptable
2669 		(item, (const uint8_t *)mask,
2670 		 (const uint8_t *)&nic_mask,
2671 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2672 		 error);
2673 	if (ret < 0)
2674 		return ret;
2675 #ifndef HAVE_MLX5DV_DR
2676 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2677 	if (spec && (spec->protocol & mask->protocol))
2678 		return rte_flow_error_set(error, ENOTSUP,
2679 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2680 					  "without MPLS support the"
2681 					  " specification cannot be used for"
2682 					  " filtering");
2683 #endif
2684 #endif
2685 	return 0;
2686 }
2687 
2688 /**
2689  * Validate Geneve item.
2690  *
2691  * @param[in] item
2692  *   Item specification.
2693  * @param[in] itemFlags
2694  *   Bit-fields that holds the items detected until now.
2695  * @param[in] enPriv
2696  *   Pointer to the private data structure.
2697  * @param[out] error
2698  *   Pointer to error structure.
2699  *
2700  * @return
2701  *   0 on success, a negative errno value otherwise and rte_errno is set.
2702  */
2703 
2704 int
2705 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2706 			       uint64_t item_flags,
2707 			       struct rte_eth_dev *dev,
2708 			       struct rte_flow_error *error)
2709 {
2710 	struct mlx5_priv *priv = dev->data->dev_private;
2711 	const struct rte_flow_item_geneve *spec = item->spec;
2712 	const struct rte_flow_item_geneve *mask = item->mask;
2713 	int ret;
2714 	uint16_t gbhdr;
2715 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2716 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2717 	const struct rte_flow_item_geneve nic_mask = {
2718 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2719 		.vni = "\xff\xff\xff",
2720 		.protocol = RTE_BE16(UINT16_MAX),
2721 	};
2722 
2723 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2724 		return rte_flow_error_set(error, ENOTSUP,
2725 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2726 					  "L3 Geneve is not enabled by device"
2727 					  " parameter and/or not configured in"
2728 					  " firmware");
2729 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2730 		return rte_flow_error_set(error, ENOTSUP,
2731 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2732 					  "multiple tunnel layers not"
2733 					  " supported");
2734 	/*
2735 	 * Verify only UDPv4 is present as defined in
2736 	 * https://tools.ietf.org/html/rfc7348
2737 	 */
2738 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2739 		return rte_flow_error_set(error, EINVAL,
2740 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2741 					  "no outer UDP layer found");
2742 	if (!mask)
2743 		mask = &rte_flow_item_geneve_mask;
2744 	ret = mlx5_flow_item_acceptable
2745 				  (item, (const uint8_t *)mask,
2746 				   (const uint8_t *)&nic_mask,
2747 				   sizeof(struct rte_flow_item_geneve),
2748 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2749 	if (ret)
2750 		return ret;
2751 	if (spec) {
2752 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2753 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2754 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2755 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2756 			return rte_flow_error_set(error, ENOTSUP,
2757 						  RTE_FLOW_ERROR_TYPE_ITEM,
2758 						  item,
2759 						  "Geneve protocol unsupported"
2760 						  " fields are being used");
2761 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2762 			return rte_flow_error_set
2763 					(error, ENOTSUP,
2764 					 RTE_FLOW_ERROR_TYPE_ITEM,
2765 					 item,
2766 					 "Unsupported Geneve options length");
2767 	}
2768 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2769 		return rte_flow_error_set
2770 				    (error, ENOTSUP,
2771 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 				     "Geneve tunnel must be fully defined");
2773 	return 0;
2774 }
2775 
2776 /**
2777  * Validate Geneve TLV option item.
2778  *
2779  * @param[in] item
2780  *   Item specification.
2781  * @param[in] last_item
2782  *   Previous validated item in the pattern items.
2783  * @param[in] geneve_item
2784  *   Previous GENEVE item specification.
2785  * @param[in] dev
2786  *   Pointer to the rte_eth_dev structure.
2787  * @param[out] error
2788  *   Pointer to error structure.
2789  *
2790  * @return
2791  *   0 on success, a negative errno value otherwise and rte_errno is set.
2792  */
2793 int
2794 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2795 				   uint64_t last_item,
2796 				   const struct rte_flow_item *geneve_item,
2797 				   struct rte_eth_dev *dev,
2798 				   struct rte_flow_error *error)
2799 {
2800 	struct mlx5_priv *priv = dev->data->dev_private;
2801 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2802 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2803 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2804 	uint8_t data_max_supported =
2805 			hca_attr->max_geneve_tlv_option_data_len * 4;
2806 	struct mlx5_dev_config *config = &priv->config;
2807 	const struct rte_flow_item_geneve *geneve_spec;
2808 	const struct rte_flow_item_geneve *geneve_mask;
2809 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2810 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2811 	unsigned int i;
2812 	unsigned int data_len;
2813 	uint8_t tlv_option_len;
2814 	uint16_t optlen_m, optlen_v;
2815 	const struct rte_flow_item_geneve_opt full_mask = {
2816 		.option_class = RTE_BE16(0xffff),
2817 		.option_type = 0xff,
2818 		.option_len = 0x1f,
2819 	};
2820 
2821 	if (!mask)
2822 		mask = &rte_flow_item_geneve_opt_mask;
2823 	if (!spec)
2824 		return rte_flow_error_set
2825 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2826 			"Geneve TLV opt class/type/length must be specified");
2827 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2828 		return rte_flow_error_set
2829 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2830 			"Geneve TLV opt length exceeeds the limit (31)");
2831 	/* Check if class type and length masks are full. */
2832 	if (full_mask.option_class != mask->option_class ||
2833 	    full_mask.option_type != mask->option_type ||
2834 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2835 		return rte_flow_error_set
2836 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2837 			"Geneve TLV opt class/type/length masks must be full");
2838 	/* Check if length is supported */
2839 	if ((uint32_t)spec->option_len >
2840 			config->hca_attr.max_geneve_tlv_option_data_len)
2841 		return rte_flow_error_set
2842 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2843 			"Geneve TLV opt length not supported");
2844 	if (config->hca_attr.max_geneve_tlv_options > 1)
2845 		DRV_LOG(DEBUG,
2846 			"max_geneve_tlv_options supports more than 1 option");
2847 	/* Check GENEVE item preceding. */
2848 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2849 		return rte_flow_error_set
2850 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2851 			"Geneve opt item must be preceded with Geneve item");
2852 	geneve_spec = geneve_item->spec;
2853 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2854 					  &rte_flow_item_geneve_mask;
2855 	/* Check if GENEVE TLV option size doesn't exceed option length */
2856 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2857 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2858 		tlv_option_len = spec->option_len & mask->option_len;
2859 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2860 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2861 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2862 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2863 		if ((optlen_v & optlen_m) <= tlv_option_len)
2864 			return rte_flow_error_set
2865 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2866 				 "GENEVE TLV option length exceeds optlen");
2867 	}
2868 	/* Check if length is 0 or data is 0. */
2869 	if (spec->data == NULL || spec->option_len == 0)
2870 		return rte_flow_error_set
2871 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2872 			"Geneve TLV opt with zero data/length not supported");
2873 	/* Check not all data & mask are 0. */
2874 	data_len = spec->option_len * 4;
2875 	if (mask->data == NULL) {
2876 		for (i = 0; i < data_len; i++)
2877 			if (spec->data[i])
2878 				break;
2879 		if (i == data_len)
2880 			return rte_flow_error_set(error, ENOTSUP,
2881 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2882 				"Can't match on Geneve option data 0");
2883 	} else {
2884 		for (i = 0; i < data_len; i++)
2885 			if (spec->data[i] & mask->data[i])
2886 				break;
2887 		if (i == data_len)
2888 			return rte_flow_error_set(error, ENOTSUP,
2889 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2890 				"Can't match on Geneve option data and mask 0");
2891 		/* Check data mask supported. */
2892 		for (i = data_max_supported; i < data_len ; i++)
2893 			if (mask->data[i])
2894 				return rte_flow_error_set(error, ENOTSUP,
2895 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2896 					"Data mask is of unsupported size");
2897 	}
2898 	/* Check GENEVE option is supported in NIC. */
2899 	if (!config->hca_attr.geneve_tlv_opt)
2900 		return rte_flow_error_set
2901 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2902 			"Geneve TLV opt not supported");
2903 	/* Check if we already have geneve option with different type/class. */
2904 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2905 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2906 	if (geneve_opt_resource != NULL)
2907 		if (geneve_opt_resource->option_class != spec->option_class ||
2908 		    geneve_opt_resource->option_type != spec->option_type ||
2909 		    geneve_opt_resource->length != spec->option_len) {
2910 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2911 			return rte_flow_error_set(error, ENOTSUP,
2912 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2913 				"Only one Geneve TLV option supported");
2914 		}
2915 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2916 	return 0;
2917 }
2918 
2919 /**
2920  * Validate MPLS item.
2921  *
2922  * @param[in] dev
2923  *   Pointer to the rte_eth_dev structure.
2924  * @param[in] item
2925  *   Item specification.
2926  * @param[in] item_flags
2927  *   Bit-fields that holds the items detected until now.
2928  * @param[in] prev_layer
2929  *   The protocol layer indicated in previous item.
2930  * @param[out] error
2931  *   Pointer to error structure.
2932  *
2933  * @return
2934  *   0 on success, a negative errno value otherwise and rte_errno is set.
2935  */
2936 int
2937 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2938 			     const struct rte_flow_item *item __rte_unused,
2939 			     uint64_t item_flags __rte_unused,
2940 			     uint64_t prev_layer __rte_unused,
2941 			     struct rte_flow_error *error)
2942 {
2943 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2944 	const struct rte_flow_item_mpls *mask = item->mask;
2945 	struct mlx5_priv *priv = dev->data->dev_private;
2946 	int ret;
2947 
2948 	if (!priv->config.mpls_en)
2949 		return rte_flow_error_set(error, ENOTSUP,
2950 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2951 					  "MPLS not supported or"
2952 					  " disabled in firmware"
2953 					  " configuration.");
2954 	/* MPLS over UDP, GRE is allowed */
2955 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2956 			    MLX5_FLOW_LAYER_GRE |
2957 			    MLX5_FLOW_LAYER_GRE_KEY)))
2958 		return rte_flow_error_set(error, EINVAL,
2959 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2960 					  "protocol filtering not compatible"
2961 					  " with MPLS layer");
2962 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2963 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2964 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2965 		return rte_flow_error_set(error, ENOTSUP,
2966 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 					  "multiple tunnel layers not"
2968 					  " supported");
2969 	if (!mask)
2970 		mask = &rte_flow_item_mpls_mask;
2971 	ret = mlx5_flow_item_acceptable
2972 		(item, (const uint8_t *)mask,
2973 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2974 		 sizeof(struct rte_flow_item_mpls),
2975 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2976 	if (ret < 0)
2977 		return ret;
2978 	return 0;
2979 #else
2980 	return rte_flow_error_set(error, ENOTSUP,
2981 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2982 				  "MPLS is not supported by Verbs, please"
2983 				  " update.");
2984 #endif
2985 }
2986 
2987 /**
2988  * Validate NVGRE item.
2989  *
2990  * @param[in] item
2991  *   Item specification.
2992  * @param[in] item_flags
2993  *   Bit flags to mark detected items.
2994  * @param[in] target_protocol
2995  *   The next protocol in the previous item.
2996  * @param[out] error
2997  *   Pointer to error structure.
2998  *
2999  * @return
3000  *   0 on success, a negative errno value otherwise and rte_errno is set.
3001  */
3002 int
3003 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3004 			      uint64_t item_flags,
3005 			      uint8_t target_protocol,
3006 			      struct rte_flow_error *error)
3007 {
3008 	const struct rte_flow_item_nvgre *mask = item->mask;
3009 	int ret;
3010 
3011 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3012 		return rte_flow_error_set(error, EINVAL,
3013 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3014 					  "protocol filtering not compatible"
3015 					  " with this GRE layer");
3016 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3017 		return rte_flow_error_set(error, ENOTSUP,
3018 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3019 					  "multiple tunnel layers not"
3020 					  " supported");
3021 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3022 		return rte_flow_error_set(error, ENOTSUP,
3023 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 					  "L3 Layer is missing");
3025 	if (!mask)
3026 		mask = &rte_flow_item_nvgre_mask;
3027 	ret = mlx5_flow_item_acceptable
3028 		(item, (const uint8_t *)mask,
3029 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3030 		 sizeof(struct rte_flow_item_nvgre),
3031 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3032 	if (ret < 0)
3033 		return ret;
3034 	return 0;
3035 }
3036 
3037 /**
3038  * Validate eCPRI item.
3039  *
3040  * @param[in] item
3041  *   Item specification.
3042  * @param[in] item_flags
3043  *   Bit-fields that holds the items detected until now.
3044  * @param[in] last_item
3045  *   Previous validated item in the pattern items.
3046  * @param[in] ether_type
3047  *   Type in the ethernet layer header (including dot1q).
3048  * @param[in] acc_mask
3049  *   Acceptable mask, if NULL default internal default mask
3050  *   will be used to check whether item fields are supported.
3051  * @param[out] error
3052  *   Pointer to error structure.
3053  *
3054  * @return
3055  *   0 on success, a negative errno value otherwise and rte_errno is set.
3056  */
3057 int
3058 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3059 			      uint64_t item_flags,
3060 			      uint64_t last_item,
3061 			      uint16_t ether_type,
3062 			      const struct rte_flow_item_ecpri *acc_mask,
3063 			      struct rte_flow_error *error)
3064 {
3065 	const struct rte_flow_item_ecpri *mask = item->mask;
3066 	const struct rte_flow_item_ecpri nic_mask = {
3067 		.hdr = {
3068 			.common = {
3069 				.u32 =
3070 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3071 					.type = 0xFF,
3072 					}).u32),
3073 			},
3074 			.dummy[0] = 0xFFFFFFFF,
3075 		},
3076 	};
3077 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3078 					MLX5_FLOW_LAYER_OUTER_VLAN);
3079 	struct rte_flow_item_ecpri mask_lo;
3080 
3081 	if (!(last_item & outer_l2_vlan) &&
3082 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3083 		return rte_flow_error_set(error, EINVAL,
3084 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3085 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3086 	if ((last_item & outer_l2_vlan) && ether_type &&
3087 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3088 		return rte_flow_error_set(error, EINVAL,
3089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3090 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3091 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3092 		return rte_flow_error_set(error, EINVAL,
3093 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3094 					  "eCPRI with tunnel is not supported right now");
3095 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3096 		return rte_flow_error_set(error, ENOTSUP,
3097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3098 					  "multiple L3 layers not supported");
3099 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3100 		return rte_flow_error_set(error, EINVAL,
3101 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3102 					  "eCPRI cannot coexist with a TCP layer");
3103 	/* In specification, eCPRI could be over UDP layer. */
3104 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3105 		return rte_flow_error_set(error, EINVAL,
3106 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 					  "eCPRI over UDP layer is not yet supported right now");
3108 	/* Mask for type field in common header could be zero. */
3109 	if (!mask)
3110 		mask = &rte_flow_item_ecpri_mask;
3111 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3112 	/* Input mask is in big-endian format. */
3113 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3114 		return rte_flow_error_set(error, EINVAL,
3115 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3116 					  "partial mask is not supported for protocol");
3117 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3118 		return rte_flow_error_set(error, EINVAL,
3119 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3120 					  "message header mask must be after a type mask");
3121 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3122 					 acc_mask ? (const uint8_t *)acc_mask
3123 						  : (const uint8_t *)&nic_mask,
3124 					 sizeof(struct rte_flow_item_ecpri),
3125 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3126 }
3127 
3128 static int
3129 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3130 		   const struct rte_flow_attr *attr __rte_unused,
3131 		   const struct rte_flow_item items[] __rte_unused,
3132 		   const struct rte_flow_action actions[] __rte_unused,
3133 		   bool external __rte_unused,
3134 		   int hairpin __rte_unused,
3135 		   struct rte_flow_error *error)
3136 {
3137 	return rte_flow_error_set(error, ENOTSUP,
3138 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3139 }
3140 
3141 static struct mlx5_flow *
3142 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3143 		  const struct rte_flow_attr *attr __rte_unused,
3144 		  const struct rte_flow_item items[] __rte_unused,
3145 		  const struct rte_flow_action actions[] __rte_unused,
3146 		  struct rte_flow_error *error)
3147 {
3148 	rte_flow_error_set(error, ENOTSUP,
3149 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3150 	return NULL;
3151 }
3152 
3153 static int
3154 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3155 		    struct mlx5_flow *dev_flow __rte_unused,
3156 		    const struct rte_flow_attr *attr __rte_unused,
3157 		    const struct rte_flow_item items[] __rte_unused,
3158 		    const struct rte_flow_action actions[] __rte_unused,
3159 		    struct rte_flow_error *error)
3160 {
3161 	return rte_flow_error_set(error, ENOTSUP,
3162 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3163 }
3164 
3165 static int
3166 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3167 		struct rte_flow *flow __rte_unused,
3168 		struct rte_flow_error *error)
3169 {
3170 	return rte_flow_error_set(error, ENOTSUP,
3171 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3172 }
3173 
3174 static void
3175 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3176 		 struct rte_flow *flow __rte_unused)
3177 {
3178 }
3179 
3180 static void
3181 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3182 		  struct rte_flow *flow __rte_unused)
3183 {
3184 }
3185 
3186 static int
3187 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3188 		struct rte_flow *flow __rte_unused,
3189 		const struct rte_flow_action *actions __rte_unused,
3190 		void *data __rte_unused,
3191 		struct rte_flow_error *error)
3192 {
3193 	return rte_flow_error_set(error, ENOTSUP,
3194 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3195 }
3196 
3197 static int
3198 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3199 		      uint32_t domains __rte_unused,
3200 		      uint32_t flags __rte_unused)
3201 {
3202 	return 0;
3203 }
3204 
3205 /* Void driver to protect from null pointer reference. */
3206 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3207 	.validate = flow_null_validate,
3208 	.prepare = flow_null_prepare,
3209 	.translate = flow_null_translate,
3210 	.apply = flow_null_apply,
3211 	.remove = flow_null_remove,
3212 	.destroy = flow_null_destroy,
3213 	.query = flow_null_query,
3214 	.sync_domain = flow_null_sync_domain,
3215 };
3216 
3217 /**
3218  * Select flow driver type according to flow attributes and device
3219  * configuration.
3220  *
3221  * @param[in] dev
3222  *   Pointer to the dev structure.
3223  * @param[in] attr
3224  *   Pointer to the flow attributes.
3225  *
3226  * @return
3227  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3228  */
3229 static enum mlx5_flow_drv_type
3230 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3231 {
3232 	struct mlx5_priv *priv = dev->data->dev_private;
3233 	/* The OS can determine first a specific flow type (DV, VERBS) */
3234 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3235 
3236 	if (type != MLX5_FLOW_TYPE_MAX)
3237 		return type;
3238 	/* If no OS specific type - continue with DV/VERBS selection */
3239 	if (attr->transfer && priv->config.dv_esw_en)
3240 		type = MLX5_FLOW_TYPE_DV;
3241 	if (!attr->transfer)
3242 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3243 						 MLX5_FLOW_TYPE_VERBS;
3244 	return type;
3245 }
3246 
3247 #define flow_get_drv_ops(type) flow_drv_ops[type]
3248 
3249 /**
3250  * Flow driver validation API. This abstracts calling driver specific functions.
3251  * The type of flow driver is determined according to flow attributes.
3252  *
3253  * @param[in] dev
3254  *   Pointer to the dev structure.
3255  * @param[in] attr
3256  *   Pointer to the flow attributes.
3257  * @param[in] items
3258  *   Pointer to the list of items.
3259  * @param[in] actions
3260  *   Pointer to the list of actions.
3261  * @param[in] external
3262  *   This flow rule is created by request external to PMD.
3263  * @param[in] hairpin
3264  *   Number of hairpin TX actions, 0 means classic flow.
3265  * @param[out] error
3266  *   Pointer to the error structure.
3267  *
3268  * @return
3269  *   0 on success, a negative errno value otherwise and rte_errno is set.
3270  */
3271 static inline int
3272 flow_drv_validate(struct rte_eth_dev *dev,
3273 		  const struct rte_flow_attr *attr,
3274 		  const struct rte_flow_item items[],
3275 		  const struct rte_flow_action actions[],
3276 		  bool external, int hairpin, struct rte_flow_error *error)
3277 {
3278 	const struct mlx5_flow_driver_ops *fops;
3279 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3280 
3281 	fops = flow_get_drv_ops(type);
3282 	return fops->validate(dev, attr, items, actions, external,
3283 			      hairpin, error);
3284 }
3285 
3286 /**
3287  * Flow driver preparation API. This abstracts calling driver specific
3288  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3289  * calculates the size of memory required for device flow, allocates the memory,
3290  * initializes the device flow and returns the pointer.
3291  *
3292  * @note
3293  *   This function initializes device flow structure such as dv or verbs in
3294  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3295  *   rest. For example, adding returning device flow to flow->dev_flow list and
3296  *   setting backward reference to the flow should be done out of this function.
3297  *   layers field is not filled either.
3298  *
3299  * @param[in] dev
3300  *   Pointer to the dev structure.
3301  * @param[in] attr
3302  *   Pointer to the flow attributes.
3303  * @param[in] items
3304  *   Pointer to the list of items.
3305  * @param[in] actions
3306  *   Pointer to the list of actions.
3307  * @param[in] flow_idx
3308  *   This memory pool index to the flow.
3309  * @param[out] error
3310  *   Pointer to the error structure.
3311  *
3312  * @return
3313  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3314  */
3315 static inline struct mlx5_flow *
3316 flow_drv_prepare(struct rte_eth_dev *dev,
3317 		 const struct rte_flow *flow,
3318 		 const struct rte_flow_attr *attr,
3319 		 const struct rte_flow_item items[],
3320 		 const struct rte_flow_action actions[],
3321 		 uint32_t flow_idx,
3322 		 struct rte_flow_error *error)
3323 {
3324 	const struct mlx5_flow_driver_ops *fops;
3325 	enum mlx5_flow_drv_type type = flow->drv_type;
3326 	struct mlx5_flow *mlx5_flow = NULL;
3327 
3328 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3329 	fops = flow_get_drv_ops(type);
3330 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3331 	if (mlx5_flow)
3332 		mlx5_flow->flow_idx = flow_idx;
3333 	return mlx5_flow;
3334 }
3335 
3336 /**
3337  * Flow driver translation API. This abstracts calling driver specific
3338  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3339  * translates a generic flow into a driver flow. flow_drv_prepare() must
3340  * precede.
3341  *
3342  * @note
3343  *   dev_flow->layers could be filled as a result of parsing during translation
3344  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3345  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3346  *   flow->actions could be overwritten even though all the expanded dev_flows
3347  *   have the same actions.
3348  *
3349  * @param[in] dev
3350  *   Pointer to the rte dev structure.
3351  * @param[in, out] dev_flow
3352  *   Pointer to the mlx5 flow.
3353  * @param[in] attr
3354  *   Pointer to the flow attributes.
3355  * @param[in] items
3356  *   Pointer to the list of items.
3357  * @param[in] actions
3358  *   Pointer to the list of actions.
3359  * @param[out] error
3360  *   Pointer to the error structure.
3361  *
3362  * @return
3363  *   0 on success, a negative errno value otherwise and rte_errno is set.
3364  */
3365 static inline int
3366 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3367 		   const struct rte_flow_attr *attr,
3368 		   const struct rte_flow_item items[],
3369 		   const struct rte_flow_action actions[],
3370 		   struct rte_flow_error *error)
3371 {
3372 	const struct mlx5_flow_driver_ops *fops;
3373 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3374 
3375 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3376 	fops = flow_get_drv_ops(type);
3377 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3378 }
3379 
3380 /**
3381  * Flow driver apply API. This abstracts calling driver specific functions.
3382  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3383  * translated driver flows on to device. flow_drv_translate() must precede.
3384  *
3385  * @param[in] dev
3386  *   Pointer to Ethernet device structure.
3387  * @param[in, out] flow
3388  *   Pointer to flow structure.
3389  * @param[out] error
3390  *   Pointer to error structure.
3391  *
3392  * @return
3393  *   0 on success, a negative errno value otherwise and rte_errno is set.
3394  */
3395 static inline int
3396 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3397 	       struct rte_flow_error *error)
3398 {
3399 	const struct mlx5_flow_driver_ops *fops;
3400 	enum mlx5_flow_drv_type type = flow->drv_type;
3401 
3402 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3403 	fops = flow_get_drv_ops(type);
3404 	return fops->apply(dev, flow, error);
3405 }
3406 
3407 /**
3408  * Flow driver destroy API. This abstracts calling driver specific functions.
3409  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3410  * on device and releases resources of the flow.
3411  *
3412  * @param[in] dev
3413  *   Pointer to Ethernet device.
3414  * @param[in, out] flow
3415  *   Pointer to flow structure.
3416  */
3417 static inline void
3418 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3419 {
3420 	const struct mlx5_flow_driver_ops *fops;
3421 	enum mlx5_flow_drv_type type = flow->drv_type;
3422 
3423 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3424 	fops = flow_get_drv_ops(type);
3425 	fops->destroy(dev, flow);
3426 }
3427 
3428 /**
3429  * Flow driver find RSS policy tbl API. This abstracts calling driver
3430  * specific functions. Parent flow (rte_flow) should have driver
3431  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3432  *
3433  * @param[in] dev
3434  *   Pointer to Ethernet device.
3435  * @param[in, out] flow
3436  *   Pointer to flow structure.
3437  * @param[in] policy
3438  *   Pointer to meter policy table.
3439  * @param[in] rss_desc
3440  *   Pointer to rss_desc
3441  */
3442 static struct mlx5_flow_meter_sub_policy *
3443 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3444 		struct rte_flow *flow,
3445 		struct mlx5_flow_meter_policy *policy,
3446 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3447 {
3448 	const struct mlx5_flow_driver_ops *fops;
3449 	enum mlx5_flow_drv_type type = flow->drv_type;
3450 
3451 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3452 	fops = flow_get_drv_ops(type);
3453 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3454 }
3455 
3456 /**
3457  * Flow driver color tag rule API. This abstracts calling driver
3458  * specific functions. Parent flow (rte_flow) should have driver
3459  * type (drv_type). It will create the color tag rules in hierarchy meter.
3460  *
3461  * @param[in] dev
3462  *   Pointer to Ethernet device.
3463  * @param[in, out] flow
3464  *   Pointer to flow structure.
3465  * @param[in] fm
3466  *   Pointer to flow meter structure.
3467  * @param[in] src_port
3468  *   The src port this extra rule should use.
3469  * @param[in] item
3470  *   The src port id match item.
3471  * @param[out] error
3472  *   Pointer to error structure.
3473  */
3474 static int
3475 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3476 		struct rte_flow *flow,
3477 		struct mlx5_flow_meter_info *fm,
3478 		int32_t src_port,
3479 		const struct rte_flow_item *item,
3480 		struct rte_flow_error *error)
3481 {
3482 	const struct mlx5_flow_driver_ops *fops;
3483 	enum mlx5_flow_drv_type type = flow->drv_type;
3484 
3485 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3486 	fops = flow_get_drv_ops(type);
3487 	return fops->meter_hierarchy_rule_create(dev, fm,
3488 						src_port, item, error);
3489 }
3490 
3491 /**
3492  * Get RSS action from the action list.
3493  *
3494  * @param[in] dev
3495  *   Pointer to Ethernet device.
3496  * @param[in] actions
3497  *   Pointer to the list of actions.
3498  * @param[in] flow
3499  *   Parent flow structure pointer.
3500  *
3501  * @return
3502  *   Pointer to the RSS action if exist, else return NULL.
3503  */
3504 static const struct rte_flow_action_rss*
3505 flow_get_rss_action(struct rte_eth_dev *dev,
3506 		    const struct rte_flow_action actions[])
3507 {
3508 	struct mlx5_priv *priv = dev->data->dev_private;
3509 	const struct rte_flow_action_rss *rss = NULL;
3510 
3511 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3512 		switch (actions->type) {
3513 		case RTE_FLOW_ACTION_TYPE_RSS:
3514 			rss = actions->conf;
3515 			break;
3516 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3517 		{
3518 			const struct rte_flow_action_sample *sample =
3519 								actions->conf;
3520 			const struct rte_flow_action *act = sample->actions;
3521 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3522 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3523 					rss = act->conf;
3524 			break;
3525 		}
3526 		case RTE_FLOW_ACTION_TYPE_METER:
3527 		{
3528 			uint32_t mtr_idx;
3529 			struct mlx5_flow_meter_info *fm;
3530 			struct mlx5_flow_meter_policy *policy;
3531 			const struct rte_flow_action_meter *mtr = actions->conf;
3532 
3533 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3534 			if (fm && !fm->def_policy) {
3535 				policy = mlx5_flow_meter_policy_find(dev,
3536 						fm->policy_id, NULL);
3537 				MLX5_ASSERT(policy);
3538 				if (policy->is_hierarchy) {
3539 					policy =
3540 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3541 									policy);
3542 					if (!policy)
3543 						return NULL;
3544 				}
3545 				if (policy->is_rss)
3546 					rss =
3547 				policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3548 			}
3549 			break;
3550 		}
3551 		default:
3552 			break;
3553 		}
3554 	}
3555 	return rss;
3556 }
3557 
3558 /**
3559  * Get ASO age action by index.
3560  *
3561  * @param[in] dev
3562  *   Pointer to the Ethernet device structure.
3563  * @param[in] age_idx
3564  *   Index to the ASO age action.
3565  *
3566  * @return
3567  *   The specified ASO age action.
3568  */
3569 struct mlx5_aso_age_action*
3570 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3571 {
3572 	uint16_t pool_idx = age_idx & UINT16_MAX;
3573 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3574 	struct mlx5_priv *priv = dev->data->dev_private;
3575 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3576 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3577 
3578 	return &pool->actions[offset - 1];
3579 }
3580 
3581 /* maps indirect action to translated direct in some actions array */
3582 struct mlx5_translated_action_handle {
3583 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3584 	int index; /**< Index in related array of rte_flow_action. */
3585 };
3586 
3587 /**
3588  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3589  * direct action if translation possible.
3590  * This functionality used to run same execution path for both direct and
3591  * indirect actions on flow create. All necessary preparations for indirect
3592  * action handling should be performed on *handle* actions list returned
3593  * from this call.
3594  *
3595  * @param[in] dev
3596  *   Pointer to Ethernet device.
3597  * @param[in] actions
3598  *   List of actions to translate.
3599  * @param[out] handle
3600  *   List to store translated indirect action object handles.
3601  * @param[in, out] indir_n
3602  *   Size of *handle* array. On return should be updated with number of
3603  *   indirect actions retrieved from the *actions* list.
3604  * @param[out] translated_actions
3605  *   List of actions where all indirect actions were translated to direct
3606  *   if possible. NULL if no translation took place.
3607  * @param[out] error
3608  *   Pointer to the error structure.
3609  *
3610  * @return
3611  *   0 on success, a negative errno value otherwise and rte_errno is set.
3612  */
3613 static int
3614 flow_action_handles_translate(struct rte_eth_dev *dev,
3615 			      const struct rte_flow_action actions[],
3616 			      struct mlx5_translated_action_handle *handle,
3617 			      int *indir_n,
3618 			      struct rte_flow_action **translated_actions,
3619 			      struct rte_flow_error *error)
3620 {
3621 	struct mlx5_priv *priv = dev->data->dev_private;
3622 	struct rte_flow_action *translated = NULL;
3623 	size_t actions_size;
3624 	int n;
3625 	int copied_n = 0;
3626 	struct mlx5_translated_action_handle *handle_end = NULL;
3627 
3628 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3629 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3630 			continue;
3631 		if (copied_n == *indir_n) {
3632 			return rte_flow_error_set
3633 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3634 				 NULL, "too many shared actions");
3635 		}
3636 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3637 			   sizeof(actions[n].conf));
3638 		handle[copied_n].index = n;
3639 		copied_n++;
3640 	}
3641 	n++;
3642 	*indir_n = copied_n;
3643 	if (!copied_n)
3644 		return 0;
3645 	actions_size = sizeof(struct rte_flow_action) * n;
3646 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3647 	if (!translated) {
3648 		rte_errno = ENOMEM;
3649 		return -ENOMEM;
3650 	}
3651 	memcpy(translated, actions, actions_size);
3652 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3653 		struct mlx5_shared_action_rss *shared_rss;
3654 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3655 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3656 		uint32_t idx = act_idx &
3657 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3658 
3659 		switch (type) {
3660 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3661 			shared_rss = mlx5_ipool_get
3662 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3663 			translated[handle->index].type =
3664 				RTE_FLOW_ACTION_TYPE_RSS;
3665 			translated[handle->index].conf =
3666 				&shared_rss->origin;
3667 			break;
3668 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3669 			translated[handle->index].type =
3670 						(enum rte_flow_action_type)
3671 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3672 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3673 			break;
3674 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3675 			if (priv->sh->flow_hit_aso_en) {
3676 				translated[handle->index].type =
3677 					(enum rte_flow_action_type)
3678 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3679 				translated[handle->index].conf =
3680 							 (void *)(uintptr_t)idx;
3681 				break;
3682 			}
3683 			/* Fall-through */
3684 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3685 			if (priv->sh->ct_aso_en) {
3686 				translated[handle->index].type =
3687 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3688 				translated[handle->index].conf =
3689 							 (void *)(uintptr_t)idx;
3690 				break;
3691 			}
3692 			/* Fall-through */
3693 		default:
3694 			mlx5_free(translated);
3695 			return rte_flow_error_set
3696 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3697 				 NULL, "invalid indirect action type");
3698 		}
3699 	}
3700 	*translated_actions = translated;
3701 	return 0;
3702 }
3703 
3704 /**
3705  * Get Shared RSS action from the action list.
3706  *
3707  * @param[in] dev
3708  *   Pointer to Ethernet device.
3709  * @param[in] shared
3710  *   Pointer to the list of actions.
3711  * @param[in] shared_n
3712  *   Actions list length.
3713  *
3714  * @return
3715  *   The MLX5 RSS action ID if exists, otherwise return 0.
3716  */
3717 static uint32_t
3718 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3719 			   struct mlx5_translated_action_handle *handle,
3720 			   int shared_n)
3721 {
3722 	struct mlx5_translated_action_handle *handle_end;
3723 	struct mlx5_priv *priv = dev->data->dev_private;
3724 	struct mlx5_shared_action_rss *shared_rss;
3725 
3726 
3727 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3728 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3729 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3730 		uint32_t idx = act_idx &
3731 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3732 		switch (type) {
3733 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3734 			shared_rss = mlx5_ipool_get
3735 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3736 									   idx);
3737 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3738 					   __ATOMIC_RELAXED);
3739 			return idx;
3740 		default:
3741 			break;
3742 		}
3743 	}
3744 	return 0;
3745 }
3746 
3747 static unsigned int
3748 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3749 {
3750 	const struct rte_flow_item *item;
3751 	unsigned int has_vlan = 0;
3752 
3753 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3754 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3755 			has_vlan = 1;
3756 			break;
3757 		}
3758 	}
3759 	if (has_vlan)
3760 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3761 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3762 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3763 			       MLX5_EXPANSION_ROOT_OUTER;
3764 }
3765 
3766 /**
3767  *  Get layer flags from the prefix flow.
3768  *
3769  *  Some flows may be split to several subflows, the prefix subflow gets the
3770  *  match items and the suffix sub flow gets the actions.
3771  *  Some actions need the user defined match item flags to get the detail for
3772  *  the action.
3773  *  This function helps the suffix flow to get the item layer flags from prefix
3774  *  subflow.
3775  *
3776  * @param[in] dev_flow
3777  *   Pointer the created preifx subflow.
3778  *
3779  * @return
3780  *   The layers get from prefix subflow.
3781  */
3782 static inline uint64_t
3783 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3784 {
3785 	uint64_t layers = 0;
3786 
3787 	/*
3788 	 * Layers bits could be localization, but usually the compiler will
3789 	 * help to do the optimization work for source code.
3790 	 * If no decap actions, use the layers directly.
3791 	 */
3792 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3793 		return dev_flow->handle->layers;
3794 	/* Convert L3 layers with decap action. */
3795 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3796 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3797 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3798 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3799 	/* Convert L4 layers with decap action.  */
3800 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3801 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3802 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3803 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3804 	return layers;
3805 }
3806 
3807 /**
3808  * Get metadata split action information.
3809  *
3810  * @param[in] actions
3811  *   Pointer to the list of actions.
3812  * @param[out] qrss
3813  *   Pointer to the return pointer.
3814  * @param[out] qrss_type
3815  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3816  *   if no QUEUE/RSS is found.
3817  * @param[out] encap_idx
3818  *   Pointer to the index of the encap action if exists, otherwise the last
3819  *   action index.
3820  *
3821  * @return
3822  *   Total number of actions.
3823  */
3824 static int
3825 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3826 				       const struct rte_flow_action **qrss,
3827 				       int *encap_idx)
3828 {
3829 	const struct rte_flow_action_raw_encap *raw_encap;
3830 	int actions_n = 0;
3831 	int raw_decap_idx = -1;
3832 
3833 	*encap_idx = -1;
3834 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3835 		switch (actions->type) {
3836 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3837 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3838 			*encap_idx = actions_n;
3839 			break;
3840 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3841 			raw_decap_idx = actions_n;
3842 			break;
3843 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3844 			raw_encap = actions->conf;
3845 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3846 				*encap_idx = raw_decap_idx != -1 ?
3847 						      raw_decap_idx : actions_n;
3848 			break;
3849 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3850 		case RTE_FLOW_ACTION_TYPE_RSS:
3851 			*qrss = actions;
3852 			break;
3853 		default:
3854 			break;
3855 		}
3856 		actions_n++;
3857 	}
3858 	if (*encap_idx == -1)
3859 		*encap_idx = actions_n;
3860 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3861 	return actions_n + 1;
3862 }
3863 
3864 /**
3865  * Check if the action will change packet.
3866  *
3867  * @param dev
3868  *   Pointer to Ethernet device.
3869  * @param[in] type
3870  *   action type.
3871  *
3872  * @return
3873  *   true if action will change packet, false otherwise.
3874  */
3875 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3876 					  enum rte_flow_action_type type)
3877 {
3878 	struct mlx5_priv *priv = dev->data->dev_private;
3879 
3880 	switch (type) {
3881 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3882 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3883 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3884 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3885 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3886 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3887 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3888 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3889 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3890 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
3891 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3892 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3893 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3894 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3895 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3896 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3897 	case RTE_FLOW_ACTION_TYPE_SET_META:
3898 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
3899 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3900 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3901 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3902 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3903 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3904 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3905 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3906 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3907 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3908 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3909 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3910 		return true;
3911 	case RTE_FLOW_ACTION_TYPE_FLAG:
3912 	case RTE_FLOW_ACTION_TYPE_MARK:
3913 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3914 			return true;
3915 		else
3916 			return false;
3917 	default:
3918 		return false;
3919 	}
3920 }
3921 
3922 /**
3923  * Check meter action from the action list.
3924  *
3925  * @param dev
3926  *   Pointer to Ethernet device.
3927  * @param[in] actions
3928  *   Pointer to the list of actions.
3929  * @param[out] has_mtr
3930  *   Pointer to the meter exist flag.
3931  * @param[out] has_modify
3932  *   Pointer to the flag showing there's packet change action.
3933  * @param[out] meter_id
3934  *   Pointer to the meter id.
3935  *
3936  * @return
3937  *   Total number of actions.
3938  */
3939 static int
3940 flow_check_meter_action(struct rte_eth_dev *dev,
3941 			const struct rte_flow_action actions[],
3942 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3943 {
3944 	const struct rte_flow_action_meter *mtr = NULL;
3945 	int actions_n = 0;
3946 
3947 	MLX5_ASSERT(has_mtr);
3948 	*has_mtr = false;
3949 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3950 		switch (actions->type) {
3951 		case RTE_FLOW_ACTION_TYPE_METER:
3952 			mtr = actions->conf;
3953 			*meter_id = mtr->mtr_id;
3954 			*has_mtr = true;
3955 			break;
3956 		default:
3957 			break;
3958 		}
3959 		if (!*has_mtr)
3960 			*has_modify |= flow_check_modify_action_type(dev,
3961 								actions->type);
3962 		actions_n++;
3963 	}
3964 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3965 	return actions_n + 1;
3966 }
3967 
3968 /**
3969  * Check if the flow should be split due to hairpin.
3970  * The reason for the split is that in current HW we can't
3971  * support encap and push-vlan on Rx, so if a flow contains
3972  * these actions we move it to Tx.
3973  *
3974  * @param dev
3975  *   Pointer to Ethernet device.
3976  * @param[in] attr
3977  *   Flow rule attributes.
3978  * @param[in] actions
3979  *   Associated actions (list terminated by the END action).
3980  *
3981  * @return
3982  *   > 0 the number of actions and the flow should be split,
3983  *   0 when no split required.
3984  */
3985 static int
3986 flow_check_hairpin_split(struct rte_eth_dev *dev,
3987 			 const struct rte_flow_attr *attr,
3988 			 const struct rte_flow_action actions[])
3989 {
3990 	int queue_action = 0;
3991 	int action_n = 0;
3992 	int split = 0;
3993 	const struct rte_flow_action_queue *queue;
3994 	const struct rte_flow_action_rss *rss;
3995 	const struct rte_flow_action_raw_encap *raw_encap;
3996 	const struct rte_eth_hairpin_conf *conf;
3997 
3998 	if (!attr->ingress)
3999 		return 0;
4000 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4001 		switch (actions->type) {
4002 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4003 			queue = actions->conf;
4004 			if (queue == NULL)
4005 				return 0;
4006 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4007 			if (conf == NULL || conf->tx_explicit != 0)
4008 				return 0;
4009 			queue_action = 1;
4010 			action_n++;
4011 			break;
4012 		case RTE_FLOW_ACTION_TYPE_RSS:
4013 			rss = actions->conf;
4014 			if (rss == NULL || rss->queue_num == 0)
4015 				return 0;
4016 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4017 			if (conf == NULL || conf->tx_explicit != 0)
4018 				return 0;
4019 			queue_action = 1;
4020 			action_n++;
4021 			break;
4022 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4023 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4024 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4025 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4026 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4027 			split++;
4028 			action_n++;
4029 			break;
4030 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4031 			raw_encap = actions->conf;
4032 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4033 				split++;
4034 			action_n++;
4035 			break;
4036 		default:
4037 			action_n++;
4038 			break;
4039 		}
4040 	}
4041 	if (split && queue_action)
4042 		return action_n;
4043 	return 0;
4044 }
4045 
4046 /* Declare flow create/destroy prototype in advance. */
4047 static uint32_t
4048 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4049 		 const struct rte_flow_attr *attr,
4050 		 const struct rte_flow_item items[],
4051 		 const struct rte_flow_action actions[],
4052 		 bool external, struct rte_flow_error *error);
4053 
4054 static void
4055 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4056 		  uint32_t flow_idx);
4057 
4058 int
4059 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
4060 		      struct mlx5_hlist_entry *entry,
4061 		      uint64_t key, void *cb_ctx __rte_unused)
4062 {
4063 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4064 		container_of(entry, typeof(*mcp_res), hlist_ent);
4065 
4066 	return mcp_res->mark_id != key;
4067 }
4068 
4069 struct mlx5_hlist_entry *
4070 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
4071 		       void *cb_ctx)
4072 {
4073 	struct rte_eth_dev *dev = list->ctx;
4074 	struct mlx5_priv *priv = dev->data->dev_private;
4075 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4076 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4077 	struct rte_flow_error *error = ctx->error;
4078 	uint32_t idx = 0;
4079 	int ret;
4080 	uint32_t mark_id = key;
4081 	struct rte_flow_attr attr = {
4082 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4083 		.ingress = 1,
4084 	};
4085 	struct mlx5_rte_flow_item_tag tag_spec = {
4086 		.data = mark_id,
4087 	};
4088 	struct rte_flow_item items[] = {
4089 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4090 	};
4091 	struct rte_flow_action_mark ftag = {
4092 		.id = mark_id,
4093 	};
4094 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4095 		.dst = REG_B,
4096 		.src = REG_NON,
4097 	};
4098 	struct rte_flow_action_jump jump = {
4099 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4100 	};
4101 	struct rte_flow_action actions[] = {
4102 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4103 	};
4104 
4105 	/* Fill the register fileds in the flow. */
4106 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4107 	if (ret < 0)
4108 		return NULL;
4109 	tag_spec.id = ret;
4110 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4111 	if (ret < 0)
4112 		return NULL;
4113 	cp_mreg.src = ret;
4114 	/* Provide the full width of FLAG specific value. */
4115 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4116 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4117 	/* Build a new flow. */
4118 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4119 		items[0] = (struct rte_flow_item){
4120 			.type = (enum rte_flow_item_type)
4121 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4122 			.spec = &tag_spec,
4123 		};
4124 		items[1] = (struct rte_flow_item){
4125 			.type = RTE_FLOW_ITEM_TYPE_END,
4126 		};
4127 		actions[0] = (struct rte_flow_action){
4128 			.type = (enum rte_flow_action_type)
4129 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4130 			.conf = &ftag,
4131 		};
4132 		actions[1] = (struct rte_flow_action){
4133 			.type = (enum rte_flow_action_type)
4134 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4135 			.conf = &cp_mreg,
4136 		};
4137 		actions[2] = (struct rte_flow_action){
4138 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4139 			.conf = &jump,
4140 		};
4141 		actions[3] = (struct rte_flow_action){
4142 			.type = RTE_FLOW_ACTION_TYPE_END,
4143 		};
4144 	} else {
4145 		/* Default rule, wildcard match. */
4146 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4147 		items[0] = (struct rte_flow_item){
4148 			.type = RTE_FLOW_ITEM_TYPE_END,
4149 		};
4150 		actions[0] = (struct rte_flow_action){
4151 			.type = (enum rte_flow_action_type)
4152 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4153 			.conf = &cp_mreg,
4154 		};
4155 		actions[1] = (struct rte_flow_action){
4156 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4157 			.conf = &jump,
4158 		};
4159 		actions[2] = (struct rte_flow_action){
4160 			.type = RTE_FLOW_ACTION_TYPE_END,
4161 		};
4162 	}
4163 	/* Build a new entry. */
4164 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4165 	if (!mcp_res) {
4166 		rte_errno = ENOMEM;
4167 		return NULL;
4168 	}
4169 	mcp_res->idx = idx;
4170 	mcp_res->mark_id = mark_id;
4171 	/*
4172 	 * The copy Flows are not included in any list. There
4173 	 * ones are referenced from other Flows and can not
4174 	 * be applied, removed, deleted in ardbitrary order
4175 	 * by list traversing.
4176 	 */
4177 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4178 					&attr, items, actions, false, error);
4179 	if (!mcp_res->rix_flow) {
4180 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4181 		return NULL;
4182 	}
4183 	return &mcp_res->hlist_ent;
4184 }
4185 
4186 /**
4187  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4188  *
4189  * As mark_id is unique, if there's already a registered flow for the mark_id,
4190  * return by increasing the reference counter of the resource. Otherwise, create
4191  * the resource (mcp_res) and flow.
4192  *
4193  * Flow looks like,
4194  *   - If ingress port is ANY and reg_c[1] is mark_id,
4195  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4196  *
4197  * For default flow (zero mark_id), flow is like,
4198  *   - If ingress port is ANY,
4199  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4200  *
4201  * @param dev
4202  *   Pointer to Ethernet device.
4203  * @param mark_id
4204  *   ID of MARK action, zero means default flow for META.
4205  * @param[out] error
4206  *   Perform verbose error reporting if not NULL.
4207  *
4208  * @return
4209  *   Associated resource on success, NULL otherwise and rte_errno is set.
4210  */
4211 static struct mlx5_flow_mreg_copy_resource *
4212 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4213 			  struct rte_flow_error *error)
4214 {
4215 	struct mlx5_priv *priv = dev->data->dev_private;
4216 	struct mlx5_hlist_entry *entry;
4217 	struct mlx5_flow_cb_ctx ctx = {
4218 		.dev = dev,
4219 		.error = error,
4220 	};
4221 
4222 	/* Check if already registered. */
4223 	MLX5_ASSERT(priv->mreg_cp_tbl);
4224 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4225 	if (!entry)
4226 		return NULL;
4227 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4228 			    hlist_ent);
4229 }
4230 
4231 void
4232 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4233 {
4234 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4235 		container_of(entry, typeof(*mcp_res), hlist_ent);
4236 	struct rte_eth_dev *dev = list->ctx;
4237 	struct mlx5_priv *priv = dev->data->dev_private;
4238 
4239 	MLX5_ASSERT(mcp_res->rix_flow);
4240 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4241 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4242 }
4243 
4244 /**
4245  * Release flow in RX_CP_TBL.
4246  *
4247  * @param dev
4248  *   Pointer to Ethernet device.
4249  * @flow
4250  *   Parent flow for wich copying is provided.
4251  */
4252 static void
4253 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4254 			  struct rte_flow *flow)
4255 {
4256 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4257 	struct mlx5_priv *priv = dev->data->dev_private;
4258 
4259 	if (!flow->rix_mreg_copy)
4260 		return;
4261 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4262 				 flow->rix_mreg_copy);
4263 	if (!mcp_res || !priv->mreg_cp_tbl)
4264 		return;
4265 	MLX5_ASSERT(mcp_res->rix_flow);
4266 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4267 	flow->rix_mreg_copy = 0;
4268 }
4269 
4270 /**
4271  * Remove the default copy action from RX_CP_TBL.
4272  *
4273  * This functions is called in the mlx5_dev_start(). No thread safe
4274  * is guaranteed.
4275  *
4276  * @param dev
4277  *   Pointer to Ethernet device.
4278  */
4279 static void
4280 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4281 {
4282 	struct mlx5_hlist_entry *entry;
4283 	struct mlx5_priv *priv = dev->data->dev_private;
4284 
4285 	/* Check if default flow is registered. */
4286 	if (!priv->mreg_cp_tbl)
4287 		return;
4288 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4289 				  MLX5_DEFAULT_COPY_ID, NULL);
4290 	if (!entry)
4291 		return;
4292 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4293 }
4294 
4295 /**
4296  * Add the default copy action in in RX_CP_TBL.
4297  *
4298  * This functions is called in the mlx5_dev_start(). No thread safe
4299  * is guaranteed.
4300  *
4301  * @param dev
4302  *   Pointer to Ethernet device.
4303  * @param[out] error
4304  *   Perform verbose error reporting if not NULL.
4305  *
4306  * @return
4307  *   0 for success, negative value otherwise and rte_errno is set.
4308  */
4309 static int
4310 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4311 				  struct rte_flow_error *error)
4312 {
4313 	struct mlx5_priv *priv = dev->data->dev_private;
4314 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4315 
4316 	/* Check whether extensive metadata feature is engaged. */
4317 	if (!priv->config.dv_flow_en ||
4318 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4319 	    !mlx5_flow_ext_mreg_supported(dev) ||
4320 	    !priv->sh->dv_regc0_mask)
4321 		return 0;
4322 	/*
4323 	 * Add default mreg copy flow may be called multiple time, but
4324 	 * only be called once in stop. Avoid register it twice.
4325 	 */
4326 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4327 		return 0;
4328 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4329 	if (!mcp_res)
4330 		return -rte_errno;
4331 	return 0;
4332 }
4333 
4334 /**
4335  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4336  *
4337  * All the flow having Q/RSS action should be split by
4338  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4339  * performs the following,
4340  *   - CQE->flow_tag := reg_c[1] (MARK)
4341  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4342  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4343  * but there should be a flow per each MARK ID set by MARK action.
4344  *
4345  * For the aforementioned reason, if there's a MARK action in flow's action
4346  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4347  * the MARK ID to CQE's flow_tag like,
4348  *   - If reg_c[1] is mark_id,
4349  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4350  *
4351  * For SET_META action which stores value in reg_c[0], as the destination is
4352  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4353  * MARK ID means the default flow. The default flow looks like,
4354  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4355  *
4356  * @param dev
4357  *   Pointer to Ethernet device.
4358  * @param flow
4359  *   Pointer to flow structure.
4360  * @param[in] actions
4361  *   Pointer to the list of actions.
4362  * @param[out] error
4363  *   Perform verbose error reporting if not NULL.
4364  *
4365  * @return
4366  *   0 on success, negative value otherwise and rte_errno is set.
4367  */
4368 static int
4369 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4370 			    struct rte_flow *flow,
4371 			    const struct rte_flow_action *actions,
4372 			    struct rte_flow_error *error)
4373 {
4374 	struct mlx5_priv *priv = dev->data->dev_private;
4375 	struct mlx5_dev_config *config = &priv->config;
4376 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4377 	const struct rte_flow_action_mark *mark;
4378 
4379 	/* Check whether extensive metadata feature is engaged. */
4380 	if (!config->dv_flow_en ||
4381 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4382 	    !mlx5_flow_ext_mreg_supported(dev) ||
4383 	    !priv->sh->dv_regc0_mask)
4384 		return 0;
4385 	/* Find MARK action. */
4386 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4387 		switch (actions->type) {
4388 		case RTE_FLOW_ACTION_TYPE_FLAG:
4389 			mcp_res = flow_mreg_add_copy_action
4390 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4391 			if (!mcp_res)
4392 				return -rte_errno;
4393 			flow->rix_mreg_copy = mcp_res->idx;
4394 			return 0;
4395 		case RTE_FLOW_ACTION_TYPE_MARK:
4396 			mark = (const struct rte_flow_action_mark *)
4397 				actions->conf;
4398 			mcp_res =
4399 				flow_mreg_add_copy_action(dev, mark->id, error);
4400 			if (!mcp_res)
4401 				return -rte_errno;
4402 			flow->rix_mreg_copy = mcp_res->idx;
4403 			return 0;
4404 		default:
4405 			break;
4406 		}
4407 	}
4408 	return 0;
4409 }
4410 
4411 #define MLX5_MAX_SPLIT_ACTIONS 24
4412 #define MLX5_MAX_SPLIT_ITEMS 24
4413 
4414 /**
4415  * Split the hairpin flow.
4416  * Since HW can't support encap and push-vlan on Rx, we move these
4417  * actions to Tx.
4418  * If the count action is after the encap then we also
4419  * move the count action. in this case the count will also measure
4420  * the outer bytes.
4421  *
4422  * @param dev
4423  *   Pointer to Ethernet device.
4424  * @param[in] actions
4425  *   Associated actions (list terminated by the END action).
4426  * @param[out] actions_rx
4427  *   Rx flow actions.
4428  * @param[out] actions_tx
4429  *   Tx flow actions..
4430  * @param[out] pattern_tx
4431  *   The pattern items for the Tx flow.
4432  * @param[out] flow_id
4433  *   The flow ID connected to this flow.
4434  *
4435  * @return
4436  *   0 on success.
4437  */
4438 static int
4439 flow_hairpin_split(struct rte_eth_dev *dev,
4440 		   const struct rte_flow_action actions[],
4441 		   struct rte_flow_action actions_rx[],
4442 		   struct rte_flow_action actions_tx[],
4443 		   struct rte_flow_item pattern_tx[],
4444 		   uint32_t flow_id)
4445 {
4446 	const struct rte_flow_action_raw_encap *raw_encap;
4447 	const struct rte_flow_action_raw_decap *raw_decap;
4448 	struct mlx5_rte_flow_action_set_tag *set_tag;
4449 	struct rte_flow_action *tag_action;
4450 	struct mlx5_rte_flow_item_tag *tag_item;
4451 	struct rte_flow_item *item;
4452 	char *addr;
4453 	int encap = 0;
4454 
4455 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4456 		switch (actions->type) {
4457 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4458 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4459 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4460 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4461 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4462 			rte_memcpy(actions_tx, actions,
4463 			       sizeof(struct rte_flow_action));
4464 			actions_tx++;
4465 			break;
4466 		case RTE_FLOW_ACTION_TYPE_COUNT:
4467 			if (encap) {
4468 				rte_memcpy(actions_tx, actions,
4469 					   sizeof(struct rte_flow_action));
4470 				actions_tx++;
4471 			} else {
4472 				rte_memcpy(actions_rx, actions,
4473 					   sizeof(struct rte_flow_action));
4474 				actions_rx++;
4475 			}
4476 			break;
4477 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4478 			raw_encap = actions->conf;
4479 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4480 				memcpy(actions_tx, actions,
4481 				       sizeof(struct rte_flow_action));
4482 				actions_tx++;
4483 				encap = 1;
4484 			} else {
4485 				rte_memcpy(actions_rx, actions,
4486 					   sizeof(struct rte_flow_action));
4487 				actions_rx++;
4488 			}
4489 			break;
4490 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4491 			raw_decap = actions->conf;
4492 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4493 				memcpy(actions_tx, actions,
4494 				       sizeof(struct rte_flow_action));
4495 				actions_tx++;
4496 			} else {
4497 				rte_memcpy(actions_rx, actions,
4498 					   sizeof(struct rte_flow_action));
4499 				actions_rx++;
4500 			}
4501 			break;
4502 		default:
4503 			rte_memcpy(actions_rx, actions,
4504 				   sizeof(struct rte_flow_action));
4505 			actions_rx++;
4506 			break;
4507 		}
4508 	}
4509 	/* Add set meta action and end action for the Rx flow. */
4510 	tag_action = actions_rx;
4511 	tag_action->type = (enum rte_flow_action_type)
4512 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4513 	actions_rx++;
4514 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4515 	actions_rx++;
4516 	set_tag = (void *)actions_rx;
4517 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4518 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4519 		.data = flow_id,
4520 	};
4521 	MLX5_ASSERT(set_tag->id > REG_NON);
4522 	tag_action->conf = set_tag;
4523 	/* Create Tx item list. */
4524 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4525 	addr = (void *)&pattern_tx[2];
4526 	item = pattern_tx;
4527 	item->type = (enum rte_flow_item_type)
4528 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4529 	tag_item = (void *)addr;
4530 	tag_item->data = flow_id;
4531 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4532 	MLX5_ASSERT(set_tag->id > REG_NON);
4533 	item->spec = tag_item;
4534 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4535 	tag_item = (void *)addr;
4536 	tag_item->data = UINT32_MAX;
4537 	tag_item->id = UINT16_MAX;
4538 	item->mask = tag_item;
4539 	item->last = NULL;
4540 	item++;
4541 	item->type = RTE_FLOW_ITEM_TYPE_END;
4542 	return 0;
4543 }
4544 
4545 /**
4546  * The last stage of splitting chain, just creates the subflow
4547  * without any modification.
4548  *
4549  * @param[in] dev
4550  *   Pointer to Ethernet device.
4551  * @param[in] flow
4552  *   Parent flow structure pointer.
4553  * @param[in, out] sub_flow
4554  *   Pointer to return the created subflow, may be NULL.
4555  * @param[in] attr
4556  *   Flow rule attributes.
4557  * @param[in] items
4558  *   Pattern specification (list terminated by the END pattern item).
4559  * @param[in] actions
4560  *   Associated actions (list terminated by the END action).
4561  * @param[in] flow_split_info
4562  *   Pointer to flow split info structure.
4563  * @param[out] error
4564  *   Perform verbose error reporting if not NULL.
4565  * @return
4566  *   0 on success, negative value otherwise
4567  */
4568 static int
4569 flow_create_split_inner(struct rte_eth_dev *dev,
4570 			struct rte_flow *flow,
4571 			struct mlx5_flow **sub_flow,
4572 			const struct rte_flow_attr *attr,
4573 			const struct rte_flow_item items[],
4574 			const struct rte_flow_action actions[],
4575 			struct mlx5_flow_split_info *flow_split_info,
4576 			struct rte_flow_error *error)
4577 {
4578 	struct mlx5_flow *dev_flow;
4579 
4580 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4581 				    flow_split_info->flow_idx, error);
4582 	if (!dev_flow)
4583 		return -rte_errno;
4584 	dev_flow->flow = flow;
4585 	dev_flow->external = flow_split_info->external;
4586 	dev_flow->skip_scale = flow_split_info->skip_scale;
4587 	/* Subflow object was created, we must include one in the list. */
4588 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4589 		      dev_flow->handle, next);
4590 	/*
4591 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4592 	 * flow may need some user defined item layer flags, and pass the
4593 	 * Metadate rxq mark flag to suffix flow as well.
4594 	 */
4595 	if (flow_split_info->prefix_layers)
4596 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4597 	if (flow_split_info->prefix_mark)
4598 		dev_flow->handle->mark = 1;
4599 	if (sub_flow)
4600 		*sub_flow = dev_flow;
4601 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4602 	dev_flow->dv.table_id = flow_split_info->table_id;
4603 #endif
4604 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4605 }
4606 
4607 /**
4608  * Get the sub policy of a meter.
4609  *
4610  * @param[in] dev
4611  *   Pointer to Ethernet device.
4612  * @param[in] flow
4613  *   Parent flow structure pointer.
4614  * @param wks
4615  *   Pointer to thread flow work space.
4616  * @param[in] attr
4617  *   Flow rule attributes.
4618  * @param[in] items
4619  *   Pattern specification (list terminated by the END pattern item).
4620  * @param[out] error
4621  *   Perform verbose error reporting if not NULL.
4622  *
4623  * @return
4624  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4625  */
4626 static struct mlx5_flow_meter_sub_policy *
4627 get_meter_sub_policy(struct rte_eth_dev *dev,
4628 		     struct rte_flow *flow,
4629 		     struct mlx5_flow_workspace *wks,
4630 		     const struct rte_flow_attr *attr,
4631 		     const struct rte_flow_item items[],
4632 		     struct rte_flow_error *error)
4633 {
4634 	struct mlx5_flow_meter_policy *policy;
4635 	struct mlx5_flow_meter_policy *final_policy;
4636 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4637 
4638 	policy = wks->policy;
4639 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4640 	if (final_policy->is_rss || final_policy->is_queue) {
4641 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4642 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4643 		uint32_t i;
4644 
4645 		/**
4646 		 * This is a tmp dev_flow,
4647 		 * no need to register any matcher for it in translate.
4648 		 */
4649 		wks->skip_matcher_reg = 1;
4650 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4651 			struct mlx5_flow dev_flow = {0};
4652 			struct mlx5_flow_handle dev_handle = { {0} };
4653 
4654 			if (final_policy->is_rss) {
4655 				const void *rss_act =
4656 					final_policy->act_cnt[i].rss->conf;
4657 				struct rte_flow_action rss_actions[2] = {
4658 					[0] = {
4659 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4660 					.conf = rss_act
4661 					},
4662 					[1] = {
4663 					.type = RTE_FLOW_ACTION_TYPE_END,
4664 					.conf = NULL
4665 					}
4666 				};
4667 
4668 				dev_flow.handle = &dev_handle;
4669 				dev_flow.ingress = attr->ingress;
4670 				dev_flow.flow = flow;
4671 				dev_flow.external = 0;
4672 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4673 				dev_flow.dv.transfer = attr->transfer;
4674 #endif
4675 				/**
4676 				 * Translate RSS action to get rss hash fields.
4677 				 */
4678 				if (flow_drv_translate(dev, &dev_flow, attr,
4679 						items, rss_actions, error))
4680 					goto exit;
4681 				rss_desc_v[i] = wks->rss_desc;
4682 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4683 				rss_desc_v[i].hash_fields =
4684 						dev_flow.hash_fields;
4685 				rss_desc_v[i].queue_num =
4686 						rss_desc_v[i].hash_fields ?
4687 						rss_desc_v[i].queue_num : 1;
4688 				rss_desc_v[i].tunnel =
4689 					!!(dev_flow.handle->layers &
4690 					MLX5_FLOW_LAYER_TUNNEL);
4691 			} else {
4692 				/* This is queue action. */
4693 				rss_desc_v[i] = wks->rss_desc;
4694 				rss_desc_v[i].key_len = 0;
4695 				rss_desc_v[i].hash_fields = 0;
4696 				rss_desc_v[i].queue =
4697 					&final_policy->act_cnt[i].queue;
4698 				rss_desc_v[i].queue_num = 1;
4699 			}
4700 			rss_desc[i] = &rss_desc_v[i];
4701 		}
4702 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4703 						flow, policy, rss_desc);
4704 	} else {
4705 		enum mlx5_meter_domain mtr_domain =
4706 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4707 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4708 					MLX5_MTR_DOMAIN_INGRESS;
4709 		sub_policy = policy->sub_policys[mtr_domain][0];
4710 	}
4711 	if (!sub_policy) {
4712 		rte_flow_error_set(error, EINVAL,
4713 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4714 			"Failed to get meter sub-policy.");
4715 		goto exit;
4716 	}
4717 exit:
4718 	return sub_policy;
4719 }
4720 
4721 /**
4722  * Split the meter flow.
4723  *
4724  * As meter flow will split to three sub flow, other than meter
4725  * action, the other actions make sense to only meter accepts
4726  * the packet. If it need to be dropped, no other additional
4727  * actions should be take.
4728  *
4729  * One kind of special action which decapsulates the L3 tunnel
4730  * header will be in the prefix sub flow, as not to take the
4731  * L3 tunnel header into account.
4732  *
4733  * @param[in] dev
4734  *   Pointer to Ethernet device.
4735  * @param[in] flow
4736  *   Parent flow structure pointer.
4737  * @param wks
4738  *   Pointer to thread flow work space.
4739  * @param[in] attr
4740  *   Flow rule attributes.
4741  * @param[in] items
4742  *   Pattern specification (list terminated by the END pattern item).
4743  * @param[out] sfx_items
4744  *   Suffix flow match items (list terminated by the END pattern item).
4745  * @param[in] actions
4746  *   Associated actions (list terminated by the END action).
4747  * @param[out] actions_sfx
4748  *   Suffix flow actions.
4749  * @param[out] actions_pre
4750  *   Prefix flow actions.
4751  * @param[out] mtr_flow_id
4752  *   Pointer to meter flow id.
4753  * @param[out] error
4754  *   Perform verbose error reporting if not NULL.
4755  *
4756  * @return
4757  *   0 on success, a negative errno value otherwise and rte_errno is set.
4758  */
4759 static int
4760 flow_meter_split_prep(struct rte_eth_dev *dev,
4761 		      struct rte_flow *flow,
4762 		      struct mlx5_flow_workspace *wks,
4763 		      const struct rte_flow_attr *attr,
4764 		      const struct rte_flow_item items[],
4765 		      struct rte_flow_item sfx_items[],
4766 		      const struct rte_flow_action actions[],
4767 		      struct rte_flow_action actions_sfx[],
4768 		      struct rte_flow_action actions_pre[],
4769 		      uint32_t *mtr_flow_id,
4770 		      struct rte_flow_error *error)
4771 {
4772 	struct mlx5_priv *priv = dev->data->dev_private;
4773 	struct mlx5_flow_meter_info *fm = wks->fm;
4774 	struct rte_flow_action *tag_action = NULL;
4775 	struct rte_flow_item *tag_item;
4776 	struct mlx5_rte_flow_action_set_tag *set_tag;
4777 	const struct rte_flow_action_raw_encap *raw_encap;
4778 	const struct rte_flow_action_raw_decap *raw_decap;
4779 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4780 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4781 	uint32_t tag_id = 0;
4782 	struct rte_flow_item *vlan_item_dst = NULL;
4783 	const struct rte_flow_item *vlan_item_src = NULL;
4784 	struct rte_flow_action *hw_mtr_action;
4785 	struct rte_flow_action *action_pre_head = NULL;
4786 	int32_t flow_src_port = priv->representor_id;
4787 	bool mtr_first;
4788 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4789 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4790 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4791 	uint32_t flow_id = 0;
4792 	uint32_t flow_id_reversed = 0;
4793 	uint8_t flow_id_bits = 0;
4794 	int shift;
4795 
4796 	/* Prepare the suffix subflow items. */
4797 	tag_item = sfx_items++;
4798 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4799 		struct mlx5_priv *port_priv;
4800 		const struct rte_flow_item_port_id *pid_v;
4801 		int item_type = items->type;
4802 
4803 		switch (item_type) {
4804 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4805 			pid_v = items->spec;
4806 			MLX5_ASSERT(pid_v);
4807 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4808 			if (!port_priv)
4809 				return rte_flow_error_set(error,
4810 						rte_errno,
4811 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4812 						pid_v,
4813 						"Failed to get port info.");
4814 			flow_src_port = port_priv->representor_id;
4815 			if (!fm->def_policy && wks->policy->is_hierarchy &&
4816 			    flow_src_port != priv->representor_id) {
4817 				if (flow_drv_mtr_hierarchy_rule_create(dev,
4818 								flow, fm,
4819 								flow_src_port,
4820 								items,
4821 								error))
4822 					return -rte_errno;
4823 			}
4824 			memcpy(sfx_items, items, sizeof(*sfx_items));
4825 			sfx_items++;
4826 			break;
4827 		case RTE_FLOW_ITEM_TYPE_VLAN:
4828 			/* Determine if copy vlan item below. */
4829 			vlan_item_src = items;
4830 			vlan_item_dst = sfx_items++;
4831 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4832 			break;
4833 		default:
4834 			break;
4835 		}
4836 	}
4837 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4838 	sfx_items++;
4839 	mtr_first = priv->sh->meter_aso_en &&
4840 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4841 	/* For ASO meter, meter must be before tag in TX direction. */
4842 	if (mtr_first) {
4843 		action_pre_head = actions_pre++;
4844 		/* Leave space for tag action. */
4845 		tag_action = actions_pre++;
4846 	}
4847 	/* Prepare the actions for prefix and suffix flow. */
4848 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4849 		struct rte_flow_action *action_cur = NULL;
4850 
4851 		switch (actions->type) {
4852 		case RTE_FLOW_ACTION_TYPE_METER:
4853 			if (mtr_first) {
4854 				action_cur = action_pre_head;
4855 			} else {
4856 				/* Leave space for tag action. */
4857 				tag_action = actions_pre++;
4858 				action_cur = actions_pre++;
4859 			}
4860 			break;
4861 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4862 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4863 			action_cur = actions_pre++;
4864 			break;
4865 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4866 			raw_encap = actions->conf;
4867 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4868 				action_cur = actions_pre++;
4869 			break;
4870 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4871 			raw_decap = actions->conf;
4872 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4873 				action_cur = actions_pre++;
4874 			break;
4875 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4876 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4877 			if (vlan_item_dst && vlan_item_src) {
4878 				memcpy(vlan_item_dst, vlan_item_src,
4879 					sizeof(*vlan_item_dst));
4880 				/*
4881 				 * Convert to internal match item, it is used
4882 				 * for vlan push and set vid.
4883 				 */
4884 				vlan_item_dst->type = (enum rte_flow_item_type)
4885 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4886 			}
4887 			break;
4888 		default:
4889 			break;
4890 		}
4891 		if (!action_cur)
4892 			action_cur = (fm->def_policy) ?
4893 					actions_sfx++ : actions_pre++;
4894 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4895 	}
4896 	/* Add end action to the actions. */
4897 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4898 	if (priv->sh->meter_aso_en) {
4899 		/**
4900 		 * For ASO meter, need to add an extra jump action explicitly,
4901 		 * to jump from meter to policer table.
4902 		 */
4903 		struct mlx5_flow_meter_sub_policy *sub_policy;
4904 		struct mlx5_flow_tbl_data_entry *tbl_data;
4905 
4906 		if (!fm->def_policy) {
4907 			sub_policy = get_meter_sub_policy(dev, flow, wks,
4908 							  attr, items, error);
4909 			if (!sub_policy)
4910 				return -rte_errno;
4911 		} else {
4912 			enum mlx5_meter_domain mtr_domain =
4913 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4914 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4915 					MLX5_MTR_DOMAIN_INGRESS;
4916 
4917 			sub_policy =
4918 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4919 		}
4920 		tbl_data = container_of(sub_policy->tbl_rsc,
4921 					struct mlx5_flow_tbl_data_entry, tbl);
4922 		hw_mtr_action = actions_pre++;
4923 		hw_mtr_action->type = (enum rte_flow_action_type)
4924 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4925 		hw_mtr_action->conf = tbl_data->jump.action;
4926 	}
4927 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4928 	actions_pre++;
4929 	if (!tag_action)
4930 		return rte_flow_error_set(error, ENOMEM,
4931 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4932 					"No tag action space.");
4933 	if (!mtr_flow_id) {
4934 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4935 		goto exit;
4936 	}
4937 	/* Only default-policy Meter creates mtr flow id. */
4938 	if (fm->def_policy) {
4939 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4940 		if (!tag_id)
4941 			return rte_flow_error_set(error, ENOMEM,
4942 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4943 					"Failed to allocate meter flow id.");
4944 		flow_id = tag_id - 1;
4945 		flow_id_bits = (!flow_id) ? 1 :
4946 				(MLX5_REG_BITS - __builtin_clz(flow_id));
4947 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4948 		    mtr_reg_bits) {
4949 			mlx5_ipool_free(fm->flow_ipool, tag_id);
4950 			return rte_flow_error_set(error, EINVAL,
4951 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4952 					"Meter flow id exceeds max limit.");
4953 		}
4954 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4955 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4956 	}
4957 	/* Build tag actions and items for meter_id/meter flow_id. */
4958 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4959 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4960 	tag_item_mask = tag_item_spec + 1;
4961 	/* Both flow_id and meter_id share the same register. */
4962 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4963 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4964 							    0, error),
4965 		.offset = mtr_id_offset,
4966 		.length = mtr_reg_bits,
4967 		.data = flow->meter,
4968 	};
4969 	/*
4970 	 * The color Reg bits used by flow_id are growing from
4971 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4972 	 */
4973 	for (shift = 0; shift < flow_id_bits; shift++)
4974 		flow_id_reversed = (flow_id_reversed << 1) |
4975 				((flow_id >> shift) & 0x1);
4976 	set_tag->data |=
4977 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4978 	tag_item_spec->id = set_tag->id;
4979 	tag_item_spec->data = set_tag->data << mtr_id_offset;
4980 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4981 	tag_action->type = (enum rte_flow_action_type)
4982 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4983 	tag_action->conf = set_tag;
4984 	tag_item->type = (enum rte_flow_item_type)
4985 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4986 	tag_item->spec = tag_item_spec;
4987 	tag_item->last = NULL;
4988 	tag_item->mask = tag_item_mask;
4989 exit:
4990 	if (mtr_flow_id)
4991 		*mtr_flow_id = tag_id;
4992 	return 0;
4993 }
4994 
4995 /**
4996  * Split action list having QUEUE/RSS for metadata register copy.
4997  *
4998  * Once Q/RSS action is detected in user's action list, the flow action
4999  * should be split in order to copy metadata registers, which will happen in
5000  * RX_CP_TBL like,
5001  *   - CQE->flow_tag := reg_c[1] (MARK)
5002  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5003  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5004  * This is because the last action of each flow must be a terminal action
5005  * (QUEUE, RSS or DROP).
5006  *
5007  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5008  * stored and kept in the mlx5_flow structure per each sub_flow.
5009  *
5010  * The Q/RSS action is replaced with,
5011  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5012  * And the following JUMP action is added at the end,
5013  *   - JUMP, to RX_CP_TBL.
5014  *
5015  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5016  * flow_create_split_metadata() routine. The flow will look like,
5017  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5018  *
5019  * @param dev
5020  *   Pointer to Ethernet device.
5021  * @param[out] split_actions
5022  *   Pointer to store split actions to jump to CP_TBL.
5023  * @param[in] actions
5024  *   Pointer to the list of original flow actions.
5025  * @param[in] qrss
5026  *   Pointer to the Q/RSS action.
5027  * @param[in] actions_n
5028  *   Number of original actions.
5029  * @param[out] error
5030  *   Perform verbose error reporting if not NULL.
5031  *
5032  * @return
5033  *   non-zero unique flow_id on success, otherwise 0 and
5034  *   error/rte_error are set.
5035  */
5036 static uint32_t
5037 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5038 			  struct rte_flow_action *split_actions,
5039 			  const struct rte_flow_action *actions,
5040 			  const struct rte_flow_action *qrss,
5041 			  int actions_n, struct rte_flow_error *error)
5042 {
5043 	struct mlx5_priv *priv = dev->data->dev_private;
5044 	struct mlx5_rte_flow_action_set_tag *set_tag;
5045 	struct rte_flow_action_jump *jump;
5046 	const int qrss_idx = qrss - actions;
5047 	uint32_t flow_id = 0;
5048 	int ret = 0;
5049 
5050 	/*
5051 	 * Given actions will be split
5052 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5053 	 * - Add jump to mreg CP_TBL.
5054 	 * As a result, there will be one more action.
5055 	 */
5056 	++actions_n;
5057 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5058 	set_tag = (void *)(split_actions + actions_n);
5059 	/*
5060 	 * If tag action is not set to void(it means we are not the meter
5061 	 * suffix flow), add the tag action. Since meter suffix flow already
5062 	 * has the tag added.
5063 	 */
5064 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5065 		/*
5066 		 * Allocate the new subflow ID. This one is unique within
5067 		 * device and not shared with representors. Otherwise,
5068 		 * we would have to resolve multi-thread access synch
5069 		 * issue. Each flow on the shared device is appended
5070 		 * with source vport identifier, so the resulting
5071 		 * flows will be unique in the shared (by master and
5072 		 * representors) domain even if they have coinciding
5073 		 * IDs.
5074 		 */
5075 		mlx5_ipool_malloc(priv->sh->ipool
5076 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5077 		if (!flow_id)
5078 			return rte_flow_error_set(error, ENOMEM,
5079 						  RTE_FLOW_ERROR_TYPE_ACTION,
5080 						  NULL, "can't allocate id "
5081 						  "for split Q/RSS subflow");
5082 		/* Internal SET_TAG action to set flow ID. */
5083 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5084 			.data = flow_id,
5085 		};
5086 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5087 		if (ret < 0)
5088 			return ret;
5089 		set_tag->id = ret;
5090 		/* Construct new actions array. */
5091 		/* Replace QUEUE/RSS action. */
5092 		split_actions[qrss_idx] = (struct rte_flow_action){
5093 			.type = (enum rte_flow_action_type)
5094 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5095 			.conf = set_tag,
5096 		};
5097 	}
5098 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5099 	jump = (void *)(set_tag + 1);
5100 	*jump = (struct rte_flow_action_jump){
5101 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5102 	};
5103 	split_actions[actions_n - 2] = (struct rte_flow_action){
5104 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5105 		.conf = jump,
5106 	};
5107 	split_actions[actions_n - 1] = (struct rte_flow_action){
5108 		.type = RTE_FLOW_ACTION_TYPE_END,
5109 	};
5110 	return flow_id;
5111 }
5112 
5113 /**
5114  * Extend the given action list for Tx metadata copy.
5115  *
5116  * Copy the given action list to the ext_actions and add flow metadata register
5117  * copy action in order to copy reg_a set by WQE to reg_c[0].
5118  *
5119  * @param[out] ext_actions
5120  *   Pointer to the extended action list.
5121  * @param[in] actions
5122  *   Pointer to the list of actions.
5123  * @param[in] actions_n
5124  *   Number of actions in the list.
5125  * @param[out] error
5126  *   Perform verbose error reporting if not NULL.
5127  * @param[in] encap_idx
5128  *   The encap action inndex.
5129  *
5130  * @return
5131  *   0 on success, negative value otherwise
5132  */
5133 static int
5134 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5135 		       struct rte_flow_action *ext_actions,
5136 		       const struct rte_flow_action *actions,
5137 		       int actions_n, struct rte_flow_error *error,
5138 		       int encap_idx)
5139 {
5140 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5141 		(struct mlx5_flow_action_copy_mreg *)
5142 			(ext_actions + actions_n + 1);
5143 	int ret;
5144 
5145 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5146 	if (ret < 0)
5147 		return ret;
5148 	cp_mreg->dst = ret;
5149 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5150 	if (ret < 0)
5151 		return ret;
5152 	cp_mreg->src = ret;
5153 	if (encap_idx != 0)
5154 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5155 	if (encap_idx == actions_n - 1) {
5156 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5157 			.type = (enum rte_flow_action_type)
5158 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5159 			.conf = cp_mreg,
5160 		};
5161 		ext_actions[actions_n] = (struct rte_flow_action){
5162 			.type = RTE_FLOW_ACTION_TYPE_END,
5163 		};
5164 	} else {
5165 		ext_actions[encap_idx] = (struct rte_flow_action){
5166 			.type = (enum rte_flow_action_type)
5167 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5168 			.conf = cp_mreg,
5169 		};
5170 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5171 				sizeof(*ext_actions) * (actions_n - encap_idx));
5172 	}
5173 	return 0;
5174 }
5175 
5176 /**
5177  * Check the match action from the action list.
5178  *
5179  * @param[in] actions
5180  *   Pointer to the list of actions.
5181  * @param[in] attr
5182  *   Flow rule attributes.
5183  * @param[in] action
5184  *   The action to be check if exist.
5185  * @param[out] match_action_pos
5186  *   Pointer to the position of the matched action if exists, otherwise is -1.
5187  * @param[out] qrss_action_pos
5188  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5189  * @param[out] modify_after_mirror
5190  *   Pointer to the flag of modify action after FDB mirroring.
5191  *
5192  * @return
5193  *   > 0 the total number of actions.
5194  *   0 if not found match action in action list.
5195  */
5196 static int
5197 flow_check_match_action(const struct rte_flow_action actions[],
5198 			const struct rte_flow_attr *attr,
5199 			enum rte_flow_action_type action,
5200 			int *match_action_pos, int *qrss_action_pos,
5201 			int *modify_after_mirror)
5202 {
5203 	const struct rte_flow_action_sample *sample;
5204 	int actions_n = 0;
5205 	uint32_t ratio = 0;
5206 	int sub_type = 0;
5207 	int flag = 0;
5208 	int fdb_mirror = 0;
5209 
5210 	*match_action_pos = -1;
5211 	*qrss_action_pos = -1;
5212 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5213 		if (actions->type == action) {
5214 			flag = 1;
5215 			*match_action_pos = actions_n;
5216 		}
5217 		switch (actions->type) {
5218 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5219 		case RTE_FLOW_ACTION_TYPE_RSS:
5220 			*qrss_action_pos = actions_n;
5221 			break;
5222 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5223 			sample = actions->conf;
5224 			ratio = sample->ratio;
5225 			sub_type = ((const struct rte_flow_action *)
5226 					(sample->actions))->type;
5227 			if (ratio == 1 && attr->transfer)
5228 				fdb_mirror = 1;
5229 			break;
5230 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5231 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5232 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5233 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5234 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5235 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5236 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5237 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5238 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5239 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5240 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5241 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5242 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5243 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5244 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5245 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5246 		case RTE_FLOW_ACTION_TYPE_FLAG:
5247 		case RTE_FLOW_ACTION_TYPE_MARK:
5248 		case RTE_FLOW_ACTION_TYPE_SET_META:
5249 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5250 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5251 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5252 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5253 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5254 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5255 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5256 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5257 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5258 		case RTE_FLOW_ACTION_TYPE_METER:
5259 			if (fdb_mirror)
5260 				*modify_after_mirror = 1;
5261 			break;
5262 		default:
5263 			break;
5264 		}
5265 		actions_n++;
5266 	}
5267 	if (flag && fdb_mirror && !*modify_after_mirror) {
5268 		/* FDB mirroring uses the destination array to implement
5269 		 * instead of FLOW_SAMPLER object.
5270 		 */
5271 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5272 			flag = 0;
5273 	}
5274 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5275 	return flag ? actions_n + 1 : 0;
5276 }
5277 
5278 #define SAMPLE_SUFFIX_ITEM 2
5279 
5280 /**
5281  * Split the sample flow.
5282  *
5283  * As sample flow will split to two sub flow, sample flow with
5284  * sample action, the other actions will move to new suffix flow.
5285  *
5286  * Also add unique tag id with tag action in the sample flow,
5287  * the same tag id will be as match in the suffix flow.
5288  *
5289  * @param dev
5290  *   Pointer to Ethernet device.
5291  * @param[in] add_tag
5292  *   Add extra tag action flag.
5293  * @param[out] sfx_items
5294  *   Suffix flow match items (list terminated by the END pattern item).
5295  * @param[in] actions
5296  *   Associated actions (list terminated by the END action).
5297  * @param[out] actions_sfx
5298  *   Suffix flow actions.
5299  * @param[out] actions_pre
5300  *   Prefix flow actions.
5301  * @param[in] actions_n
5302  *  The total number of actions.
5303  * @param[in] sample_action_pos
5304  *   The sample action position.
5305  * @param[in] qrss_action_pos
5306  *   The Queue/RSS action position.
5307  * @param[in] jump_table
5308  *   Add extra jump action flag.
5309  * @param[out] error
5310  *   Perform verbose error reporting if not NULL.
5311  *
5312  * @return
5313  *   0 on success, or unique flow_id, a negative errno value
5314  *   otherwise and rte_errno is set.
5315  */
5316 static int
5317 flow_sample_split_prep(struct rte_eth_dev *dev,
5318 		       int add_tag,
5319 		       struct rte_flow_item sfx_items[],
5320 		       const struct rte_flow_action actions[],
5321 		       struct rte_flow_action actions_sfx[],
5322 		       struct rte_flow_action actions_pre[],
5323 		       int actions_n,
5324 		       int sample_action_pos,
5325 		       int qrss_action_pos,
5326 		       int jump_table,
5327 		       struct rte_flow_error *error)
5328 {
5329 	struct mlx5_priv *priv = dev->data->dev_private;
5330 	struct mlx5_rte_flow_action_set_tag *set_tag;
5331 	struct mlx5_rte_flow_item_tag *tag_spec;
5332 	struct mlx5_rte_flow_item_tag *tag_mask;
5333 	struct rte_flow_action_jump *jump_action;
5334 	uint32_t tag_id = 0;
5335 	int index;
5336 	int append_index = 0;
5337 	int ret;
5338 
5339 	if (sample_action_pos < 0)
5340 		return rte_flow_error_set(error, EINVAL,
5341 					  RTE_FLOW_ERROR_TYPE_ACTION,
5342 					  NULL, "invalid position of sample "
5343 					  "action in list");
5344 	/* Prepare the actions for prefix and suffix flow. */
5345 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5346 		index = qrss_action_pos;
5347 		/* Put the preceding the Queue/RSS action into prefix flow. */
5348 		if (index != 0)
5349 			memcpy(actions_pre, actions,
5350 			       sizeof(struct rte_flow_action) * index);
5351 		/* Put others preceding the sample action into prefix flow. */
5352 		if (sample_action_pos > index + 1)
5353 			memcpy(actions_pre + index, actions + index + 1,
5354 			       sizeof(struct rte_flow_action) *
5355 			       (sample_action_pos - index - 1));
5356 		index = sample_action_pos - 1;
5357 		/* Put Queue/RSS action into Suffix flow. */
5358 		memcpy(actions_sfx, actions + qrss_action_pos,
5359 		       sizeof(struct rte_flow_action));
5360 		actions_sfx++;
5361 	} else {
5362 		index = sample_action_pos;
5363 		if (index != 0)
5364 			memcpy(actions_pre, actions,
5365 			       sizeof(struct rte_flow_action) * index);
5366 	}
5367 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5368 	 * For CX6DX and above, metadata registers Cx preserve their value,
5369 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5370 	 */
5371 	if (add_tag) {
5372 		/* Prepare the prefix tag action. */
5373 		append_index++;
5374 		set_tag = (void *)(actions_pre + actions_n + append_index);
5375 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5376 		if (ret < 0)
5377 			return ret;
5378 		mlx5_ipool_malloc(priv->sh->ipool
5379 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5380 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5381 			.id = ret,
5382 			.data = tag_id,
5383 		};
5384 		/* Prepare the suffix subflow items. */
5385 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5386 		tag_spec->data = tag_id;
5387 		tag_spec->id = set_tag->id;
5388 		tag_mask = tag_spec + 1;
5389 		tag_mask->data = UINT32_MAX;
5390 		sfx_items[0] = (struct rte_flow_item){
5391 			.type = (enum rte_flow_item_type)
5392 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5393 			.spec = tag_spec,
5394 			.last = NULL,
5395 			.mask = tag_mask,
5396 		};
5397 		sfx_items[1] = (struct rte_flow_item){
5398 			.type = (enum rte_flow_item_type)
5399 				RTE_FLOW_ITEM_TYPE_END,
5400 		};
5401 		/* Prepare the tag action in prefix subflow. */
5402 		actions_pre[index++] =
5403 			(struct rte_flow_action){
5404 			.type = (enum rte_flow_action_type)
5405 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5406 			.conf = set_tag,
5407 		};
5408 	}
5409 	memcpy(actions_pre + index, actions + sample_action_pos,
5410 	       sizeof(struct rte_flow_action));
5411 	index += 1;
5412 	/* For the modify action after the sample action in E-Switch mirroring,
5413 	 * Add the extra jump action in prefix subflow and jump into the next
5414 	 * table, then do the modify action in the new table.
5415 	 */
5416 	if (jump_table) {
5417 		/* Prepare the prefix jump action. */
5418 		append_index++;
5419 		jump_action = (void *)(actions_pre + actions_n + append_index);
5420 		jump_action->group = jump_table;
5421 		actions_pre[index++] =
5422 			(struct rte_flow_action){
5423 			.type = (enum rte_flow_action_type)
5424 				RTE_FLOW_ACTION_TYPE_JUMP,
5425 			.conf = jump_action,
5426 		};
5427 	}
5428 	actions_pre[index] = (struct rte_flow_action){
5429 		.type = (enum rte_flow_action_type)
5430 			RTE_FLOW_ACTION_TYPE_END,
5431 	};
5432 	/* Put the actions after sample into Suffix flow. */
5433 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5434 	       sizeof(struct rte_flow_action) *
5435 	       (actions_n - sample_action_pos - 1));
5436 	return tag_id;
5437 }
5438 
5439 /**
5440  * The splitting for metadata feature.
5441  *
5442  * - Q/RSS action on NIC Rx should be split in order to pass by
5443  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5444  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5445  *
5446  * - All the actions on NIC Tx should have a mreg copy action to
5447  *   copy reg_a from WQE to reg_c[0].
5448  *
5449  * @param dev
5450  *   Pointer to Ethernet device.
5451  * @param[in] flow
5452  *   Parent flow structure pointer.
5453  * @param[in] attr
5454  *   Flow rule attributes.
5455  * @param[in] items
5456  *   Pattern specification (list terminated by the END pattern item).
5457  * @param[in] actions
5458  *   Associated actions (list terminated by the END action).
5459  * @param[in] flow_split_info
5460  *   Pointer to flow split info structure.
5461  * @param[out] error
5462  *   Perform verbose error reporting if not NULL.
5463  * @return
5464  *   0 on success, negative value otherwise
5465  */
5466 static int
5467 flow_create_split_metadata(struct rte_eth_dev *dev,
5468 			   struct rte_flow *flow,
5469 			   const struct rte_flow_attr *attr,
5470 			   const struct rte_flow_item items[],
5471 			   const struct rte_flow_action actions[],
5472 			   struct mlx5_flow_split_info *flow_split_info,
5473 			   struct rte_flow_error *error)
5474 {
5475 	struct mlx5_priv *priv = dev->data->dev_private;
5476 	struct mlx5_dev_config *config = &priv->config;
5477 	const struct rte_flow_action *qrss = NULL;
5478 	struct rte_flow_action *ext_actions = NULL;
5479 	struct mlx5_flow *dev_flow = NULL;
5480 	uint32_t qrss_id = 0;
5481 	int mtr_sfx = 0;
5482 	size_t act_size;
5483 	int actions_n;
5484 	int encap_idx;
5485 	int ret;
5486 
5487 	/* Check whether extensive metadata feature is engaged. */
5488 	if (!config->dv_flow_en ||
5489 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5490 	    !mlx5_flow_ext_mreg_supported(dev))
5491 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5492 					       actions, flow_split_info, error);
5493 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5494 							   &encap_idx);
5495 	if (qrss) {
5496 		/* Exclude hairpin flows from splitting. */
5497 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5498 			const struct rte_flow_action_queue *queue;
5499 
5500 			queue = qrss->conf;
5501 			if (mlx5_rxq_get_type(dev, queue->index) ==
5502 			    MLX5_RXQ_TYPE_HAIRPIN)
5503 				qrss = NULL;
5504 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5505 			const struct rte_flow_action_rss *rss;
5506 
5507 			rss = qrss->conf;
5508 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5509 			    MLX5_RXQ_TYPE_HAIRPIN)
5510 				qrss = NULL;
5511 		}
5512 	}
5513 	if (qrss) {
5514 		/* Check if it is in meter suffix table. */
5515 		mtr_sfx = attr->group == (attr->transfer ?
5516 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5517 			  MLX5_FLOW_TABLE_LEVEL_METER);
5518 		/*
5519 		 * Q/RSS action on NIC Rx should be split in order to pass by
5520 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5521 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5522 		 */
5523 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5524 			   sizeof(struct rte_flow_action_set_tag) +
5525 			   sizeof(struct rte_flow_action_jump);
5526 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5527 					  SOCKET_ID_ANY);
5528 		if (!ext_actions)
5529 			return rte_flow_error_set(error, ENOMEM,
5530 						  RTE_FLOW_ERROR_TYPE_ACTION,
5531 						  NULL, "no memory to split "
5532 						  "metadata flow");
5533 		/*
5534 		 * If we are the suffix flow of meter, tag already exist.
5535 		 * Set the tag action to void.
5536 		 */
5537 		if (mtr_sfx)
5538 			ext_actions[qrss - actions].type =
5539 						RTE_FLOW_ACTION_TYPE_VOID;
5540 		else
5541 			ext_actions[qrss - actions].type =
5542 						(enum rte_flow_action_type)
5543 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5544 		/*
5545 		 * Create the new actions list with removed Q/RSS action
5546 		 * and appended set tag and jump to register copy table
5547 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5548 		 * in advance, because it is needed for set tag action.
5549 		 */
5550 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5551 						    qrss, actions_n, error);
5552 		if (!mtr_sfx && !qrss_id) {
5553 			ret = -rte_errno;
5554 			goto exit;
5555 		}
5556 	} else if (attr->egress && !attr->transfer) {
5557 		/*
5558 		 * All the actions on NIC Tx should have a metadata register
5559 		 * copy action to copy reg_a from WQE to reg_c[meta]
5560 		 */
5561 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5562 			   sizeof(struct mlx5_flow_action_copy_mreg);
5563 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5564 					  SOCKET_ID_ANY);
5565 		if (!ext_actions)
5566 			return rte_flow_error_set(error, ENOMEM,
5567 						  RTE_FLOW_ERROR_TYPE_ACTION,
5568 						  NULL, "no memory to split "
5569 						  "metadata flow");
5570 		/* Create the action list appended with copy register. */
5571 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5572 					     actions_n, error, encap_idx);
5573 		if (ret < 0)
5574 			goto exit;
5575 	}
5576 	/* Add the unmodified original or prefix subflow. */
5577 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5578 				      items, ext_actions ? ext_actions :
5579 				      actions, flow_split_info, error);
5580 	if (ret < 0)
5581 		goto exit;
5582 	MLX5_ASSERT(dev_flow);
5583 	if (qrss) {
5584 		const struct rte_flow_attr q_attr = {
5585 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5586 			.ingress = 1,
5587 		};
5588 		/* Internal PMD action to set register. */
5589 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5590 			.data = qrss_id,
5591 			.id = REG_NON,
5592 		};
5593 		struct rte_flow_item q_items[] = {
5594 			{
5595 				.type = (enum rte_flow_item_type)
5596 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5597 				.spec = &q_tag_spec,
5598 				.last = NULL,
5599 				.mask = NULL,
5600 			},
5601 			{
5602 				.type = RTE_FLOW_ITEM_TYPE_END,
5603 			},
5604 		};
5605 		struct rte_flow_action q_actions[] = {
5606 			{
5607 				.type = qrss->type,
5608 				.conf = qrss->conf,
5609 			},
5610 			{
5611 				.type = RTE_FLOW_ACTION_TYPE_END,
5612 			},
5613 		};
5614 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5615 
5616 		/*
5617 		 * Configure the tag item only if there is no meter subflow.
5618 		 * Since tag is already marked in the meter suffix subflow
5619 		 * we can just use the meter suffix items as is.
5620 		 */
5621 		if (qrss_id) {
5622 			/* Not meter subflow. */
5623 			MLX5_ASSERT(!mtr_sfx);
5624 			/*
5625 			 * Put unique id in prefix flow due to it is destroyed
5626 			 * after suffix flow and id will be freed after there
5627 			 * is no actual flows with this id and identifier
5628 			 * reallocation becomes possible (for example, for
5629 			 * other flows in other threads).
5630 			 */
5631 			dev_flow->handle->split_flow_id = qrss_id;
5632 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5633 						   error);
5634 			if (ret < 0)
5635 				goto exit;
5636 			q_tag_spec.id = ret;
5637 		}
5638 		dev_flow = NULL;
5639 		/* Add suffix subflow to execute Q/RSS. */
5640 		flow_split_info->prefix_layers = layers;
5641 		flow_split_info->prefix_mark = 0;
5642 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5643 					      &q_attr, mtr_sfx ? items :
5644 					      q_items, q_actions,
5645 					      flow_split_info, error);
5646 		if (ret < 0)
5647 			goto exit;
5648 		/* qrss ID should be freed if failed. */
5649 		qrss_id = 0;
5650 		MLX5_ASSERT(dev_flow);
5651 	}
5652 
5653 exit:
5654 	/*
5655 	 * We do not destroy the partially created sub_flows in case of error.
5656 	 * These ones are included into parent flow list and will be destroyed
5657 	 * by flow_drv_destroy.
5658 	 */
5659 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5660 			qrss_id);
5661 	mlx5_free(ext_actions);
5662 	return ret;
5663 }
5664 
5665 /**
5666  * Create meter internal drop flow with the original pattern.
5667  *
5668  * @param dev
5669  *   Pointer to Ethernet device.
5670  * @param[in] flow
5671  *   Parent flow structure pointer.
5672  * @param[in] attr
5673  *   Flow rule attributes.
5674  * @param[in] items
5675  *   Pattern specification (list terminated by the END pattern item).
5676  * @param[in] flow_split_info
5677  *   Pointer to flow split info structure.
5678  * @param[in] fm
5679  *   Pointer to flow meter structure.
5680  * @param[out] error
5681  *   Perform verbose error reporting if not NULL.
5682  * @return
5683  *   0 on success, negative value otherwise
5684  */
5685 static uint32_t
5686 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5687 			struct rte_flow *flow,
5688 			const struct rte_flow_attr *attr,
5689 			const struct rte_flow_item items[],
5690 			struct mlx5_flow_split_info *flow_split_info,
5691 			struct mlx5_flow_meter_info *fm,
5692 			struct rte_flow_error *error)
5693 {
5694 	struct mlx5_flow *dev_flow = NULL;
5695 	struct rte_flow_attr drop_attr = *attr;
5696 	struct rte_flow_action drop_actions[3];
5697 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5698 
5699 	MLX5_ASSERT(fm->drop_cnt);
5700 	drop_actions[0].type =
5701 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5702 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5703 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5704 	drop_actions[1].conf = NULL;
5705 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5706 	drop_actions[2].conf = NULL;
5707 	drop_split_info.external = false;
5708 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5709 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5710 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5711 	return flow_create_split_inner(dev, flow, &dev_flow,
5712 				&drop_attr, items, drop_actions,
5713 				&drop_split_info, error);
5714 }
5715 
5716 /**
5717  * The splitting for meter feature.
5718  *
5719  * - The meter flow will be split to two flows as prefix and
5720  *   suffix flow. The packets make sense only it pass the prefix
5721  *   meter action.
5722  *
5723  * - Reg_C_5 is used for the packet to match betweend prefix and
5724  *   suffix flow.
5725  *
5726  * @param dev
5727  *   Pointer to Ethernet device.
5728  * @param[in] flow
5729  *   Parent flow structure pointer.
5730  * @param[in] attr
5731  *   Flow rule attributes.
5732  * @param[in] items
5733  *   Pattern specification (list terminated by the END pattern item).
5734  * @param[in] actions
5735  *   Associated actions (list terminated by the END action).
5736  * @param[in] flow_split_info
5737  *   Pointer to flow split info structure.
5738  * @param[out] error
5739  *   Perform verbose error reporting if not NULL.
5740  * @return
5741  *   0 on success, negative value otherwise
5742  */
5743 static int
5744 flow_create_split_meter(struct rte_eth_dev *dev,
5745 			struct rte_flow *flow,
5746 			const struct rte_flow_attr *attr,
5747 			const struct rte_flow_item items[],
5748 			const struct rte_flow_action actions[],
5749 			struct mlx5_flow_split_info *flow_split_info,
5750 			struct rte_flow_error *error)
5751 {
5752 	struct mlx5_priv *priv = dev->data->dev_private;
5753 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5754 	struct rte_flow_action *sfx_actions = NULL;
5755 	struct rte_flow_action *pre_actions = NULL;
5756 	struct rte_flow_item *sfx_items = NULL;
5757 	struct mlx5_flow *dev_flow = NULL;
5758 	struct rte_flow_attr sfx_attr = *attr;
5759 	struct mlx5_flow_meter_info *fm = NULL;
5760 	uint8_t skip_scale_restore;
5761 	bool has_mtr = false;
5762 	bool has_modify = false;
5763 	bool set_mtr_reg = true;
5764 	bool is_mtr_hierarchy = false;
5765 	uint32_t meter_id = 0;
5766 	uint32_t mtr_idx = 0;
5767 	uint32_t mtr_flow_id = 0;
5768 	size_t act_size;
5769 	size_t item_size;
5770 	int actions_n = 0;
5771 	int ret = 0;
5772 
5773 	if (priv->mtr_en)
5774 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5775 						    &has_modify, &meter_id);
5776 	if (has_mtr) {
5777 		if (flow->meter) {
5778 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5779 			if (!fm)
5780 				return rte_flow_error_set(error, EINVAL,
5781 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5782 						NULL, "Meter not found.");
5783 		} else {
5784 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5785 			if (!fm)
5786 				return rte_flow_error_set(error, EINVAL,
5787 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5788 						NULL, "Meter not found.");
5789 			ret = mlx5_flow_meter_attach(priv, fm,
5790 						     &sfx_attr, error);
5791 			if (ret)
5792 				return -rte_errno;
5793 			flow->meter = mtr_idx;
5794 		}
5795 		MLX5_ASSERT(wks);
5796 		wks->fm = fm;
5797 		if (!fm->def_policy) {
5798 			wks->policy = mlx5_flow_meter_policy_find(dev,
5799 								  fm->policy_id,
5800 								  NULL);
5801 			MLX5_ASSERT(wks->policy);
5802 			if (wks->policy->is_hierarchy) {
5803 				wks->final_policy =
5804 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
5805 								wks->policy);
5806 				if (!wks->final_policy)
5807 					return rte_flow_error_set(error,
5808 					EINVAL,
5809 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5810 				"Failed to find terminal policy of hierarchy.");
5811 				is_mtr_hierarchy = true;
5812 			}
5813 		}
5814 		/*
5815 		 * If it isn't default-policy Meter, and
5816 		 * 1. There's no action in flow to change
5817 		 *    packet (modify/encap/decap etc.), OR
5818 		 * 2. No drop count needed for this meter.
5819 		 * 3. It's not meter hierarchy.
5820 		 * Then no need to use regC to save meter id anymore.
5821 		 */
5822 		if (!fm->def_policy && !is_mtr_hierarchy &&
5823 		    (!has_modify || !fm->drop_cnt))
5824 			set_mtr_reg = false;
5825 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5826 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5827 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5828 		/* Suffix items: tag, vlan, port id, end. */
5829 #define METER_SUFFIX_ITEM 4
5830 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5831 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5832 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5833 					  0, SOCKET_ID_ANY);
5834 		if (!sfx_actions)
5835 			return rte_flow_error_set(error, ENOMEM,
5836 						  RTE_FLOW_ERROR_TYPE_ACTION,
5837 						  NULL, "no memory to split "
5838 						  "meter flow");
5839 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5840 			     act_size);
5841 		/* There's no suffix flow for meter of non-default policy. */
5842 		if (!fm->def_policy)
5843 			pre_actions = sfx_actions + 1;
5844 		else
5845 			pre_actions = sfx_actions + actions_n;
5846 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
5847 					    items, sfx_items, actions,
5848 					    sfx_actions, pre_actions,
5849 					    (set_mtr_reg ? &mtr_flow_id : NULL),
5850 					    error);
5851 		if (ret) {
5852 			ret = -rte_errno;
5853 			goto exit;
5854 		}
5855 		/* Add the prefix subflow. */
5856 		flow_split_info->prefix_mark = 0;
5857 		skip_scale_restore = flow_split_info->skip_scale;
5858 		flow_split_info->skip_scale |=
5859 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5860 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5861 					      attr, items, pre_actions,
5862 					      flow_split_info, error);
5863 		flow_split_info->skip_scale = skip_scale_restore;
5864 		if (ret) {
5865 			if (mtr_flow_id)
5866 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5867 			ret = -rte_errno;
5868 			goto exit;
5869 		}
5870 		if (mtr_flow_id) {
5871 			dev_flow->handle->split_flow_id = mtr_flow_id;
5872 			dev_flow->handle->is_meter_flow_id = 1;
5873 		}
5874 		if (!fm->def_policy) {
5875 			if (!set_mtr_reg && fm->drop_cnt)
5876 				ret =
5877 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5878 							&sfx_attr, items,
5879 							flow_split_info,
5880 							fm, error);
5881 			goto exit;
5882 		}
5883 		/* Setting the sfx group atrr. */
5884 		sfx_attr.group = sfx_attr.transfer ?
5885 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5886 				 MLX5_FLOW_TABLE_LEVEL_METER;
5887 		flow_split_info->prefix_layers =
5888 				flow_get_prefix_layer_flags(dev_flow);
5889 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5890 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5891 	}
5892 	/* Add the prefix subflow. */
5893 	ret = flow_create_split_metadata(dev, flow,
5894 					 &sfx_attr, sfx_items ?
5895 					 sfx_items : items,
5896 					 sfx_actions ? sfx_actions : actions,
5897 					 flow_split_info, error);
5898 exit:
5899 	if (sfx_actions)
5900 		mlx5_free(sfx_actions);
5901 	return ret;
5902 }
5903 
5904 /**
5905  * The splitting for sample feature.
5906  *
5907  * Once Sample action is detected in the action list, the flow actions should
5908  * be split into prefix sub flow and suffix sub flow.
5909  *
5910  * The original items remain in the prefix sub flow, all actions preceding the
5911  * sample action and the sample action itself will be copied to the prefix
5912  * sub flow, the actions following the sample action will be copied to the
5913  * suffix sub flow, Queue action always be located in the suffix sub flow.
5914  *
5915  * In order to make the packet from prefix sub flow matches with suffix sub
5916  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5917  * flow uses tag item with the unique flow id.
5918  *
5919  * @param dev
5920  *   Pointer to Ethernet device.
5921  * @param[in] flow
5922  *   Parent flow structure pointer.
5923  * @param[in] attr
5924  *   Flow rule attributes.
5925  * @param[in] items
5926  *   Pattern specification (list terminated by the END pattern item).
5927  * @param[in] actions
5928  *   Associated actions (list terminated by the END action).
5929  * @param[in] flow_split_info
5930  *   Pointer to flow split info structure.
5931  * @param[out] error
5932  *   Perform verbose error reporting if not NULL.
5933  * @return
5934  *   0 on success, negative value otherwise
5935  */
5936 static int
5937 flow_create_split_sample(struct rte_eth_dev *dev,
5938 			 struct rte_flow *flow,
5939 			 const struct rte_flow_attr *attr,
5940 			 const struct rte_flow_item items[],
5941 			 const struct rte_flow_action actions[],
5942 			 struct mlx5_flow_split_info *flow_split_info,
5943 			 struct rte_flow_error *error)
5944 {
5945 	struct mlx5_priv *priv = dev->data->dev_private;
5946 	struct rte_flow_action *sfx_actions = NULL;
5947 	struct rte_flow_action *pre_actions = NULL;
5948 	struct rte_flow_item *sfx_items = NULL;
5949 	struct mlx5_flow *dev_flow = NULL;
5950 	struct rte_flow_attr sfx_attr = *attr;
5951 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5952 	struct mlx5_flow_dv_sample_resource *sample_res;
5953 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5954 	struct mlx5_flow_tbl_resource *sfx_tbl;
5955 #endif
5956 	size_t act_size;
5957 	size_t item_size;
5958 	uint32_t fdb_tx = 0;
5959 	int32_t tag_id = 0;
5960 	int actions_n = 0;
5961 	int sample_action_pos;
5962 	int qrss_action_pos;
5963 	int add_tag = 0;
5964 	int modify_after_mirror = 0;
5965 	uint16_t jump_table = 0;
5966 	const uint32_t next_ft_step = 1;
5967 	int ret = 0;
5968 
5969 	if (priv->sampler_en)
5970 		actions_n = flow_check_match_action(actions, attr,
5971 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5972 					&sample_action_pos, &qrss_action_pos,
5973 					&modify_after_mirror);
5974 	if (actions_n) {
5975 		/* The prefix actions must includes sample, tag, end. */
5976 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5977 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5978 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5979 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5980 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5981 					  item_size), 0, SOCKET_ID_ANY);
5982 		if (!sfx_actions)
5983 			return rte_flow_error_set(error, ENOMEM,
5984 						  RTE_FLOW_ERROR_TYPE_ACTION,
5985 						  NULL, "no memory to split "
5986 						  "sample flow");
5987 		/* The representor_id is UINT16_MAX for uplink. */
5988 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
5989 		/*
5990 		 * When reg_c_preserve is set, metadata registers Cx preserve
5991 		 * their value even through packet duplication.
5992 		 */
5993 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5994 		if (add_tag)
5995 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5996 					+ act_size);
5997 		if (modify_after_mirror)
5998 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5999 				     next_ft_step;
6000 		pre_actions = sfx_actions + actions_n;
6001 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6002 						actions, sfx_actions,
6003 						pre_actions, actions_n,
6004 						sample_action_pos,
6005 						qrss_action_pos, jump_table,
6006 						error);
6007 		if (tag_id < 0 || (add_tag && !tag_id)) {
6008 			ret = -rte_errno;
6009 			goto exit;
6010 		}
6011 		if (modify_after_mirror)
6012 			flow_split_info->skip_scale =
6013 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6014 		/* Add the prefix subflow. */
6015 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6016 					      items, pre_actions,
6017 					      flow_split_info, error);
6018 		if (ret) {
6019 			ret = -rte_errno;
6020 			goto exit;
6021 		}
6022 		dev_flow->handle->split_flow_id = tag_id;
6023 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6024 		if (!modify_after_mirror) {
6025 			/* Set the sfx group attr. */
6026 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6027 						dev_flow->dv.sample_res;
6028 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6029 						sample_res->normal_path_tbl;
6030 			sfx_tbl_data = container_of(sfx_tbl,
6031 						struct mlx5_flow_tbl_data_entry,
6032 						tbl);
6033 			sfx_attr.group = sfx_attr.transfer ?
6034 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6035 		} else {
6036 			MLX5_ASSERT(attr->transfer);
6037 			sfx_attr.group = jump_table;
6038 		}
6039 		flow_split_info->prefix_layers =
6040 				flow_get_prefix_layer_flags(dev_flow);
6041 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6042 		/* Suffix group level already be scaled with factor, set
6043 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6044 		 * again in translation.
6045 		 */
6046 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6047 #endif
6048 	}
6049 	/* Add the suffix subflow. */
6050 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6051 				      sfx_items ? sfx_items : items,
6052 				      sfx_actions ? sfx_actions : actions,
6053 				      flow_split_info, error);
6054 exit:
6055 	if (sfx_actions)
6056 		mlx5_free(sfx_actions);
6057 	return ret;
6058 }
6059 
6060 /**
6061  * Split the flow to subflow set. The splitters might be linked
6062  * in the chain, like this:
6063  * flow_create_split_outer() calls:
6064  *   flow_create_split_meter() calls:
6065  *     flow_create_split_metadata(meter_subflow_0) calls:
6066  *       flow_create_split_inner(metadata_subflow_0)
6067  *       flow_create_split_inner(metadata_subflow_1)
6068  *       flow_create_split_inner(metadata_subflow_2)
6069  *     flow_create_split_metadata(meter_subflow_1) calls:
6070  *       flow_create_split_inner(metadata_subflow_0)
6071  *       flow_create_split_inner(metadata_subflow_1)
6072  *       flow_create_split_inner(metadata_subflow_2)
6073  *
6074  * This provide flexible way to add new levels of flow splitting.
6075  * The all of successfully created subflows are included to the
6076  * parent flow dev_flow list.
6077  *
6078  * @param dev
6079  *   Pointer to Ethernet device.
6080  * @param[in] flow
6081  *   Parent flow structure pointer.
6082  * @param[in] attr
6083  *   Flow rule attributes.
6084  * @param[in] items
6085  *   Pattern specification (list terminated by the END pattern item).
6086  * @param[in] actions
6087  *   Associated actions (list terminated by the END action).
6088  * @param[in] flow_split_info
6089  *   Pointer to flow split info structure.
6090  * @param[out] error
6091  *   Perform verbose error reporting if not NULL.
6092  * @return
6093  *   0 on success, negative value otherwise
6094  */
6095 static int
6096 flow_create_split_outer(struct rte_eth_dev *dev,
6097 			struct rte_flow *flow,
6098 			const struct rte_flow_attr *attr,
6099 			const struct rte_flow_item items[],
6100 			const struct rte_flow_action actions[],
6101 			struct mlx5_flow_split_info *flow_split_info,
6102 			struct rte_flow_error *error)
6103 {
6104 	int ret;
6105 
6106 	ret = flow_create_split_sample(dev, flow, attr, items,
6107 				       actions, flow_split_info, error);
6108 	MLX5_ASSERT(ret <= 0);
6109 	return ret;
6110 }
6111 
6112 static inline struct mlx5_flow_tunnel *
6113 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6114 {
6115 	struct mlx5_flow_tunnel *tunnel;
6116 
6117 #pragma GCC diagnostic push
6118 #pragma GCC diagnostic ignored "-Wcast-qual"
6119 	tunnel = (typeof(tunnel))flow->tunnel;
6120 #pragma GCC diagnostic pop
6121 
6122 	return tunnel;
6123 }
6124 
6125 /**
6126  * Adjust flow RSS workspace if needed.
6127  *
6128  * @param wks
6129  *   Pointer to thread flow work space.
6130  * @param rss_desc
6131  *   Pointer to RSS descriptor.
6132  * @param[in] nrssq_num
6133  *   New RSS queue number.
6134  *
6135  * @return
6136  *   0 on success, -1 otherwise and rte_errno is set.
6137  */
6138 static int
6139 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6140 			  struct mlx5_flow_rss_desc *rss_desc,
6141 			  uint32_t nrssq_num)
6142 {
6143 	if (likely(nrssq_num <= wks->rssq_num))
6144 		return 0;
6145 	rss_desc->queue = realloc(rss_desc->queue,
6146 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6147 	if (!rss_desc->queue) {
6148 		rte_errno = ENOMEM;
6149 		return -1;
6150 	}
6151 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6152 	return 0;
6153 }
6154 
6155 /**
6156  * Create a flow and add it to @p list.
6157  *
6158  * @param dev
6159  *   Pointer to Ethernet device.
6160  * @param list
6161  *   Pointer to a TAILQ flow list. If this parameter NULL,
6162  *   no list insertion occurred, flow is just created,
6163  *   this is caller's responsibility to track the
6164  *   created flow.
6165  * @param[in] attr
6166  *   Flow rule attributes.
6167  * @param[in] items
6168  *   Pattern specification (list terminated by the END pattern item).
6169  * @param[in] actions
6170  *   Associated actions (list terminated by the END action).
6171  * @param[in] external
6172  *   This flow rule is created by request external to PMD.
6173  * @param[out] error
6174  *   Perform verbose error reporting if not NULL.
6175  *
6176  * @return
6177  *   A flow index on success, 0 otherwise and rte_errno is set.
6178  */
6179 static uint32_t
6180 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6181 		 const struct rte_flow_attr *attr,
6182 		 const struct rte_flow_item items[],
6183 		 const struct rte_flow_action original_actions[],
6184 		 bool external, struct rte_flow_error *error)
6185 {
6186 	struct mlx5_priv *priv = dev->data->dev_private;
6187 	struct rte_flow *flow = NULL;
6188 	struct mlx5_flow *dev_flow;
6189 	const struct rte_flow_action_rss *rss = NULL;
6190 	struct mlx5_translated_action_handle
6191 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6192 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6193 	union {
6194 		struct mlx5_flow_expand_rss buf;
6195 		uint8_t buffer[2048];
6196 	} expand_buffer;
6197 	union {
6198 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6199 		uint8_t buffer[2048];
6200 	} actions_rx;
6201 	union {
6202 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6203 		uint8_t buffer[2048];
6204 	} actions_hairpin_tx;
6205 	union {
6206 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6207 		uint8_t buffer[2048];
6208 	} items_tx;
6209 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6210 	struct mlx5_flow_rss_desc *rss_desc;
6211 	const struct rte_flow_action *p_actions_rx;
6212 	uint32_t i;
6213 	uint32_t idx = 0;
6214 	int hairpin_flow;
6215 	struct rte_flow_attr attr_tx = { .priority = 0 };
6216 	const struct rte_flow_action *actions;
6217 	struct rte_flow_action *translated_actions = NULL;
6218 	struct mlx5_flow_tunnel *tunnel;
6219 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6220 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6221 	struct mlx5_flow_split_info flow_split_info = {
6222 		.external = !!external,
6223 		.skip_scale = 0,
6224 		.flow_idx = 0,
6225 		.prefix_mark = 0,
6226 		.prefix_layers = 0,
6227 		.table_id = 0
6228 	};
6229 	int ret;
6230 
6231 	MLX5_ASSERT(wks);
6232 	rss_desc = &wks->rss_desc;
6233 	ret = flow_action_handles_translate(dev, original_actions,
6234 					    indir_actions,
6235 					    &indir_actions_n,
6236 					    &translated_actions, error);
6237 	if (ret < 0) {
6238 		MLX5_ASSERT(translated_actions == NULL);
6239 		return 0;
6240 	}
6241 	actions = translated_actions ? translated_actions : original_actions;
6242 	p_actions_rx = actions;
6243 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6244 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6245 				external, hairpin_flow, error);
6246 	if (ret < 0)
6247 		goto error_before_hairpin_split;
6248 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6249 	if (!flow) {
6250 		rte_errno = ENOMEM;
6251 		goto error_before_hairpin_split;
6252 	}
6253 	if (hairpin_flow > 0) {
6254 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6255 			rte_errno = EINVAL;
6256 			goto error_before_hairpin_split;
6257 		}
6258 		flow_hairpin_split(dev, actions, actions_rx.actions,
6259 				   actions_hairpin_tx.actions, items_tx.items,
6260 				   idx);
6261 		p_actions_rx = actions_rx.actions;
6262 	}
6263 	flow_split_info.flow_idx = idx;
6264 	flow->drv_type = flow_get_drv_type(dev, attr);
6265 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6266 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6267 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6268 	/* RSS Action only works on NIC RX domain */
6269 	if (attr->ingress && !attr->transfer)
6270 		rss = flow_get_rss_action(dev, p_actions_rx);
6271 	if (rss) {
6272 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6273 			return 0;
6274 		/*
6275 		 * The following information is required by
6276 		 * mlx5_flow_hashfields_adjust() in advance.
6277 		 */
6278 		rss_desc->level = rss->level;
6279 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6280 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6281 	}
6282 	flow->dev_handles = 0;
6283 	if (rss && rss->types) {
6284 		unsigned int graph_root;
6285 
6286 		graph_root = find_graph_root(items, rss->level);
6287 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6288 					   items, rss->types,
6289 					   mlx5_support_expansion, graph_root);
6290 		MLX5_ASSERT(ret > 0 &&
6291 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6292 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6293 			for (i = 0; i < buf->entries; ++i)
6294 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6295 		}
6296 	} else {
6297 		buf->entries = 1;
6298 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6299 	}
6300 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6301 						      indir_actions_n);
6302 	for (i = 0; i < buf->entries; ++i) {
6303 		/* Initialize flow split data. */
6304 		flow_split_info.prefix_layers = 0;
6305 		flow_split_info.prefix_mark = 0;
6306 		flow_split_info.skip_scale = 0;
6307 		/*
6308 		 * The splitter may create multiple dev_flows,
6309 		 * depending on configuration. In the simplest
6310 		 * case it just creates unmodified original flow.
6311 		 */
6312 		ret = flow_create_split_outer(dev, flow, attr,
6313 					      buf->entry[i].pattern,
6314 					      p_actions_rx, &flow_split_info,
6315 					      error);
6316 		if (ret < 0)
6317 			goto error;
6318 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6319 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6320 							   p_actions_rx,
6321 							   idx,
6322 							   wks->flows[0].tunnel,
6323 							   &default_miss_ctx,
6324 							   error);
6325 			if (ret < 0) {
6326 				mlx5_free(default_miss_ctx.queue);
6327 				goto error;
6328 			}
6329 		}
6330 	}
6331 	/* Create the tx flow. */
6332 	if (hairpin_flow) {
6333 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6334 		attr_tx.ingress = 0;
6335 		attr_tx.egress = 1;
6336 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6337 					 actions_hairpin_tx.actions,
6338 					 idx, error);
6339 		if (!dev_flow)
6340 			goto error;
6341 		dev_flow->flow = flow;
6342 		dev_flow->external = 0;
6343 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6344 			      dev_flow->handle, next);
6345 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6346 					 items_tx.items,
6347 					 actions_hairpin_tx.actions, error);
6348 		if (ret < 0)
6349 			goto error;
6350 	}
6351 	/*
6352 	 * Update the metadata register copy table. If extensive
6353 	 * metadata feature is enabled and registers are supported
6354 	 * we might create the extra rte_flow for each unique
6355 	 * MARK/FLAG action ID.
6356 	 *
6357 	 * The table is updated for ingress Flows only, because
6358 	 * the egress Flows belong to the different device and
6359 	 * copy table should be updated in peer NIC Rx domain.
6360 	 */
6361 	if (attr->ingress &&
6362 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6363 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6364 		if (ret)
6365 			goto error;
6366 	}
6367 	/*
6368 	 * If the flow is external (from application) OR device is started,
6369 	 * OR mreg discover, then apply immediately.
6370 	 */
6371 	if (external || dev->data->dev_started ||
6372 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6373 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6374 		ret = flow_drv_apply(dev, flow, error);
6375 		if (ret < 0)
6376 			goto error;
6377 	}
6378 	flow->type = type;
6379 	flow_rxq_flags_set(dev, flow);
6380 	rte_free(translated_actions);
6381 	tunnel = flow_tunnel_from_rule(wks->flows);
6382 	if (tunnel) {
6383 		flow->tunnel = 1;
6384 		flow->tunnel_id = tunnel->tunnel_id;
6385 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6386 		mlx5_free(default_miss_ctx.queue);
6387 	}
6388 	mlx5_flow_pop_thread_workspace();
6389 	return idx;
6390 error:
6391 	MLX5_ASSERT(flow);
6392 	ret = rte_errno; /* Save rte_errno before cleanup. */
6393 	flow_mreg_del_copy_action(dev, flow);
6394 	flow_drv_destroy(dev, flow);
6395 	if (rss_desc->shared_rss)
6396 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6397 			mlx5_ipool_get
6398 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6399 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6400 	mlx5_ipool_free(priv->flows[type], idx);
6401 	rte_errno = ret; /* Restore rte_errno. */
6402 	ret = rte_errno;
6403 	rte_errno = ret;
6404 	mlx5_flow_pop_thread_workspace();
6405 error_before_hairpin_split:
6406 	rte_free(translated_actions);
6407 	return 0;
6408 }
6409 
6410 /**
6411  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6412  * incoming packets to table 1.
6413  *
6414  * Other flow rules, requested for group n, will be created in
6415  * e-switch table n+1.
6416  * Jump action to e-switch group n will be created to group n+1.
6417  *
6418  * Used when working in switchdev mode, to utilise advantages of table 1
6419  * and above.
6420  *
6421  * @param dev
6422  *   Pointer to Ethernet device.
6423  *
6424  * @return
6425  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6426  */
6427 struct rte_flow *
6428 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6429 {
6430 	const struct rte_flow_attr attr = {
6431 		.group = 0,
6432 		.priority = 0,
6433 		.ingress = 1,
6434 		.egress = 0,
6435 		.transfer = 1,
6436 	};
6437 	const struct rte_flow_item pattern = {
6438 		.type = RTE_FLOW_ITEM_TYPE_END,
6439 	};
6440 	struct rte_flow_action_jump jump = {
6441 		.group = 1,
6442 	};
6443 	const struct rte_flow_action actions[] = {
6444 		{
6445 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6446 			.conf = &jump,
6447 		},
6448 		{
6449 			.type = RTE_FLOW_ACTION_TYPE_END,
6450 		},
6451 	};
6452 	struct rte_flow_error error;
6453 
6454 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6455 						   &attr, &pattern,
6456 						   actions, false, &error);
6457 }
6458 
6459 /**
6460  * Validate a flow supported by the NIC.
6461  *
6462  * @see rte_flow_validate()
6463  * @see rte_flow_ops
6464  */
6465 int
6466 mlx5_flow_validate(struct rte_eth_dev *dev,
6467 		   const struct rte_flow_attr *attr,
6468 		   const struct rte_flow_item items[],
6469 		   const struct rte_flow_action original_actions[],
6470 		   struct rte_flow_error *error)
6471 {
6472 	int hairpin_flow;
6473 	struct mlx5_translated_action_handle
6474 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6475 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6476 	const struct rte_flow_action *actions;
6477 	struct rte_flow_action *translated_actions = NULL;
6478 	int ret = flow_action_handles_translate(dev, original_actions,
6479 						indir_actions,
6480 						&indir_actions_n,
6481 						&translated_actions, error);
6482 
6483 	if (ret)
6484 		return ret;
6485 	actions = translated_actions ? translated_actions : original_actions;
6486 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6487 	ret = flow_drv_validate(dev, attr, items, actions,
6488 				true, hairpin_flow, error);
6489 	rte_free(translated_actions);
6490 	return ret;
6491 }
6492 
6493 /**
6494  * Create a flow.
6495  *
6496  * @see rte_flow_create()
6497  * @see rte_flow_ops
6498  */
6499 struct rte_flow *
6500 mlx5_flow_create(struct rte_eth_dev *dev,
6501 		 const struct rte_flow_attr *attr,
6502 		 const struct rte_flow_item items[],
6503 		 const struct rte_flow_action actions[],
6504 		 struct rte_flow_error *error)
6505 {
6506 	/*
6507 	 * If the device is not started yet, it is not allowed to created a
6508 	 * flow from application. PMD default flows and traffic control flows
6509 	 * are not affected.
6510 	 */
6511 	if (unlikely(!dev->data->dev_started)) {
6512 		DRV_LOG(DEBUG, "port %u is not started when "
6513 			"inserting a flow", dev->data->port_id);
6514 		rte_flow_error_set(error, ENODEV,
6515 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6516 				   NULL,
6517 				   "port not started");
6518 		return NULL;
6519 	}
6520 
6521 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6522 						   attr, items, actions,
6523 						   true, error);
6524 }
6525 
6526 /**
6527  * Destroy a flow in a list.
6528  *
6529  * @param dev
6530  *   Pointer to Ethernet device.
6531  * @param[in] flow_idx
6532  *   Index of flow to destroy.
6533  */
6534 static void
6535 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6536 		  uint32_t flow_idx)
6537 {
6538 	struct mlx5_priv *priv = dev->data->dev_private;
6539 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6540 
6541 	if (!flow)
6542 		return;
6543 	MLX5_ASSERT(flow->type == type);
6544 	/*
6545 	 * Update RX queue flags only if port is started, otherwise it is
6546 	 * already clean.
6547 	 */
6548 	if (dev->data->dev_started)
6549 		flow_rxq_flags_trim(dev, flow);
6550 	flow_drv_destroy(dev, flow);
6551 	if (flow->tunnel) {
6552 		struct mlx5_flow_tunnel *tunnel;
6553 
6554 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6555 		RTE_VERIFY(tunnel);
6556 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6557 			mlx5_flow_tunnel_free(dev, tunnel);
6558 	}
6559 	flow_mreg_del_copy_action(dev, flow);
6560 	mlx5_ipool_free(priv->flows[type], flow_idx);
6561 }
6562 
6563 /**
6564  * Destroy all flows.
6565  *
6566  * @param dev
6567  *   Pointer to Ethernet device.
6568  * @param type
6569  *   Flow type to be flushed.
6570  * @param active
6571  *   If flushing is called avtively.
6572  */
6573 void
6574 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6575 		     bool active)
6576 {
6577 	struct mlx5_priv *priv = dev->data->dev_private;
6578 	uint32_t num_flushed = 0, fidx = 1;
6579 	struct rte_flow *flow;
6580 
6581 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6582 		flow_list_destroy(dev, type, fidx);
6583 		num_flushed++;
6584 	}
6585 	if (active) {
6586 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6587 			dev->data->port_id, num_flushed);
6588 	}
6589 }
6590 
6591 /**
6592  * Stop all default actions for flows.
6593  *
6594  * @param dev
6595  *   Pointer to Ethernet device.
6596  */
6597 void
6598 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6599 {
6600 	flow_mreg_del_default_copy_action(dev);
6601 	flow_rxq_flags_clear(dev);
6602 }
6603 
6604 /**
6605  * Start all default actions for flows.
6606  *
6607  * @param dev
6608  *   Pointer to Ethernet device.
6609  * @return
6610  *   0 on success, a negative errno value otherwise and rte_errno is set.
6611  */
6612 int
6613 mlx5_flow_start_default(struct rte_eth_dev *dev)
6614 {
6615 	struct rte_flow_error error;
6616 
6617 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6618 	return flow_mreg_add_default_copy_action(dev, &error);
6619 }
6620 
6621 /**
6622  * Release key of thread specific flow workspace data.
6623  */
6624 void
6625 flow_release_workspace(void *data)
6626 {
6627 	struct mlx5_flow_workspace *wks = data;
6628 	struct mlx5_flow_workspace *next;
6629 
6630 	while (wks) {
6631 		next = wks->next;
6632 		free(wks->rss_desc.queue);
6633 		free(wks);
6634 		wks = next;
6635 	}
6636 }
6637 
6638 /**
6639  * Get thread specific current flow workspace.
6640  *
6641  * @return pointer to thread specific flow workspace data, NULL on error.
6642  */
6643 struct mlx5_flow_workspace*
6644 mlx5_flow_get_thread_workspace(void)
6645 {
6646 	struct mlx5_flow_workspace *data;
6647 
6648 	data = mlx5_flow_os_get_specific_workspace();
6649 	MLX5_ASSERT(data && data->inuse);
6650 	if (!data || !data->inuse)
6651 		DRV_LOG(ERR, "flow workspace not initialized.");
6652 	return data;
6653 }
6654 
6655 /**
6656  * Allocate and init new flow workspace.
6657  *
6658  * @return pointer to flow workspace data, NULL on error.
6659  */
6660 static struct mlx5_flow_workspace*
6661 flow_alloc_thread_workspace(void)
6662 {
6663 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6664 
6665 	if (!data) {
6666 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6667 			"memory.");
6668 		return NULL;
6669 	}
6670 	data->rss_desc.queue = calloc(1,
6671 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6672 	if (!data->rss_desc.queue)
6673 		goto err;
6674 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6675 	return data;
6676 err:
6677 	if (data->rss_desc.queue)
6678 		free(data->rss_desc.queue);
6679 	free(data);
6680 	return NULL;
6681 }
6682 
6683 /**
6684  * Get new thread specific flow workspace.
6685  *
6686  * If current workspace inuse, create new one and set as current.
6687  *
6688  * @return pointer to thread specific flow workspace data, NULL on error.
6689  */
6690 static struct mlx5_flow_workspace*
6691 mlx5_flow_push_thread_workspace(void)
6692 {
6693 	struct mlx5_flow_workspace *curr;
6694 	struct mlx5_flow_workspace *data;
6695 
6696 	curr = mlx5_flow_os_get_specific_workspace();
6697 	if (!curr) {
6698 		data = flow_alloc_thread_workspace();
6699 		if (!data)
6700 			return NULL;
6701 	} else if (!curr->inuse) {
6702 		data = curr;
6703 	} else if (curr->next) {
6704 		data = curr->next;
6705 	} else {
6706 		data = flow_alloc_thread_workspace();
6707 		if (!data)
6708 			return NULL;
6709 		curr->next = data;
6710 		data->prev = curr;
6711 	}
6712 	data->inuse = 1;
6713 	data->flow_idx = 0;
6714 	/* Set as current workspace */
6715 	if (mlx5_flow_os_set_specific_workspace(data))
6716 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6717 	return data;
6718 }
6719 
6720 /**
6721  * Close current thread specific flow workspace.
6722  *
6723  * If previous workspace available, set it as current.
6724  *
6725  * @return pointer to thread specific flow workspace data, NULL on error.
6726  */
6727 static void
6728 mlx5_flow_pop_thread_workspace(void)
6729 {
6730 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6731 
6732 	if (!data)
6733 		return;
6734 	if (!data->inuse) {
6735 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6736 		return;
6737 	}
6738 	data->inuse = 0;
6739 	if (!data->prev)
6740 		return;
6741 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6742 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6743 }
6744 
6745 /**
6746  * Verify the flow list is empty
6747  *
6748  * @param dev
6749  *  Pointer to Ethernet device.
6750  *
6751  * @return the number of flows not released.
6752  */
6753 int
6754 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
6755 {
6756 	struct mlx5_priv *priv = dev->data->dev_private;
6757 	struct rte_flow *flow;
6758 	uint32_t idx = 0;
6759 	int ret = 0, i;
6760 
6761 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
6762 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
6763 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
6764 				dev->data->port_id, (void *)flow);
6765 			ret++;
6766 		}
6767 	}
6768 	return ret;
6769 }
6770 
6771 /**
6772  * Enable default hairpin egress flow.
6773  *
6774  * @param dev
6775  *   Pointer to Ethernet device.
6776  * @param queue
6777  *   The queue index.
6778  *
6779  * @return
6780  *   0 on success, a negative errno value otherwise and rte_errno is set.
6781  */
6782 int
6783 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6784 			    uint32_t queue)
6785 {
6786 	const struct rte_flow_attr attr = {
6787 		.egress = 1,
6788 		.priority = 0,
6789 	};
6790 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6791 		.queue = queue,
6792 	};
6793 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6794 		.queue = UINT32_MAX,
6795 	};
6796 	struct rte_flow_item items[] = {
6797 		{
6798 			.type = (enum rte_flow_item_type)
6799 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6800 			.spec = &queue_spec,
6801 			.last = NULL,
6802 			.mask = &queue_mask,
6803 		},
6804 		{
6805 			.type = RTE_FLOW_ITEM_TYPE_END,
6806 		},
6807 	};
6808 	struct rte_flow_action_jump jump = {
6809 		.group = MLX5_HAIRPIN_TX_TABLE,
6810 	};
6811 	struct rte_flow_action actions[2];
6812 	uint32_t flow_idx;
6813 	struct rte_flow_error error;
6814 
6815 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6816 	actions[0].conf = &jump;
6817 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6818 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6819 				    &attr, items, actions, false, &error);
6820 	if (!flow_idx) {
6821 		DRV_LOG(DEBUG,
6822 			"Failed to create ctrl flow: rte_errno(%d),"
6823 			" type(%d), message(%s)",
6824 			rte_errno, error.type,
6825 			error.message ? error.message : " (no stated reason)");
6826 		return -rte_errno;
6827 	}
6828 	return 0;
6829 }
6830 
6831 /**
6832  * Enable a control flow configured from the control plane.
6833  *
6834  * @param dev
6835  *   Pointer to Ethernet device.
6836  * @param eth_spec
6837  *   An Ethernet flow spec to apply.
6838  * @param eth_mask
6839  *   An Ethernet flow mask to apply.
6840  * @param vlan_spec
6841  *   A VLAN flow spec to apply.
6842  * @param vlan_mask
6843  *   A VLAN flow mask to apply.
6844  *
6845  * @return
6846  *   0 on success, a negative errno value otherwise and rte_errno is set.
6847  */
6848 int
6849 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6850 		    struct rte_flow_item_eth *eth_spec,
6851 		    struct rte_flow_item_eth *eth_mask,
6852 		    struct rte_flow_item_vlan *vlan_spec,
6853 		    struct rte_flow_item_vlan *vlan_mask)
6854 {
6855 	struct mlx5_priv *priv = dev->data->dev_private;
6856 	const struct rte_flow_attr attr = {
6857 		.ingress = 1,
6858 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6859 	};
6860 	struct rte_flow_item items[] = {
6861 		{
6862 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6863 			.spec = eth_spec,
6864 			.last = NULL,
6865 			.mask = eth_mask,
6866 		},
6867 		{
6868 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6869 					      RTE_FLOW_ITEM_TYPE_END,
6870 			.spec = vlan_spec,
6871 			.last = NULL,
6872 			.mask = vlan_mask,
6873 		},
6874 		{
6875 			.type = RTE_FLOW_ITEM_TYPE_END,
6876 		},
6877 	};
6878 	uint16_t queue[priv->reta_idx_n];
6879 	struct rte_flow_action_rss action_rss = {
6880 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6881 		.level = 0,
6882 		.types = priv->rss_conf.rss_hf,
6883 		.key_len = priv->rss_conf.rss_key_len,
6884 		.queue_num = priv->reta_idx_n,
6885 		.key = priv->rss_conf.rss_key,
6886 		.queue = queue,
6887 	};
6888 	struct rte_flow_action actions[] = {
6889 		{
6890 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6891 			.conf = &action_rss,
6892 		},
6893 		{
6894 			.type = RTE_FLOW_ACTION_TYPE_END,
6895 		},
6896 	};
6897 	uint32_t flow_idx;
6898 	struct rte_flow_error error;
6899 	unsigned int i;
6900 
6901 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6902 		return 0;
6903 	}
6904 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6905 		action_rss.types = 0;
6906 	for (i = 0; i != priv->reta_idx_n; ++i)
6907 		queue[i] = (*priv->reta_idx)[i];
6908 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6909 				    &attr, items, actions, false, &error);
6910 	if (!flow_idx)
6911 		return -rte_errno;
6912 	return 0;
6913 }
6914 
6915 /**
6916  * Enable a flow control configured from the control plane.
6917  *
6918  * @param dev
6919  *   Pointer to Ethernet device.
6920  * @param eth_spec
6921  *   An Ethernet flow spec to apply.
6922  * @param eth_mask
6923  *   An Ethernet flow mask to apply.
6924  *
6925  * @return
6926  *   0 on success, a negative errno value otherwise and rte_errno is set.
6927  */
6928 int
6929 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6930 	       struct rte_flow_item_eth *eth_spec,
6931 	       struct rte_flow_item_eth *eth_mask)
6932 {
6933 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6934 }
6935 
6936 /**
6937  * Create default miss flow rule matching lacp traffic
6938  *
6939  * @param dev
6940  *   Pointer to Ethernet device.
6941  * @param eth_spec
6942  *   An Ethernet flow spec to apply.
6943  *
6944  * @return
6945  *   0 on success, a negative errno value otherwise and rte_errno is set.
6946  */
6947 int
6948 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6949 {
6950 	/*
6951 	 * The LACP matching is done by only using ether type since using
6952 	 * a multicast dst mac causes kernel to give low priority to this flow.
6953 	 */
6954 	static const struct rte_flow_item_eth lacp_spec = {
6955 		.type = RTE_BE16(0x8809),
6956 	};
6957 	static const struct rte_flow_item_eth lacp_mask = {
6958 		.type = 0xffff,
6959 	};
6960 	const struct rte_flow_attr attr = {
6961 		.ingress = 1,
6962 	};
6963 	struct rte_flow_item items[] = {
6964 		{
6965 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6966 			.spec = &lacp_spec,
6967 			.mask = &lacp_mask,
6968 		},
6969 		{
6970 			.type = RTE_FLOW_ITEM_TYPE_END,
6971 		},
6972 	};
6973 	struct rte_flow_action actions[] = {
6974 		{
6975 			.type = (enum rte_flow_action_type)
6976 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6977 		},
6978 		{
6979 			.type = RTE_FLOW_ACTION_TYPE_END,
6980 		},
6981 	};
6982 	struct rte_flow_error error;
6983 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6984 					&attr, items, actions,
6985 					false, &error);
6986 
6987 	if (!flow_idx)
6988 		return -rte_errno;
6989 	return 0;
6990 }
6991 
6992 /**
6993  * Destroy a flow.
6994  *
6995  * @see rte_flow_destroy()
6996  * @see rte_flow_ops
6997  */
6998 int
6999 mlx5_flow_destroy(struct rte_eth_dev *dev,
7000 		  struct rte_flow *flow,
7001 		  struct rte_flow_error *error __rte_unused)
7002 {
7003 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7004 				(uintptr_t)(void *)flow);
7005 	return 0;
7006 }
7007 
7008 /**
7009  * Destroy all flows.
7010  *
7011  * @see rte_flow_flush()
7012  * @see rte_flow_ops
7013  */
7014 int
7015 mlx5_flow_flush(struct rte_eth_dev *dev,
7016 		struct rte_flow_error *error __rte_unused)
7017 {
7018 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7019 	return 0;
7020 }
7021 
7022 /**
7023  * Isolated mode.
7024  *
7025  * @see rte_flow_isolate()
7026  * @see rte_flow_ops
7027  */
7028 int
7029 mlx5_flow_isolate(struct rte_eth_dev *dev,
7030 		  int enable,
7031 		  struct rte_flow_error *error)
7032 {
7033 	struct mlx5_priv *priv = dev->data->dev_private;
7034 
7035 	if (dev->data->dev_started) {
7036 		rte_flow_error_set(error, EBUSY,
7037 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7038 				   NULL,
7039 				   "port must be stopped first");
7040 		return -rte_errno;
7041 	}
7042 	priv->isolated = !!enable;
7043 	if (enable)
7044 		dev->dev_ops = &mlx5_dev_ops_isolate;
7045 	else
7046 		dev->dev_ops = &mlx5_dev_ops;
7047 
7048 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7049 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7050 
7051 	return 0;
7052 }
7053 
7054 /**
7055  * Query a flow.
7056  *
7057  * @see rte_flow_query()
7058  * @see rte_flow_ops
7059  */
7060 static int
7061 flow_drv_query(struct rte_eth_dev *dev,
7062 	       uint32_t flow_idx,
7063 	       const struct rte_flow_action *actions,
7064 	       void *data,
7065 	       struct rte_flow_error *error)
7066 {
7067 	struct mlx5_priv *priv = dev->data->dev_private;
7068 	const struct mlx5_flow_driver_ops *fops;
7069 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7070 					       flow_idx);
7071 	enum mlx5_flow_drv_type ftype;
7072 
7073 	if (!flow) {
7074 		return rte_flow_error_set(error, ENOENT,
7075 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7076 			  NULL,
7077 			  "invalid flow handle");
7078 	}
7079 	ftype = flow->drv_type;
7080 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7081 	fops = flow_get_drv_ops(ftype);
7082 
7083 	return fops->query(dev, flow, actions, data, error);
7084 }
7085 
7086 /**
7087  * Query a flow.
7088  *
7089  * @see rte_flow_query()
7090  * @see rte_flow_ops
7091  */
7092 int
7093 mlx5_flow_query(struct rte_eth_dev *dev,
7094 		struct rte_flow *flow,
7095 		const struct rte_flow_action *actions,
7096 		void *data,
7097 		struct rte_flow_error *error)
7098 {
7099 	int ret;
7100 
7101 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7102 			     error);
7103 	if (ret < 0)
7104 		return ret;
7105 	return 0;
7106 }
7107 
7108 /**
7109  * Get rte_flow callbacks.
7110  *
7111  * @param dev
7112  *   Pointer to Ethernet device structure.
7113  * @param ops
7114  *   Pointer to operation-specific structure.
7115  *
7116  * @return 0
7117  */
7118 int
7119 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7120 		  const struct rte_flow_ops **ops)
7121 {
7122 	*ops = &mlx5_flow_ops;
7123 	return 0;
7124 }
7125 
7126 /**
7127  * Validate meter policy actions.
7128  * Dispatcher for action type specific validation.
7129  *
7130  * @param[in] dev
7131  *   Pointer to the Ethernet device structure.
7132  * @param[in] action
7133  *   The meter policy action object to validate.
7134  * @param[in] attr
7135  *   Attributes of flow to determine steering domain.
7136  * @param[out] is_rss
7137  *   Is RSS or not.
7138  * @param[out] domain_bitmap
7139  *   Domain bitmap.
7140  * @param[out] is_def_policy
7141  *   Is default policy or not.
7142  * @param[out] error
7143  *   Perform verbose error reporting if not NULL. Initialized in case of
7144  *   error only.
7145  *
7146  * @return
7147  *   0 on success, otherwise negative errno value.
7148  */
7149 int
7150 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7151 			const struct rte_flow_action *actions[RTE_COLORS],
7152 			struct rte_flow_attr *attr,
7153 			bool *is_rss,
7154 			uint8_t *domain_bitmap,
7155 			bool *is_def_policy,
7156 			struct rte_mtr_error *error)
7157 {
7158 	const struct mlx5_flow_driver_ops *fops;
7159 
7160 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7161 	return fops->validate_mtr_acts(dev, actions, attr,
7162 			is_rss, domain_bitmap, is_def_policy, error);
7163 }
7164 
7165 /**
7166  * Destroy the meter table set.
7167  *
7168  * @param[in] dev
7169  *   Pointer to Ethernet device.
7170  * @param[in] mtr_policy
7171  *   Meter policy struct.
7172  */
7173 void
7174 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7175 		      struct mlx5_flow_meter_policy *mtr_policy)
7176 {
7177 	const struct mlx5_flow_driver_ops *fops;
7178 
7179 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7180 	fops->destroy_mtr_acts(dev, mtr_policy);
7181 }
7182 
7183 /**
7184  * Create policy action, lock free,
7185  * (mutex should be acquired by caller).
7186  * Dispatcher for action type specific call.
7187  *
7188  * @param[in] dev
7189  *   Pointer to the Ethernet device structure.
7190  * @param[in] mtr_policy
7191  *   Meter policy struct.
7192  * @param[in] action
7193  *   Action specification used to create meter actions.
7194  * @param[out] error
7195  *   Perform verbose error reporting if not NULL. Initialized in case of
7196  *   error only.
7197  *
7198  * @return
7199  *   0 on success, otherwise negative errno value.
7200  */
7201 int
7202 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7203 		      struct mlx5_flow_meter_policy *mtr_policy,
7204 		      const struct rte_flow_action *actions[RTE_COLORS],
7205 		      struct rte_mtr_error *error)
7206 {
7207 	const struct mlx5_flow_driver_ops *fops;
7208 
7209 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7210 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7211 }
7212 
7213 /**
7214  * Create policy rules, lock free,
7215  * (mutex should be acquired by caller).
7216  * Dispatcher for action type specific call.
7217  *
7218  * @param[in] dev
7219  *   Pointer to the Ethernet device structure.
7220  * @param[in] mtr_policy
7221  *   Meter policy struct.
7222  *
7223  * @return
7224  *   0 on success, -1 otherwise.
7225  */
7226 int
7227 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7228 			     struct mlx5_flow_meter_policy *mtr_policy)
7229 {
7230 	const struct mlx5_flow_driver_ops *fops;
7231 
7232 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7233 	return fops->create_policy_rules(dev, mtr_policy);
7234 }
7235 
7236 /**
7237  * Destroy policy rules, lock free,
7238  * (mutex should be acquired by caller).
7239  * Dispatcher for action type specific call.
7240  *
7241  * @param[in] dev
7242  *   Pointer to the Ethernet device structure.
7243  * @param[in] mtr_policy
7244  *   Meter policy struct.
7245  */
7246 void
7247 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7248 			     struct mlx5_flow_meter_policy *mtr_policy)
7249 {
7250 	const struct mlx5_flow_driver_ops *fops;
7251 
7252 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7253 	fops->destroy_policy_rules(dev, mtr_policy);
7254 }
7255 
7256 /**
7257  * Destroy the default policy table set.
7258  *
7259  * @param[in] dev
7260  *   Pointer to Ethernet device.
7261  */
7262 void
7263 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7264 {
7265 	const struct mlx5_flow_driver_ops *fops;
7266 
7267 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7268 	fops->destroy_def_policy(dev);
7269 }
7270 
7271 /**
7272  * Destroy the default policy table set.
7273  *
7274  * @param[in] dev
7275  *   Pointer to Ethernet device.
7276  *
7277  * @return
7278  *   0 on success, -1 otherwise.
7279  */
7280 int
7281 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7282 {
7283 	const struct mlx5_flow_driver_ops *fops;
7284 
7285 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7286 	return fops->create_def_policy(dev);
7287 }
7288 
7289 /**
7290  * Create the needed meter and suffix tables.
7291  *
7292  * @param[in] dev
7293  *   Pointer to Ethernet device.
7294  *
7295  * @return
7296  *   0 on success, -1 otherwise.
7297  */
7298 int
7299 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7300 			struct mlx5_flow_meter_info *fm,
7301 			uint32_t mtr_idx,
7302 			uint8_t domain_bitmap)
7303 {
7304 	const struct mlx5_flow_driver_ops *fops;
7305 
7306 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7307 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7308 }
7309 
7310 /**
7311  * Destroy the meter table set.
7312  *
7313  * @param[in] dev
7314  *   Pointer to Ethernet device.
7315  * @param[in] tbl
7316  *   Pointer to the meter table set.
7317  */
7318 void
7319 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7320 			   struct mlx5_flow_meter_info *fm)
7321 {
7322 	const struct mlx5_flow_driver_ops *fops;
7323 
7324 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7325 	fops->destroy_mtr_tbls(dev, fm);
7326 }
7327 
7328 /**
7329  * Destroy the global meter drop table.
7330  *
7331  * @param[in] dev
7332  *   Pointer to Ethernet device.
7333  */
7334 void
7335 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7336 {
7337 	const struct mlx5_flow_driver_ops *fops;
7338 
7339 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7340 	fops->destroy_mtr_drop_tbls(dev);
7341 }
7342 
7343 /**
7344  * Destroy the sub policy table with RX queue.
7345  *
7346  * @param[in] dev
7347  *   Pointer to Ethernet device.
7348  * @param[in] mtr_policy
7349  *   Pointer to meter policy table.
7350  */
7351 void
7352 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7353 		struct mlx5_flow_meter_policy *mtr_policy)
7354 {
7355 	const struct mlx5_flow_driver_ops *fops;
7356 
7357 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7358 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7359 }
7360 
7361 /**
7362  * Allocate the needed aso flow meter id.
7363  *
7364  * @param[in] dev
7365  *   Pointer to Ethernet device.
7366  *
7367  * @return
7368  *   Index to aso flow meter on success, NULL otherwise.
7369  */
7370 uint32_t
7371 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7372 {
7373 	const struct mlx5_flow_driver_ops *fops;
7374 
7375 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7376 	return fops->create_meter(dev);
7377 }
7378 
7379 /**
7380  * Free the aso flow meter id.
7381  *
7382  * @param[in] dev
7383  *   Pointer to Ethernet device.
7384  * @param[in] mtr_idx
7385  *  Index to aso flow meter to be free.
7386  *
7387  * @return
7388  *   0 on success.
7389  */
7390 void
7391 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7392 {
7393 	const struct mlx5_flow_driver_ops *fops;
7394 
7395 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7396 	fops->free_meter(dev, mtr_idx);
7397 }
7398 
7399 /**
7400  * Allocate a counter.
7401  *
7402  * @param[in] dev
7403  *   Pointer to Ethernet device structure.
7404  *
7405  * @return
7406  *   Index to allocated counter  on success, 0 otherwise.
7407  */
7408 uint32_t
7409 mlx5_counter_alloc(struct rte_eth_dev *dev)
7410 {
7411 	const struct mlx5_flow_driver_ops *fops;
7412 	struct rte_flow_attr attr = { .transfer = 0 };
7413 
7414 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7415 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7416 		return fops->counter_alloc(dev);
7417 	}
7418 	DRV_LOG(ERR,
7419 		"port %u counter allocate is not supported.",
7420 		 dev->data->port_id);
7421 	return 0;
7422 }
7423 
7424 /**
7425  * Free a counter.
7426  *
7427  * @param[in] dev
7428  *   Pointer to Ethernet device structure.
7429  * @param[in] cnt
7430  *   Index to counter to be free.
7431  */
7432 void
7433 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7434 {
7435 	const struct mlx5_flow_driver_ops *fops;
7436 	struct rte_flow_attr attr = { .transfer = 0 };
7437 
7438 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7439 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7440 		fops->counter_free(dev, cnt);
7441 		return;
7442 	}
7443 	DRV_LOG(ERR,
7444 		"port %u counter free is not supported.",
7445 		 dev->data->port_id);
7446 }
7447 
7448 /**
7449  * Query counter statistics.
7450  *
7451  * @param[in] dev
7452  *   Pointer to Ethernet device structure.
7453  * @param[in] cnt
7454  *   Index to counter to query.
7455  * @param[in] clear
7456  *   Set to clear counter statistics.
7457  * @param[out] pkts
7458  *   The counter hits packets number to save.
7459  * @param[out] bytes
7460  *   The counter hits bytes number to save.
7461  *
7462  * @return
7463  *   0 on success, a negative errno value otherwise.
7464  */
7465 int
7466 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7467 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7468 {
7469 	const struct mlx5_flow_driver_ops *fops;
7470 	struct rte_flow_attr attr = { .transfer = 0 };
7471 
7472 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7473 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7474 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7475 	}
7476 	DRV_LOG(ERR,
7477 		"port %u counter query is not supported.",
7478 		 dev->data->port_id);
7479 	return -ENOTSUP;
7480 }
7481 
7482 /**
7483  * Allocate a new memory for the counter values wrapped by all the needed
7484  * management.
7485  *
7486  * @param[in] sh
7487  *   Pointer to mlx5_dev_ctx_shared object.
7488  *
7489  * @return
7490  *   0 on success, a negative errno value otherwise.
7491  */
7492 static int
7493 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7494 {
7495 	struct mlx5_devx_mkey_attr mkey_attr;
7496 	struct mlx5_counter_stats_mem_mng *mem_mng;
7497 	volatile struct flow_counter_stats *raw_data;
7498 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7499 	int size = (sizeof(struct flow_counter_stats) *
7500 			MLX5_COUNTERS_PER_POOL +
7501 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7502 			sizeof(struct mlx5_counter_stats_mem_mng);
7503 	size_t pgsize = rte_mem_page_size();
7504 	uint8_t *mem;
7505 	int i;
7506 
7507 	if (pgsize == (size_t)-1) {
7508 		DRV_LOG(ERR, "Failed to get mem page size");
7509 		rte_errno = ENOMEM;
7510 		return -ENOMEM;
7511 	}
7512 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7513 	if (!mem) {
7514 		rte_errno = ENOMEM;
7515 		return -ENOMEM;
7516 	}
7517 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7518 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7519 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7520 						 IBV_ACCESS_LOCAL_WRITE);
7521 	if (!mem_mng->umem) {
7522 		rte_errno = errno;
7523 		mlx5_free(mem);
7524 		return -rte_errno;
7525 	}
7526 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7527 	mkey_attr.addr = (uintptr_t)mem;
7528 	mkey_attr.size = size;
7529 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7530 	mkey_attr.pd = sh->pdn;
7531 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7532 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7533 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7534 	if (!mem_mng->dm) {
7535 		mlx5_os_umem_dereg(mem_mng->umem);
7536 		rte_errno = errno;
7537 		mlx5_free(mem);
7538 		return -rte_errno;
7539 	}
7540 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7541 	raw_data = (volatile struct flow_counter_stats *)mem;
7542 	for (i = 0; i < raws_n; ++i) {
7543 		mem_mng->raws[i].mem_mng = mem_mng;
7544 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7545 	}
7546 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7547 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7548 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7549 				 next);
7550 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7551 	sh->cmng.mem_mng = mem_mng;
7552 	return 0;
7553 }
7554 
7555 /**
7556  * Set the statistic memory to the new counter pool.
7557  *
7558  * @param[in] sh
7559  *   Pointer to mlx5_dev_ctx_shared object.
7560  * @param[in] pool
7561  *   Pointer to the pool to set the statistic memory.
7562  *
7563  * @return
7564  *   0 on success, a negative errno value otherwise.
7565  */
7566 static int
7567 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7568 			       struct mlx5_flow_counter_pool *pool)
7569 {
7570 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7571 	/* Resize statistic memory once used out. */
7572 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7573 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7574 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7575 		return -1;
7576 	}
7577 	rte_spinlock_lock(&pool->sl);
7578 	pool->raw = cmng->mem_mng->raws + pool->index %
7579 		    MLX5_CNT_CONTAINER_RESIZE;
7580 	rte_spinlock_unlock(&pool->sl);
7581 	pool->raw_hw = NULL;
7582 	return 0;
7583 }
7584 
7585 #define MLX5_POOL_QUERY_FREQ_US 1000000
7586 
7587 /**
7588  * Set the periodic procedure for triggering asynchronous batch queries for all
7589  * the counter pools.
7590  *
7591  * @param[in] sh
7592  *   Pointer to mlx5_dev_ctx_shared object.
7593  */
7594 void
7595 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7596 {
7597 	uint32_t pools_n, us;
7598 
7599 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7600 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7601 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7602 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7603 		sh->cmng.query_thread_on = 0;
7604 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7605 	} else {
7606 		sh->cmng.query_thread_on = 1;
7607 	}
7608 }
7609 
7610 /**
7611  * The periodic procedure for triggering asynchronous batch queries for all the
7612  * counter pools. This function is probably called by the host thread.
7613  *
7614  * @param[in] arg
7615  *   The parameter for the alarm process.
7616  */
7617 void
7618 mlx5_flow_query_alarm(void *arg)
7619 {
7620 	struct mlx5_dev_ctx_shared *sh = arg;
7621 	int ret;
7622 	uint16_t pool_index = sh->cmng.pool_index;
7623 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7624 	struct mlx5_flow_counter_pool *pool;
7625 	uint16_t n_valid;
7626 
7627 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7628 		goto set_alarm;
7629 	rte_spinlock_lock(&cmng->pool_update_sl);
7630 	pool = cmng->pools[pool_index];
7631 	n_valid = cmng->n_valid;
7632 	rte_spinlock_unlock(&cmng->pool_update_sl);
7633 	/* Set the statistic memory to the new created pool. */
7634 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7635 		goto set_alarm;
7636 	if (pool->raw_hw)
7637 		/* There is a pool query in progress. */
7638 		goto set_alarm;
7639 	pool->raw_hw =
7640 		LIST_FIRST(&sh->cmng.free_stat_raws);
7641 	if (!pool->raw_hw)
7642 		/* No free counter statistics raw memory. */
7643 		goto set_alarm;
7644 	/*
7645 	 * Identify the counters released between query trigger and query
7646 	 * handle more efficiently. The counter released in this gap period
7647 	 * should wait for a new round of query as the new arrived packets
7648 	 * will not be taken into account.
7649 	 */
7650 	pool->query_gen++;
7651 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7652 					       MLX5_COUNTERS_PER_POOL,
7653 					       NULL, NULL,
7654 					       pool->raw_hw->mem_mng->dm->id,
7655 					       (void *)(uintptr_t)
7656 					       pool->raw_hw->data,
7657 					       sh->devx_comp,
7658 					       (uint64_t)(uintptr_t)pool);
7659 	if (ret) {
7660 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7661 			" %d", pool->min_dcs->id);
7662 		pool->raw_hw = NULL;
7663 		goto set_alarm;
7664 	}
7665 	LIST_REMOVE(pool->raw_hw, next);
7666 	sh->cmng.pending_queries++;
7667 	pool_index++;
7668 	if (pool_index >= n_valid)
7669 		pool_index = 0;
7670 set_alarm:
7671 	sh->cmng.pool_index = pool_index;
7672 	mlx5_set_query_alarm(sh);
7673 }
7674 
7675 /**
7676  * Check and callback event for new aged flow in the counter pool
7677  *
7678  * @param[in] sh
7679  *   Pointer to mlx5_dev_ctx_shared object.
7680  * @param[in] pool
7681  *   Pointer to Current counter pool.
7682  */
7683 static void
7684 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7685 		   struct mlx5_flow_counter_pool *pool)
7686 {
7687 	struct mlx5_priv *priv;
7688 	struct mlx5_flow_counter *cnt;
7689 	struct mlx5_age_info *age_info;
7690 	struct mlx5_age_param *age_param;
7691 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7692 	struct mlx5_counter_stats_raw *prev = pool->raw;
7693 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7694 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7695 	uint16_t expected = AGE_CANDIDATE;
7696 	uint32_t i;
7697 
7698 	pool->time_of_last_age_check = curr_time;
7699 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7700 		cnt = MLX5_POOL_GET_CNT(pool, i);
7701 		age_param = MLX5_CNT_TO_AGE(cnt);
7702 		if (__atomic_load_n(&age_param->state,
7703 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7704 			continue;
7705 		if (cur->data[i].hits != prev->data[i].hits) {
7706 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7707 					 __ATOMIC_RELAXED);
7708 			continue;
7709 		}
7710 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7711 				       time_delta,
7712 				       __ATOMIC_RELAXED) <= age_param->timeout)
7713 			continue;
7714 		/**
7715 		 * Hold the lock first, or if between the
7716 		 * state AGE_TMOUT and tailq operation the
7717 		 * release happened, the release procedure
7718 		 * may delete a non-existent tailq node.
7719 		 */
7720 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7721 		age_info = GET_PORT_AGE_INFO(priv);
7722 		rte_spinlock_lock(&age_info->aged_sl);
7723 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7724 						AGE_TMOUT, false,
7725 						__ATOMIC_RELAXED,
7726 						__ATOMIC_RELAXED)) {
7727 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7728 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7729 		}
7730 		rte_spinlock_unlock(&age_info->aged_sl);
7731 	}
7732 	mlx5_age_event_prepare(sh);
7733 }
7734 
7735 /**
7736  * Handler for the HW respond about ready values from an asynchronous batch
7737  * query. This function is probably called by the host thread.
7738  *
7739  * @param[in] sh
7740  *   The pointer to the shared device context.
7741  * @param[in] async_id
7742  *   The Devx async ID.
7743  * @param[in] status
7744  *   The status of the completion.
7745  */
7746 void
7747 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7748 				  uint64_t async_id, int status)
7749 {
7750 	struct mlx5_flow_counter_pool *pool =
7751 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7752 	struct mlx5_counter_stats_raw *raw_to_free;
7753 	uint8_t query_gen = pool->query_gen ^ 1;
7754 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7755 	enum mlx5_counter_type cnt_type =
7756 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7757 				MLX5_COUNTER_TYPE_ORIGIN;
7758 
7759 	if (unlikely(status)) {
7760 		raw_to_free = pool->raw_hw;
7761 	} else {
7762 		raw_to_free = pool->raw;
7763 		if (pool->is_aged)
7764 			mlx5_flow_aging_check(sh, pool);
7765 		rte_spinlock_lock(&pool->sl);
7766 		pool->raw = pool->raw_hw;
7767 		rte_spinlock_unlock(&pool->sl);
7768 		/* Be sure the new raw counters data is updated in memory. */
7769 		rte_io_wmb();
7770 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7771 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7772 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7773 				     &pool->counters[query_gen], next);
7774 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7775 		}
7776 	}
7777 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7778 	pool->raw_hw = NULL;
7779 	sh->cmng.pending_queries--;
7780 }
7781 
7782 static int
7783 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7784 		    const struct flow_grp_info *grp_info,
7785 		    struct rte_flow_error *error)
7786 {
7787 	if (grp_info->transfer && grp_info->external &&
7788 	    grp_info->fdb_def_rule) {
7789 		if (group == UINT32_MAX)
7790 			return rte_flow_error_set
7791 						(error, EINVAL,
7792 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7793 						 NULL,
7794 						 "group index not supported");
7795 		*table = group + 1;
7796 	} else {
7797 		*table = group;
7798 	}
7799 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7800 	return 0;
7801 }
7802 
7803 /**
7804  * Translate the rte_flow group index to HW table value.
7805  *
7806  * If tunnel offload is disabled, all group ids converted to flow table
7807  * id using the standard method.
7808  * If tunnel offload is enabled, group id can be converted using the
7809  * standard or tunnel conversion method. Group conversion method
7810  * selection depends on flags in `grp_info` parameter:
7811  * - Internal (grp_info.external == 0) groups conversion uses the
7812  *   standard method.
7813  * - Group ids in JUMP action converted with the tunnel conversion.
7814  * - Group id in rule attribute conversion depends on a rule type and
7815  *   group id value:
7816  *   ** non zero group attributes converted with the tunnel method
7817  *   ** zero group attribute in non-tunnel rule is converted using the
7818  *      standard method - there's only one root table
7819  *   ** zero group attribute in steer tunnel rule is converted with the
7820  *      standard method - single root table
7821  *   ** zero group attribute in match tunnel rule is a special OvS
7822  *      case: that value is used for portability reasons. That group
7823  *      id is converted with the tunnel conversion method.
7824  *
7825  * @param[in] dev
7826  *   Port device
7827  * @param[in] tunnel
7828  *   PMD tunnel offload object
7829  * @param[in] group
7830  *   rte_flow group index value.
7831  * @param[out] table
7832  *   HW table value.
7833  * @param[in] grp_info
7834  *   flags used for conversion
7835  * @param[out] error
7836  *   Pointer to error structure.
7837  *
7838  * @return
7839  *   0 on success, a negative errno value otherwise and rte_errno is set.
7840  */
7841 int
7842 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7843 			 const struct mlx5_flow_tunnel *tunnel,
7844 			 uint32_t group, uint32_t *table,
7845 			 const struct flow_grp_info *grp_info,
7846 			 struct rte_flow_error *error)
7847 {
7848 	int ret;
7849 	bool standard_translation;
7850 
7851 	if (!grp_info->skip_scale && grp_info->external &&
7852 	    group < MLX5_MAX_TABLES_EXTERNAL)
7853 		group *= MLX5_FLOW_TABLE_FACTOR;
7854 	if (is_tunnel_offload_active(dev)) {
7855 		standard_translation = !grp_info->external ||
7856 					grp_info->std_tbl_fix;
7857 	} else {
7858 		standard_translation = true;
7859 	}
7860 	DRV_LOG(DEBUG,
7861 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7862 		dev->data->port_id, group, grp_info->transfer,
7863 		grp_info->external, grp_info->fdb_def_rule,
7864 		standard_translation ? "STANDARD" : "TUNNEL");
7865 	if (standard_translation)
7866 		ret = flow_group_to_table(dev->data->port_id, group, table,
7867 					  grp_info, error);
7868 	else
7869 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7870 						      table, error);
7871 
7872 	return ret;
7873 }
7874 
7875 /**
7876  * Discover availability of metadata reg_c's.
7877  *
7878  * Iteratively use test flows to check availability.
7879  *
7880  * @param[in] dev
7881  *   Pointer to the Ethernet device structure.
7882  *
7883  * @return
7884  *   0 on success, a negative errno value otherwise and rte_errno is set.
7885  */
7886 int
7887 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7888 {
7889 	struct mlx5_priv *priv = dev->data->dev_private;
7890 	struct mlx5_dev_config *config = &priv->config;
7891 	enum modify_reg idx;
7892 	int n = 0;
7893 
7894 	/* reg_c[0] and reg_c[1] are reserved. */
7895 	config->flow_mreg_c[n++] = REG_C_0;
7896 	config->flow_mreg_c[n++] = REG_C_1;
7897 	/* Discover availability of other reg_c's. */
7898 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7899 		struct rte_flow_attr attr = {
7900 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7901 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7902 			.ingress = 1,
7903 		};
7904 		struct rte_flow_item items[] = {
7905 			[0] = {
7906 				.type = RTE_FLOW_ITEM_TYPE_END,
7907 			},
7908 		};
7909 		struct rte_flow_action actions[] = {
7910 			[0] = {
7911 				.type = (enum rte_flow_action_type)
7912 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7913 				.conf = &(struct mlx5_flow_action_copy_mreg){
7914 					.src = REG_C_1,
7915 					.dst = idx,
7916 				},
7917 			},
7918 			[1] = {
7919 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7920 				.conf = &(struct rte_flow_action_jump){
7921 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7922 				},
7923 			},
7924 			[2] = {
7925 				.type = RTE_FLOW_ACTION_TYPE_END,
7926 			},
7927 		};
7928 		uint32_t flow_idx;
7929 		struct rte_flow *flow;
7930 		struct rte_flow_error error;
7931 
7932 		if (!config->dv_flow_en)
7933 			break;
7934 		/* Create internal flow, validation skips copy action. */
7935 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
7936 					items, actions, false, &error);
7937 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7938 				      flow_idx);
7939 		if (!flow)
7940 			continue;
7941 		config->flow_mreg_c[n++] = idx;
7942 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
7943 	}
7944 	for (; n < MLX5_MREG_C_NUM; ++n)
7945 		config->flow_mreg_c[n] = REG_NON;
7946 	return 0;
7947 }
7948 
7949 int
7950 save_dump_file(const uint8_t *data, uint32_t size,
7951 	uint32_t type, uint32_t id, void *arg, FILE *file)
7952 {
7953 	char line[BUF_SIZE];
7954 	uint32_t out = 0;
7955 	uint32_t k;
7956 	uint32_t actions_num;
7957 	struct rte_flow_query_count *count;
7958 
7959 	memset(line, 0, BUF_SIZE);
7960 	switch (type) {
7961 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
7962 		actions_num = *(uint32_t *)(arg);
7963 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
7964 				type, id, actions_num);
7965 		break;
7966 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
7967 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
7968 				type, id);
7969 		break;
7970 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
7971 		count = (struct rte_flow_query_count *)arg;
7972 		fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
7973 				id, count->hits, count->bytes);
7974 		return 0;
7975 	default:
7976 		return -1;
7977 	}
7978 
7979 	for (k = 0; k < size; k++) {
7980 		/* Make sure we do not overrun the line buffer length. */
7981 		if (out >= BUF_SIZE - 4) {
7982 			line[out] = '\0';
7983 			break;
7984 		}
7985 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
7986 				(data[k]) & 0xff);
7987 	}
7988 	fprintf(file, "%s\n", line);
7989 	return 0;
7990 }
7991 
7992 int
7993 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
7994 	struct rte_flow_query_count *count, struct rte_flow_error *error)
7995 {
7996 	struct rte_flow_action action[2];
7997 	enum mlx5_flow_drv_type ftype;
7998 	const struct mlx5_flow_driver_ops *fops;
7999 
8000 	if (!flow) {
8001 		return rte_flow_error_set(error, ENOENT,
8002 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8003 				NULL,
8004 				"invalid flow handle");
8005 	}
8006 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8007 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8008 	if (flow->counter) {
8009 		memset(count, 0, sizeof(struct rte_flow_query_count));
8010 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8011 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8012 						ftype < MLX5_FLOW_TYPE_MAX);
8013 		fops = flow_get_drv_ops(ftype);
8014 		return fops->query(dev, flow, action, count, error);
8015 	}
8016 	return -1;
8017 }
8018 
8019 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8020 /**
8021  * Dump flow ipool data to file
8022  *
8023  * @param[in] dev
8024  *   The pointer to Ethernet device.
8025  * @param[in] file
8026  *   A pointer to a file for output.
8027  * @param[out] error
8028  *   Perform verbose error reporting if not NULL. PMDs initialize this
8029  *   structure in case of error only.
8030  * @return
8031  *   0 on success, a negative value otherwise.
8032  */
8033 int
8034 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8035 	struct rte_flow *flow, FILE *file,
8036 	struct rte_flow_error *error)
8037 {
8038 	struct mlx5_priv *priv = dev->data->dev_private;
8039 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8040 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8041 	uint32_t handle_idx;
8042 	struct mlx5_flow_handle *dh;
8043 	struct rte_flow_query_count count;
8044 	uint32_t actions_num;
8045 	const uint8_t *data;
8046 	size_t size;
8047 	uint32_t id;
8048 	uint32_t type;
8049 
8050 	if (!flow) {
8051 		return rte_flow_error_set(error, ENOENT,
8052 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8053 			NULL,
8054 			"invalid flow handle");
8055 	}
8056 	handle_idx = flow->dev_handles;
8057 	while (handle_idx) {
8058 		dh = mlx5_ipool_get(priv->sh->ipool
8059 			[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8060 		if (!dh)
8061 			continue;
8062 		handle_idx = dh->next.next;
8063 		id = (uint32_t)(uintptr_t)dh->drv_flow;
8064 
8065 		/* query counter */
8066 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8067 		if (!mlx5_flow_query_counter(dev, flow, &count, error))
8068 			save_dump_file(NULL, 0, type,
8069 					id, (void *)&count, file);
8070 
8071 		/* Get modify_hdr and encap_decap buf from ipools. */
8072 		encap_decap = NULL;
8073 		modify_hdr = dh->dvh.modify_hdr;
8074 
8075 		if (dh->dvh.rix_encap_decap) {
8076 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8077 						[MLX5_IPOOL_DECAP_ENCAP],
8078 						dh->dvh.rix_encap_decap);
8079 		}
8080 		if (modify_hdr) {
8081 			data = (const uint8_t *)modify_hdr->actions;
8082 			size = (size_t)(modify_hdr->actions_num) * 8;
8083 			actions_num = modify_hdr->actions_num;
8084 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8085 			save_dump_file(data, size, type, id,
8086 					(void *)(&actions_num), file);
8087 		}
8088 		if (encap_decap) {
8089 			data = encap_decap->buf;
8090 			size = encap_decap->size;
8091 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8092 			save_dump_file(data, size, type,
8093 						id, NULL, file);
8094 		}
8095 	}
8096 	return 0;
8097 }
8098 #endif
8099 
8100 /**
8101  * Dump flow raw hw data to file
8102  *
8103  * @param[in] dev
8104  *    The pointer to Ethernet device.
8105  * @param[in] file
8106  *   A pointer to a file for output.
8107  * @param[out] error
8108  *   Perform verbose error reporting if not NULL. PMDs initialize this
8109  *   structure in case of error only.
8110  * @return
8111  *   0 on success, a nagative value otherwise.
8112  */
8113 int
8114 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8115 		   FILE *file,
8116 		   struct rte_flow_error *error __rte_unused)
8117 {
8118 	struct mlx5_priv *priv = dev->data->dev_private;
8119 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8120 	uint32_t handle_idx;
8121 	int ret;
8122 	struct mlx5_flow_handle *dh;
8123 	struct rte_flow *flow;
8124 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8125 	uint32_t idx;
8126 #endif
8127 
8128 	if (!priv->config.dv_flow_en) {
8129 		if (fputs("device dv flow disabled\n", file) <= 0)
8130 			return -errno;
8131 		return -ENOTSUP;
8132 	}
8133 
8134 	/* dump all */
8135 	if (!flow_idx) {
8136 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8137 		MLX5_IPOOL_FOREACH(priv->flows[MLX5_FLOW_TYPE_GEN], idx, flow)
8138 			mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8139 #endif
8140 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8141 					sh->rx_domain,
8142 					sh->tx_domain, file);
8143 	}
8144 	/* dump one */
8145 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8146 			(uintptr_t)(void *)flow_idx);
8147 	if (!flow)
8148 		return -ENOENT;
8149 
8150 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8151 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8152 #endif
8153 	handle_idx = flow->dev_handles;
8154 	while (handle_idx) {
8155 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8156 				handle_idx);
8157 		if (!dh)
8158 			return -ENOENT;
8159 		if (dh->drv_flow) {
8160 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8161 					file);
8162 			if (ret)
8163 				return -ENOENT;
8164 		}
8165 		handle_idx = dh->next.next;
8166 	}
8167 	return 0;
8168 }
8169 
8170 /**
8171  * Get aged-out flows.
8172  *
8173  * @param[in] dev
8174  *   Pointer to the Ethernet device structure.
8175  * @param[in] context
8176  *   The address of an array of pointers to the aged-out flows contexts.
8177  * @param[in] nb_countexts
8178  *   The length of context array pointers.
8179  * @param[out] error
8180  *   Perform verbose error reporting if not NULL. Initialized in case of
8181  *   error only.
8182  *
8183  * @return
8184  *   how many contexts get in success, otherwise negative errno value.
8185  *   if nb_contexts is 0, return the amount of all aged contexts.
8186  *   if nb_contexts is not 0 , return the amount of aged flows reported
8187  *   in the context array.
8188  */
8189 int
8190 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8191 			uint32_t nb_contexts, struct rte_flow_error *error)
8192 {
8193 	const struct mlx5_flow_driver_ops *fops;
8194 	struct rte_flow_attr attr = { .transfer = 0 };
8195 
8196 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8197 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8198 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8199 						    error);
8200 	}
8201 	DRV_LOG(ERR,
8202 		"port %u get aged flows is not supported.",
8203 		 dev->data->port_id);
8204 	return -ENOTSUP;
8205 }
8206 
8207 /* Wrapper for driver action_validate op callback */
8208 static int
8209 flow_drv_action_validate(struct rte_eth_dev *dev,
8210 			 const struct rte_flow_indir_action_conf *conf,
8211 			 const struct rte_flow_action *action,
8212 			 const struct mlx5_flow_driver_ops *fops,
8213 			 struct rte_flow_error *error)
8214 {
8215 	static const char err_msg[] = "indirect action validation unsupported";
8216 
8217 	if (!fops->action_validate) {
8218 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8219 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8220 				   NULL, err_msg);
8221 		return -rte_errno;
8222 	}
8223 	return fops->action_validate(dev, conf, action, error);
8224 }
8225 
8226 /**
8227  * Destroys the shared action by handle.
8228  *
8229  * @param dev
8230  *   Pointer to Ethernet device structure.
8231  * @param[in] handle
8232  *   Handle for the indirect action object to be destroyed.
8233  * @param[out] error
8234  *   Perform verbose error reporting if not NULL. PMDs initialize this
8235  *   structure in case of error only.
8236  *
8237  * @return
8238  *   0 on success, a negative errno value otherwise and rte_errno is set.
8239  *
8240  * @note: wrapper for driver action_create op callback.
8241  */
8242 static int
8243 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8244 			   struct rte_flow_action_handle *handle,
8245 			   struct rte_flow_error *error)
8246 {
8247 	static const char err_msg[] = "indirect action destruction unsupported";
8248 	struct rte_flow_attr attr = { .transfer = 0 };
8249 	const struct mlx5_flow_driver_ops *fops =
8250 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8251 
8252 	if (!fops->action_destroy) {
8253 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8254 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8255 				   NULL, err_msg);
8256 		return -rte_errno;
8257 	}
8258 	return fops->action_destroy(dev, handle, error);
8259 }
8260 
8261 /* Wrapper for driver action_destroy op callback */
8262 static int
8263 flow_drv_action_update(struct rte_eth_dev *dev,
8264 		       struct rte_flow_action_handle *handle,
8265 		       const void *update,
8266 		       const struct mlx5_flow_driver_ops *fops,
8267 		       struct rte_flow_error *error)
8268 {
8269 	static const char err_msg[] = "indirect action update unsupported";
8270 
8271 	if (!fops->action_update) {
8272 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8273 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8274 				   NULL, err_msg);
8275 		return -rte_errno;
8276 	}
8277 	return fops->action_update(dev, handle, update, error);
8278 }
8279 
8280 /* Wrapper for driver action_destroy op callback */
8281 static int
8282 flow_drv_action_query(struct rte_eth_dev *dev,
8283 		      const struct rte_flow_action_handle *handle,
8284 		      void *data,
8285 		      const struct mlx5_flow_driver_ops *fops,
8286 		      struct rte_flow_error *error)
8287 {
8288 	static const char err_msg[] = "indirect action query unsupported";
8289 
8290 	if (!fops->action_query) {
8291 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8292 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8293 				   NULL, err_msg);
8294 		return -rte_errno;
8295 	}
8296 	return fops->action_query(dev, handle, data, error);
8297 }
8298 
8299 /**
8300  * Create indirect action for reuse in multiple flow rules.
8301  *
8302  * @param dev
8303  *   Pointer to Ethernet device structure.
8304  * @param conf
8305  *   Pointer to indirect action object configuration.
8306  * @param[in] action
8307  *   Action configuration for indirect action object creation.
8308  * @param[out] error
8309  *   Perform verbose error reporting if not NULL. PMDs initialize this
8310  *   structure in case of error only.
8311  * @return
8312  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8313  */
8314 static struct rte_flow_action_handle *
8315 mlx5_action_handle_create(struct rte_eth_dev *dev,
8316 			  const struct rte_flow_indir_action_conf *conf,
8317 			  const struct rte_flow_action *action,
8318 			  struct rte_flow_error *error)
8319 {
8320 	static const char err_msg[] = "indirect action creation unsupported";
8321 	struct rte_flow_attr attr = { .transfer = 0 };
8322 	const struct mlx5_flow_driver_ops *fops =
8323 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8324 
8325 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8326 		return NULL;
8327 	if (!fops->action_create) {
8328 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8329 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8330 				   NULL, err_msg);
8331 		return NULL;
8332 	}
8333 	return fops->action_create(dev, conf, action, error);
8334 }
8335 
8336 /**
8337  * Updates inplace the indirect action configuration pointed by *handle*
8338  * with the configuration provided as *update* argument.
8339  * The update of the indirect action configuration effects all flow rules
8340  * reusing the action via handle.
8341  *
8342  * @param dev
8343  *   Pointer to Ethernet device structure.
8344  * @param[in] handle
8345  *   Handle for the indirect action to be updated.
8346  * @param[in] update
8347  *   Action specification used to modify the action pointed by handle.
8348  *   *update* could be of same type with the action pointed by the *handle*
8349  *   handle argument, or some other structures like a wrapper, depending on
8350  *   the indirect action type.
8351  * @param[out] error
8352  *   Perform verbose error reporting if not NULL. PMDs initialize this
8353  *   structure in case of error only.
8354  *
8355  * @return
8356  *   0 on success, a negative errno value otherwise and rte_errno is set.
8357  */
8358 static int
8359 mlx5_action_handle_update(struct rte_eth_dev *dev,
8360 		struct rte_flow_action_handle *handle,
8361 		const void *update,
8362 		struct rte_flow_error *error)
8363 {
8364 	struct rte_flow_attr attr = { .transfer = 0 };
8365 	const struct mlx5_flow_driver_ops *fops =
8366 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8367 	int ret;
8368 
8369 	ret = flow_drv_action_validate(dev, NULL,
8370 			(const struct rte_flow_action *)update, fops, error);
8371 	if (ret)
8372 		return ret;
8373 	return flow_drv_action_update(dev, handle, update, fops,
8374 				      error);
8375 }
8376 
8377 /**
8378  * Query the indirect action by handle.
8379  *
8380  * This function allows retrieving action-specific data such as counters.
8381  * Data is gathered by special action which may be present/referenced in
8382  * more than one flow rule definition.
8383  *
8384  * see @RTE_FLOW_ACTION_TYPE_COUNT
8385  *
8386  * @param dev
8387  *   Pointer to Ethernet device structure.
8388  * @param[in] handle
8389  *   Handle for the indirect action to query.
8390  * @param[in, out] data
8391  *   Pointer to storage for the associated query data type.
8392  * @param[out] error
8393  *   Perform verbose error reporting if not NULL. PMDs initialize this
8394  *   structure in case of error only.
8395  *
8396  * @return
8397  *   0 on success, a negative errno value otherwise and rte_errno is set.
8398  */
8399 static int
8400 mlx5_action_handle_query(struct rte_eth_dev *dev,
8401 			 const struct rte_flow_action_handle *handle,
8402 			 void *data,
8403 			 struct rte_flow_error *error)
8404 {
8405 	struct rte_flow_attr attr = { .transfer = 0 };
8406 	const struct mlx5_flow_driver_ops *fops =
8407 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8408 
8409 	return flow_drv_action_query(dev, handle, data, fops, error);
8410 }
8411 
8412 /**
8413  * Destroy all indirect actions (shared RSS).
8414  *
8415  * @param dev
8416  *   Pointer to Ethernet device.
8417  *
8418  * @return
8419  *   0 on success, a negative errno value otherwise and rte_errno is set.
8420  */
8421 int
8422 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8423 {
8424 	struct rte_flow_error error;
8425 	struct mlx5_priv *priv = dev->data->dev_private;
8426 	struct mlx5_shared_action_rss *shared_rss;
8427 	int ret = 0;
8428 	uint32_t idx;
8429 
8430 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8431 		      priv->rss_shared_actions, idx, shared_rss, next) {
8432 		ret |= mlx5_action_handle_destroy(dev,
8433 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8434 	}
8435 	return ret;
8436 }
8437 
8438 #ifndef HAVE_MLX5DV_DR
8439 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8440 #else
8441 #define MLX5_DOMAIN_SYNC_FLOW \
8442 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8443 #endif
8444 
8445 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8446 {
8447 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8448 	const struct mlx5_flow_driver_ops *fops;
8449 	int ret;
8450 	struct rte_flow_attr attr = { .transfer = 0 };
8451 
8452 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8453 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8454 	if (ret > 0)
8455 		ret = -ret;
8456 	return ret;
8457 }
8458 
8459 const struct mlx5_flow_tunnel *
8460 mlx5_get_tof(const struct rte_flow_item *item,
8461 	     const struct rte_flow_action *action,
8462 	     enum mlx5_tof_rule_type *rule_type)
8463 {
8464 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8465 		if (item->type == (typeof(item->type))
8466 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8467 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8468 			return flow_items_to_tunnel(item);
8469 		}
8470 	}
8471 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8472 		if (action->type == (typeof(action->type))
8473 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8474 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8475 			return flow_actions_to_tunnel(action);
8476 		}
8477 	}
8478 	return NULL;
8479 }
8480 
8481 /**
8482  * tunnel offload functionalilty is defined for DV environment only
8483  */
8484 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8485 __extension__
8486 union tunnel_offload_mark {
8487 	uint32_t val;
8488 	struct {
8489 		uint32_t app_reserve:8;
8490 		uint32_t table_id:15;
8491 		uint32_t transfer:1;
8492 		uint32_t _unused_:8;
8493 	};
8494 };
8495 
8496 static bool
8497 mlx5_access_tunnel_offload_db
8498 	(struct rte_eth_dev *dev,
8499 	 bool (*match)(struct rte_eth_dev *,
8500 		       struct mlx5_flow_tunnel *, const void *),
8501 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8502 	 void (*miss)(struct rte_eth_dev *, void *),
8503 	 void *ctx, bool lock_op);
8504 
8505 static int
8506 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8507 			     struct rte_flow *flow,
8508 			     const struct rte_flow_attr *attr,
8509 			     const struct rte_flow_action *app_actions,
8510 			     uint32_t flow_idx,
8511 			     const struct mlx5_flow_tunnel *tunnel,
8512 			     struct tunnel_default_miss_ctx *ctx,
8513 			     struct rte_flow_error *error)
8514 {
8515 	struct mlx5_priv *priv = dev->data->dev_private;
8516 	struct mlx5_flow *dev_flow;
8517 	struct rte_flow_attr miss_attr = *attr;
8518 	const struct rte_flow_item miss_items[2] = {
8519 		{
8520 			.type = RTE_FLOW_ITEM_TYPE_ETH,
8521 			.spec = NULL,
8522 			.last = NULL,
8523 			.mask = NULL
8524 		},
8525 		{
8526 			.type = RTE_FLOW_ITEM_TYPE_END,
8527 			.spec = NULL,
8528 			.last = NULL,
8529 			.mask = NULL
8530 		}
8531 	};
8532 	union tunnel_offload_mark mark_id;
8533 	struct rte_flow_action_mark miss_mark;
8534 	struct rte_flow_action miss_actions[3] = {
8535 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8536 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
8537 	};
8538 	const struct rte_flow_action_jump *jump_data;
8539 	uint32_t i, flow_table = 0; /* prevent compilation warning */
8540 	struct flow_grp_info grp_info = {
8541 		.external = 1,
8542 		.transfer = attr->transfer,
8543 		.fdb_def_rule = !!priv->fdb_def_rule,
8544 		.std_tbl_fix = 0,
8545 	};
8546 	int ret;
8547 
8548 	if (!attr->transfer) {
8549 		uint32_t q_size;
8550 
8551 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8552 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8553 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8554 					 0, SOCKET_ID_ANY);
8555 		if (!ctx->queue)
8556 			return rte_flow_error_set
8557 				(error, ENOMEM,
8558 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8559 				NULL, "invalid default miss RSS");
8560 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8561 		ctx->action_rss.level = 0,
8562 		ctx->action_rss.types = priv->rss_conf.rss_hf,
8563 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8564 		ctx->action_rss.queue_num = priv->reta_idx_n,
8565 		ctx->action_rss.key = priv->rss_conf.rss_key,
8566 		ctx->action_rss.queue = ctx->queue;
8567 		if (!priv->reta_idx_n || !priv->rxqs_n)
8568 			return rte_flow_error_set
8569 				(error, EINVAL,
8570 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8571 				NULL, "invalid port configuration");
8572 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8573 			ctx->action_rss.types = 0;
8574 		for (i = 0; i != priv->reta_idx_n; ++i)
8575 			ctx->queue[i] = (*priv->reta_idx)[i];
8576 	} else {
8577 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8578 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8579 	}
8580 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8581 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8582 	jump_data = app_actions->conf;
8583 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8584 	miss_attr.group = jump_data->group;
8585 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8586 				       &flow_table, &grp_info, error);
8587 	if (ret)
8588 		return rte_flow_error_set(error, EINVAL,
8589 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8590 					  NULL, "invalid tunnel id");
8591 	mark_id.app_reserve = 0;
8592 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8593 	mark_id.transfer = !!attr->transfer;
8594 	mark_id._unused_ = 0;
8595 	miss_mark.id = mark_id.val;
8596 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8597 				    miss_items, miss_actions, flow_idx, error);
8598 	if (!dev_flow)
8599 		return -rte_errno;
8600 	dev_flow->flow = flow;
8601 	dev_flow->external = true;
8602 	dev_flow->tunnel = tunnel;
8603 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8604 	/* Subflow object was created, we must include one in the list. */
8605 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8606 		      dev_flow->handle, next);
8607 	DRV_LOG(DEBUG,
8608 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8609 		dev->data->port_id, tunnel->app_tunnel.type,
8610 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8611 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8612 				  miss_actions, error);
8613 	if (!ret)
8614 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8615 						  error);
8616 
8617 	return ret;
8618 }
8619 
8620 static const struct mlx5_flow_tbl_data_entry  *
8621 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8622 {
8623 	struct mlx5_priv *priv = dev->data->dev_private;
8624 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8625 	struct mlx5_hlist_entry *he;
8626 	union tunnel_offload_mark mbits = { .val = mark };
8627 	union mlx5_flow_tbl_key table_key = {
8628 		{
8629 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
8630 			.id = 0,
8631 			.reserved = 0,
8632 			.dummy = 0,
8633 			.is_fdb = !!mbits.transfer,
8634 			.is_egress = 0,
8635 		}
8636 	};
8637 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8638 	return he ?
8639 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8640 }
8641 
8642 static void
8643 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8644 				   struct mlx5_hlist_entry *entry)
8645 {
8646 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8647 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8648 
8649 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8650 			tunnel_flow_tbl_to_id(tte->flow_table));
8651 	mlx5_free(tte);
8652 }
8653 
8654 static int
8655 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8656 				  struct mlx5_hlist_entry *entry,
8657 				  uint64_t key, void *cb_ctx __rte_unused)
8658 {
8659 	union tunnel_tbl_key tbl = {
8660 		.val = key,
8661 	};
8662 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8663 
8664 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8665 }
8666 
8667 static struct mlx5_hlist_entry *
8668 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8669 				   void *ctx __rte_unused)
8670 {
8671 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8672 	struct tunnel_tbl_entry *tte;
8673 	union tunnel_tbl_key tbl = {
8674 		.val = key,
8675 	};
8676 
8677 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8678 			  sizeof(*tte), 0,
8679 			  SOCKET_ID_ANY);
8680 	if (!tte)
8681 		goto err;
8682 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8683 			  &tte->flow_table);
8684 	if (tte->flow_table >= MLX5_MAX_TABLES) {
8685 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8686 			tte->flow_table);
8687 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8688 				tte->flow_table);
8689 		goto err;
8690 	} else if (!tte->flow_table) {
8691 		goto err;
8692 	}
8693 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8694 	tte->tunnel_id = tbl.tunnel_id;
8695 	tte->group = tbl.group;
8696 	return &tte->hash;
8697 err:
8698 	if (tte)
8699 		mlx5_free(tte);
8700 	return NULL;
8701 }
8702 
8703 static uint32_t
8704 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8705 				const struct mlx5_flow_tunnel *tunnel,
8706 				uint32_t group, uint32_t *table,
8707 				struct rte_flow_error *error)
8708 {
8709 	struct mlx5_hlist_entry *he;
8710 	struct tunnel_tbl_entry *tte;
8711 	union tunnel_tbl_key key = {
8712 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8713 		.group = group
8714 	};
8715 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8716 	struct mlx5_hlist *group_hash;
8717 
8718 	group_hash = tunnel ? tunnel->groups : thub->groups;
8719 	he = mlx5_hlist_register(group_hash, key.val, NULL);
8720 	if (!he)
8721 		return rte_flow_error_set(error, EINVAL,
8722 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8723 					  NULL,
8724 					  "tunnel group index not supported");
8725 	tte = container_of(he, typeof(*tte), hash);
8726 	*table = tte->flow_table;
8727 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8728 		dev->data->port_id, key.tunnel_id, group, *table);
8729 	return 0;
8730 }
8731 
8732 static void
8733 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8734 		      struct mlx5_flow_tunnel *tunnel)
8735 {
8736 	struct mlx5_priv *priv = dev->data->dev_private;
8737 	struct mlx5_indexed_pool *ipool;
8738 
8739 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8740 		dev->data->port_id, tunnel->tunnel_id);
8741 	LIST_REMOVE(tunnel, chain);
8742 	mlx5_hlist_destroy(tunnel->groups);
8743 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8744 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
8745 }
8746 
8747 static bool
8748 mlx5_access_tunnel_offload_db
8749 	(struct rte_eth_dev *dev,
8750 	 bool (*match)(struct rte_eth_dev *,
8751 		       struct mlx5_flow_tunnel *, const void *),
8752 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8753 	 void (*miss)(struct rte_eth_dev *, void *),
8754 	 void *ctx, bool lock_op)
8755 {
8756 	bool verdict = false;
8757 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8758 	struct mlx5_flow_tunnel *tunnel;
8759 
8760 	rte_spinlock_lock(&thub->sl);
8761 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8762 		verdict = match(dev, tunnel, (const void *)ctx);
8763 		if (verdict)
8764 			break;
8765 	}
8766 	if (!lock_op)
8767 		rte_spinlock_unlock(&thub->sl);
8768 	if (verdict && hit)
8769 		hit(dev, tunnel, ctx);
8770 	if (!verdict && miss)
8771 		miss(dev, ctx);
8772 	if (lock_op)
8773 		rte_spinlock_unlock(&thub->sl);
8774 
8775 	return verdict;
8776 }
8777 
8778 struct tunnel_db_find_tunnel_id_ctx {
8779 	uint32_t tunnel_id;
8780 	struct mlx5_flow_tunnel *tunnel;
8781 };
8782 
8783 static bool
8784 find_tunnel_id_match(struct rte_eth_dev *dev,
8785 		     struct mlx5_flow_tunnel *tunnel, const void *x)
8786 {
8787 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8788 
8789 	RTE_SET_USED(dev);
8790 	return tunnel->tunnel_id == ctx->tunnel_id;
8791 }
8792 
8793 static void
8794 find_tunnel_id_hit(struct rte_eth_dev *dev,
8795 		   struct mlx5_flow_tunnel *tunnel, void *x)
8796 {
8797 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8798 	RTE_SET_USED(dev);
8799 	ctx->tunnel = tunnel;
8800 }
8801 
8802 static struct mlx5_flow_tunnel *
8803 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8804 {
8805 	struct tunnel_db_find_tunnel_id_ctx ctx = {
8806 		.tunnel_id = id,
8807 	};
8808 
8809 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8810 				      find_tunnel_id_hit, NULL, &ctx, true);
8811 
8812 	return ctx.tunnel;
8813 }
8814 
8815 static struct mlx5_flow_tunnel *
8816 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8817 			  const struct rte_flow_tunnel *app_tunnel)
8818 {
8819 	struct mlx5_priv *priv = dev->data->dev_private;
8820 	struct mlx5_indexed_pool *ipool;
8821 	struct mlx5_flow_tunnel *tunnel;
8822 	uint32_t id;
8823 
8824 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8825 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
8826 	if (!tunnel)
8827 		return NULL;
8828 	if (id >= MLX5_MAX_TUNNELS) {
8829 		mlx5_ipool_free(ipool, id);
8830 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8831 		return NULL;
8832 	}
8833 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8834 					   mlx5_flow_tunnel_grp2tbl_create_cb,
8835 					   mlx5_flow_tunnel_grp2tbl_match_cb,
8836 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
8837 	if (!tunnel->groups) {
8838 		mlx5_ipool_free(ipool, id);
8839 		return NULL;
8840 	}
8841 	tunnel->groups->ctx = priv->sh;
8842 	/* initiate new PMD tunnel */
8843 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8844 	tunnel->tunnel_id = id;
8845 	tunnel->action.type = (typeof(tunnel->action.type))
8846 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8847 	tunnel->action.conf = tunnel;
8848 	tunnel->item.type = (typeof(tunnel->item.type))
8849 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8850 	tunnel->item.spec = tunnel;
8851 	tunnel->item.last = NULL;
8852 	tunnel->item.mask = NULL;
8853 
8854 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8855 		dev->data->port_id, tunnel->tunnel_id);
8856 
8857 	return tunnel;
8858 }
8859 
8860 struct tunnel_db_get_tunnel_ctx {
8861 	const struct rte_flow_tunnel *app_tunnel;
8862 	struct mlx5_flow_tunnel *tunnel;
8863 };
8864 
8865 static bool get_tunnel_match(struct rte_eth_dev *dev,
8866 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8867 {
8868 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
8869 
8870 	RTE_SET_USED(dev);
8871 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8872 		       sizeof(*ctx->app_tunnel));
8873 }
8874 
8875 static void get_tunnel_hit(struct rte_eth_dev *dev,
8876 			   struct mlx5_flow_tunnel *tunnel, void *x)
8877 {
8878 	/* called under tunnel spinlock protection */
8879 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8880 
8881 	RTE_SET_USED(dev);
8882 	tunnel->refctn++;
8883 	ctx->tunnel = tunnel;
8884 }
8885 
8886 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8887 {
8888 	/* called under tunnel spinlock protection */
8889 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8890 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8891 
8892 	rte_spinlock_unlock(&thub->sl);
8893 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8894 	rte_spinlock_lock(&thub->sl);
8895 	if (ctx->tunnel) {
8896 		ctx->tunnel->refctn = 1;
8897 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8898 	}
8899 }
8900 
8901 
8902 static int
8903 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8904 		     const struct rte_flow_tunnel *app_tunnel,
8905 		     struct mlx5_flow_tunnel **tunnel)
8906 {
8907 	struct tunnel_db_get_tunnel_ctx ctx = {
8908 		.app_tunnel = app_tunnel,
8909 	};
8910 
8911 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8912 				      get_tunnel_miss, &ctx, true);
8913 	*tunnel = ctx.tunnel;
8914 	return ctx.tunnel ? 0 : -ENOMEM;
8915 }
8916 
8917 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8918 {
8919 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8920 
8921 	if (!thub)
8922 		return;
8923 	if (!LIST_EMPTY(&thub->tunnels))
8924 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
8925 	mlx5_hlist_destroy(thub->groups);
8926 	mlx5_free(thub);
8927 }
8928 
8929 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8930 {
8931 	int err;
8932 	struct mlx5_flow_tunnel_hub *thub;
8933 
8934 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8935 			   0, SOCKET_ID_ANY);
8936 	if (!thub)
8937 		return -ENOMEM;
8938 	LIST_INIT(&thub->tunnels);
8939 	rte_spinlock_init(&thub->sl);
8940 	thub->groups = mlx5_hlist_create("flow groups",
8941 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
8942 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8943 					 mlx5_flow_tunnel_grp2tbl_match_cb,
8944 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
8945 	if (!thub->groups) {
8946 		err = -rte_errno;
8947 		goto err;
8948 	}
8949 	thub->groups->ctx = sh;
8950 	sh->tunnel_hub = thub;
8951 
8952 	return 0;
8953 
8954 err:
8955 	if (thub->groups)
8956 		mlx5_hlist_destroy(thub->groups);
8957 	if (thub)
8958 		mlx5_free(thub);
8959 	return err;
8960 }
8961 
8962 static inline bool
8963 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8964 			  struct rte_flow_tunnel *tunnel,
8965 			  const char *err_msg)
8966 {
8967 	err_msg = NULL;
8968 	if (!is_tunnel_offload_active(dev)) {
8969 		err_msg = "tunnel offload was not activated";
8970 		goto out;
8971 	} else if (!tunnel) {
8972 		err_msg = "no application tunnel";
8973 		goto out;
8974 	}
8975 
8976 	switch (tunnel->type) {
8977 	default:
8978 		err_msg = "unsupported tunnel type";
8979 		goto out;
8980 	case RTE_FLOW_ITEM_TYPE_VXLAN:
8981 		break;
8982 	}
8983 
8984 out:
8985 	return !err_msg;
8986 }
8987 
8988 static int
8989 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8990 		    struct rte_flow_tunnel *app_tunnel,
8991 		    struct rte_flow_action **actions,
8992 		    uint32_t *num_of_actions,
8993 		    struct rte_flow_error *error)
8994 {
8995 	int ret;
8996 	struct mlx5_flow_tunnel *tunnel;
8997 	const char *err_msg = NULL;
8998 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8999 
9000 	if (!verdict)
9001 		return rte_flow_error_set(error, EINVAL,
9002 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9003 					  err_msg);
9004 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9005 	if (ret < 0) {
9006 		return rte_flow_error_set(error, ret,
9007 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9008 					  "failed to initialize pmd tunnel");
9009 	}
9010 	*actions = &tunnel->action;
9011 	*num_of_actions = 1;
9012 	return 0;
9013 }
9014 
9015 static int
9016 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9017 		       struct rte_flow_tunnel *app_tunnel,
9018 		       struct rte_flow_item **items,
9019 		       uint32_t *num_of_items,
9020 		       struct rte_flow_error *error)
9021 {
9022 	int ret;
9023 	struct mlx5_flow_tunnel *tunnel;
9024 	const char *err_msg = NULL;
9025 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9026 
9027 	if (!verdict)
9028 		return rte_flow_error_set(error, EINVAL,
9029 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9030 					  err_msg);
9031 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9032 	if (ret < 0) {
9033 		return rte_flow_error_set(error, ret,
9034 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9035 					  "failed to initialize pmd tunnel");
9036 	}
9037 	*items = &tunnel->item;
9038 	*num_of_items = 1;
9039 	return 0;
9040 }
9041 
9042 struct tunnel_db_element_release_ctx {
9043 	struct rte_flow_item *items;
9044 	struct rte_flow_action *actions;
9045 	uint32_t num_elements;
9046 	struct rte_flow_error *error;
9047 	int ret;
9048 };
9049 
9050 static bool
9051 tunnel_element_release_match(struct rte_eth_dev *dev,
9052 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9053 {
9054 	const struct tunnel_db_element_release_ctx *ctx = x;
9055 
9056 	RTE_SET_USED(dev);
9057 	if (ctx->num_elements != 1)
9058 		return false;
9059 	else if (ctx->items)
9060 		return ctx->items == &tunnel->item;
9061 	else if (ctx->actions)
9062 		return ctx->actions == &tunnel->action;
9063 
9064 	return false;
9065 }
9066 
9067 static void
9068 tunnel_element_release_hit(struct rte_eth_dev *dev,
9069 			   struct mlx5_flow_tunnel *tunnel, void *x)
9070 {
9071 	struct tunnel_db_element_release_ctx *ctx = x;
9072 	ctx->ret = 0;
9073 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9074 		mlx5_flow_tunnel_free(dev, tunnel);
9075 }
9076 
9077 static void
9078 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9079 {
9080 	struct tunnel_db_element_release_ctx *ctx = x;
9081 	RTE_SET_USED(dev);
9082 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9083 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9084 				      "invalid argument");
9085 }
9086 
9087 static int
9088 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9089 		       struct rte_flow_item *pmd_items,
9090 		       uint32_t num_items, struct rte_flow_error *err)
9091 {
9092 	struct tunnel_db_element_release_ctx ctx = {
9093 		.items = pmd_items,
9094 		.actions = NULL,
9095 		.num_elements = num_items,
9096 		.error = err,
9097 	};
9098 
9099 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9100 				      tunnel_element_release_hit,
9101 				      tunnel_element_release_miss, &ctx, false);
9102 
9103 	return ctx.ret;
9104 }
9105 
9106 static int
9107 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9108 			 struct rte_flow_action *pmd_actions,
9109 			 uint32_t num_actions, struct rte_flow_error *err)
9110 {
9111 	struct tunnel_db_element_release_ctx ctx = {
9112 		.items = NULL,
9113 		.actions = pmd_actions,
9114 		.num_elements = num_actions,
9115 		.error = err,
9116 	};
9117 
9118 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9119 				      tunnel_element_release_hit,
9120 				      tunnel_element_release_miss, &ctx, false);
9121 
9122 	return ctx.ret;
9123 }
9124 
9125 static int
9126 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9127 				  struct rte_mbuf *m,
9128 				  struct rte_flow_restore_info *info,
9129 				  struct rte_flow_error *err)
9130 {
9131 	uint64_t ol_flags = m->ol_flags;
9132 	const struct mlx5_flow_tbl_data_entry *tble;
9133 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
9134 
9135 	if (!is_tunnel_offload_active(dev)) {
9136 		info->flags = 0;
9137 		return 0;
9138 	}
9139 
9140 	if ((ol_flags & mask) != mask)
9141 		goto err;
9142 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9143 	if (!tble) {
9144 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9145 			dev->data->port_id, m->hash.fdir.hi);
9146 		goto err;
9147 	}
9148 	MLX5_ASSERT(tble->tunnel);
9149 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9150 	info->group_id = tble->group_id;
9151 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9152 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9153 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9154 
9155 	return 0;
9156 
9157 err:
9158 	return rte_flow_error_set(err, EINVAL,
9159 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9160 				  "failed to get restore info");
9161 }
9162 
9163 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9164 static int
9165 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9166 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9167 			   __rte_unused struct rte_flow_action **actions,
9168 			   __rte_unused uint32_t *num_of_actions,
9169 			   __rte_unused struct rte_flow_error *error)
9170 {
9171 	return -ENOTSUP;
9172 }
9173 
9174 static int
9175 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9176 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9177 		       __rte_unused struct rte_flow_item **items,
9178 		       __rte_unused uint32_t *num_of_items,
9179 		       __rte_unused struct rte_flow_error *error)
9180 {
9181 	return -ENOTSUP;
9182 }
9183 
9184 static int
9185 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9186 			      __rte_unused struct rte_flow_item *pmd_items,
9187 			      __rte_unused uint32_t num_items,
9188 			      __rte_unused struct rte_flow_error *err)
9189 {
9190 	return -ENOTSUP;
9191 }
9192 
9193 static int
9194 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9195 				__rte_unused struct rte_flow_action *pmd_action,
9196 				__rte_unused uint32_t num_actions,
9197 				__rte_unused struct rte_flow_error *err)
9198 {
9199 	return -ENOTSUP;
9200 }
9201 
9202 static int
9203 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9204 				  __rte_unused struct rte_mbuf *m,
9205 				  __rte_unused struct rte_flow_restore_info *i,
9206 				  __rte_unused struct rte_flow_error *err)
9207 {
9208 	return -ENOTSUP;
9209 }
9210 
9211 static int
9212 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9213 			     __rte_unused struct rte_flow *flow,
9214 			     __rte_unused const struct rte_flow_attr *attr,
9215 			     __rte_unused const struct rte_flow_action *actions,
9216 			     __rte_unused uint32_t flow_idx,
9217 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9218 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9219 			     __rte_unused struct rte_flow_error *error)
9220 {
9221 	return -ENOTSUP;
9222 }
9223 
9224 static struct mlx5_flow_tunnel *
9225 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9226 		    __rte_unused uint32_t id)
9227 {
9228 	return NULL;
9229 }
9230 
9231 static void
9232 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9233 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9234 {
9235 }
9236 
9237 static uint32_t
9238 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9239 				__rte_unused const struct mlx5_flow_tunnel *t,
9240 				__rte_unused uint32_t group,
9241 				__rte_unused uint32_t *table,
9242 				struct rte_flow_error *error)
9243 {
9244 	return rte_flow_error_set(error, ENOTSUP,
9245 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9246 				  "tunnel offload requires DV support");
9247 }
9248 
9249 void
9250 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9251 			__rte_unused  uint16_t port_id)
9252 {
9253 }
9254 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9255 
9256 static void
9257 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9258 {
9259 	int ret;
9260 	struct rte_flow_error error;
9261 
9262 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9263 		char *item_name;
9264 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9265 				    sizeof(item_name),
9266 				    (void *)(uintptr_t)item->type, &error);
9267 		if (ret > 0)
9268 			printf("%s ", item_name);
9269 		else
9270 			printf("%d\n", (int)item->type);
9271 	}
9272 	printf("END\n");
9273 }
9274