xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision a9b6ea45bed6acb2d95c93bc177f9d5cb53275ef)
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 RTE_ETH_RSS_* definitions).
102 	 */
103 	uint64_t node_flags;
104 	/**<
105 	 *  Bit-fields that define how the node is used in the expansion.
106 	 * (see MLX5_EXPANSION_NODE_* definitions).
107 	 */
108 };
109 
110 /* Optional expand field. The expansion alg will not go deeper. */
111 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
112 
113 /* The node is not added implicitly as expansion to the flow pattern.
114  * If the node type does not match the flow pattern item type, the
115  * expansion alg will go deeper to its next items.
116  * In the current implementation, the list of next nodes indexes can
117  * have up to one node with this flag set and it has to be the last
118  * node index (before the list terminator).
119  */
120 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
121 
122 /** Object returned by mlx5_flow_expand_rss(). */
123 struct mlx5_flow_expand_rss {
124 	uint32_t entries;
125 	/**< Number of entries @p patterns and @p priorities. */
126 	struct {
127 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
128 		uint32_t priority; /**< Priority offset for each expansion. */
129 	} entry[];
130 };
131 
132 static void
133 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
134 
135 static const struct mlx5_flow_expand_node *
136 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
137 		unsigned int item_idx,
138 		const struct mlx5_flow_expand_node graph[],
139 		const struct mlx5_flow_expand_node *node);
140 
141 static bool
142 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
143 {
144 	switch (item->type) {
145 	case RTE_FLOW_ITEM_TYPE_ETH:
146 	case RTE_FLOW_ITEM_TYPE_VLAN:
147 	case RTE_FLOW_ITEM_TYPE_IPV4:
148 	case RTE_FLOW_ITEM_TYPE_IPV6:
149 	case RTE_FLOW_ITEM_TYPE_UDP:
150 	case RTE_FLOW_ITEM_TYPE_TCP:
151 	case RTE_FLOW_ITEM_TYPE_VXLAN:
152 	case RTE_FLOW_ITEM_TYPE_NVGRE:
153 	case RTE_FLOW_ITEM_TYPE_GRE:
154 	case RTE_FLOW_ITEM_TYPE_GENEVE:
155 	case RTE_FLOW_ITEM_TYPE_MPLS:
156 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
157 	case RTE_FLOW_ITEM_TYPE_GRE_KEY:
158 	case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
159 	case RTE_FLOW_ITEM_TYPE_GTP:
160 		return true;
161 	default:
162 		break;
163 	}
164 	return false;
165 }
166 
167 static enum rte_flow_item_type
168 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
169 {
170 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
171 	uint16_t ether_type = 0;
172 	uint16_t ether_type_m;
173 	uint8_t ip_next_proto = 0;
174 	uint8_t ip_next_proto_m;
175 
176 	if (item == NULL || item->spec == NULL)
177 		return ret;
178 	switch (item->type) {
179 	case RTE_FLOW_ITEM_TYPE_ETH:
180 		if (item->mask)
181 			ether_type_m = ((const struct rte_flow_item_eth *)
182 						(item->mask))->type;
183 		else
184 			ether_type_m = rte_flow_item_eth_mask.type;
185 		if (ether_type_m != RTE_BE16(0xFFFF))
186 			break;
187 		ether_type = ((const struct rte_flow_item_eth *)
188 				(item->spec))->type;
189 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
190 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
191 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
192 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
193 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
194 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
195 		else
196 			ret = RTE_FLOW_ITEM_TYPE_END;
197 		break;
198 	case RTE_FLOW_ITEM_TYPE_VLAN:
199 		if (item->mask)
200 			ether_type_m = ((const struct rte_flow_item_vlan *)
201 						(item->mask))->inner_type;
202 		else
203 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
204 		if (ether_type_m != RTE_BE16(0xFFFF))
205 			break;
206 		ether_type = ((const struct rte_flow_item_vlan *)
207 				(item->spec))->inner_type;
208 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
209 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
210 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
211 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
212 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
213 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
214 		else
215 			ret = RTE_FLOW_ITEM_TYPE_END;
216 		break;
217 	case RTE_FLOW_ITEM_TYPE_IPV4:
218 		if (item->mask)
219 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
220 					(item->mask))->hdr.next_proto_id;
221 		else
222 			ip_next_proto_m =
223 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
224 		if (ip_next_proto_m != 0xFF)
225 			break;
226 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
227 				(item->spec))->hdr.next_proto_id;
228 		if (ip_next_proto == IPPROTO_UDP)
229 			ret = RTE_FLOW_ITEM_TYPE_UDP;
230 		else if (ip_next_proto == IPPROTO_TCP)
231 			ret = RTE_FLOW_ITEM_TYPE_TCP;
232 		else if (ip_next_proto == IPPROTO_IP)
233 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
234 		else if (ip_next_proto == IPPROTO_IPV6)
235 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
236 		else
237 			ret = RTE_FLOW_ITEM_TYPE_END;
238 		break;
239 	case RTE_FLOW_ITEM_TYPE_IPV6:
240 		if (item->mask)
241 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
242 						(item->mask))->hdr.proto;
243 		else
244 			ip_next_proto_m =
245 				rte_flow_item_ipv6_mask.hdr.proto;
246 		if (ip_next_proto_m != 0xFF)
247 			break;
248 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
249 				(item->spec))->hdr.proto;
250 		if (ip_next_proto == IPPROTO_UDP)
251 			ret = RTE_FLOW_ITEM_TYPE_UDP;
252 		else if (ip_next_proto == IPPROTO_TCP)
253 			ret = RTE_FLOW_ITEM_TYPE_TCP;
254 		else if (ip_next_proto == IPPROTO_IP)
255 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
256 		else if (ip_next_proto == IPPROTO_IPV6)
257 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
258 		else
259 			ret = RTE_FLOW_ITEM_TYPE_END;
260 		break;
261 	default:
262 		ret = RTE_FLOW_ITEM_TYPE_VOID;
263 		break;
264 	}
265 	return ret;
266 }
267 
268 static const int *
269 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
270 		const int *next_node)
271 {
272 	const struct mlx5_flow_expand_node *node = NULL;
273 	const int *next = next_node;
274 
275 	while (next && *next) {
276 		/*
277 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
278 		 * flag set, because they were not found in the flow pattern.
279 		 */
280 		node = &graph[*next];
281 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
282 			break;
283 		next = node->next;
284 	}
285 	return next;
286 }
287 
288 #define MLX5_RSS_EXP_ELT_N 16
289 
290 /**
291  * Expand RSS flows into several possible flows according to the RSS hash
292  * fields requested and the driver capabilities.
293  *
294  * @param[out] buf
295  *   Buffer to store the result expansion.
296  * @param[in] size
297  *   Buffer size in bytes. If 0, @p buf can be NULL.
298  * @param[in] pattern
299  *   User flow pattern.
300  * @param[in] types
301  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
302  * @param[in] graph
303  *   Input graph to expand @p pattern according to @p types.
304  * @param[in] graph_root_index
305  *   Index of root node in @p graph, typically 0.
306  *
307  * @return
308  *   A positive value representing the size of @p buf in bytes regardless of
309  *   @p size on success, a negative errno value otherwise and rte_errno is
310  *   set, the following errors are defined:
311  *
312  *   -E2BIG: graph-depth @p graph is too deep.
313  *   -EINVAL: @p size has not enough space for expanded pattern.
314  */
315 static int
316 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
317 		     const struct rte_flow_item *pattern, uint64_t types,
318 		     const struct mlx5_flow_expand_node graph[],
319 		     int graph_root_index)
320 {
321 	const struct rte_flow_item *item;
322 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
323 	const int *next_node;
324 	const int *stack[MLX5_RSS_EXP_ELT_N];
325 	int stack_pos = 0;
326 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
327 	unsigned int i, item_idx, last_expand_item_idx = 0;
328 	size_t lsize;
329 	size_t user_pattern_size = 0;
330 	void *addr = NULL;
331 	const struct mlx5_flow_expand_node *next = NULL;
332 	struct rte_flow_item missed_item;
333 	int missed = 0;
334 	int elt = 0;
335 	const struct rte_flow_item *last_expand_item = NULL;
336 
337 	memset(&missed_item, 0, sizeof(missed_item));
338 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
339 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
340 	if (lsize > size)
341 		return -EINVAL;
342 	buf->entry[0].priority = 0;
343 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
344 	buf->entries = 0;
345 	addr = buf->entry[0].pattern;
346 	for (item = pattern, item_idx = 0;
347 			item->type != RTE_FLOW_ITEM_TYPE_END;
348 			item++, item_idx++) {
349 		if (!mlx5_flow_is_rss_expandable_item(item)) {
350 			user_pattern_size += sizeof(*item);
351 			continue;
352 		}
353 		last_expand_item = item;
354 		last_expand_item_idx = item_idx;
355 		i = 0;
356 		while (node->next && node->next[i]) {
357 			next = &graph[node->next[i]];
358 			if (next->type == item->type)
359 				break;
360 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
361 				node = next;
362 				i = 0;
363 			} else {
364 				++i;
365 			}
366 		}
367 		if (next)
368 			node = next;
369 		user_pattern_size += sizeof(*item);
370 	}
371 	user_pattern_size += sizeof(*item); /* Handle END item. */
372 	lsize += user_pattern_size;
373 	if (lsize > size)
374 		return -EINVAL;
375 	/* Copy the user pattern in the first entry of the buffer. */
376 	rte_memcpy(addr, pattern, user_pattern_size);
377 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
378 	buf->entries = 1;
379 	/* Start expanding. */
380 	memset(flow_items, 0, sizeof(flow_items));
381 	user_pattern_size -= sizeof(*item);
382 	/*
383 	 * Check if the last valid item has spec set, need complete pattern,
384 	 * and the pattern can be used for expansion.
385 	 */
386 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
387 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
388 		/* Item type END indicates expansion is not required. */
389 		return lsize;
390 	}
391 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
392 		next = NULL;
393 		missed = 1;
394 		i = 0;
395 		while (node->next && node->next[i]) {
396 			next = &graph[node->next[i]];
397 			if (next->type == missed_item.type) {
398 				flow_items[0].type = missed_item.type;
399 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
400 				break;
401 			}
402 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
403 				node = next;
404 				i = 0;
405 			} else {
406 				++i;
407 			}
408 			next = NULL;
409 		}
410 	}
411 	if (next && missed) {
412 		elt = 2; /* missed item + item end. */
413 		node = next;
414 		lsize += elt * sizeof(*item) + user_pattern_size;
415 		if (lsize > size)
416 			return -EINVAL;
417 		if (node->rss_types & types) {
418 			buf->entry[buf->entries].priority = 1;
419 			buf->entry[buf->entries].pattern = addr;
420 			buf->entries++;
421 			rte_memcpy(addr, buf->entry[0].pattern,
422 				   user_pattern_size);
423 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
424 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
425 			addr = (void *)(((uintptr_t)addr) +
426 					elt * sizeof(*item));
427 		}
428 	} else if (last_expand_item != NULL) {
429 		node = mlx5_flow_expand_rss_adjust_node(pattern,
430 				last_expand_item_idx, graph, node);
431 	}
432 	memset(flow_items, 0, sizeof(flow_items));
433 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
434 			node->next);
435 	stack[stack_pos] = next_node;
436 	node = next_node ? &graph[*next_node] : NULL;
437 	while (node) {
438 		flow_items[stack_pos].type = node->type;
439 		if (node->rss_types & types) {
440 			size_t n;
441 			/*
442 			 * compute the number of items to copy from the
443 			 * expansion and copy it.
444 			 * When the stack_pos is 0, there are 1 element in it,
445 			 * plus the addition END item.
446 			 */
447 			elt = stack_pos + 2;
448 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
449 			lsize += elt * sizeof(*item) + user_pattern_size;
450 			if (lsize > size)
451 				return -EINVAL;
452 			n = elt * sizeof(*item);
453 			buf->entry[buf->entries].priority =
454 				stack_pos + 1 + missed;
455 			buf->entry[buf->entries].pattern = addr;
456 			buf->entries++;
457 			rte_memcpy(addr, buf->entry[0].pattern,
458 				   user_pattern_size);
459 			addr = (void *)(((uintptr_t)addr) +
460 					user_pattern_size);
461 			rte_memcpy(addr, &missed_item,
462 				   missed * sizeof(*item));
463 			addr = (void *)(((uintptr_t)addr) +
464 				missed * sizeof(*item));
465 			rte_memcpy(addr, flow_items, n);
466 			addr = (void *)(((uintptr_t)addr) + n);
467 		}
468 		/* Go deeper. */
469 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
470 				node->next) {
471 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
472 					node->next);
473 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
474 				rte_errno = E2BIG;
475 				return -rte_errno;
476 			}
477 			stack[stack_pos] = next_node;
478 		} else if (*(next_node + 1)) {
479 			/* Follow up with the next possibility. */
480 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
481 					++next_node);
482 		} else if (!stack_pos) {
483 			/*
484 			 * Completing the traverse over the different paths.
485 			 * The next_node is advanced to the terminator.
486 			 */
487 			++next_node;
488 		} else {
489 			/* Move to the next path. */
490 			while (stack_pos) {
491 				next_node = stack[--stack_pos];
492 				next_node++;
493 				if (*next_node)
494 					break;
495 			}
496 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
497 					next_node);
498 			stack[stack_pos] = next_node;
499 		}
500 		node = next_node && *next_node ? &graph[*next_node] : NULL;
501 	};
502 	return lsize;
503 }
504 
505 enum mlx5_expansion {
506 	MLX5_EXPANSION_ROOT,
507 	MLX5_EXPANSION_ROOT_OUTER,
508 	MLX5_EXPANSION_OUTER_ETH,
509 	MLX5_EXPANSION_OUTER_VLAN,
510 	MLX5_EXPANSION_OUTER_IPV4,
511 	MLX5_EXPANSION_OUTER_IPV4_UDP,
512 	MLX5_EXPANSION_OUTER_IPV4_TCP,
513 	MLX5_EXPANSION_OUTER_IPV6,
514 	MLX5_EXPANSION_OUTER_IPV6_UDP,
515 	MLX5_EXPANSION_OUTER_IPV6_TCP,
516 	MLX5_EXPANSION_VXLAN,
517 	MLX5_EXPANSION_STD_VXLAN,
518 	MLX5_EXPANSION_L3_VXLAN,
519 	MLX5_EXPANSION_VXLAN_GPE,
520 	MLX5_EXPANSION_GRE,
521 	MLX5_EXPANSION_NVGRE,
522 	MLX5_EXPANSION_GRE_KEY,
523 	MLX5_EXPANSION_MPLS,
524 	MLX5_EXPANSION_ETH,
525 	MLX5_EXPANSION_VLAN,
526 	MLX5_EXPANSION_IPV4,
527 	MLX5_EXPANSION_IPV4_UDP,
528 	MLX5_EXPANSION_IPV4_TCP,
529 	MLX5_EXPANSION_IPV6,
530 	MLX5_EXPANSION_IPV6_UDP,
531 	MLX5_EXPANSION_IPV6_TCP,
532 	MLX5_EXPANSION_IPV6_FRAG_EXT,
533 	MLX5_EXPANSION_GTP
534 };
535 
536 /** Supported expansion of items. */
537 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
538 	[MLX5_EXPANSION_ROOT] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
540 						  MLX5_EXPANSION_IPV4,
541 						  MLX5_EXPANSION_IPV6),
542 		.type = RTE_FLOW_ITEM_TYPE_END,
543 	},
544 	[MLX5_EXPANSION_ROOT_OUTER] = {
545 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
546 						  MLX5_EXPANSION_OUTER_IPV4,
547 						  MLX5_EXPANSION_OUTER_IPV6),
548 		.type = RTE_FLOW_ITEM_TYPE_END,
549 	},
550 	[MLX5_EXPANSION_OUTER_ETH] = {
551 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
552 		.type = RTE_FLOW_ITEM_TYPE_ETH,
553 		.rss_types = 0,
554 	},
555 	[MLX5_EXPANSION_OUTER_VLAN] = {
556 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
557 						  MLX5_EXPANSION_OUTER_IPV6),
558 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
559 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
560 	},
561 	[MLX5_EXPANSION_OUTER_IPV4] = {
562 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
563 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
564 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
565 			 MLX5_EXPANSION_GRE,
566 			 MLX5_EXPANSION_NVGRE,
567 			 MLX5_EXPANSION_IPV4,
568 			 MLX5_EXPANSION_IPV6),
569 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
570 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
571 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
572 	},
573 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
574 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
575 						  MLX5_EXPANSION_VXLAN_GPE,
576 						  MLX5_EXPANSION_MPLS,
577 						  MLX5_EXPANSION_GTP),
578 		.type = RTE_FLOW_ITEM_TYPE_UDP,
579 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
580 	},
581 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
582 		.type = RTE_FLOW_ITEM_TYPE_TCP,
583 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
584 	},
585 	[MLX5_EXPANSION_OUTER_IPV6] = {
586 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
587 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
588 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
589 			 MLX5_EXPANSION_IPV4,
590 			 MLX5_EXPANSION_IPV6,
591 			 MLX5_EXPANSION_GRE,
592 			 MLX5_EXPANSION_NVGRE),
593 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
594 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
595 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
596 	},
597 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
598 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
599 						  MLX5_EXPANSION_VXLAN_GPE,
600 						  MLX5_EXPANSION_MPLS,
601 						  MLX5_EXPANSION_GTP),
602 		.type = RTE_FLOW_ITEM_TYPE_UDP,
603 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
604 	},
605 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
606 		.type = RTE_FLOW_ITEM_TYPE_TCP,
607 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
608 	},
609 	[MLX5_EXPANSION_VXLAN] = {
610 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
611 						  MLX5_EXPANSION_IPV4,
612 						  MLX5_EXPANSION_IPV6),
613 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
614 	},
615 	[MLX5_EXPANSION_STD_VXLAN] = {
616 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
617 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
618 	},
619 	[MLX5_EXPANSION_L3_VXLAN] = {
620 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
621 					MLX5_EXPANSION_IPV6),
622 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
623 	},
624 	[MLX5_EXPANSION_VXLAN_GPE] = {
625 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
626 						  MLX5_EXPANSION_IPV4,
627 						  MLX5_EXPANSION_IPV6),
628 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
629 	},
630 	[MLX5_EXPANSION_GRE] = {
631 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
632 						  MLX5_EXPANSION_IPV6,
633 						  MLX5_EXPANSION_GRE_KEY,
634 						  MLX5_EXPANSION_MPLS),
635 		.type = RTE_FLOW_ITEM_TYPE_GRE,
636 	},
637 	[MLX5_EXPANSION_GRE_KEY] = {
638 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
639 						  MLX5_EXPANSION_IPV6,
640 						  MLX5_EXPANSION_MPLS),
641 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
642 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
643 	},
644 	[MLX5_EXPANSION_NVGRE] = {
645 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
646 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
647 	},
648 	[MLX5_EXPANSION_MPLS] = {
649 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
650 						  MLX5_EXPANSION_IPV6,
651 						  MLX5_EXPANSION_ETH),
652 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
653 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
654 	},
655 	[MLX5_EXPANSION_ETH] = {
656 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
657 		.type = RTE_FLOW_ITEM_TYPE_ETH,
658 	},
659 	[MLX5_EXPANSION_VLAN] = {
660 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
661 						  MLX5_EXPANSION_IPV6),
662 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
663 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
664 	},
665 	[MLX5_EXPANSION_IPV4] = {
666 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
667 						  MLX5_EXPANSION_IPV4_TCP),
668 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
669 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
670 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
671 	},
672 	[MLX5_EXPANSION_IPV4_UDP] = {
673 		.type = RTE_FLOW_ITEM_TYPE_UDP,
674 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
675 	},
676 	[MLX5_EXPANSION_IPV4_TCP] = {
677 		.type = RTE_FLOW_ITEM_TYPE_TCP,
678 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
679 	},
680 	[MLX5_EXPANSION_IPV6] = {
681 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
682 						  MLX5_EXPANSION_IPV6_TCP,
683 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
684 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
685 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
686 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
687 	},
688 	[MLX5_EXPANSION_IPV6_UDP] = {
689 		.type = RTE_FLOW_ITEM_TYPE_UDP,
690 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
691 	},
692 	[MLX5_EXPANSION_IPV6_TCP] = {
693 		.type = RTE_FLOW_ITEM_TYPE_TCP,
694 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
695 	},
696 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
697 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
698 	},
699 	[MLX5_EXPANSION_GTP] = {
700 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
701 						  MLX5_EXPANSION_IPV6),
702 		.type = RTE_FLOW_ITEM_TYPE_GTP,
703 	},
704 };
705 
706 static struct rte_flow_action_handle *
707 mlx5_action_handle_create(struct rte_eth_dev *dev,
708 			  const struct rte_flow_indir_action_conf *conf,
709 			  const struct rte_flow_action *action,
710 			  struct rte_flow_error *error);
711 static int mlx5_action_handle_destroy
712 				(struct rte_eth_dev *dev,
713 				 struct rte_flow_action_handle *handle,
714 				 struct rte_flow_error *error);
715 static int mlx5_action_handle_update
716 				(struct rte_eth_dev *dev,
717 				 struct rte_flow_action_handle *handle,
718 				 const void *update,
719 				 struct rte_flow_error *error);
720 static int mlx5_action_handle_query
721 				(struct rte_eth_dev *dev,
722 				 const struct rte_flow_action_handle *handle,
723 				 void *data,
724 				 struct rte_flow_error *error);
725 static int
726 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
727 		    struct rte_flow_tunnel *app_tunnel,
728 		    struct rte_flow_action **actions,
729 		    uint32_t *num_of_actions,
730 		    struct rte_flow_error *error);
731 static int
732 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
733 		       struct rte_flow_tunnel *app_tunnel,
734 		       struct rte_flow_item **items,
735 		       uint32_t *num_of_items,
736 		       struct rte_flow_error *error);
737 static int
738 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
739 			      struct rte_flow_item *pmd_items,
740 			      uint32_t num_items, struct rte_flow_error *err);
741 static int
742 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
743 				struct rte_flow_action *pmd_actions,
744 				uint32_t num_actions,
745 				struct rte_flow_error *err);
746 static int
747 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
748 				  struct rte_mbuf *m,
749 				  struct rte_flow_restore_info *info,
750 				  struct rte_flow_error *err);
751 static struct rte_flow_item_flex_handle *
752 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
753 			   const struct rte_flow_item_flex_conf *conf,
754 			   struct rte_flow_error *error);
755 static int
756 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
757 			    const struct rte_flow_item_flex_handle *handle,
758 			    struct rte_flow_error *error);
759 
760 static const struct rte_flow_ops mlx5_flow_ops = {
761 	.validate = mlx5_flow_validate,
762 	.create = mlx5_flow_create,
763 	.destroy = mlx5_flow_destroy,
764 	.flush = mlx5_flow_flush,
765 	.isolate = mlx5_flow_isolate,
766 	.query = mlx5_flow_query,
767 	.dev_dump = mlx5_flow_dev_dump,
768 	.get_aged_flows = mlx5_flow_get_aged_flows,
769 	.action_handle_create = mlx5_action_handle_create,
770 	.action_handle_destroy = mlx5_action_handle_destroy,
771 	.action_handle_update = mlx5_action_handle_update,
772 	.action_handle_query = mlx5_action_handle_query,
773 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
774 	.tunnel_match = mlx5_flow_tunnel_match,
775 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
776 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
777 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
778 	.flex_item_create = mlx5_flow_flex_item_create,
779 	.flex_item_release = mlx5_flow_flex_item_release,
780 };
781 
782 /* Tunnel information. */
783 struct mlx5_flow_tunnel_info {
784 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
785 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
786 };
787 
788 static struct mlx5_flow_tunnel_info tunnels_info[] = {
789 	{
790 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
791 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
792 	},
793 	{
794 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
795 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
796 	},
797 	{
798 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
799 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
800 	},
801 	{
802 		.tunnel = MLX5_FLOW_LAYER_GRE,
803 		.ptype = RTE_PTYPE_TUNNEL_GRE,
804 	},
805 	{
806 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
807 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
808 	},
809 	{
810 		.tunnel = MLX5_FLOW_LAYER_MPLS,
811 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
812 	},
813 	{
814 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
815 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
816 	},
817 	{
818 		.tunnel = MLX5_FLOW_LAYER_IPIP,
819 		.ptype = RTE_PTYPE_TUNNEL_IP,
820 	},
821 	{
822 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
823 		.ptype = RTE_PTYPE_TUNNEL_IP,
824 	},
825 	{
826 		.tunnel = MLX5_FLOW_LAYER_GTP,
827 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
828 	},
829 };
830 
831 
832 
833 /**
834  * Translate tag ID to register.
835  *
836  * @param[in] dev
837  *   Pointer to the Ethernet device structure.
838  * @param[in] feature
839  *   The feature that request the register.
840  * @param[in] id
841  *   The request register ID.
842  * @param[out] error
843  *   Error description in case of any.
844  *
845  * @return
846  *   The request register on success, a negative errno
847  *   value otherwise and rte_errno is set.
848  */
849 int
850 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
851 		     enum mlx5_feature_name feature,
852 		     uint32_t id,
853 		     struct rte_flow_error *error)
854 {
855 	struct mlx5_priv *priv = dev->data->dev_private;
856 	struct mlx5_dev_config *config = &priv->config;
857 	enum modify_reg start_reg;
858 	bool skip_mtr_reg = false;
859 
860 	switch (feature) {
861 	case MLX5_HAIRPIN_RX:
862 		return REG_B;
863 	case MLX5_HAIRPIN_TX:
864 		return REG_A;
865 	case MLX5_METADATA_RX:
866 		switch (config->dv_xmeta_en) {
867 		case MLX5_XMETA_MODE_LEGACY:
868 			return REG_B;
869 		case MLX5_XMETA_MODE_META16:
870 			return REG_C_0;
871 		case MLX5_XMETA_MODE_META32:
872 			return REG_C_1;
873 		}
874 		break;
875 	case MLX5_METADATA_TX:
876 		return REG_A;
877 	case MLX5_METADATA_FDB:
878 		switch (config->dv_xmeta_en) {
879 		case MLX5_XMETA_MODE_LEGACY:
880 			return REG_NON;
881 		case MLX5_XMETA_MODE_META16:
882 			return REG_C_0;
883 		case MLX5_XMETA_MODE_META32:
884 			return REG_C_1;
885 		}
886 		break;
887 	case MLX5_FLOW_MARK:
888 		switch (config->dv_xmeta_en) {
889 		case MLX5_XMETA_MODE_LEGACY:
890 			return REG_NON;
891 		case MLX5_XMETA_MODE_META16:
892 			return REG_C_1;
893 		case MLX5_XMETA_MODE_META32:
894 			return REG_C_0;
895 		}
896 		break;
897 	case MLX5_MTR_ID:
898 		/*
899 		 * If meter color and meter id share one register, flow match
900 		 * should use the meter color register for match.
901 		 */
902 		if (priv->mtr_reg_share)
903 			return priv->mtr_color_reg;
904 		else
905 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
906 			       REG_C_3;
907 	case MLX5_MTR_COLOR:
908 	case MLX5_ASO_FLOW_HIT:
909 	case MLX5_ASO_CONNTRACK:
910 	case MLX5_SAMPLE_ID:
911 		/* All features use the same REG_C. */
912 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
913 		return priv->mtr_color_reg;
914 	case MLX5_COPY_MARK:
915 		/*
916 		 * Metadata COPY_MARK register using is in meter suffix sub
917 		 * flow while with meter. It's safe to share the same register.
918 		 */
919 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
920 	case MLX5_APP_TAG:
921 		/*
922 		 * If meter is enable, it will engage the register for color
923 		 * match and flow match. If meter color match is not using the
924 		 * REG_C_2, need to skip the REG_C_x be used by meter color
925 		 * match.
926 		 * If meter is disable, free to use all available registers.
927 		 */
928 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
929 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
930 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
931 		if (id > (uint32_t)(REG_C_7 - start_reg))
932 			return rte_flow_error_set(error, EINVAL,
933 						  RTE_FLOW_ERROR_TYPE_ITEM,
934 						  NULL, "invalid tag id");
935 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
936 			return rte_flow_error_set(error, ENOTSUP,
937 						  RTE_FLOW_ERROR_TYPE_ITEM,
938 						  NULL, "unsupported tag id");
939 		/*
940 		 * This case means meter is using the REG_C_x great than 2.
941 		 * Take care not to conflict with meter color REG_C_x.
942 		 * If the available index REG_C_y >= REG_C_x, skip the
943 		 * color register.
944 		 */
945 		if (skip_mtr_reg && priv->sh->flow_mreg_c
946 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
947 			if (id >= (uint32_t)(REG_C_7 - start_reg))
948 				return rte_flow_error_set(error, EINVAL,
949 						       RTE_FLOW_ERROR_TYPE_ITEM,
950 							NULL, "invalid tag id");
951 			if (priv->sh->flow_mreg_c
952 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
953 				return priv->sh->flow_mreg_c
954 					       [id + 1 + start_reg - REG_C_0];
955 			return rte_flow_error_set(error, ENOTSUP,
956 						  RTE_FLOW_ERROR_TYPE_ITEM,
957 						  NULL, "unsupported tag id");
958 		}
959 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
960 	}
961 	MLX5_ASSERT(false);
962 	return rte_flow_error_set(error, EINVAL,
963 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
964 				  NULL, "invalid feature name");
965 }
966 
967 /**
968  * Check extensive flow metadata register support.
969  *
970  * @param dev
971  *   Pointer to rte_eth_dev structure.
972  *
973  * @return
974  *   True if device supports extensive flow metadata register, otherwise false.
975  */
976 bool
977 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
978 {
979 	struct mlx5_priv *priv = dev->data->dev_private;
980 
981 	/*
982 	 * Having available reg_c can be regarded inclusively as supporting
983 	 * extensive flow metadata register, which could mean,
984 	 * - metadata register copy action by modify header.
985 	 * - 16 modify header actions is supported.
986 	 * - reg_c's are preserved across different domain (FDB and NIC) on
987 	 *   packet loopback by flow lookup miss.
988 	 */
989 	return priv->sh->flow_mreg_c[2] != REG_NON;
990 }
991 
992 /**
993  * Get the lowest priority.
994  *
995  * @param[in] dev
996  *   Pointer to the Ethernet device structure.
997  * @param[in] attributes
998  *   Pointer to device flow rule attributes.
999  *
1000  * @return
1001  *   The value of lowest priority of flow.
1002  */
1003 uint32_t
1004 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1005 			  const struct rte_flow_attr *attr)
1006 {
1007 	struct mlx5_priv *priv = dev->data->dev_private;
1008 
1009 	if (!attr->group && !attr->transfer)
1010 		return priv->sh->flow_max_priority - 2;
1011 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1012 }
1013 
1014 /**
1015  * Calculate matcher priority of the flow.
1016  *
1017  * @param[in] dev
1018  *   Pointer to the Ethernet device structure.
1019  * @param[in] attr
1020  *   Pointer to device flow rule attributes.
1021  * @param[in] subpriority
1022  *   The priority based on the items.
1023  * @param[in] external
1024  *   Flow is user flow.
1025  * @return
1026  *   The matcher priority of the flow.
1027  */
1028 uint16_t
1029 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1030 			  const struct rte_flow_attr *attr,
1031 			  uint32_t subpriority, bool external)
1032 {
1033 	uint16_t priority = (uint16_t)attr->priority;
1034 	struct mlx5_priv *priv = dev->data->dev_private;
1035 
1036 	if (!attr->group && !attr->transfer) {
1037 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1038 			priority = priv->sh->flow_max_priority - 1;
1039 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1040 	} else if (!external && attr->transfer && attr->group == 0 &&
1041 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1042 		return (priv->sh->flow_max_priority - 1) * 3;
1043 	}
1044 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1045 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1046 	return priority * 3 + subpriority;
1047 }
1048 
1049 /**
1050  * Verify the @p item specifications (spec, last, mask) are compatible with the
1051  * NIC capabilities.
1052  *
1053  * @param[in] item
1054  *   Item specification.
1055  * @param[in] mask
1056  *   @p item->mask or flow default bit-masks.
1057  * @param[in] nic_mask
1058  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1059  * @param[in] size
1060  *   Bit-masks size in bytes.
1061  * @param[in] range_accepted
1062  *   True if range of values is accepted for specific fields, false otherwise.
1063  * @param[out] error
1064  *   Pointer to error structure.
1065  *
1066  * @return
1067  *   0 on success, a negative errno value otherwise and rte_errno is set.
1068  */
1069 int
1070 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1071 			  const uint8_t *mask,
1072 			  const uint8_t *nic_mask,
1073 			  unsigned int size,
1074 			  bool range_accepted,
1075 			  struct rte_flow_error *error)
1076 {
1077 	unsigned int i;
1078 
1079 	MLX5_ASSERT(nic_mask);
1080 	for (i = 0; i < size; ++i)
1081 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1082 			return rte_flow_error_set(error, ENOTSUP,
1083 						  RTE_FLOW_ERROR_TYPE_ITEM,
1084 						  item,
1085 						  "mask enables non supported"
1086 						  " bits");
1087 	if (!item->spec && (item->mask || item->last))
1088 		return rte_flow_error_set(error, EINVAL,
1089 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1090 					  "mask/last without a spec is not"
1091 					  " supported");
1092 	if (item->spec && item->last && !range_accepted) {
1093 		uint8_t spec[size];
1094 		uint8_t last[size];
1095 		unsigned int i;
1096 		int ret;
1097 
1098 		for (i = 0; i < size; ++i) {
1099 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1100 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1101 		}
1102 		ret = memcmp(spec, last, size);
1103 		if (ret != 0)
1104 			return rte_flow_error_set(error, EINVAL,
1105 						  RTE_FLOW_ERROR_TYPE_ITEM,
1106 						  item,
1107 						  "range is not valid");
1108 	}
1109 	return 0;
1110 }
1111 
1112 /**
1113  * Adjust the hash fields according to the @p flow information.
1114  *
1115  * @param[in] dev_flow.
1116  *   Pointer to the mlx5_flow.
1117  * @param[in] tunnel
1118  *   1 when the hash field is for a tunnel item.
1119  * @param[in] layer_types
1120  *   RTE_ETH_RSS_* types.
1121  * @param[in] hash_fields
1122  *   Item hash fields.
1123  *
1124  * @return
1125  *   The hash fields that should be used.
1126  */
1127 uint64_t
1128 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1129 			    int tunnel __rte_unused, uint64_t layer_types,
1130 			    uint64_t hash_fields)
1131 {
1132 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1133 	int rss_request_inner = rss_desc->level >= 2;
1134 
1135 	/* Check RSS hash level for tunnel. */
1136 	if (tunnel && rss_request_inner)
1137 		hash_fields |= IBV_RX_HASH_INNER;
1138 	else if (tunnel || rss_request_inner)
1139 		return 0;
1140 #endif
1141 	/* Check if requested layer matches RSS hash fields. */
1142 	if (!(rss_desc->types & layer_types))
1143 		return 0;
1144 	return hash_fields;
1145 }
1146 
1147 /**
1148  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1149  * if several tunnel rules are used on this queue, the tunnel ptype will be
1150  * cleared.
1151  *
1152  * @param rxq_ctrl
1153  *   Rx queue to update.
1154  */
1155 static void
1156 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1157 {
1158 	unsigned int i;
1159 	uint32_t tunnel_ptype = 0;
1160 
1161 	/* Look up for the ptype to use. */
1162 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1163 		if (!rxq_ctrl->flow_tunnels_n[i])
1164 			continue;
1165 		if (!tunnel_ptype) {
1166 			tunnel_ptype = tunnels_info[i].ptype;
1167 		} else {
1168 			tunnel_ptype = 0;
1169 			break;
1170 		}
1171 	}
1172 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1173 }
1174 
1175 /**
1176  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1177  * flow.
1178  *
1179  * @param[in] dev
1180  *   Pointer to the Ethernet device structure.
1181  * @param[in] dev_handle
1182  *   Pointer to device flow handle structure.
1183  */
1184 void
1185 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1186 		       struct mlx5_flow_handle *dev_handle)
1187 {
1188 	struct mlx5_priv *priv = dev->data->dev_private;
1189 	const int mark = dev_handle->mark;
1190 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1191 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1192 	unsigned int i;
1193 
1194 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1195 		struct mlx5_hrxq *hrxq;
1196 
1197 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1198 			      dev_handle->rix_hrxq);
1199 		if (hrxq)
1200 			ind_tbl = hrxq->ind_table;
1201 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1202 		struct mlx5_shared_action_rss *shared_rss;
1203 
1204 		shared_rss = mlx5_ipool_get
1205 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1206 			 dev_handle->rix_srss);
1207 		if (shared_rss)
1208 			ind_tbl = shared_rss->ind_tbl;
1209 	}
1210 	if (!ind_tbl)
1211 		return;
1212 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1213 		int idx = ind_tbl->queues[i];
1214 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1215 
1216 		MLX5_ASSERT(rxq_ctrl != NULL);
1217 		if (rxq_ctrl == NULL)
1218 			continue;
1219 		/*
1220 		 * To support metadata register copy on Tx loopback,
1221 		 * this must be always enabled (metadata may arive
1222 		 * from other port - not from local flows only.
1223 		 */
1224 		if (priv->config.dv_flow_en &&
1225 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1226 		    mlx5_flow_ext_mreg_supported(dev)) {
1227 			rxq_ctrl->rxq.mark = 1;
1228 			rxq_ctrl->flow_mark_n = 1;
1229 		} else if (mark) {
1230 			rxq_ctrl->rxq.mark = 1;
1231 			rxq_ctrl->flow_mark_n++;
1232 		}
1233 		if (tunnel) {
1234 			unsigned int j;
1235 
1236 			/* Increase the counter matching the flow. */
1237 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1238 				if ((tunnels_info[j].tunnel &
1239 				     dev_handle->layers) ==
1240 				    tunnels_info[j].tunnel) {
1241 					rxq_ctrl->flow_tunnels_n[j]++;
1242 					break;
1243 				}
1244 			}
1245 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1246 		}
1247 	}
1248 }
1249 
1250 /**
1251  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1252  *
1253  * @param[in] dev
1254  *   Pointer to the Ethernet device structure.
1255  * @param[in] flow
1256  *   Pointer to flow structure.
1257  */
1258 static void
1259 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1260 {
1261 	struct mlx5_priv *priv = dev->data->dev_private;
1262 	uint32_t handle_idx;
1263 	struct mlx5_flow_handle *dev_handle;
1264 
1265 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1266 		       handle_idx, dev_handle, next)
1267 		flow_drv_rxq_flags_set(dev, dev_handle);
1268 }
1269 
1270 /**
1271  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1272  * device flow if no other flow uses it with the same kind of request.
1273  *
1274  * @param dev
1275  *   Pointer to Ethernet device.
1276  * @param[in] dev_handle
1277  *   Pointer to the device flow handle structure.
1278  */
1279 static void
1280 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1281 			struct mlx5_flow_handle *dev_handle)
1282 {
1283 	struct mlx5_priv *priv = dev->data->dev_private;
1284 	const int mark = dev_handle->mark;
1285 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1286 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1287 	unsigned int i;
1288 
1289 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1290 		struct mlx5_hrxq *hrxq;
1291 
1292 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1293 			      dev_handle->rix_hrxq);
1294 		if (hrxq)
1295 			ind_tbl = hrxq->ind_table;
1296 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1297 		struct mlx5_shared_action_rss *shared_rss;
1298 
1299 		shared_rss = mlx5_ipool_get
1300 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1301 			 dev_handle->rix_srss);
1302 		if (shared_rss)
1303 			ind_tbl = shared_rss->ind_tbl;
1304 	}
1305 	if (!ind_tbl)
1306 		return;
1307 	MLX5_ASSERT(dev->data->dev_started);
1308 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1309 		int idx = ind_tbl->queues[i];
1310 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1311 
1312 		MLX5_ASSERT(rxq_ctrl != NULL);
1313 		if (rxq_ctrl == NULL)
1314 			continue;
1315 		if (priv->config.dv_flow_en &&
1316 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1317 		    mlx5_flow_ext_mreg_supported(dev)) {
1318 			rxq_ctrl->rxq.mark = 1;
1319 			rxq_ctrl->flow_mark_n = 1;
1320 		} else if (mark) {
1321 			rxq_ctrl->flow_mark_n--;
1322 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1323 		}
1324 		if (tunnel) {
1325 			unsigned int j;
1326 
1327 			/* Decrease the counter matching the flow. */
1328 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1329 				if ((tunnels_info[j].tunnel &
1330 				     dev_handle->layers) ==
1331 				    tunnels_info[j].tunnel) {
1332 					rxq_ctrl->flow_tunnels_n[j]--;
1333 					break;
1334 				}
1335 			}
1336 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1337 		}
1338 	}
1339 }
1340 
1341 /**
1342  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1343  * @p flow if no other flow uses it with the same kind of request.
1344  *
1345  * @param dev
1346  *   Pointer to Ethernet device.
1347  * @param[in] flow
1348  *   Pointer to the flow.
1349  */
1350 static void
1351 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1352 {
1353 	struct mlx5_priv *priv = dev->data->dev_private;
1354 	uint32_t handle_idx;
1355 	struct mlx5_flow_handle *dev_handle;
1356 
1357 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1358 		       handle_idx, dev_handle, next)
1359 		flow_drv_rxq_flags_trim(dev, dev_handle);
1360 }
1361 
1362 /**
1363  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1364  *
1365  * @param dev
1366  *   Pointer to Ethernet device.
1367  */
1368 static void
1369 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1370 {
1371 	struct mlx5_priv *priv = dev->data->dev_private;
1372 	unsigned int i;
1373 
1374 	for (i = 0; i != priv->rxqs_n; ++i) {
1375 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1376 		unsigned int j;
1377 
1378 		if (rxq == NULL || rxq->ctrl == NULL)
1379 			continue;
1380 		rxq->ctrl->flow_mark_n = 0;
1381 		rxq->ctrl->rxq.mark = 0;
1382 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1383 			rxq->ctrl->flow_tunnels_n[j] = 0;
1384 		rxq->ctrl->rxq.tunnel = 0;
1385 	}
1386 }
1387 
1388 /**
1389  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1390  *
1391  * @param[in] dev
1392  *   Pointer to the Ethernet device structure.
1393  */
1394 void
1395 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1396 {
1397 	struct mlx5_priv *priv = dev->data->dev_private;
1398 	unsigned int i;
1399 
1400 	for (i = 0; i != priv->rxqs_n; ++i) {
1401 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1402 		struct mlx5_rxq_data *data;
1403 
1404 		if (rxq == NULL || rxq->ctrl == NULL)
1405 			continue;
1406 		data = &rxq->ctrl->rxq;
1407 		if (!rte_flow_dynf_metadata_avail()) {
1408 			data->dynf_meta = 0;
1409 			data->flow_meta_mask = 0;
1410 			data->flow_meta_offset = -1;
1411 			data->flow_meta_port_mask = 0;
1412 		} else {
1413 			data->dynf_meta = 1;
1414 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1415 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1416 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1417 		}
1418 	}
1419 }
1420 
1421 /*
1422  * return a pointer to the desired action in the list of actions.
1423  *
1424  * @param[in] actions
1425  *   The list of actions to search the action in.
1426  * @param[in] action
1427  *   The action to find.
1428  *
1429  * @return
1430  *   Pointer to the action in the list, if found. NULL otherwise.
1431  */
1432 const struct rte_flow_action *
1433 mlx5_flow_find_action(const struct rte_flow_action *actions,
1434 		      enum rte_flow_action_type action)
1435 {
1436 	if (actions == NULL)
1437 		return NULL;
1438 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1439 		if (actions->type == action)
1440 			return actions;
1441 	return NULL;
1442 }
1443 
1444 /*
1445  * Validate the flag action.
1446  *
1447  * @param[in] action_flags
1448  *   Bit-fields that holds the actions detected until now.
1449  * @param[in] attr
1450  *   Attributes of flow that includes this action.
1451  * @param[out] error
1452  *   Pointer to error structure.
1453  *
1454  * @return
1455  *   0 on success, a negative errno value otherwise and rte_errno is set.
1456  */
1457 int
1458 mlx5_flow_validate_action_flag(uint64_t action_flags,
1459 			       const struct rte_flow_attr *attr,
1460 			       struct rte_flow_error *error)
1461 {
1462 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1463 		return rte_flow_error_set(error, EINVAL,
1464 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1465 					  "can't mark and flag in same flow");
1466 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1467 		return rte_flow_error_set(error, EINVAL,
1468 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1469 					  "can't have 2 flag"
1470 					  " actions in same flow");
1471 	if (attr->egress)
1472 		return rte_flow_error_set(error, ENOTSUP,
1473 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1474 					  "flag action not supported for "
1475 					  "egress");
1476 	return 0;
1477 }
1478 
1479 /*
1480  * Validate the mark action.
1481  *
1482  * @param[in] action
1483  *   Pointer to the queue action.
1484  * @param[in] action_flags
1485  *   Bit-fields that holds the actions detected until now.
1486  * @param[in] attr
1487  *   Attributes of flow that includes this action.
1488  * @param[out] error
1489  *   Pointer to error structure.
1490  *
1491  * @return
1492  *   0 on success, a negative errno value otherwise and rte_errno is set.
1493  */
1494 int
1495 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1496 			       uint64_t action_flags,
1497 			       const struct rte_flow_attr *attr,
1498 			       struct rte_flow_error *error)
1499 {
1500 	const struct rte_flow_action_mark *mark = action->conf;
1501 
1502 	if (!mark)
1503 		return rte_flow_error_set(error, EINVAL,
1504 					  RTE_FLOW_ERROR_TYPE_ACTION,
1505 					  action,
1506 					  "configuration cannot be null");
1507 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1510 					  &mark->id,
1511 					  "mark id must in 0 <= id < "
1512 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1513 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1514 		return rte_flow_error_set(error, EINVAL,
1515 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1516 					  "can't flag and mark in same flow");
1517 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1518 		return rte_flow_error_set(error, EINVAL,
1519 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1520 					  "can't have 2 mark actions in same"
1521 					  " flow");
1522 	if (attr->egress)
1523 		return rte_flow_error_set(error, ENOTSUP,
1524 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1525 					  "mark action not supported for "
1526 					  "egress");
1527 	return 0;
1528 }
1529 
1530 /*
1531  * Validate the drop action.
1532  *
1533  * @param[in] action_flags
1534  *   Bit-fields that holds the actions detected until now.
1535  * @param[in] attr
1536  *   Attributes of flow that includes this action.
1537  * @param[out] error
1538  *   Pointer to error structure.
1539  *
1540  * @return
1541  *   0 on success, a negative errno value otherwise and rte_errno is set.
1542  */
1543 int
1544 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1545 			       const struct rte_flow_attr *attr,
1546 			       struct rte_flow_error *error)
1547 {
1548 	if (attr->egress)
1549 		return rte_flow_error_set(error, ENOTSUP,
1550 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1551 					  "drop action not supported for "
1552 					  "egress");
1553 	return 0;
1554 }
1555 
1556 /*
1557  * Validate the queue action.
1558  *
1559  * @param[in] action
1560  *   Pointer to the queue action.
1561  * @param[in] action_flags
1562  *   Bit-fields that holds the actions detected until now.
1563  * @param[in] dev
1564  *   Pointer to the Ethernet device structure.
1565  * @param[in] attr
1566  *   Attributes of flow that includes this action.
1567  * @param[out] error
1568  *   Pointer to error structure.
1569  *
1570  * @return
1571  *   0 on success, a negative errno value otherwise and rte_errno is set.
1572  */
1573 int
1574 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1575 				uint64_t action_flags,
1576 				struct rte_eth_dev *dev,
1577 				const struct rte_flow_attr *attr,
1578 				struct rte_flow_error *error)
1579 {
1580 	struct mlx5_priv *priv = dev->data->dev_private;
1581 	const struct rte_flow_action_queue *queue = action->conf;
1582 
1583 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1584 		return rte_flow_error_set(error, EINVAL,
1585 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1586 					  "can't have 2 fate actions in"
1587 					  " same flow");
1588 	if (!priv->rxqs_n)
1589 		return rte_flow_error_set(error, EINVAL,
1590 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1591 					  NULL, "No Rx queues configured");
1592 	if (queue->index >= priv->rxqs_n)
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1595 					  &queue->index,
1596 					  "queue index out of range");
1597 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1598 		return rte_flow_error_set(error, EINVAL,
1599 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1600 					  &queue->index,
1601 					  "queue is not configured");
1602 	if (attr->egress)
1603 		return rte_flow_error_set(error, ENOTSUP,
1604 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1605 					  "queue action not supported for "
1606 					  "egress");
1607 	return 0;
1608 }
1609 
1610 /**
1611  * Validate queue numbers for device RSS.
1612  *
1613  * @param[in] dev
1614  *   Configured device.
1615  * @param[in] queues
1616  *   Array of queue numbers.
1617  * @param[in] queues_n
1618  *   Size of the @p queues array.
1619  * @param[out] error
1620  *   On error, filled with a textual error description.
1621  * @param[out] queue
1622  *   On error, filled with an offending queue index in @p queues array.
1623  *
1624  * @return
1625  *   0 on success, a negative errno code on error.
1626  */
1627 static int
1628 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1629 			 const uint16_t *queues, uint32_t queues_n,
1630 			 const char **error, uint32_t *queue_idx)
1631 {
1632 	const struct mlx5_priv *priv = dev->data->dev_private;
1633 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1634 	uint32_t i;
1635 
1636 	for (i = 0; i != queues_n; ++i) {
1637 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1638 								   queues[i]);
1639 
1640 		if (queues[i] >= priv->rxqs_n) {
1641 			*error = "queue index out of range";
1642 			*queue_idx = i;
1643 			return -EINVAL;
1644 		}
1645 		if (rxq_ctrl == NULL) {
1646 			*error =  "queue is not configured";
1647 			*queue_idx = i;
1648 			return -EINVAL;
1649 		}
1650 		if (i == 0)
1651 			rxq_type = rxq_ctrl->type;
1652 		if (rxq_type != rxq_ctrl->type) {
1653 			*error = "combining hairpin and regular RSS queues is not supported";
1654 			*queue_idx = i;
1655 			return -ENOTSUP;
1656 		}
1657 	}
1658 	return 0;
1659 }
1660 
1661 /*
1662  * Validate the rss action.
1663  *
1664  * @param[in] dev
1665  *   Pointer to the Ethernet device structure.
1666  * @param[in] action
1667  *   Pointer to the queue action.
1668  * @param[out] error
1669  *   Pointer to error structure.
1670  *
1671  * @return
1672  *   0 on success, a negative errno value otherwise and rte_errno is set.
1673  */
1674 int
1675 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1676 			 const struct rte_flow_action *action,
1677 			 struct rte_flow_error *error)
1678 {
1679 	struct mlx5_priv *priv = dev->data->dev_private;
1680 	const struct rte_flow_action_rss *rss = action->conf;
1681 	int ret;
1682 	const char *message;
1683 	uint32_t queue_idx;
1684 
1685 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1686 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1687 		return rte_flow_error_set(error, ENOTSUP,
1688 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1689 					  &rss->func,
1690 					  "RSS hash function not supported");
1691 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1692 	if (rss->level > 2)
1693 #else
1694 	if (rss->level > 1)
1695 #endif
1696 		return rte_flow_error_set(error, ENOTSUP,
1697 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1698 					  &rss->level,
1699 					  "tunnel RSS is not supported");
1700 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1701 	if (rss->key_len == 0 && rss->key != NULL)
1702 		return rte_flow_error_set(error, ENOTSUP,
1703 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1704 					  &rss->key_len,
1705 					  "RSS hash key length 0");
1706 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1707 		return rte_flow_error_set(error, ENOTSUP,
1708 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1709 					  &rss->key_len,
1710 					  "RSS hash key too small");
1711 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1712 		return rte_flow_error_set(error, ENOTSUP,
1713 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1714 					  &rss->key_len,
1715 					  "RSS hash key too large");
1716 	if (rss->queue_num > priv->config.ind_table_max_size)
1717 		return rte_flow_error_set(error, ENOTSUP,
1718 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1719 					  &rss->queue_num,
1720 					  "number of queues too large");
1721 	if (rss->types & MLX5_RSS_HF_MASK)
1722 		return rte_flow_error_set(error, ENOTSUP,
1723 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1724 					  &rss->types,
1725 					  "some RSS protocols are not"
1726 					  " supported");
1727 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1728 	    !(rss->types & RTE_ETH_RSS_IP))
1729 		return rte_flow_error_set(error, EINVAL,
1730 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1731 					  "L3 partial RSS requested but L3 RSS"
1732 					  " type not specified");
1733 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1734 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1735 		return rte_flow_error_set(error, EINVAL,
1736 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1737 					  "L4 partial RSS requested but L4 RSS"
1738 					  " type not specified");
1739 	if (!priv->rxqs_n)
1740 		return rte_flow_error_set(error, EINVAL,
1741 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1742 					  NULL, "No Rx queues configured");
1743 	if (!rss->queue_num)
1744 		return rte_flow_error_set(error, EINVAL,
1745 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1746 					  NULL, "No queues configured");
1747 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1748 				       &message, &queue_idx);
1749 	if (ret != 0) {
1750 		return rte_flow_error_set(error, -ret,
1751 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1752 					  &rss->queue[queue_idx], message);
1753 	}
1754 	return 0;
1755 }
1756 
1757 /*
1758  * Validate the rss action.
1759  *
1760  * @param[in] action
1761  *   Pointer to the queue action.
1762  * @param[in] action_flags
1763  *   Bit-fields that holds the actions detected until now.
1764  * @param[in] dev
1765  *   Pointer to the Ethernet device structure.
1766  * @param[in] attr
1767  *   Attributes of flow that includes this action.
1768  * @param[in] item_flags
1769  *   Items that were detected.
1770  * @param[out] error
1771  *   Pointer to error structure.
1772  *
1773  * @return
1774  *   0 on success, a negative errno value otherwise and rte_errno is set.
1775  */
1776 int
1777 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1778 			      uint64_t action_flags,
1779 			      struct rte_eth_dev *dev,
1780 			      const struct rte_flow_attr *attr,
1781 			      uint64_t item_flags,
1782 			      struct rte_flow_error *error)
1783 {
1784 	const struct rte_flow_action_rss *rss = action->conf;
1785 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1786 	int ret;
1787 
1788 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1791 					  "can't have 2 fate actions"
1792 					  " in same flow");
1793 	ret = mlx5_validate_action_rss(dev, action, error);
1794 	if (ret)
1795 		return ret;
1796 	if (attr->egress)
1797 		return rte_flow_error_set(error, ENOTSUP,
1798 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1799 					  "rss action not supported for "
1800 					  "egress");
1801 	if (rss->level > 1 && !tunnel)
1802 		return rte_flow_error_set(error, EINVAL,
1803 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1804 					  "inner RSS is not supported for "
1805 					  "non-tunnel flows");
1806 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1807 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1808 		return rte_flow_error_set(error, EINVAL,
1809 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1810 					  "RSS on eCPRI is not supported now");
1811 	}
1812 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1813 	    !(item_flags &
1814 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1815 	    rss->level > 1)
1816 		return rte_flow_error_set(error, EINVAL,
1817 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1818 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1819 	return 0;
1820 }
1821 
1822 /*
1823  * Validate the default miss action.
1824  *
1825  * @param[in] action_flags
1826  *   Bit-fields that holds the actions detected until now.
1827  * @param[out] error
1828  *   Pointer to error structure.
1829  *
1830  * @return
1831  *   0 on success, a negative errno value otherwise and rte_errno is set.
1832  */
1833 int
1834 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1835 				const struct rte_flow_attr *attr,
1836 				struct rte_flow_error *error)
1837 {
1838 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1839 		return rte_flow_error_set(error, EINVAL,
1840 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1841 					  "can't have 2 fate actions in"
1842 					  " same flow");
1843 	if (attr->egress)
1844 		return rte_flow_error_set(error, ENOTSUP,
1845 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1846 					  "default miss action not supported "
1847 					  "for egress");
1848 	if (attr->group)
1849 		return rte_flow_error_set(error, ENOTSUP,
1850 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1851 					  "only group 0 is supported");
1852 	if (attr->transfer)
1853 		return rte_flow_error_set(error, ENOTSUP,
1854 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1855 					  NULL, "transfer is not supported");
1856 	return 0;
1857 }
1858 
1859 /*
1860  * Validate the count action.
1861  *
1862  * @param[in] dev
1863  *   Pointer to the Ethernet device structure.
1864  * @param[in] attr
1865  *   Attributes of flow that includes this action.
1866  * @param[out] error
1867  *   Pointer to error structure.
1868  *
1869  * @return
1870  *   0 on success, a negative errno value otherwise and rte_errno is set.
1871  */
1872 int
1873 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1874 				const struct rte_flow_attr *attr,
1875 				struct rte_flow_error *error)
1876 {
1877 	if (attr->egress)
1878 		return rte_flow_error_set(error, ENOTSUP,
1879 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1880 					  "count action not supported for "
1881 					  "egress");
1882 	return 0;
1883 }
1884 
1885 /*
1886  * Validate the ASO CT action.
1887  *
1888  * @param[in] dev
1889  *   Pointer to the Ethernet device structure.
1890  * @param[in] conntrack
1891  *   Pointer to the CT action profile.
1892  * @param[out] error
1893  *   Pointer to error structure.
1894  *
1895  * @return
1896  *   0 on success, a negative errno value otherwise and rte_errno is set.
1897  */
1898 int
1899 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1900 			const struct rte_flow_action_conntrack *conntrack,
1901 			struct rte_flow_error *error)
1902 {
1903 	RTE_SET_USED(dev);
1904 
1905 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1906 		return rte_flow_error_set(error, EINVAL,
1907 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1908 					  "Invalid CT state");
1909 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1910 		return rte_flow_error_set(error, EINVAL,
1911 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1912 					  "Invalid last TCP packet flag");
1913 	return 0;
1914 }
1915 
1916 /**
1917  * Verify the @p attributes will be correctly understood by the NIC and store
1918  * them in the @p flow if everything is correct.
1919  *
1920  * @param[in] dev
1921  *   Pointer to the Ethernet device structure.
1922  * @param[in] attributes
1923  *   Pointer to flow attributes
1924  * @param[out] error
1925  *   Pointer to error structure.
1926  *
1927  * @return
1928  *   0 on success, a negative errno value otherwise and rte_errno is set.
1929  */
1930 int
1931 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1932 			      const struct rte_flow_attr *attributes,
1933 			      struct rte_flow_error *error)
1934 {
1935 	struct mlx5_priv *priv = dev->data->dev_private;
1936 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1937 
1938 	if (attributes->group)
1939 		return rte_flow_error_set(error, ENOTSUP,
1940 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1941 					  NULL, "groups is not supported");
1942 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1943 	    attributes->priority >= priority_max)
1944 		return rte_flow_error_set(error, ENOTSUP,
1945 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1946 					  NULL, "priority out of range");
1947 	if (attributes->egress)
1948 		return rte_flow_error_set(error, ENOTSUP,
1949 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1950 					  "egress is not supported");
1951 	if (attributes->transfer && !priv->config.dv_esw_en)
1952 		return rte_flow_error_set(error, ENOTSUP,
1953 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1954 					  NULL, "transfer is not supported");
1955 	if (!attributes->ingress)
1956 		return rte_flow_error_set(error, EINVAL,
1957 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1958 					  NULL,
1959 					  "ingress attribute is mandatory");
1960 	return 0;
1961 }
1962 
1963 /**
1964  * Validate ICMP6 item.
1965  *
1966  * @param[in] item
1967  *   Item specification.
1968  * @param[in] item_flags
1969  *   Bit-fields that holds the items detected until now.
1970  * @param[in] ext_vlan_sup
1971  *   Whether extended VLAN features are supported or not.
1972  * @param[out] error
1973  *   Pointer to error structure.
1974  *
1975  * @return
1976  *   0 on success, a negative errno value otherwise and rte_errno is set.
1977  */
1978 int
1979 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1980 			       uint64_t item_flags,
1981 			       uint8_t target_protocol,
1982 			       struct rte_flow_error *error)
1983 {
1984 	const struct rte_flow_item_icmp6 *mask = item->mask;
1985 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1986 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1987 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1988 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1989 				      MLX5_FLOW_LAYER_OUTER_L4;
1990 	int ret;
1991 
1992 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1993 		return rte_flow_error_set(error, EINVAL,
1994 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1995 					  "protocol filtering not compatible"
1996 					  " with ICMP6 layer");
1997 	if (!(item_flags & l3m))
1998 		return rte_flow_error_set(error, EINVAL,
1999 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2000 					  "IPv6 is mandatory to filter on"
2001 					  " ICMP6");
2002 	if (item_flags & l4m)
2003 		return rte_flow_error_set(error, EINVAL,
2004 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2005 					  "multiple L4 layers not supported");
2006 	if (!mask)
2007 		mask = &rte_flow_item_icmp6_mask;
2008 	ret = mlx5_flow_item_acceptable
2009 		(item, (const uint8_t *)mask,
2010 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2011 		 sizeof(struct rte_flow_item_icmp6),
2012 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2013 	if (ret < 0)
2014 		return ret;
2015 	return 0;
2016 }
2017 
2018 /**
2019  * Validate ICMP item.
2020  *
2021  * @param[in] item
2022  *   Item specification.
2023  * @param[in] item_flags
2024  *   Bit-fields that holds the items detected until now.
2025  * @param[out] error
2026  *   Pointer to error structure.
2027  *
2028  * @return
2029  *   0 on success, a negative errno value otherwise and rte_errno is set.
2030  */
2031 int
2032 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2033 			     uint64_t item_flags,
2034 			     uint8_t target_protocol,
2035 			     struct rte_flow_error *error)
2036 {
2037 	const struct rte_flow_item_icmp *mask = item->mask;
2038 	const struct rte_flow_item_icmp nic_mask = {
2039 		.hdr.icmp_type = 0xff,
2040 		.hdr.icmp_code = 0xff,
2041 		.hdr.icmp_ident = RTE_BE16(0xffff),
2042 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2043 	};
2044 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2045 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2046 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2047 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2048 				      MLX5_FLOW_LAYER_OUTER_L4;
2049 	int ret;
2050 
2051 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2052 		return rte_flow_error_set(error, EINVAL,
2053 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2054 					  "protocol filtering not compatible"
2055 					  " with ICMP layer");
2056 	if (!(item_flags & l3m))
2057 		return rte_flow_error_set(error, EINVAL,
2058 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2059 					  "IPv4 is mandatory to filter"
2060 					  " on ICMP");
2061 	if (item_flags & l4m)
2062 		return rte_flow_error_set(error, EINVAL,
2063 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2064 					  "multiple L4 layers not supported");
2065 	if (!mask)
2066 		mask = &nic_mask;
2067 	ret = mlx5_flow_item_acceptable
2068 		(item, (const uint8_t *)mask,
2069 		 (const uint8_t *)&nic_mask,
2070 		 sizeof(struct rte_flow_item_icmp),
2071 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2072 	if (ret < 0)
2073 		return ret;
2074 	return 0;
2075 }
2076 
2077 /**
2078  * Validate Ethernet item.
2079  *
2080  * @param[in] item
2081  *   Item specification.
2082  * @param[in] item_flags
2083  *   Bit-fields that holds the items detected until now.
2084  * @param[out] error
2085  *   Pointer to error structure.
2086  *
2087  * @return
2088  *   0 on success, a negative errno value otherwise and rte_errno is set.
2089  */
2090 int
2091 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2092 			    uint64_t item_flags, bool ext_vlan_sup,
2093 			    struct rte_flow_error *error)
2094 {
2095 	const struct rte_flow_item_eth *mask = item->mask;
2096 	const struct rte_flow_item_eth nic_mask = {
2097 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2098 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2099 		.type = RTE_BE16(0xffff),
2100 		.has_vlan = ext_vlan_sup ? 1 : 0,
2101 	};
2102 	int ret;
2103 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2104 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2105 				       MLX5_FLOW_LAYER_OUTER_L2;
2106 
2107 	if (item_flags & ethm)
2108 		return rte_flow_error_set(error, ENOTSUP,
2109 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2110 					  "multiple L2 layers not supported");
2111 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2112 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2113 		return rte_flow_error_set(error, EINVAL,
2114 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2115 					  "L2 layer should not follow "
2116 					  "L3 layers");
2117 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2118 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2119 		return rte_flow_error_set(error, EINVAL,
2120 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2121 					  "L2 layer should not follow VLAN");
2122 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2123 		return rte_flow_error_set(error, EINVAL,
2124 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2125 					  "L2 layer should not follow GTP");
2126 	if (!mask)
2127 		mask = &rte_flow_item_eth_mask;
2128 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2129 					(const uint8_t *)&nic_mask,
2130 					sizeof(struct rte_flow_item_eth),
2131 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2132 	return ret;
2133 }
2134 
2135 /**
2136  * Validate VLAN item.
2137  *
2138  * @param[in] item
2139  *   Item specification.
2140  * @param[in] item_flags
2141  *   Bit-fields that holds the items detected until now.
2142  * @param[in] dev
2143  *   Ethernet device flow is being created on.
2144  * @param[out] error
2145  *   Pointer to error structure.
2146  *
2147  * @return
2148  *   0 on success, a negative errno value otherwise and rte_errno is set.
2149  */
2150 int
2151 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2152 			     uint64_t item_flags,
2153 			     struct rte_eth_dev *dev,
2154 			     struct rte_flow_error *error)
2155 {
2156 	const struct rte_flow_item_vlan *spec = item->spec;
2157 	const struct rte_flow_item_vlan *mask = item->mask;
2158 	const struct rte_flow_item_vlan nic_mask = {
2159 		.tci = RTE_BE16(UINT16_MAX),
2160 		.inner_type = RTE_BE16(UINT16_MAX),
2161 	};
2162 	uint16_t vlan_tag = 0;
2163 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2164 	int ret;
2165 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2166 					MLX5_FLOW_LAYER_INNER_L4) :
2167 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2168 					MLX5_FLOW_LAYER_OUTER_L4);
2169 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2170 					MLX5_FLOW_LAYER_OUTER_VLAN;
2171 
2172 	if (item_flags & vlanm)
2173 		return rte_flow_error_set(error, EINVAL,
2174 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2175 					  "multiple VLAN layers not supported");
2176 	else if ((item_flags & l34m) != 0)
2177 		return rte_flow_error_set(error, EINVAL,
2178 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2179 					  "VLAN cannot follow L3/L4 layer");
2180 	if (!mask)
2181 		mask = &rte_flow_item_vlan_mask;
2182 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2183 					(const uint8_t *)&nic_mask,
2184 					sizeof(struct rte_flow_item_vlan),
2185 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2186 	if (ret)
2187 		return ret;
2188 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2189 		struct mlx5_priv *priv = dev->data->dev_private;
2190 
2191 		if (priv->vmwa_context) {
2192 			/*
2193 			 * Non-NULL context means we have a virtual machine
2194 			 * and SR-IOV enabled, we have to create VLAN interface
2195 			 * to make hypervisor to setup E-Switch vport
2196 			 * context correctly. We avoid creating the multiple
2197 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2198 			 */
2199 			return rte_flow_error_set(error, EINVAL,
2200 						  RTE_FLOW_ERROR_TYPE_ITEM,
2201 						  item,
2202 						  "VLAN tag mask is not"
2203 						  " supported in virtual"
2204 						  " environment");
2205 		}
2206 	}
2207 	if (spec) {
2208 		vlan_tag = spec->tci;
2209 		vlan_tag &= mask->tci;
2210 	}
2211 	/*
2212 	 * From verbs perspective an empty VLAN is equivalent
2213 	 * to a packet without VLAN layer.
2214 	 */
2215 	if (!vlan_tag)
2216 		return rte_flow_error_set(error, EINVAL,
2217 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2218 					  item->spec,
2219 					  "VLAN cannot be empty");
2220 	return 0;
2221 }
2222 
2223 /**
2224  * Validate IPV4 item.
2225  *
2226  * @param[in] item
2227  *   Item specification.
2228  * @param[in] item_flags
2229  *   Bit-fields that holds the items detected until now.
2230  * @param[in] last_item
2231  *   Previous validated item in the pattern items.
2232  * @param[in] ether_type
2233  *   Type in the ethernet layer header (including dot1q).
2234  * @param[in] acc_mask
2235  *   Acceptable mask, if NULL default internal default mask
2236  *   will be used to check whether item fields are supported.
2237  * @param[in] range_accepted
2238  *   True if range of values is accepted for specific fields, false otherwise.
2239  * @param[out] error
2240  *   Pointer to error structure.
2241  *
2242  * @return
2243  *   0 on success, a negative errno value otherwise and rte_errno is set.
2244  */
2245 int
2246 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2247 			     uint64_t item_flags,
2248 			     uint64_t last_item,
2249 			     uint16_t ether_type,
2250 			     const struct rte_flow_item_ipv4 *acc_mask,
2251 			     bool range_accepted,
2252 			     struct rte_flow_error *error)
2253 {
2254 	const struct rte_flow_item_ipv4 *mask = item->mask;
2255 	const struct rte_flow_item_ipv4 *spec = item->spec;
2256 	const struct rte_flow_item_ipv4 nic_mask = {
2257 		.hdr = {
2258 			.src_addr = RTE_BE32(0xffffffff),
2259 			.dst_addr = RTE_BE32(0xffffffff),
2260 			.type_of_service = 0xff,
2261 			.next_proto_id = 0xff,
2262 		},
2263 	};
2264 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2265 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2266 				      MLX5_FLOW_LAYER_OUTER_L3;
2267 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2268 				      MLX5_FLOW_LAYER_OUTER_L4;
2269 	int ret;
2270 	uint8_t next_proto = 0xFF;
2271 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2272 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2273 				  MLX5_FLOW_LAYER_INNER_VLAN);
2274 
2275 	if ((last_item & l2_vlan) && ether_type &&
2276 	    ether_type != RTE_ETHER_TYPE_IPV4)
2277 		return rte_flow_error_set(error, EINVAL,
2278 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2279 					  "IPv4 cannot follow L2/VLAN layer "
2280 					  "which ether type is not IPv4");
2281 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2282 		if (mask && spec)
2283 			next_proto = mask->hdr.next_proto_id &
2284 				     spec->hdr.next_proto_id;
2285 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2286 			return rte_flow_error_set(error, EINVAL,
2287 						  RTE_FLOW_ERROR_TYPE_ITEM,
2288 						  item,
2289 						  "multiple tunnel "
2290 						  "not supported");
2291 	}
2292 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2293 		return rte_flow_error_set(error, EINVAL,
2294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 					  "wrong tunnel type - IPv6 specified "
2296 					  "but IPv4 item provided");
2297 	if (item_flags & l3m)
2298 		return rte_flow_error_set(error, ENOTSUP,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "multiple L3 layers not supported");
2301 	else if (item_flags & l4m)
2302 		return rte_flow_error_set(error, EINVAL,
2303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2304 					  "L3 cannot follow an L4 layer.");
2305 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2306 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2307 		return rte_flow_error_set(error, EINVAL,
2308 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2309 					  "L3 cannot follow an NVGRE layer.");
2310 	if (!mask)
2311 		mask = &rte_flow_item_ipv4_mask;
2312 	else if (mask->hdr.next_proto_id != 0 &&
2313 		 mask->hdr.next_proto_id != 0xff)
2314 		return rte_flow_error_set(error, EINVAL,
2315 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2316 					  "partial mask is not supported"
2317 					  " for protocol");
2318 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2319 					acc_mask ? (const uint8_t *)acc_mask
2320 						 : (const uint8_t *)&nic_mask,
2321 					sizeof(struct rte_flow_item_ipv4),
2322 					range_accepted, error);
2323 	if (ret < 0)
2324 		return ret;
2325 	return 0;
2326 }
2327 
2328 /**
2329  * Validate IPV6 item.
2330  *
2331  * @param[in] item
2332  *   Item specification.
2333  * @param[in] item_flags
2334  *   Bit-fields that holds the items detected until now.
2335  * @param[in] last_item
2336  *   Previous validated item in the pattern items.
2337  * @param[in] ether_type
2338  *   Type in the ethernet layer header (including dot1q).
2339  * @param[in] acc_mask
2340  *   Acceptable mask, if NULL default internal default mask
2341  *   will be used to check whether item fields are supported.
2342  * @param[out] error
2343  *   Pointer to error structure.
2344  *
2345  * @return
2346  *   0 on success, a negative errno value otherwise and rte_errno is set.
2347  */
2348 int
2349 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2350 			     uint64_t item_flags,
2351 			     uint64_t last_item,
2352 			     uint16_t ether_type,
2353 			     const struct rte_flow_item_ipv6 *acc_mask,
2354 			     struct rte_flow_error *error)
2355 {
2356 	const struct rte_flow_item_ipv6 *mask = item->mask;
2357 	const struct rte_flow_item_ipv6 *spec = item->spec;
2358 	const struct rte_flow_item_ipv6 nic_mask = {
2359 		.hdr = {
2360 			.src_addr =
2361 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2362 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2363 			.dst_addr =
2364 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2365 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2366 			.vtc_flow = RTE_BE32(0xffffffff),
2367 			.proto = 0xff,
2368 		},
2369 	};
2370 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2371 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2372 				      MLX5_FLOW_LAYER_OUTER_L3;
2373 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2374 				      MLX5_FLOW_LAYER_OUTER_L4;
2375 	int ret;
2376 	uint8_t next_proto = 0xFF;
2377 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2378 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2379 				  MLX5_FLOW_LAYER_INNER_VLAN);
2380 
2381 	if ((last_item & l2_vlan) && ether_type &&
2382 	    ether_type != RTE_ETHER_TYPE_IPV6)
2383 		return rte_flow_error_set(error, EINVAL,
2384 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2385 					  "IPv6 cannot follow L2/VLAN layer "
2386 					  "which ether type is not IPv6");
2387 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2388 		next_proto = spec->hdr.proto;
2389 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2390 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2391 			return rte_flow_error_set(error, EINVAL,
2392 						  RTE_FLOW_ERROR_TYPE_ITEM,
2393 						  item,
2394 						  "multiple tunnel "
2395 						  "not supported");
2396 	}
2397 	if (next_proto == IPPROTO_HOPOPTS  ||
2398 	    next_proto == IPPROTO_ROUTING  ||
2399 	    next_proto == IPPROTO_FRAGMENT ||
2400 	    next_proto == IPPROTO_ESP	   ||
2401 	    next_proto == IPPROTO_AH	   ||
2402 	    next_proto == IPPROTO_DSTOPTS)
2403 		return rte_flow_error_set(error, EINVAL,
2404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2405 					  "IPv6 proto (next header) should "
2406 					  "not be set as extension header");
2407 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2408 		return rte_flow_error_set(error, EINVAL,
2409 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2410 					  "wrong tunnel type - IPv4 specified "
2411 					  "but IPv6 item provided");
2412 	if (item_flags & l3m)
2413 		return rte_flow_error_set(error, ENOTSUP,
2414 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2415 					  "multiple L3 layers not supported");
2416 	else if (item_flags & l4m)
2417 		return rte_flow_error_set(error, EINVAL,
2418 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 					  "L3 cannot follow an L4 layer.");
2420 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2421 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2422 		return rte_flow_error_set(error, EINVAL,
2423 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2424 					  "L3 cannot follow an NVGRE layer.");
2425 	if (!mask)
2426 		mask = &rte_flow_item_ipv6_mask;
2427 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2428 					acc_mask ? (const uint8_t *)acc_mask
2429 						 : (const uint8_t *)&nic_mask,
2430 					sizeof(struct rte_flow_item_ipv6),
2431 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2432 	if (ret < 0)
2433 		return ret;
2434 	return 0;
2435 }
2436 
2437 /**
2438  * Validate UDP item.
2439  *
2440  * @param[in] item
2441  *   Item specification.
2442  * @param[in] item_flags
2443  *   Bit-fields that holds the items detected until now.
2444  * @param[in] target_protocol
2445  *   The next protocol in the previous item.
2446  * @param[in] flow_mask
2447  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2448  * @param[out] error
2449  *   Pointer to error structure.
2450  *
2451  * @return
2452  *   0 on success, a negative errno value otherwise and rte_errno is set.
2453  */
2454 int
2455 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2456 			    uint64_t item_flags,
2457 			    uint8_t target_protocol,
2458 			    struct rte_flow_error *error)
2459 {
2460 	const struct rte_flow_item_udp *mask = item->mask;
2461 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2462 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2463 				      MLX5_FLOW_LAYER_OUTER_L3;
2464 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2465 				      MLX5_FLOW_LAYER_OUTER_L4;
2466 	int ret;
2467 
2468 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2469 		return rte_flow_error_set(error, EINVAL,
2470 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 					  "protocol filtering not compatible"
2472 					  " with UDP layer");
2473 	if (!(item_flags & l3m))
2474 		return rte_flow_error_set(error, EINVAL,
2475 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2476 					  "L3 is mandatory to filter on L4");
2477 	if (item_flags & l4m)
2478 		return rte_flow_error_set(error, EINVAL,
2479 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2480 					  "multiple L4 layers not supported");
2481 	if (!mask)
2482 		mask = &rte_flow_item_udp_mask;
2483 	ret = mlx5_flow_item_acceptable
2484 		(item, (const uint8_t *)mask,
2485 		 (const uint8_t *)&rte_flow_item_udp_mask,
2486 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2487 		 error);
2488 	if (ret < 0)
2489 		return ret;
2490 	return 0;
2491 }
2492 
2493 /**
2494  * Validate TCP item.
2495  *
2496  * @param[in] item
2497  *   Item specification.
2498  * @param[in] item_flags
2499  *   Bit-fields that holds the items detected until now.
2500  * @param[in] target_protocol
2501  *   The next protocol in the previous item.
2502  * @param[out] error
2503  *   Pointer to error structure.
2504  *
2505  * @return
2506  *   0 on success, a negative errno value otherwise and rte_errno is set.
2507  */
2508 int
2509 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2510 			    uint64_t item_flags,
2511 			    uint8_t target_protocol,
2512 			    const struct rte_flow_item_tcp *flow_mask,
2513 			    struct rte_flow_error *error)
2514 {
2515 	const struct rte_flow_item_tcp *mask = item->mask;
2516 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2517 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2518 				      MLX5_FLOW_LAYER_OUTER_L3;
2519 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2520 				      MLX5_FLOW_LAYER_OUTER_L4;
2521 	int ret;
2522 
2523 	MLX5_ASSERT(flow_mask);
2524 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2525 		return rte_flow_error_set(error, EINVAL,
2526 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2527 					  "protocol filtering not compatible"
2528 					  " with TCP layer");
2529 	if (!(item_flags & l3m))
2530 		return rte_flow_error_set(error, EINVAL,
2531 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2532 					  "L3 is mandatory to filter on L4");
2533 	if (item_flags & l4m)
2534 		return rte_flow_error_set(error, EINVAL,
2535 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2536 					  "multiple L4 layers not supported");
2537 	if (!mask)
2538 		mask = &rte_flow_item_tcp_mask;
2539 	ret = mlx5_flow_item_acceptable
2540 		(item, (const uint8_t *)mask,
2541 		 (const uint8_t *)flow_mask,
2542 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2543 		 error);
2544 	if (ret < 0)
2545 		return ret;
2546 	return 0;
2547 }
2548 
2549 /**
2550  * Validate VXLAN item.
2551  *
2552  * @param[in] dev
2553  *   Pointer to the Ethernet device structure.
2554  * @param[in] udp_dport
2555  *   UDP destination port
2556  * @param[in] item
2557  *   Item specification.
2558  * @param[in] item_flags
2559  *   Bit-fields that holds the items detected until now.
2560  * @param[in] attr
2561  *   Flow rule attributes.
2562  * @param[out] error
2563  *   Pointer to error structure.
2564  *
2565  * @return
2566  *   0 on success, a negative errno value otherwise and rte_errno is set.
2567  */
2568 int
2569 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2570 			      uint16_t udp_dport,
2571 			      const struct rte_flow_item *item,
2572 			      uint64_t item_flags,
2573 			      const struct rte_flow_attr *attr,
2574 			      struct rte_flow_error *error)
2575 {
2576 	const struct rte_flow_item_vxlan *spec = item->spec;
2577 	const struct rte_flow_item_vxlan *mask = item->mask;
2578 	int ret;
2579 	struct mlx5_priv *priv = dev->data->dev_private;
2580 	union vni {
2581 		uint32_t vlan_id;
2582 		uint8_t vni[4];
2583 	} id = { .vlan_id = 0, };
2584 	const struct rte_flow_item_vxlan nic_mask = {
2585 		.vni = "\xff\xff\xff",
2586 		.rsvd1 = 0xff,
2587 	};
2588 	const struct rte_flow_item_vxlan *valid_mask;
2589 
2590 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2591 		return rte_flow_error_set(error, ENOTSUP,
2592 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2593 					  "multiple tunnel layers not"
2594 					  " supported");
2595 	valid_mask = &rte_flow_item_vxlan_mask;
2596 	/*
2597 	 * Verify only UDPv4 is present as defined in
2598 	 * https://tools.ietf.org/html/rfc7348
2599 	 */
2600 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2601 		return rte_flow_error_set(error, EINVAL,
2602 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2603 					  "no outer UDP layer found");
2604 	if (!mask)
2605 		mask = &rte_flow_item_vxlan_mask;
2606 
2607 	if (priv->sh->steering_format_version !=
2608 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2609 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2610 		/* FDB domain & NIC domain non-zero group */
2611 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2612 			valid_mask = &nic_mask;
2613 		/* Group zero in NIC domain */
2614 		if (!attr->group && !attr->transfer &&
2615 		    priv->sh->tunnel_header_0_1)
2616 			valid_mask = &nic_mask;
2617 	}
2618 	ret = mlx5_flow_item_acceptable
2619 		(item, (const uint8_t *)mask,
2620 		 (const uint8_t *)valid_mask,
2621 		 sizeof(struct rte_flow_item_vxlan),
2622 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2623 	if (ret < 0)
2624 		return ret;
2625 	if (spec) {
2626 		memcpy(&id.vni[1], spec->vni, 3);
2627 		memcpy(&id.vni[1], mask->vni, 3);
2628 	}
2629 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2630 		return rte_flow_error_set(error, ENOTSUP,
2631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2632 					  "VXLAN tunnel must be fully defined");
2633 	return 0;
2634 }
2635 
2636 /**
2637  * Validate VXLAN_GPE item.
2638  *
2639  * @param[in] item
2640  *   Item specification.
2641  * @param[in] item_flags
2642  *   Bit-fields that holds the items detected until now.
2643  * @param[in] priv
2644  *   Pointer to the private data structure.
2645  * @param[in] target_protocol
2646  *   The next protocol in the previous item.
2647  * @param[out] error
2648  *   Pointer to error structure.
2649  *
2650  * @return
2651  *   0 on success, a negative errno value otherwise and rte_errno is set.
2652  */
2653 int
2654 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2655 				  uint64_t item_flags,
2656 				  struct rte_eth_dev *dev,
2657 				  struct rte_flow_error *error)
2658 {
2659 	struct mlx5_priv *priv = dev->data->dev_private;
2660 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2661 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2662 	int ret;
2663 	union vni {
2664 		uint32_t vlan_id;
2665 		uint8_t vni[4];
2666 	} id = { .vlan_id = 0, };
2667 
2668 	if (!priv->config.l3_vxlan_en)
2669 		return rte_flow_error_set(error, ENOTSUP,
2670 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2671 					  "L3 VXLAN is not enabled by device"
2672 					  " parameter and/or not configured in"
2673 					  " firmware");
2674 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2675 		return rte_flow_error_set(error, ENOTSUP,
2676 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2677 					  "multiple tunnel layers not"
2678 					  " supported");
2679 	/*
2680 	 * Verify only UDPv4 is present as defined in
2681 	 * https://tools.ietf.org/html/rfc7348
2682 	 */
2683 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2684 		return rte_flow_error_set(error, EINVAL,
2685 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2686 					  "no outer UDP layer found");
2687 	if (!mask)
2688 		mask = &rte_flow_item_vxlan_gpe_mask;
2689 	ret = mlx5_flow_item_acceptable
2690 		(item, (const uint8_t *)mask,
2691 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2692 		 sizeof(struct rte_flow_item_vxlan_gpe),
2693 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2694 	if (ret < 0)
2695 		return ret;
2696 	if (spec) {
2697 		if (spec->protocol)
2698 			return rte_flow_error_set(error, ENOTSUP,
2699 						  RTE_FLOW_ERROR_TYPE_ITEM,
2700 						  item,
2701 						  "VxLAN-GPE protocol"
2702 						  " not supported");
2703 		memcpy(&id.vni[1], spec->vni, 3);
2704 		memcpy(&id.vni[1], mask->vni, 3);
2705 	}
2706 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2707 		return rte_flow_error_set(error, ENOTSUP,
2708 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2709 					  "VXLAN-GPE tunnel must be fully"
2710 					  " defined");
2711 	return 0;
2712 }
2713 /**
2714  * Validate GRE Key item.
2715  *
2716  * @param[in] item
2717  *   Item specification.
2718  * @param[in] item_flags
2719  *   Bit flags to mark detected items.
2720  * @param[in] gre_item
2721  *   Pointer to gre_item
2722  * @param[out] error
2723  *   Pointer to error structure.
2724  *
2725  * @return
2726  *   0 on success, a negative errno value otherwise and rte_errno is set.
2727  */
2728 int
2729 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2730 				uint64_t item_flags,
2731 				const struct rte_flow_item *gre_item,
2732 				struct rte_flow_error *error)
2733 {
2734 	const rte_be32_t *mask = item->mask;
2735 	int ret = 0;
2736 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2737 	const struct rte_flow_item_gre *gre_spec;
2738 	const struct rte_flow_item_gre *gre_mask;
2739 
2740 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2741 		return rte_flow_error_set(error, ENOTSUP,
2742 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2743 					  "Multiple GRE key not support");
2744 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2745 		return rte_flow_error_set(error, ENOTSUP,
2746 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2747 					  "No preceding GRE header");
2748 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2749 		return rte_flow_error_set(error, ENOTSUP,
2750 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2751 					  "GRE key following a wrong item");
2752 	gre_mask = gre_item->mask;
2753 	if (!gre_mask)
2754 		gre_mask = &rte_flow_item_gre_mask;
2755 	gre_spec = gre_item->spec;
2756 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2757 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2758 		return rte_flow_error_set(error, EINVAL,
2759 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2760 					  "Key bit must be on");
2761 
2762 	if (!mask)
2763 		mask = &gre_key_default_mask;
2764 	ret = mlx5_flow_item_acceptable
2765 		(item, (const uint8_t *)mask,
2766 		 (const uint8_t *)&gre_key_default_mask,
2767 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2768 	return ret;
2769 }
2770 
2771 /**
2772  * Validate GRE item.
2773  *
2774  * @param[in] item
2775  *   Item specification.
2776  * @param[in] item_flags
2777  *   Bit flags to mark detected items.
2778  * @param[in] target_protocol
2779  *   The next protocol in the previous item.
2780  * @param[out] error
2781  *   Pointer to error structure.
2782  *
2783  * @return
2784  *   0 on success, a negative errno value otherwise and rte_errno is set.
2785  */
2786 int
2787 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2788 			    uint64_t item_flags,
2789 			    uint8_t target_protocol,
2790 			    struct rte_flow_error *error)
2791 {
2792 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2793 	const struct rte_flow_item_gre *mask = item->mask;
2794 	int ret;
2795 	const struct rte_flow_item_gre nic_mask = {
2796 		.c_rsvd0_ver = RTE_BE16(0xB000),
2797 		.protocol = RTE_BE16(UINT16_MAX),
2798 	};
2799 
2800 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2801 		return rte_flow_error_set(error, EINVAL,
2802 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2803 					  "protocol filtering not compatible"
2804 					  " with this GRE layer");
2805 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2806 		return rte_flow_error_set(error, ENOTSUP,
2807 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2808 					  "multiple tunnel layers not"
2809 					  " supported");
2810 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2811 		return rte_flow_error_set(error, ENOTSUP,
2812 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2813 					  "L3 Layer is missing");
2814 	if (!mask)
2815 		mask = &rte_flow_item_gre_mask;
2816 	ret = mlx5_flow_item_acceptable
2817 		(item, (const uint8_t *)mask,
2818 		 (const uint8_t *)&nic_mask,
2819 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2820 		 error);
2821 	if (ret < 0)
2822 		return ret;
2823 #ifndef HAVE_MLX5DV_DR
2824 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2825 	if (spec && (spec->protocol & mask->protocol))
2826 		return rte_flow_error_set(error, ENOTSUP,
2827 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2828 					  "without MPLS support the"
2829 					  " specification cannot be used for"
2830 					  " filtering");
2831 #endif
2832 #endif
2833 	return 0;
2834 }
2835 
2836 /**
2837  * Validate Geneve item.
2838  *
2839  * @param[in] item
2840  *   Item specification.
2841  * @param[in] itemFlags
2842  *   Bit-fields that holds the items detected until now.
2843  * @param[in] enPriv
2844  *   Pointer to the private data structure.
2845  * @param[out] error
2846  *   Pointer to error structure.
2847  *
2848  * @return
2849  *   0 on success, a negative errno value otherwise and rte_errno is set.
2850  */
2851 
2852 int
2853 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2854 			       uint64_t item_flags,
2855 			       struct rte_eth_dev *dev,
2856 			       struct rte_flow_error *error)
2857 {
2858 	struct mlx5_priv *priv = dev->data->dev_private;
2859 	const struct rte_flow_item_geneve *spec = item->spec;
2860 	const struct rte_flow_item_geneve *mask = item->mask;
2861 	int ret;
2862 	uint16_t gbhdr;
2863 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2864 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2865 	const struct rte_flow_item_geneve nic_mask = {
2866 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2867 		.vni = "\xff\xff\xff",
2868 		.protocol = RTE_BE16(UINT16_MAX),
2869 	};
2870 
2871 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2872 		return rte_flow_error_set(error, ENOTSUP,
2873 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 					  "L3 Geneve is not enabled by device"
2875 					  " parameter and/or not configured in"
2876 					  " firmware");
2877 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2878 		return rte_flow_error_set(error, ENOTSUP,
2879 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2880 					  "multiple tunnel layers not"
2881 					  " supported");
2882 	/*
2883 	 * Verify only UDPv4 is present as defined in
2884 	 * https://tools.ietf.org/html/rfc7348
2885 	 */
2886 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2887 		return rte_flow_error_set(error, EINVAL,
2888 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2889 					  "no outer UDP layer found");
2890 	if (!mask)
2891 		mask = &rte_flow_item_geneve_mask;
2892 	ret = mlx5_flow_item_acceptable
2893 				  (item, (const uint8_t *)mask,
2894 				   (const uint8_t *)&nic_mask,
2895 				   sizeof(struct rte_flow_item_geneve),
2896 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2897 	if (ret)
2898 		return ret;
2899 	if (spec) {
2900 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2901 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2902 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2903 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2904 			return rte_flow_error_set(error, ENOTSUP,
2905 						  RTE_FLOW_ERROR_TYPE_ITEM,
2906 						  item,
2907 						  "Geneve protocol unsupported"
2908 						  " fields are being used");
2909 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2910 			return rte_flow_error_set
2911 					(error, ENOTSUP,
2912 					 RTE_FLOW_ERROR_TYPE_ITEM,
2913 					 item,
2914 					 "Unsupported Geneve options length");
2915 	}
2916 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2917 		return rte_flow_error_set
2918 				    (error, ENOTSUP,
2919 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2920 				     "Geneve tunnel must be fully defined");
2921 	return 0;
2922 }
2923 
2924 /**
2925  * Validate Geneve TLV option item.
2926  *
2927  * @param[in] item
2928  *   Item specification.
2929  * @param[in] last_item
2930  *   Previous validated item in the pattern items.
2931  * @param[in] geneve_item
2932  *   Previous GENEVE item specification.
2933  * @param[in] dev
2934  *   Pointer to the rte_eth_dev structure.
2935  * @param[out] error
2936  *   Pointer to error structure.
2937  *
2938  * @return
2939  *   0 on success, a negative errno value otherwise and rte_errno is set.
2940  */
2941 int
2942 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2943 				   uint64_t last_item,
2944 				   const struct rte_flow_item *geneve_item,
2945 				   struct rte_eth_dev *dev,
2946 				   struct rte_flow_error *error)
2947 {
2948 	struct mlx5_priv *priv = dev->data->dev_private;
2949 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2950 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2951 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2952 	uint8_t data_max_supported =
2953 			hca_attr->max_geneve_tlv_option_data_len * 4;
2954 	struct mlx5_dev_config *config = &priv->config;
2955 	const struct rte_flow_item_geneve *geneve_spec;
2956 	const struct rte_flow_item_geneve *geneve_mask;
2957 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2958 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2959 	unsigned int i;
2960 	unsigned int data_len;
2961 	uint8_t tlv_option_len;
2962 	uint16_t optlen_m, optlen_v;
2963 	const struct rte_flow_item_geneve_opt full_mask = {
2964 		.option_class = RTE_BE16(0xffff),
2965 		.option_type = 0xff,
2966 		.option_len = 0x1f,
2967 	};
2968 
2969 	if (!mask)
2970 		mask = &rte_flow_item_geneve_opt_mask;
2971 	if (!spec)
2972 		return rte_flow_error_set
2973 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2974 			"Geneve TLV opt class/type/length must be specified");
2975 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2976 		return rte_flow_error_set
2977 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2978 			"Geneve TLV opt length exceeeds the limit (31)");
2979 	/* Check if class type and length masks are full. */
2980 	if (full_mask.option_class != mask->option_class ||
2981 	    full_mask.option_type != mask->option_type ||
2982 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2983 		return rte_flow_error_set
2984 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2985 			"Geneve TLV opt class/type/length masks must be full");
2986 	/* Check if length is supported */
2987 	if ((uint32_t)spec->option_len >
2988 			config->hca_attr.max_geneve_tlv_option_data_len)
2989 		return rte_flow_error_set
2990 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2991 			"Geneve TLV opt length not supported");
2992 	if (config->hca_attr.max_geneve_tlv_options > 1)
2993 		DRV_LOG(DEBUG,
2994 			"max_geneve_tlv_options supports more than 1 option");
2995 	/* Check GENEVE item preceding. */
2996 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2997 		return rte_flow_error_set
2998 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2999 			"Geneve opt item must be preceded with Geneve item");
3000 	geneve_spec = geneve_item->spec;
3001 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3002 					  &rte_flow_item_geneve_mask;
3003 	/* Check if GENEVE TLV option size doesn't exceed option length */
3004 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3005 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3006 		tlv_option_len = spec->option_len & mask->option_len;
3007 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3008 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3009 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3010 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3011 		if ((optlen_v & optlen_m) <= tlv_option_len)
3012 			return rte_flow_error_set
3013 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3014 				 "GENEVE TLV option length exceeds optlen");
3015 	}
3016 	/* Check if length is 0 or data is 0. */
3017 	if (spec->data == NULL || spec->option_len == 0)
3018 		return rte_flow_error_set
3019 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3020 			"Geneve TLV opt with zero data/length not supported");
3021 	/* Check not all data & mask are 0. */
3022 	data_len = spec->option_len * 4;
3023 	if (mask->data == NULL) {
3024 		for (i = 0; i < data_len; i++)
3025 			if (spec->data[i])
3026 				break;
3027 		if (i == data_len)
3028 			return rte_flow_error_set(error, ENOTSUP,
3029 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3030 				"Can't match on Geneve option data 0");
3031 	} else {
3032 		for (i = 0; i < data_len; i++)
3033 			if (spec->data[i] & mask->data[i])
3034 				break;
3035 		if (i == data_len)
3036 			return rte_flow_error_set(error, ENOTSUP,
3037 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3038 				"Can't match on Geneve option data and mask 0");
3039 		/* Check data mask supported. */
3040 		for (i = data_max_supported; i < data_len ; i++)
3041 			if (mask->data[i])
3042 				return rte_flow_error_set(error, ENOTSUP,
3043 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3044 					"Data mask is of unsupported size");
3045 	}
3046 	/* Check GENEVE option is supported in NIC. */
3047 	if (!config->hca_attr.geneve_tlv_opt)
3048 		return rte_flow_error_set
3049 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3050 			"Geneve TLV opt not supported");
3051 	/* Check if we already have geneve option with different type/class. */
3052 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3053 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3054 	if (geneve_opt_resource != NULL)
3055 		if (geneve_opt_resource->option_class != spec->option_class ||
3056 		    geneve_opt_resource->option_type != spec->option_type ||
3057 		    geneve_opt_resource->length != spec->option_len) {
3058 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3059 			return rte_flow_error_set(error, ENOTSUP,
3060 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3061 				"Only one Geneve TLV option supported");
3062 		}
3063 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3064 	return 0;
3065 }
3066 
3067 /**
3068  * Validate MPLS item.
3069  *
3070  * @param[in] dev
3071  *   Pointer to the rte_eth_dev structure.
3072  * @param[in] item
3073  *   Item specification.
3074  * @param[in] item_flags
3075  *   Bit-fields that holds the items detected until now.
3076  * @param[in] prev_layer
3077  *   The protocol layer indicated in previous item.
3078  * @param[out] error
3079  *   Pointer to error structure.
3080  *
3081  * @return
3082  *   0 on success, a negative errno value otherwise and rte_errno is set.
3083  */
3084 int
3085 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3086 			     const struct rte_flow_item *item __rte_unused,
3087 			     uint64_t item_flags __rte_unused,
3088 			     uint64_t prev_layer __rte_unused,
3089 			     struct rte_flow_error *error)
3090 {
3091 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3092 	const struct rte_flow_item_mpls *mask = item->mask;
3093 	struct mlx5_priv *priv = dev->data->dev_private;
3094 	int ret;
3095 
3096 	if (!priv->config.mpls_en)
3097 		return rte_flow_error_set(error, ENOTSUP,
3098 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3099 					  "MPLS not supported or"
3100 					  " disabled in firmware"
3101 					  " configuration.");
3102 	/* MPLS over UDP, GRE is allowed */
3103 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3104 			    MLX5_FLOW_LAYER_GRE |
3105 			    MLX5_FLOW_LAYER_GRE_KEY)))
3106 		return rte_flow_error_set(error, EINVAL,
3107 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3108 					  "protocol filtering not compatible"
3109 					  " with MPLS layer");
3110 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3111 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3112 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3113 		return rte_flow_error_set(error, ENOTSUP,
3114 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3115 					  "multiple tunnel layers not"
3116 					  " supported");
3117 	if (!mask)
3118 		mask = &rte_flow_item_mpls_mask;
3119 	ret = mlx5_flow_item_acceptable
3120 		(item, (const uint8_t *)mask,
3121 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3122 		 sizeof(struct rte_flow_item_mpls),
3123 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3124 	if (ret < 0)
3125 		return ret;
3126 	return 0;
3127 #else
3128 	return rte_flow_error_set(error, ENOTSUP,
3129 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3130 				  "MPLS is not supported by Verbs, please"
3131 				  " update.");
3132 #endif
3133 }
3134 
3135 /**
3136  * Validate NVGRE item.
3137  *
3138  * @param[in] item
3139  *   Item specification.
3140  * @param[in] item_flags
3141  *   Bit flags to mark detected items.
3142  * @param[in] target_protocol
3143  *   The next protocol in the previous item.
3144  * @param[out] error
3145  *   Pointer to error structure.
3146  *
3147  * @return
3148  *   0 on success, a negative errno value otherwise and rte_errno is set.
3149  */
3150 int
3151 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3152 			      uint64_t item_flags,
3153 			      uint8_t target_protocol,
3154 			      struct rte_flow_error *error)
3155 {
3156 	const struct rte_flow_item_nvgre *mask = item->mask;
3157 	int ret;
3158 
3159 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3160 		return rte_flow_error_set(error, EINVAL,
3161 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3162 					  "protocol filtering not compatible"
3163 					  " with this GRE layer");
3164 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3165 		return rte_flow_error_set(error, ENOTSUP,
3166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3167 					  "multiple tunnel layers not"
3168 					  " supported");
3169 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3170 		return rte_flow_error_set(error, ENOTSUP,
3171 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3172 					  "L3 Layer is missing");
3173 	if (!mask)
3174 		mask = &rte_flow_item_nvgre_mask;
3175 	ret = mlx5_flow_item_acceptable
3176 		(item, (const uint8_t *)mask,
3177 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3178 		 sizeof(struct rte_flow_item_nvgre),
3179 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3180 	if (ret < 0)
3181 		return ret;
3182 	return 0;
3183 }
3184 
3185 /**
3186  * Validate eCPRI item.
3187  *
3188  * @param[in] item
3189  *   Item specification.
3190  * @param[in] item_flags
3191  *   Bit-fields that holds the items detected until now.
3192  * @param[in] last_item
3193  *   Previous validated item in the pattern items.
3194  * @param[in] ether_type
3195  *   Type in the ethernet layer header (including dot1q).
3196  * @param[in] acc_mask
3197  *   Acceptable mask, if NULL default internal default mask
3198  *   will be used to check whether item fields are supported.
3199  * @param[out] error
3200  *   Pointer to error structure.
3201  *
3202  * @return
3203  *   0 on success, a negative errno value otherwise and rte_errno is set.
3204  */
3205 int
3206 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3207 			      uint64_t item_flags,
3208 			      uint64_t last_item,
3209 			      uint16_t ether_type,
3210 			      const struct rte_flow_item_ecpri *acc_mask,
3211 			      struct rte_flow_error *error)
3212 {
3213 	const struct rte_flow_item_ecpri *mask = item->mask;
3214 	const struct rte_flow_item_ecpri nic_mask = {
3215 		.hdr = {
3216 			.common = {
3217 				.u32 =
3218 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3219 					.type = 0xFF,
3220 					}).u32),
3221 			},
3222 			.dummy[0] = 0xFFFFFFFF,
3223 		},
3224 	};
3225 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3226 					MLX5_FLOW_LAYER_OUTER_VLAN);
3227 	struct rte_flow_item_ecpri mask_lo;
3228 
3229 	if (!(last_item & outer_l2_vlan) &&
3230 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3231 		return rte_flow_error_set(error, EINVAL,
3232 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3233 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3234 	if ((last_item & outer_l2_vlan) && ether_type &&
3235 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3236 		return rte_flow_error_set(error, EINVAL,
3237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3238 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3239 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3240 		return rte_flow_error_set(error, EINVAL,
3241 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3242 					  "eCPRI with tunnel is not supported right now");
3243 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3244 		return rte_flow_error_set(error, ENOTSUP,
3245 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3246 					  "multiple L3 layers not supported");
3247 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3248 		return rte_flow_error_set(error, EINVAL,
3249 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3250 					  "eCPRI cannot coexist with a TCP layer");
3251 	/* In specification, eCPRI could be over UDP layer. */
3252 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3253 		return rte_flow_error_set(error, EINVAL,
3254 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3255 					  "eCPRI over UDP layer is not yet supported right now");
3256 	/* Mask for type field in common header could be zero. */
3257 	if (!mask)
3258 		mask = &rte_flow_item_ecpri_mask;
3259 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3260 	/* Input mask is in big-endian format. */
3261 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3262 		return rte_flow_error_set(error, EINVAL,
3263 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3264 					  "partial mask is not supported for protocol");
3265 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3266 		return rte_flow_error_set(error, EINVAL,
3267 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3268 					  "message header mask must be after a type mask");
3269 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3270 					 acc_mask ? (const uint8_t *)acc_mask
3271 						  : (const uint8_t *)&nic_mask,
3272 					 sizeof(struct rte_flow_item_ecpri),
3273 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3274 }
3275 
3276 static int
3277 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3278 		   const struct rte_flow_attr *attr __rte_unused,
3279 		   const struct rte_flow_item items[] __rte_unused,
3280 		   const struct rte_flow_action actions[] __rte_unused,
3281 		   bool external __rte_unused,
3282 		   int hairpin __rte_unused,
3283 		   struct rte_flow_error *error)
3284 {
3285 	return rte_flow_error_set(error, ENOTSUP,
3286 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3287 }
3288 
3289 static struct mlx5_flow *
3290 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3291 		  const struct rte_flow_attr *attr __rte_unused,
3292 		  const struct rte_flow_item items[] __rte_unused,
3293 		  const struct rte_flow_action actions[] __rte_unused,
3294 		  struct rte_flow_error *error)
3295 {
3296 	rte_flow_error_set(error, ENOTSUP,
3297 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3298 	return NULL;
3299 }
3300 
3301 static int
3302 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3303 		    struct mlx5_flow *dev_flow __rte_unused,
3304 		    const struct rte_flow_attr *attr __rte_unused,
3305 		    const struct rte_flow_item items[] __rte_unused,
3306 		    const struct rte_flow_action actions[] __rte_unused,
3307 		    struct rte_flow_error *error)
3308 {
3309 	return rte_flow_error_set(error, ENOTSUP,
3310 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3311 }
3312 
3313 static int
3314 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3315 		struct rte_flow *flow __rte_unused,
3316 		struct rte_flow_error *error)
3317 {
3318 	return rte_flow_error_set(error, ENOTSUP,
3319 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3320 }
3321 
3322 static void
3323 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3324 		 struct rte_flow *flow __rte_unused)
3325 {
3326 }
3327 
3328 static void
3329 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3330 		  struct rte_flow *flow __rte_unused)
3331 {
3332 }
3333 
3334 static int
3335 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3336 		struct rte_flow *flow __rte_unused,
3337 		const struct rte_flow_action *actions __rte_unused,
3338 		void *data __rte_unused,
3339 		struct rte_flow_error *error)
3340 {
3341 	return rte_flow_error_set(error, ENOTSUP,
3342 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3343 }
3344 
3345 static int
3346 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3347 		      uint32_t domains __rte_unused,
3348 		      uint32_t flags __rte_unused)
3349 {
3350 	return 0;
3351 }
3352 
3353 /* Void driver to protect from null pointer reference. */
3354 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3355 	.validate = flow_null_validate,
3356 	.prepare = flow_null_prepare,
3357 	.translate = flow_null_translate,
3358 	.apply = flow_null_apply,
3359 	.remove = flow_null_remove,
3360 	.destroy = flow_null_destroy,
3361 	.query = flow_null_query,
3362 	.sync_domain = flow_null_sync_domain,
3363 };
3364 
3365 /**
3366  * Select flow driver type according to flow attributes and device
3367  * configuration.
3368  *
3369  * @param[in] dev
3370  *   Pointer to the dev structure.
3371  * @param[in] attr
3372  *   Pointer to the flow attributes.
3373  *
3374  * @return
3375  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3376  */
3377 static enum mlx5_flow_drv_type
3378 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3379 {
3380 	struct mlx5_priv *priv = dev->data->dev_private;
3381 	/* The OS can determine first a specific flow type (DV, VERBS) */
3382 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3383 
3384 	if (type != MLX5_FLOW_TYPE_MAX)
3385 		return type;
3386 	/* If no OS specific type - continue with DV/VERBS selection */
3387 	if (attr->transfer && priv->config.dv_esw_en)
3388 		type = MLX5_FLOW_TYPE_DV;
3389 	if (!attr->transfer)
3390 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3391 						 MLX5_FLOW_TYPE_VERBS;
3392 	return type;
3393 }
3394 
3395 #define flow_get_drv_ops(type) flow_drv_ops[type]
3396 
3397 /**
3398  * Flow driver validation API. This abstracts calling driver specific functions.
3399  * The type of flow driver is determined according to flow attributes.
3400  *
3401  * @param[in] dev
3402  *   Pointer to the dev structure.
3403  * @param[in] attr
3404  *   Pointer to the flow attributes.
3405  * @param[in] items
3406  *   Pointer to the list of items.
3407  * @param[in] actions
3408  *   Pointer to the list of actions.
3409  * @param[in] external
3410  *   This flow rule is created by request external to PMD.
3411  * @param[in] hairpin
3412  *   Number of hairpin TX actions, 0 means classic flow.
3413  * @param[out] error
3414  *   Pointer to the error structure.
3415  *
3416  * @return
3417  *   0 on success, a negative errno value otherwise and rte_errno is set.
3418  */
3419 static inline int
3420 flow_drv_validate(struct rte_eth_dev *dev,
3421 		  const struct rte_flow_attr *attr,
3422 		  const struct rte_flow_item items[],
3423 		  const struct rte_flow_action actions[],
3424 		  bool external, int hairpin, struct rte_flow_error *error)
3425 {
3426 	const struct mlx5_flow_driver_ops *fops;
3427 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3428 
3429 	fops = flow_get_drv_ops(type);
3430 	return fops->validate(dev, attr, items, actions, external,
3431 			      hairpin, error);
3432 }
3433 
3434 /**
3435  * Flow driver preparation API. This abstracts calling driver specific
3436  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3437  * calculates the size of memory required for device flow, allocates the memory,
3438  * initializes the device flow and returns the pointer.
3439  *
3440  * @note
3441  *   This function initializes device flow structure such as dv or verbs in
3442  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3443  *   rest. For example, adding returning device flow to flow->dev_flow list and
3444  *   setting backward reference to the flow should be done out of this function.
3445  *   layers field is not filled either.
3446  *
3447  * @param[in] dev
3448  *   Pointer to the dev structure.
3449  * @param[in] attr
3450  *   Pointer to the flow attributes.
3451  * @param[in] items
3452  *   Pointer to the list of items.
3453  * @param[in] actions
3454  *   Pointer to the list of actions.
3455  * @param[in] flow_idx
3456  *   This memory pool index to the flow.
3457  * @param[out] error
3458  *   Pointer to the error structure.
3459  *
3460  * @return
3461  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3462  */
3463 static inline struct mlx5_flow *
3464 flow_drv_prepare(struct rte_eth_dev *dev,
3465 		 const struct rte_flow *flow,
3466 		 const struct rte_flow_attr *attr,
3467 		 const struct rte_flow_item items[],
3468 		 const struct rte_flow_action actions[],
3469 		 uint32_t flow_idx,
3470 		 struct rte_flow_error *error)
3471 {
3472 	const struct mlx5_flow_driver_ops *fops;
3473 	enum mlx5_flow_drv_type type = flow->drv_type;
3474 	struct mlx5_flow *mlx5_flow = NULL;
3475 
3476 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3477 	fops = flow_get_drv_ops(type);
3478 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3479 	if (mlx5_flow)
3480 		mlx5_flow->flow_idx = flow_idx;
3481 	return mlx5_flow;
3482 }
3483 
3484 /**
3485  * Flow driver translation API. This abstracts calling driver specific
3486  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3487  * translates a generic flow into a driver flow. flow_drv_prepare() must
3488  * precede.
3489  *
3490  * @note
3491  *   dev_flow->layers could be filled as a result of parsing during translation
3492  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3493  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3494  *   flow->actions could be overwritten even though all the expanded dev_flows
3495  *   have the same actions.
3496  *
3497  * @param[in] dev
3498  *   Pointer to the rte dev structure.
3499  * @param[in, out] dev_flow
3500  *   Pointer to the mlx5 flow.
3501  * @param[in] attr
3502  *   Pointer to the flow attributes.
3503  * @param[in] items
3504  *   Pointer to the list of items.
3505  * @param[in] actions
3506  *   Pointer to the list of actions.
3507  * @param[out] error
3508  *   Pointer to the error structure.
3509  *
3510  * @return
3511  *   0 on success, a negative errno value otherwise and rte_errno is set.
3512  */
3513 static inline int
3514 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3515 		   const struct rte_flow_attr *attr,
3516 		   const struct rte_flow_item items[],
3517 		   const struct rte_flow_action actions[],
3518 		   struct rte_flow_error *error)
3519 {
3520 	const struct mlx5_flow_driver_ops *fops;
3521 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3522 
3523 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3524 	fops = flow_get_drv_ops(type);
3525 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3526 }
3527 
3528 /**
3529  * Flow driver apply API. This abstracts calling driver specific functions.
3530  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3531  * translated driver flows on to device. flow_drv_translate() must precede.
3532  *
3533  * @param[in] dev
3534  *   Pointer to Ethernet device structure.
3535  * @param[in, out] flow
3536  *   Pointer to flow structure.
3537  * @param[out] error
3538  *   Pointer to error structure.
3539  *
3540  * @return
3541  *   0 on success, a negative errno value otherwise and rte_errno is set.
3542  */
3543 static inline int
3544 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3545 	       struct rte_flow_error *error)
3546 {
3547 	const struct mlx5_flow_driver_ops *fops;
3548 	enum mlx5_flow_drv_type type = flow->drv_type;
3549 
3550 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3551 	fops = flow_get_drv_ops(type);
3552 	return fops->apply(dev, flow, error);
3553 }
3554 
3555 /**
3556  * Flow driver destroy API. This abstracts calling driver specific functions.
3557  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3558  * on device and releases resources of the flow.
3559  *
3560  * @param[in] dev
3561  *   Pointer to Ethernet device.
3562  * @param[in, out] flow
3563  *   Pointer to flow structure.
3564  */
3565 static inline void
3566 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3567 {
3568 	const struct mlx5_flow_driver_ops *fops;
3569 	enum mlx5_flow_drv_type type = flow->drv_type;
3570 
3571 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3572 	fops = flow_get_drv_ops(type);
3573 	fops->destroy(dev, flow);
3574 }
3575 
3576 /**
3577  * Flow driver find RSS policy tbl API. This abstracts calling driver
3578  * specific functions. Parent flow (rte_flow) should have driver
3579  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3580  *
3581  * @param[in] dev
3582  *   Pointer to Ethernet device.
3583  * @param[in, out] flow
3584  *   Pointer to flow structure.
3585  * @param[in] policy
3586  *   Pointer to meter policy table.
3587  * @param[in] rss_desc
3588  *   Pointer to rss_desc
3589  */
3590 static struct mlx5_flow_meter_sub_policy *
3591 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3592 		struct rte_flow *flow,
3593 		struct mlx5_flow_meter_policy *policy,
3594 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3595 {
3596 	const struct mlx5_flow_driver_ops *fops;
3597 	enum mlx5_flow_drv_type type = flow->drv_type;
3598 
3599 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3600 	fops = flow_get_drv_ops(type);
3601 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3602 }
3603 
3604 /**
3605  * Flow driver color tag rule API. This abstracts calling driver
3606  * specific functions. Parent flow (rte_flow) should have driver
3607  * type (drv_type). It will create the color tag rules in hierarchy meter.
3608  *
3609  * @param[in] dev
3610  *   Pointer to Ethernet device.
3611  * @param[in, out] flow
3612  *   Pointer to flow structure.
3613  * @param[in] fm
3614  *   Pointer to flow meter structure.
3615  * @param[in] src_port
3616  *   The src port this extra rule should use.
3617  * @param[in] item
3618  *   The src port id match item.
3619  * @param[out] error
3620  *   Pointer to error structure.
3621  */
3622 static int
3623 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3624 		struct rte_flow *flow,
3625 		struct mlx5_flow_meter_info *fm,
3626 		int32_t src_port,
3627 		const struct rte_flow_item *item,
3628 		struct rte_flow_error *error)
3629 {
3630 	const struct mlx5_flow_driver_ops *fops;
3631 	enum mlx5_flow_drv_type type = flow->drv_type;
3632 
3633 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3634 	fops = flow_get_drv_ops(type);
3635 	return fops->meter_hierarchy_rule_create(dev, fm,
3636 						src_port, item, error);
3637 }
3638 
3639 /**
3640  * Get RSS action from the action list.
3641  *
3642  * @param[in] dev
3643  *   Pointer to Ethernet device.
3644  * @param[in] actions
3645  *   Pointer to the list of actions.
3646  * @param[in] flow
3647  *   Parent flow structure pointer.
3648  *
3649  * @return
3650  *   Pointer to the RSS action if exist, else return NULL.
3651  */
3652 static const struct rte_flow_action_rss*
3653 flow_get_rss_action(struct rte_eth_dev *dev,
3654 		    const struct rte_flow_action actions[])
3655 {
3656 	struct mlx5_priv *priv = dev->data->dev_private;
3657 	const struct rte_flow_action_rss *rss = NULL;
3658 	struct mlx5_meter_policy_action_container *acg;
3659 	struct mlx5_meter_policy_action_container *acy;
3660 
3661 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3662 		switch (actions->type) {
3663 		case RTE_FLOW_ACTION_TYPE_RSS:
3664 			rss = actions->conf;
3665 			break;
3666 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3667 		{
3668 			const struct rte_flow_action_sample *sample =
3669 								actions->conf;
3670 			const struct rte_flow_action *act = sample->actions;
3671 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3672 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3673 					rss = act->conf;
3674 			break;
3675 		}
3676 		case RTE_FLOW_ACTION_TYPE_METER:
3677 		{
3678 			uint32_t mtr_idx;
3679 			struct mlx5_flow_meter_info *fm;
3680 			struct mlx5_flow_meter_policy *policy;
3681 			const struct rte_flow_action_meter *mtr = actions->conf;
3682 
3683 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3684 			if (fm && !fm->def_policy) {
3685 				policy = mlx5_flow_meter_policy_find(dev,
3686 						fm->policy_id, NULL);
3687 				MLX5_ASSERT(policy);
3688 				if (policy->is_hierarchy) {
3689 					policy =
3690 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3691 									policy);
3692 					if (!policy)
3693 						return NULL;
3694 				}
3695 				if (policy->is_rss) {
3696 					acg =
3697 					&policy->act_cnt[RTE_COLOR_GREEN];
3698 					acy =
3699 					&policy->act_cnt[RTE_COLOR_YELLOW];
3700 					if (acg->fate_action ==
3701 					    MLX5_FLOW_FATE_SHARED_RSS)
3702 						rss = acg->rss->conf;
3703 					else if (acy->fate_action ==
3704 						 MLX5_FLOW_FATE_SHARED_RSS)
3705 						rss = acy->rss->conf;
3706 				}
3707 			}
3708 			break;
3709 		}
3710 		default:
3711 			break;
3712 		}
3713 	}
3714 	return rss;
3715 }
3716 
3717 /**
3718  * Get ASO age action by index.
3719  *
3720  * @param[in] dev
3721  *   Pointer to the Ethernet device structure.
3722  * @param[in] age_idx
3723  *   Index to the ASO age action.
3724  *
3725  * @return
3726  *   The specified ASO age action.
3727  */
3728 struct mlx5_aso_age_action*
3729 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3730 {
3731 	uint16_t pool_idx = age_idx & UINT16_MAX;
3732 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3733 	struct mlx5_priv *priv = dev->data->dev_private;
3734 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3735 	struct mlx5_aso_age_pool *pool;
3736 
3737 	rte_rwlock_read_lock(&mng->resize_rwl);
3738 	pool = mng->pools[pool_idx];
3739 	rte_rwlock_read_unlock(&mng->resize_rwl);
3740 	return &pool->actions[offset - 1];
3741 }
3742 
3743 /* maps indirect action to translated direct in some actions array */
3744 struct mlx5_translated_action_handle {
3745 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3746 	int index; /**< Index in related array of rte_flow_action. */
3747 };
3748 
3749 /**
3750  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3751  * direct action if translation possible.
3752  * This functionality used to run same execution path for both direct and
3753  * indirect actions on flow create. All necessary preparations for indirect
3754  * action handling should be performed on *handle* actions list returned
3755  * from this call.
3756  *
3757  * @param[in] dev
3758  *   Pointer to Ethernet device.
3759  * @param[in] actions
3760  *   List of actions to translate.
3761  * @param[out] handle
3762  *   List to store translated indirect action object handles.
3763  * @param[in, out] indir_n
3764  *   Size of *handle* array. On return should be updated with number of
3765  *   indirect actions retrieved from the *actions* list.
3766  * @param[out] translated_actions
3767  *   List of actions where all indirect actions were translated to direct
3768  *   if possible. NULL if no translation took place.
3769  * @param[out] error
3770  *   Pointer to the error structure.
3771  *
3772  * @return
3773  *   0 on success, a negative errno value otherwise and rte_errno is set.
3774  */
3775 static int
3776 flow_action_handles_translate(struct rte_eth_dev *dev,
3777 			      const struct rte_flow_action actions[],
3778 			      struct mlx5_translated_action_handle *handle,
3779 			      int *indir_n,
3780 			      struct rte_flow_action **translated_actions,
3781 			      struct rte_flow_error *error)
3782 {
3783 	struct mlx5_priv *priv = dev->data->dev_private;
3784 	struct rte_flow_action *translated = NULL;
3785 	size_t actions_size;
3786 	int n;
3787 	int copied_n = 0;
3788 	struct mlx5_translated_action_handle *handle_end = NULL;
3789 
3790 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3791 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3792 			continue;
3793 		if (copied_n == *indir_n) {
3794 			return rte_flow_error_set
3795 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3796 				 NULL, "too many shared actions");
3797 		}
3798 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3799 			   sizeof(actions[n].conf));
3800 		handle[copied_n].index = n;
3801 		copied_n++;
3802 	}
3803 	n++;
3804 	*indir_n = copied_n;
3805 	if (!copied_n)
3806 		return 0;
3807 	actions_size = sizeof(struct rte_flow_action) * n;
3808 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3809 	if (!translated) {
3810 		rte_errno = ENOMEM;
3811 		return -ENOMEM;
3812 	}
3813 	memcpy(translated, actions, actions_size);
3814 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3815 		struct mlx5_shared_action_rss *shared_rss;
3816 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3817 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3818 		uint32_t idx = act_idx &
3819 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3820 
3821 		switch (type) {
3822 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3823 			shared_rss = mlx5_ipool_get
3824 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3825 			translated[handle->index].type =
3826 				RTE_FLOW_ACTION_TYPE_RSS;
3827 			translated[handle->index].conf =
3828 				&shared_rss->origin;
3829 			break;
3830 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3831 			translated[handle->index].type =
3832 						(enum rte_flow_action_type)
3833 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3834 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3835 			break;
3836 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3837 			if (priv->sh->flow_hit_aso_en) {
3838 				translated[handle->index].type =
3839 					(enum rte_flow_action_type)
3840 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3841 				translated[handle->index].conf =
3842 							 (void *)(uintptr_t)idx;
3843 				break;
3844 			}
3845 			/* Fall-through */
3846 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3847 			if (priv->sh->ct_aso_en) {
3848 				translated[handle->index].type =
3849 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3850 				translated[handle->index].conf =
3851 							 (void *)(uintptr_t)idx;
3852 				break;
3853 			}
3854 			/* Fall-through */
3855 		default:
3856 			mlx5_free(translated);
3857 			return rte_flow_error_set
3858 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3859 				 NULL, "invalid indirect action type");
3860 		}
3861 	}
3862 	*translated_actions = translated;
3863 	return 0;
3864 }
3865 
3866 /**
3867  * Get Shared RSS action from the action list.
3868  *
3869  * @param[in] dev
3870  *   Pointer to Ethernet device.
3871  * @param[in] shared
3872  *   Pointer to the list of actions.
3873  * @param[in] shared_n
3874  *   Actions list length.
3875  *
3876  * @return
3877  *   The MLX5 RSS action ID if exists, otherwise return 0.
3878  */
3879 static uint32_t
3880 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3881 			   struct mlx5_translated_action_handle *handle,
3882 			   int shared_n)
3883 {
3884 	struct mlx5_translated_action_handle *handle_end;
3885 	struct mlx5_priv *priv = dev->data->dev_private;
3886 	struct mlx5_shared_action_rss *shared_rss;
3887 
3888 
3889 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3890 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3891 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3892 		uint32_t idx = act_idx &
3893 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3894 		switch (type) {
3895 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3896 			shared_rss = mlx5_ipool_get
3897 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3898 									   idx);
3899 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3900 					   __ATOMIC_RELAXED);
3901 			return idx;
3902 		default:
3903 			break;
3904 		}
3905 	}
3906 	return 0;
3907 }
3908 
3909 static unsigned int
3910 find_graph_root(uint32_t rss_level)
3911 {
3912 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3913 			       MLX5_EXPANSION_ROOT_OUTER;
3914 }
3915 
3916 /**
3917  *  Get layer flags from the prefix flow.
3918  *
3919  *  Some flows may be split to several subflows, the prefix subflow gets the
3920  *  match items and the suffix sub flow gets the actions.
3921  *  Some actions need the user defined match item flags to get the detail for
3922  *  the action.
3923  *  This function helps the suffix flow to get the item layer flags from prefix
3924  *  subflow.
3925  *
3926  * @param[in] dev_flow
3927  *   Pointer the created preifx subflow.
3928  *
3929  * @return
3930  *   The layers get from prefix subflow.
3931  */
3932 static inline uint64_t
3933 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3934 {
3935 	uint64_t layers = 0;
3936 
3937 	/*
3938 	 * Layers bits could be localization, but usually the compiler will
3939 	 * help to do the optimization work for source code.
3940 	 * If no decap actions, use the layers directly.
3941 	 */
3942 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3943 		return dev_flow->handle->layers;
3944 	/* Convert L3 layers with decap action. */
3945 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3946 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3947 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3948 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3949 	/* Convert L4 layers with decap action.  */
3950 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3951 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3952 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3953 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3954 	return layers;
3955 }
3956 
3957 /**
3958  * Get metadata split action information.
3959  *
3960  * @param[in] actions
3961  *   Pointer to the list of actions.
3962  * @param[out] qrss
3963  *   Pointer to the return pointer.
3964  * @param[out] qrss_type
3965  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3966  *   if no QUEUE/RSS is found.
3967  * @param[out] encap_idx
3968  *   Pointer to the index of the encap action if exists, otherwise the last
3969  *   action index.
3970  *
3971  * @return
3972  *   Total number of actions.
3973  */
3974 static int
3975 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3976 				       const struct rte_flow_action **qrss,
3977 				       int *encap_idx)
3978 {
3979 	const struct rte_flow_action_raw_encap *raw_encap;
3980 	int actions_n = 0;
3981 	int raw_decap_idx = -1;
3982 
3983 	*encap_idx = -1;
3984 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3985 		switch (actions->type) {
3986 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3987 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3988 			*encap_idx = actions_n;
3989 			break;
3990 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3991 			raw_decap_idx = actions_n;
3992 			break;
3993 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3994 			raw_encap = actions->conf;
3995 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3996 				*encap_idx = raw_decap_idx != -1 ?
3997 						      raw_decap_idx : actions_n;
3998 			break;
3999 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4000 		case RTE_FLOW_ACTION_TYPE_RSS:
4001 			*qrss = actions;
4002 			break;
4003 		default:
4004 			break;
4005 		}
4006 		actions_n++;
4007 	}
4008 	if (*encap_idx == -1)
4009 		*encap_idx = actions_n;
4010 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4011 	return actions_n + 1;
4012 }
4013 
4014 /**
4015  * Check if the action will change packet.
4016  *
4017  * @param dev
4018  *   Pointer to Ethernet device.
4019  * @param[in] type
4020  *   action type.
4021  *
4022  * @return
4023  *   true if action will change packet, false otherwise.
4024  */
4025 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4026 					  enum rte_flow_action_type type)
4027 {
4028 	struct mlx5_priv *priv = dev->data->dev_private;
4029 
4030 	switch (type) {
4031 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4032 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4033 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4034 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4035 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4036 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4037 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4038 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4039 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4040 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4041 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4042 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4043 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4044 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4045 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4046 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4047 	case RTE_FLOW_ACTION_TYPE_SET_META:
4048 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4049 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4050 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4051 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4052 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4053 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4054 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4055 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4056 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4057 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4058 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4059 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4060 		return true;
4061 	case RTE_FLOW_ACTION_TYPE_FLAG:
4062 	case RTE_FLOW_ACTION_TYPE_MARK:
4063 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4064 			return true;
4065 		else
4066 			return false;
4067 	default:
4068 		return false;
4069 	}
4070 }
4071 
4072 /**
4073  * Check meter action from the action list.
4074  *
4075  * @param dev
4076  *   Pointer to Ethernet device.
4077  * @param[in] actions
4078  *   Pointer to the list of actions.
4079  * @param[out] has_mtr
4080  *   Pointer to the meter exist flag.
4081  * @param[out] has_modify
4082  *   Pointer to the flag showing there's packet change action.
4083  * @param[out] meter_id
4084  *   Pointer to the meter id.
4085  *
4086  * @return
4087  *   Total number of actions.
4088  */
4089 static int
4090 flow_check_meter_action(struct rte_eth_dev *dev,
4091 			const struct rte_flow_action actions[],
4092 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4093 {
4094 	const struct rte_flow_action_meter *mtr = NULL;
4095 	int actions_n = 0;
4096 
4097 	MLX5_ASSERT(has_mtr);
4098 	*has_mtr = false;
4099 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4100 		switch (actions->type) {
4101 		case RTE_FLOW_ACTION_TYPE_METER:
4102 			mtr = actions->conf;
4103 			*meter_id = mtr->mtr_id;
4104 			*has_mtr = true;
4105 			break;
4106 		default:
4107 			break;
4108 		}
4109 		if (!*has_mtr)
4110 			*has_modify |= flow_check_modify_action_type(dev,
4111 								actions->type);
4112 		actions_n++;
4113 	}
4114 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4115 	return actions_n + 1;
4116 }
4117 
4118 /**
4119  * Check if the flow should be split due to hairpin.
4120  * The reason for the split is that in current HW we can't
4121  * support encap and push-vlan on Rx, so if a flow contains
4122  * these actions we move it to Tx.
4123  *
4124  * @param dev
4125  *   Pointer to Ethernet device.
4126  * @param[in] attr
4127  *   Flow rule attributes.
4128  * @param[in] actions
4129  *   Associated actions (list terminated by the END action).
4130  *
4131  * @return
4132  *   > 0 the number of actions and the flow should be split,
4133  *   0 when no split required.
4134  */
4135 static int
4136 flow_check_hairpin_split(struct rte_eth_dev *dev,
4137 			 const struct rte_flow_attr *attr,
4138 			 const struct rte_flow_action actions[])
4139 {
4140 	int queue_action = 0;
4141 	int action_n = 0;
4142 	int split = 0;
4143 	const struct rte_flow_action_queue *queue;
4144 	const struct rte_flow_action_rss *rss;
4145 	const struct rte_flow_action_raw_encap *raw_encap;
4146 	const struct rte_eth_hairpin_conf *conf;
4147 
4148 	if (!attr->ingress)
4149 		return 0;
4150 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4151 		switch (actions->type) {
4152 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4153 			queue = actions->conf;
4154 			if (queue == NULL)
4155 				return 0;
4156 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4157 			if (conf == NULL || conf->tx_explicit != 0)
4158 				return 0;
4159 			queue_action = 1;
4160 			action_n++;
4161 			break;
4162 		case RTE_FLOW_ACTION_TYPE_RSS:
4163 			rss = actions->conf;
4164 			if (rss == NULL || rss->queue_num == 0)
4165 				return 0;
4166 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4167 			if (conf == NULL || conf->tx_explicit != 0)
4168 				return 0;
4169 			queue_action = 1;
4170 			action_n++;
4171 			break;
4172 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4173 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4174 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4175 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4176 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4177 			split++;
4178 			action_n++;
4179 			break;
4180 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4181 			raw_encap = actions->conf;
4182 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4183 				split++;
4184 			action_n++;
4185 			break;
4186 		default:
4187 			action_n++;
4188 			break;
4189 		}
4190 	}
4191 	if (split && queue_action)
4192 		return action_n;
4193 	return 0;
4194 }
4195 
4196 /* Declare flow create/destroy prototype in advance. */
4197 static uint32_t
4198 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4199 		 const struct rte_flow_attr *attr,
4200 		 const struct rte_flow_item items[],
4201 		 const struct rte_flow_action actions[],
4202 		 bool external, struct rte_flow_error *error);
4203 
4204 static void
4205 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4206 		  uint32_t flow_idx);
4207 
4208 int
4209 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4210 		      struct mlx5_list_entry *entry, void *cb_ctx)
4211 {
4212 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4213 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4214 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4215 
4216 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4217 }
4218 
4219 struct mlx5_list_entry *
4220 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4221 {
4222 	struct rte_eth_dev *dev = tool_ctx;
4223 	struct mlx5_priv *priv = dev->data->dev_private;
4224 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4225 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4226 	struct rte_flow_error *error = ctx->error;
4227 	uint32_t idx = 0;
4228 	int ret;
4229 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4230 	struct rte_flow_attr attr = {
4231 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4232 		.ingress = 1,
4233 	};
4234 	struct mlx5_rte_flow_item_tag tag_spec = {
4235 		.data = mark_id,
4236 	};
4237 	struct rte_flow_item items[] = {
4238 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4239 	};
4240 	struct rte_flow_action_mark ftag = {
4241 		.id = mark_id,
4242 	};
4243 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4244 		.dst = REG_B,
4245 		.src = REG_NON,
4246 	};
4247 	struct rte_flow_action_jump jump = {
4248 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4249 	};
4250 	struct rte_flow_action actions[] = {
4251 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4252 	};
4253 
4254 	/* Fill the register fileds in the flow. */
4255 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4256 	if (ret < 0)
4257 		return NULL;
4258 	tag_spec.id = ret;
4259 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4260 	if (ret < 0)
4261 		return NULL;
4262 	cp_mreg.src = ret;
4263 	/* Provide the full width of FLAG specific value. */
4264 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4265 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4266 	/* Build a new flow. */
4267 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4268 		items[0] = (struct rte_flow_item){
4269 			.type = (enum rte_flow_item_type)
4270 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4271 			.spec = &tag_spec,
4272 		};
4273 		items[1] = (struct rte_flow_item){
4274 			.type = RTE_FLOW_ITEM_TYPE_END,
4275 		};
4276 		actions[0] = (struct rte_flow_action){
4277 			.type = (enum rte_flow_action_type)
4278 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4279 			.conf = &ftag,
4280 		};
4281 		actions[1] = (struct rte_flow_action){
4282 			.type = (enum rte_flow_action_type)
4283 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4284 			.conf = &cp_mreg,
4285 		};
4286 		actions[2] = (struct rte_flow_action){
4287 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4288 			.conf = &jump,
4289 		};
4290 		actions[3] = (struct rte_flow_action){
4291 			.type = RTE_FLOW_ACTION_TYPE_END,
4292 		};
4293 	} else {
4294 		/* Default rule, wildcard match. */
4295 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4296 		items[0] = (struct rte_flow_item){
4297 			.type = RTE_FLOW_ITEM_TYPE_END,
4298 		};
4299 		actions[0] = (struct rte_flow_action){
4300 			.type = (enum rte_flow_action_type)
4301 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4302 			.conf = &cp_mreg,
4303 		};
4304 		actions[1] = (struct rte_flow_action){
4305 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4306 			.conf = &jump,
4307 		};
4308 		actions[2] = (struct rte_flow_action){
4309 			.type = RTE_FLOW_ACTION_TYPE_END,
4310 		};
4311 	}
4312 	/* Build a new entry. */
4313 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4314 	if (!mcp_res) {
4315 		rte_errno = ENOMEM;
4316 		return NULL;
4317 	}
4318 	mcp_res->idx = idx;
4319 	mcp_res->mark_id = mark_id;
4320 	/*
4321 	 * The copy Flows are not included in any list. There
4322 	 * ones are referenced from other Flows and can not
4323 	 * be applied, removed, deleted in ardbitrary order
4324 	 * by list traversing.
4325 	 */
4326 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4327 					&attr, items, actions, false, error);
4328 	if (!mcp_res->rix_flow) {
4329 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4330 		return NULL;
4331 	}
4332 	return &mcp_res->hlist_ent;
4333 }
4334 
4335 struct mlx5_list_entry *
4336 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4337 		      void *cb_ctx __rte_unused)
4338 {
4339 	struct rte_eth_dev *dev = tool_ctx;
4340 	struct mlx5_priv *priv = dev->data->dev_private;
4341 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4342 	uint32_t idx = 0;
4343 
4344 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4345 	if (!mcp_res) {
4346 		rte_errno = ENOMEM;
4347 		return NULL;
4348 	}
4349 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4350 	mcp_res->idx = idx;
4351 	return &mcp_res->hlist_ent;
4352 }
4353 
4354 void
4355 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4356 {
4357 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4358 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4359 	struct rte_eth_dev *dev = tool_ctx;
4360 	struct mlx5_priv *priv = dev->data->dev_private;
4361 
4362 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4363 }
4364 
4365 /**
4366  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4367  *
4368  * As mark_id is unique, if there's already a registered flow for the mark_id,
4369  * return by increasing the reference counter of the resource. Otherwise, create
4370  * the resource (mcp_res) and flow.
4371  *
4372  * Flow looks like,
4373  *   - If ingress port is ANY and reg_c[1] is mark_id,
4374  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4375  *
4376  * For default flow (zero mark_id), flow is like,
4377  *   - If ingress port is ANY,
4378  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4379  *
4380  * @param dev
4381  *   Pointer to Ethernet device.
4382  * @param mark_id
4383  *   ID of MARK action, zero means default flow for META.
4384  * @param[out] error
4385  *   Perform verbose error reporting if not NULL.
4386  *
4387  * @return
4388  *   Associated resource on success, NULL otherwise and rte_errno is set.
4389  */
4390 static struct mlx5_flow_mreg_copy_resource *
4391 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4392 			  struct rte_flow_error *error)
4393 {
4394 	struct mlx5_priv *priv = dev->data->dev_private;
4395 	struct mlx5_list_entry *entry;
4396 	struct mlx5_flow_cb_ctx ctx = {
4397 		.dev = dev,
4398 		.error = error,
4399 		.data = &mark_id,
4400 	};
4401 
4402 	/* Check if already registered. */
4403 	MLX5_ASSERT(priv->mreg_cp_tbl);
4404 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4405 	if (!entry)
4406 		return NULL;
4407 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4408 			    hlist_ent);
4409 }
4410 
4411 void
4412 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4413 {
4414 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4415 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4416 	struct rte_eth_dev *dev = tool_ctx;
4417 	struct mlx5_priv *priv = dev->data->dev_private;
4418 
4419 	MLX5_ASSERT(mcp_res->rix_flow);
4420 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4421 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4422 }
4423 
4424 /**
4425  * Release flow in RX_CP_TBL.
4426  *
4427  * @param dev
4428  *   Pointer to Ethernet device.
4429  * @flow
4430  *   Parent flow for wich copying is provided.
4431  */
4432 static void
4433 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4434 			  struct rte_flow *flow)
4435 {
4436 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4437 	struct mlx5_priv *priv = dev->data->dev_private;
4438 
4439 	if (!flow->rix_mreg_copy)
4440 		return;
4441 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4442 				 flow->rix_mreg_copy);
4443 	if (!mcp_res || !priv->mreg_cp_tbl)
4444 		return;
4445 	MLX5_ASSERT(mcp_res->rix_flow);
4446 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4447 	flow->rix_mreg_copy = 0;
4448 }
4449 
4450 /**
4451  * Remove the default copy action from RX_CP_TBL.
4452  *
4453  * This functions is called in the mlx5_dev_start(). No thread safe
4454  * is guaranteed.
4455  *
4456  * @param dev
4457  *   Pointer to Ethernet device.
4458  */
4459 static void
4460 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4461 {
4462 	struct mlx5_list_entry *entry;
4463 	struct mlx5_priv *priv = dev->data->dev_private;
4464 	struct mlx5_flow_cb_ctx ctx;
4465 	uint32_t mark_id;
4466 
4467 	/* Check if default flow is registered. */
4468 	if (!priv->mreg_cp_tbl)
4469 		return;
4470 	mark_id = MLX5_DEFAULT_COPY_ID;
4471 	ctx.data = &mark_id;
4472 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4473 	if (!entry)
4474 		return;
4475 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4476 }
4477 
4478 /**
4479  * Add the default copy action in in RX_CP_TBL.
4480  *
4481  * This functions is called in the mlx5_dev_start(). No thread safe
4482  * is guaranteed.
4483  *
4484  * @param dev
4485  *   Pointer to Ethernet device.
4486  * @param[out] error
4487  *   Perform verbose error reporting if not NULL.
4488  *
4489  * @return
4490  *   0 for success, negative value otherwise and rte_errno is set.
4491  */
4492 static int
4493 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4494 				  struct rte_flow_error *error)
4495 {
4496 	struct mlx5_priv *priv = dev->data->dev_private;
4497 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4498 	struct mlx5_flow_cb_ctx ctx;
4499 	uint32_t mark_id;
4500 
4501 	/* Check whether extensive metadata feature is engaged. */
4502 	if (!priv->config.dv_flow_en ||
4503 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4504 	    !mlx5_flow_ext_mreg_supported(dev) ||
4505 	    !priv->sh->dv_regc0_mask)
4506 		return 0;
4507 	/*
4508 	 * Add default mreg copy flow may be called multiple time, but
4509 	 * only be called once in stop. Avoid register it twice.
4510 	 */
4511 	mark_id = MLX5_DEFAULT_COPY_ID;
4512 	ctx.data = &mark_id;
4513 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4514 		return 0;
4515 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4516 	if (!mcp_res)
4517 		return -rte_errno;
4518 	return 0;
4519 }
4520 
4521 /**
4522  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4523  *
4524  * All the flow having Q/RSS action should be split by
4525  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4526  * performs the following,
4527  *   - CQE->flow_tag := reg_c[1] (MARK)
4528  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4529  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4530  * but there should be a flow per each MARK ID set by MARK action.
4531  *
4532  * For the aforementioned reason, if there's a MARK action in flow's action
4533  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4534  * the MARK ID to CQE's flow_tag like,
4535  *   - If reg_c[1] is mark_id,
4536  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4537  *
4538  * For SET_META action which stores value in reg_c[0], as the destination is
4539  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4540  * MARK ID means the default flow. The default flow looks like,
4541  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4542  *
4543  * @param dev
4544  *   Pointer to Ethernet device.
4545  * @param flow
4546  *   Pointer to flow structure.
4547  * @param[in] actions
4548  *   Pointer to the list of actions.
4549  * @param[out] error
4550  *   Perform verbose error reporting if not NULL.
4551  *
4552  * @return
4553  *   0 on success, negative value otherwise and rte_errno is set.
4554  */
4555 static int
4556 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4557 			    struct rte_flow *flow,
4558 			    const struct rte_flow_action *actions,
4559 			    struct rte_flow_error *error)
4560 {
4561 	struct mlx5_priv *priv = dev->data->dev_private;
4562 	struct mlx5_dev_config *config = &priv->config;
4563 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4564 	const struct rte_flow_action_mark *mark;
4565 
4566 	/* Check whether extensive metadata feature is engaged. */
4567 	if (!config->dv_flow_en ||
4568 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4569 	    !mlx5_flow_ext_mreg_supported(dev) ||
4570 	    !priv->sh->dv_regc0_mask)
4571 		return 0;
4572 	/* Find MARK action. */
4573 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4574 		switch (actions->type) {
4575 		case RTE_FLOW_ACTION_TYPE_FLAG:
4576 			mcp_res = flow_mreg_add_copy_action
4577 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4578 			if (!mcp_res)
4579 				return -rte_errno;
4580 			flow->rix_mreg_copy = mcp_res->idx;
4581 			return 0;
4582 		case RTE_FLOW_ACTION_TYPE_MARK:
4583 			mark = (const struct rte_flow_action_mark *)
4584 				actions->conf;
4585 			mcp_res =
4586 				flow_mreg_add_copy_action(dev, mark->id, error);
4587 			if (!mcp_res)
4588 				return -rte_errno;
4589 			flow->rix_mreg_copy = mcp_res->idx;
4590 			return 0;
4591 		default:
4592 			break;
4593 		}
4594 	}
4595 	return 0;
4596 }
4597 
4598 #define MLX5_MAX_SPLIT_ACTIONS 24
4599 #define MLX5_MAX_SPLIT_ITEMS 24
4600 
4601 /**
4602  * Split the hairpin flow.
4603  * Since HW can't support encap and push-vlan on Rx, we move these
4604  * actions to Tx.
4605  * If the count action is after the encap then we also
4606  * move the count action. in this case the count will also measure
4607  * the outer bytes.
4608  *
4609  * @param dev
4610  *   Pointer to Ethernet device.
4611  * @param[in] actions
4612  *   Associated actions (list terminated by the END action).
4613  * @param[out] actions_rx
4614  *   Rx flow actions.
4615  * @param[out] actions_tx
4616  *   Tx flow actions..
4617  * @param[out] pattern_tx
4618  *   The pattern items for the Tx flow.
4619  * @param[out] flow_id
4620  *   The flow ID connected to this flow.
4621  *
4622  * @return
4623  *   0 on success.
4624  */
4625 static int
4626 flow_hairpin_split(struct rte_eth_dev *dev,
4627 		   const struct rte_flow_action actions[],
4628 		   struct rte_flow_action actions_rx[],
4629 		   struct rte_flow_action actions_tx[],
4630 		   struct rte_flow_item pattern_tx[],
4631 		   uint32_t flow_id)
4632 {
4633 	const struct rte_flow_action_raw_encap *raw_encap;
4634 	const struct rte_flow_action_raw_decap *raw_decap;
4635 	struct mlx5_rte_flow_action_set_tag *set_tag;
4636 	struct rte_flow_action *tag_action;
4637 	struct mlx5_rte_flow_item_tag *tag_item;
4638 	struct rte_flow_item *item;
4639 	char *addr;
4640 	int encap = 0;
4641 
4642 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4643 		switch (actions->type) {
4644 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4645 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4646 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4647 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4648 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4649 			rte_memcpy(actions_tx, actions,
4650 			       sizeof(struct rte_flow_action));
4651 			actions_tx++;
4652 			break;
4653 		case RTE_FLOW_ACTION_TYPE_COUNT:
4654 			if (encap) {
4655 				rte_memcpy(actions_tx, actions,
4656 					   sizeof(struct rte_flow_action));
4657 				actions_tx++;
4658 			} else {
4659 				rte_memcpy(actions_rx, actions,
4660 					   sizeof(struct rte_flow_action));
4661 				actions_rx++;
4662 			}
4663 			break;
4664 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4665 			raw_encap = actions->conf;
4666 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4667 				memcpy(actions_tx, actions,
4668 				       sizeof(struct rte_flow_action));
4669 				actions_tx++;
4670 				encap = 1;
4671 			} else {
4672 				rte_memcpy(actions_rx, actions,
4673 					   sizeof(struct rte_flow_action));
4674 				actions_rx++;
4675 			}
4676 			break;
4677 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4678 			raw_decap = actions->conf;
4679 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4680 				memcpy(actions_tx, actions,
4681 				       sizeof(struct rte_flow_action));
4682 				actions_tx++;
4683 			} else {
4684 				rte_memcpy(actions_rx, actions,
4685 					   sizeof(struct rte_flow_action));
4686 				actions_rx++;
4687 			}
4688 			break;
4689 		default:
4690 			rte_memcpy(actions_rx, actions,
4691 				   sizeof(struct rte_flow_action));
4692 			actions_rx++;
4693 			break;
4694 		}
4695 	}
4696 	/* Add set meta action and end action for the Rx flow. */
4697 	tag_action = actions_rx;
4698 	tag_action->type = (enum rte_flow_action_type)
4699 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4700 	actions_rx++;
4701 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4702 	actions_rx++;
4703 	set_tag = (void *)actions_rx;
4704 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4705 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4706 		.data = flow_id,
4707 	};
4708 	MLX5_ASSERT(set_tag->id > REG_NON);
4709 	tag_action->conf = set_tag;
4710 	/* Create Tx item list. */
4711 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4712 	addr = (void *)&pattern_tx[2];
4713 	item = pattern_tx;
4714 	item->type = (enum rte_flow_item_type)
4715 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4716 	tag_item = (void *)addr;
4717 	tag_item->data = flow_id;
4718 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4719 	MLX5_ASSERT(set_tag->id > REG_NON);
4720 	item->spec = tag_item;
4721 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4722 	tag_item = (void *)addr;
4723 	tag_item->data = UINT32_MAX;
4724 	tag_item->id = UINT16_MAX;
4725 	item->mask = tag_item;
4726 	item->last = NULL;
4727 	item++;
4728 	item->type = RTE_FLOW_ITEM_TYPE_END;
4729 	return 0;
4730 }
4731 
4732 /**
4733  * The last stage of splitting chain, just creates the subflow
4734  * without any modification.
4735  *
4736  * @param[in] dev
4737  *   Pointer to Ethernet device.
4738  * @param[in] flow
4739  *   Parent flow structure pointer.
4740  * @param[in, out] sub_flow
4741  *   Pointer to return the created subflow, may be NULL.
4742  * @param[in] attr
4743  *   Flow rule attributes.
4744  * @param[in] items
4745  *   Pattern specification (list terminated by the END pattern item).
4746  * @param[in] actions
4747  *   Associated actions (list terminated by the END action).
4748  * @param[in] flow_split_info
4749  *   Pointer to flow split info structure.
4750  * @param[out] error
4751  *   Perform verbose error reporting if not NULL.
4752  * @return
4753  *   0 on success, negative value otherwise
4754  */
4755 static int
4756 flow_create_split_inner(struct rte_eth_dev *dev,
4757 			struct rte_flow *flow,
4758 			struct mlx5_flow **sub_flow,
4759 			const struct rte_flow_attr *attr,
4760 			const struct rte_flow_item items[],
4761 			const struct rte_flow_action actions[],
4762 			struct mlx5_flow_split_info *flow_split_info,
4763 			struct rte_flow_error *error)
4764 {
4765 	struct mlx5_flow *dev_flow;
4766 
4767 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4768 				    flow_split_info->flow_idx, error);
4769 	if (!dev_flow)
4770 		return -rte_errno;
4771 	dev_flow->flow = flow;
4772 	dev_flow->external = flow_split_info->external;
4773 	dev_flow->skip_scale = flow_split_info->skip_scale;
4774 	/* Subflow object was created, we must include one in the list. */
4775 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4776 		      dev_flow->handle, next);
4777 	/*
4778 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4779 	 * flow may need some user defined item layer flags, and pass the
4780 	 * Metadate rxq mark flag to suffix flow as well.
4781 	 */
4782 	if (flow_split_info->prefix_layers)
4783 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4784 	if (flow_split_info->prefix_mark)
4785 		dev_flow->handle->mark = 1;
4786 	if (sub_flow)
4787 		*sub_flow = dev_flow;
4788 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4789 	dev_flow->dv.table_id = flow_split_info->table_id;
4790 #endif
4791 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4792 }
4793 
4794 /**
4795  * Get the sub policy of a meter.
4796  *
4797  * @param[in] dev
4798  *   Pointer to Ethernet device.
4799  * @param[in] flow
4800  *   Parent flow structure pointer.
4801  * @param wks
4802  *   Pointer to thread flow work space.
4803  * @param[in] attr
4804  *   Flow rule attributes.
4805  * @param[in] items
4806  *   Pattern specification (list terminated by the END pattern item).
4807  * @param[out] error
4808  *   Perform verbose error reporting if not NULL.
4809  *
4810  * @return
4811  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4812  */
4813 static struct mlx5_flow_meter_sub_policy *
4814 get_meter_sub_policy(struct rte_eth_dev *dev,
4815 		     struct rte_flow *flow,
4816 		     struct mlx5_flow_workspace *wks,
4817 		     const struct rte_flow_attr *attr,
4818 		     const struct rte_flow_item items[],
4819 		     struct rte_flow_error *error)
4820 {
4821 	struct mlx5_flow_meter_policy *policy;
4822 	struct mlx5_flow_meter_policy *final_policy;
4823 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4824 
4825 	policy = wks->policy;
4826 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4827 	if (final_policy->is_rss || final_policy->is_queue) {
4828 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4829 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4830 		uint32_t i;
4831 
4832 		/*
4833 		 * This is a tmp dev_flow,
4834 		 * no need to register any matcher for it in translate.
4835 		 */
4836 		wks->skip_matcher_reg = 1;
4837 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4838 			struct mlx5_flow dev_flow = {0};
4839 			struct mlx5_flow_handle dev_handle = { {0} };
4840 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4841 
4842 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4843 				const struct rte_flow_action_rss *rss_act =
4844 					final_policy->act_cnt[i].rss->conf;
4845 				struct rte_flow_action rss_actions[2] = {
4846 					[0] = {
4847 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4848 					.conf = rss_act,
4849 					},
4850 					[1] = {
4851 					.type = RTE_FLOW_ACTION_TYPE_END,
4852 					.conf = NULL,
4853 					}
4854 				};
4855 
4856 				dev_flow.handle = &dev_handle;
4857 				dev_flow.ingress = attr->ingress;
4858 				dev_flow.flow = flow;
4859 				dev_flow.external = 0;
4860 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4861 				dev_flow.dv.transfer = attr->transfer;
4862 #endif
4863 				/**
4864 				 * Translate RSS action to get rss hash fields.
4865 				 */
4866 				if (flow_drv_translate(dev, &dev_flow, attr,
4867 						items, rss_actions, error))
4868 					goto exit;
4869 				rss_desc_v[i] = wks->rss_desc;
4870 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4871 				rss_desc_v[i].hash_fields =
4872 						dev_flow.hash_fields;
4873 				rss_desc_v[i].queue_num =
4874 						rss_desc_v[i].hash_fields ?
4875 						rss_desc_v[i].queue_num : 1;
4876 				rss_desc_v[i].tunnel =
4877 						!!(dev_flow.handle->layers &
4878 						   MLX5_FLOW_LAYER_TUNNEL);
4879 				/* Use the RSS queues in the containers. */
4880 				rss_desc_v[i].queue =
4881 					(uint16_t *)(uintptr_t)rss_act->queue;
4882 				rss_desc[i] = &rss_desc_v[i];
4883 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4884 				/* This is queue action. */
4885 				rss_desc_v[i] = wks->rss_desc;
4886 				rss_desc_v[i].key_len = 0;
4887 				rss_desc_v[i].hash_fields = 0;
4888 				rss_desc_v[i].queue =
4889 					&final_policy->act_cnt[i].queue;
4890 				rss_desc_v[i].queue_num = 1;
4891 				rss_desc[i] = &rss_desc_v[i];
4892 			} else {
4893 				rss_desc[i] = NULL;
4894 			}
4895 		}
4896 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4897 						flow, policy, rss_desc);
4898 	} else {
4899 		enum mlx5_meter_domain mtr_domain =
4900 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4901 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4902 						MLX5_MTR_DOMAIN_INGRESS);
4903 		sub_policy = policy->sub_policys[mtr_domain][0];
4904 	}
4905 	if (!sub_policy)
4906 		rte_flow_error_set(error, EINVAL,
4907 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4908 				   "Failed to get meter sub-policy.");
4909 exit:
4910 	return sub_policy;
4911 }
4912 
4913 /**
4914  * Split the meter flow.
4915  *
4916  * As meter flow will split to three sub flow, other than meter
4917  * action, the other actions make sense to only meter accepts
4918  * the packet. If it need to be dropped, no other additional
4919  * actions should be take.
4920  *
4921  * One kind of special action which decapsulates the L3 tunnel
4922  * header will be in the prefix sub flow, as not to take the
4923  * L3 tunnel header into account.
4924  *
4925  * @param[in] dev
4926  *   Pointer to Ethernet device.
4927  * @param[in] flow
4928  *   Parent flow structure pointer.
4929  * @param wks
4930  *   Pointer to thread flow work space.
4931  * @param[in] attr
4932  *   Flow rule attributes.
4933  * @param[in] items
4934  *   Pattern specification (list terminated by the END pattern item).
4935  * @param[out] sfx_items
4936  *   Suffix flow match items (list terminated by the END pattern item).
4937  * @param[in] actions
4938  *   Associated actions (list terminated by the END action).
4939  * @param[out] actions_sfx
4940  *   Suffix flow actions.
4941  * @param[out] actions_pre
4942  *   Prefix flow actions.
4943  * @param[out] mtr_flow_id
4944  *   Pointer to meter flow id.
4945  * @param[out] error
4946  *   Perform verbose error reporting if not NULL.
4947  *
4948  * @return
4949  *   0 on success, a negative errno value otherwise and rte_errno is set.
4950  */
4951 static int
4952 flow_meter_split_prep(struct rte_eth_dev *dev,
4953 		      struct rte_flow *flow,
4954 		      struct mlx5_flow_workspace *wks,
4955 		      const struct rte_flow_attr *attr,
4956 		      const struct rte_flow_item items[],
4957 		      struct rte_flow_item sfx_items[],
4958 		      const struct rte_flow_action actions[],
4959 		      struct rte_flow_action actions_sfx[],
4960 		      struct rte_flow_action actions_pre[],
4961 		      uint32_t *mtr_flow_id,
4962 		      struct rte_flow_error *error)
4963 {
4964 	struct mlx5_priv *priv = dev->data->dev_private;
4965 	struct mlx5_flow_meter_info *fm = wks->fm;
4966 	struct rte_flow_action *tag_action = NULL;
4967 	struct rte_flow_item *tag_item;
4968 	struct mlx5_rte_flow_action_set_tag *set_tag;
4969 	const struct rte_flow_action_raw_encap *raw_encap;
4970 	const struct rte_flow_action_raw_decap *raw_decap;
4971 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4972 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4973 	uint32_t tag_id = 0;
4974 	struct rte_flow_item *vlan_item_dst = NULL;
4975 	const struct rte_flow_item *vlan_item_src = NULL;
4976 	struct rte_flow_action *hw_mtr_action;
4977 	struct rte_flow_action *action_pre_head = NULL;
4978 	int32_t flow_src_port = priv->representor_id;
4979 	bool mtr_first;
4980 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4981 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4982 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4983 	uint32_t flow_id = 0;
4984 	uint32_t flow_id_reversed = 0;
4985 	uint8_t flow_id_bits = 0;
4986 	int shift;
4987 
4988 	/* Prepare the suffix subflow items. */
4989 	tag_item = sfx_items++;
4990 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4991 		struct mlx5_priv *port_priv;
4992 		const struct rte_flow_item_port_id *pid_v;
4993 		int item_type = items->type;
4994 
4995 		switch (item_type) {
4996 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4997 			pid_v = items->spec;
4998 			MLX5_ASSERT(pid_v);
4999 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5000 			if (!port_priv)
5001 				return rte_flow_error_set(error,
5002 						rte_errno,
5003 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5004 						pid_v,
5005 						"Failed to get port info.");
5006 			flow_src_port = port_priv->representor_id;
5007 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5008 			    flow_src_port != priv->representor_id) {
5009 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5010 								flow, fm,
5011 								flow_src_port,
5012 								items,
5013 								error))
5014 					return -rte_errno;
5015 			}
5016 			memcpy(sfx_items, items, sizeof(*sfx_items));
5017 			sfx_items++;
5018 			break;
5019 		case RTE_FLOW_ITEM_TYPE_VLAN:
5020 			/* Determine if copy vlan item below. */
5021 			vlan_item_src = items;
5022 			vlan_item_dst = sfx_items++;
5023 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5024 			break;
5025 		default:
5026 			break;
5027 		}
5028 	}
5029 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5030 	sfx_items++;
5031 	mtr_first = priv->sh->meter_aso_en &&
5032 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5033 	/* For ASO meter, meter must be before tag in TX direction. */
5034 	if (mtr_first) {
5035 		action_pre_head = actions_pre++;
5036 		/* Leave space for tag action. */
5037 		tag_action = actions_pre++;
5038 	}
5039 	/* Prepare the actions for prefix and suffix flow. */
5040 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5041 		struct rte_flow_action *action_cur = NULL;
5042 
5043 		switch (actions->type) {
5044 		case RTE_FLOW_ACTION_TYPE_METER:
5045 			if (mtr_first) {
5046 				action_cur = action_pre_head;
5047 			} else {
5048 				/* Leave space for tag action. */
5049 				tag_action = actions_pre++;
5050 				action_cur = actions_pre++;
5051 			}
5052 			break;
5053 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5054 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5055 			action_cur = actions_pre++;
5056 			break;
5057 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5058 			raw_encap = actions->conf;
5059 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5060 				action_cur = actions_pre++;
5061 			break;
5062 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5063 			raw_decap = actions->conf;
5064 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5065 				action_cur = actions_pre++;
5066 			break;
5067 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5068 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5069 			if (vlan_item_dst && vlan_item_src) {
5070 				memcpy(vlan_item_dst, vlan_item_src,
5071 					sizeof(*vlan_item_dst));
5072 				/*
5073 				 * Convert to internal match item, it is used
5074 				 * for vlan push and set vid.
5075 				 */
5076 				vlan_item_dst->type = (enum rte_flow_item_type)
5077 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5078 			}
5079 			break;
5080 		default:
5081 			break;
5082 		}
5083 		if (!action_cur)
5084 			action_cur = (fm->def_policy) ?
5085 					actions_sfx++ : actions_pre++;
5086 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5087 	}
5088 	/* Add end action to the actions. */
5089 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5090 	if (priv->sh->meter_aso_en) {
5091 		/**
5092 		 * For ASO meter, need to add an extra jump action explicitly,
5093 		 * to jump from meter to policer table.
5094 		 */
5095 		struct mlx5_flow_meter_sub_policy *sub_policy;
5096 		struct mlx5_flow_tbl_data_entry *tbl_data;
5097 
5098 		if (!fm->def_policy) {
5099 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5100 							  attr, items, error);
5101 			if (!sub_policy)
5102 				return -rte_errno;
5103 		} else {
5104 			enum mlx5_meter_domain mtr_domain =
5105 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5106 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5107 						MLX5_MTR_DOMAIN_INGRESS);
5108 
5109 			sub_policy =
5110 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5111 		}
5112 		tbl_data = container_of(sub_policy->tbl_rsc,
5113 					struct mlx5_flow_tbl_data_entry, tbl);
5114 		hw_mtr_action = actions_pre++;
5115 		hw_mtr_action->type = (enum rte_flow_action_type)
5116 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5117 		hw_mtr_action->conf = tbl_data->jump.action;
5118 	}
5119 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5120 	actions_pre++;
5121 	if (!tag_action)
5122 		return rte_flow_error_set(error, ENOMEM,
5123 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5124 					  NULL, "No tag action space.");
5125 	if (!mtr_flow_id) {
5126 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5127 		goto exit;
5128 	}
5129 	/* Only default-policy Meter creates mtr flow id. */
5130 	if (fm->def_policy) {
5131 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5132 		if (!tag_id)
5133 			return rte_flow_error_set(error, ENOMEM,
5134 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5135 					"Failed to allocate meter flow id.");
5136 		flow_id = tag_id - 1;
5137 		flow_id_bits = (!flow_id) ? 1 :
5138 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5139 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5140 		    mtr_reg_bits) {
5141 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5142 			return rte_flow_error_set(error, EINVAL,
5143 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5144 					"Meter flow id exceeds max limit.");
5145 		}
5146 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5147 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5148 	}
5149 	/* Build tag actions and items for meter_id/meter flow_id. */
5150 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5151 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5152 	tag_item_mask = tag_item_spec + 1;
5153 	/* Both flow_id and meter_id share the same register. */
5154 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5155 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5156 							    0, error),
5157 		.offset = mtr_id_offset,
5158 		.length = mtr_reg_bits,
5159 		.data = flow->meter,
5160 	};
5161 	/*
5162 	 * The color Reg bits used by flow_id are growing from
5163 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5164 	 */
5165 	for (shift = 0; shift < flow_id_bits; shift++)
5166 		flow_id_reversed = (flow_id_reversed << 1) |
5167 				((flow_id >> shift) & 0x1);
5168 	set_tag->data |=
5169 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5170 	tag_item_spec->id = set_tag->id;
5171 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5172 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5173 	tag_action->type = (enum rte_flow_action_type)
5174 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5175 	tag_action->conf = set_tag;
5176 	tag_item->type = (enum rte_flow_item_type)
5177 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5178 	tag_item->spec = tag_item_spec;
5179 	tag_item->last = NULL;
5180 	tag_item->mask = tag_item_mask;
5181 exit:
5182 	if (mtr_flow_id)
5183 		*mtr_flow_id = tag_id;
5184 	return 0;
5185 }
5186 
5187 /**
5188  * Split action list having QUEUE/RSS for metadata register copy.
5189  *
5190  * Once Q/RSS action is detected in user's action list, the flow action
5191  * should be split in order to copy metadata registers, which will happen in
5192  * RX_CP_TBL like,
5193  *   - CQE->flow_tag := reg_c[1] (MARK)
5194  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5195  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5196  * This is because the last action of each flow must be a terminal action
5197  * (QUEUE, RSS or DROP).
5198  *
5199  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5200  * stored and kept in the mlx5_flow structure per each sub_flow.
5201  *
5202  * The Q/RSS action is replaced with,
5203  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5204  * And the following JUMP action is added at the end,
5205  *   - JUMP, to RX_CP_TBL.
5206  *
5207  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5208  * flow_create_split_metadata() routine. The flow will look like,
5209  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5210  *
5211  * @param dev
5212  *   Pointer to Ethernet device.
5213  * @param[out] split_actions
5214  *   Pointer to store split actions to jump to CP_TBL.
5215  * @param[in] actions
5216  *   Pointer to the list of original flow actions.
5217  * @param[in] qrss
5218  *   Pointer to the Q/RSS action.
5219  * @param[in] actions_n
5220  *   Number of original actions.
5221  * @param[out] error
5222  *   Perform verbose error reporting if not NULL.
5223  *
5224  * @return
5225  *   non-zero unique flow_id on success, otherwise 0 and
5226  *   error/rte_error are set.
5227  */
5228 static uint32_t
5229 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5230 			  struct rte_flow_action *split_actions,
5231 			  const struct rte_flow_action *actions,
5232 			  const struct rte_flow_action *qrss,
5233 			  int actions_n, struct rte_flow_error *error)
5234 {
5235 	struct mlx5_priv *priv = dev->data->dev_private;
5236 	struct mlx5_rte_flow_action_set_tag *set_tag;
5237 	struct rte_flow_action_jump *jump;
5238 	const int qrss_idx = qrss - actions;
5239 	uint32_t flow_id = 0;
5240 	int ret = 0;
5241 
5242 	/*
5243 	 * Given actions will be split
5244 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5245 	 * - Add jump to mreg CP_TBL.
5246 	 * As a result, there will be one more action.
5247 	 */
5248 	++actions_n;
5249 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5250 	set_tag = (void *)(split_actions + actions_n);
5251 	/*
5252 	 * If tag action is not set to void(it means we are not the meter
5253 	 * suffix flow), add the tag action. Since meter suffix flow already
5254 	 * has the tag added.
5255 	 */
5256 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5257 		/*
5258 		 * Allocate the new subflow ID. This one is unique within
5259 		 * device and not shared with representors. Otherwise,
5260 		 * we would have to resolve multi-thread access synch
5261 		 * issue. Each flow on the shared device is appended
5262 		 * with source vport identifier, so the resulting
5263 		 * flows will be unique in the shared (by master and
5264 		 * representors) domain even if they have coinciding
5265 		 * IDs.
5266 		 */
5267 		mlx5_ipool_malloc(priv->sh->ipool
5268 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5269 		if (!flow_id)
5270 			return rte_flow_error_set(error, ENOMEM,
5271 						  RTE_FLOW_ERROR_TYPE_ACTION,
5272 						  NULL, "can't allocate id "
5273 						  "for split Q/RSS subflow");
5274 		/* Internal SET_TAG action to set flow ID. */
5275 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5276 			.data = flow_id,
5277 		};
5278 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5279 		if (ret < 0)
5280 			return ret;
5281 		set_tag->id = ret;
5282 		/* Construct new actions array. */
5283 		/* Replace QUEUE/RSS action. */
5284 		split_actions[qrss_idx] = (struct rte_flow_action){
5285 			.type = (enum rte_flow_action_type)
5286 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5287 			.conf = set_tag,
5288 		};
5289 	}
5290 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5291 	jump = (void *)(set_tag + 1);
5292 	*jump = (struct rte_flow_action_jump){
5293 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5294 	};
5295 	split_actions[actions_n - 2] = (struct rte_flow_action){
5296 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5297 		.conf = jump,
5298 	};
5299 	split_actions[actions_n - 1] = (struct rte_flow_action){
5300 		.type = RTE_FLOW_ACTION_TYPE_END,
5301 	};
5302 	return flow_id;
5303 }
5304 
5305 /**
5306  * Extend the given action list for Tx metadata copy.
5307  *
5308  * Copy the given action list to the ext_actions and add flow metadata register
5309  * copy action in order to copy reg_a set by WQE to reg_c[0].
5310  *
5311  * @param[out] ext_actions
5312  *   Pointer to the extended action list.
5313  * @param[in] actions
5314  *   Pointer to the list of actions.
5315  * @param[in] actions_n
5316  *   Number of actions in the list.
5317  * @param[out] error
5318  *   Perform verbose error reporting if not NULL.
5319  * @param[in] encap_idx
5320  *   The encap action inndex.
5321  *
5322  * @return
5323  *   0 on success, negative value otherwise
5324  */
5325 static int
5326 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5327 		       struct rte_flow_action *ext_actions,
5328 		       const struct rte_flow_action *actions,
5329 		       int actions_n, struct rte_flow_error *error,
5330 		       int encap_idx)
5331 {
5332 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5333 		(struct mlx5_flow_action_copy_mreg *)
5334 			(ext_actions + actions_n + 1);
5335 	int ret;
5336 
5337 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5338 	if (ret < 0)
5339 		return ret;
5340 	cp_mreg->dst = ret;
5341 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5342 	if (ret < 0)
5343 		return ret;
5344 	cp_mreg->src = ret;
5345 	if (encap_idx != 0)
5346 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5347 	if (encap_idx == actions_n - 1) {
5348 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5349 			.type = (enum rte_flow_action_type)
5350 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5351 			.conf = cp_mreg,
5352 		};
5353 		ext_actions[actions_n] = (struct rte_flow_action){
5354 			.type = RTE_FLOW_ACTION_TYPE_END,
5355 		};
5356 	} else {
5357 		ext_actions[encap_idx] = (struct rte_flow_action){
5358 			.type = (enum rte_flow_action_type)
5359 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5360 			.conf = cp_mreg,
5361 		};
5362 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5363 				sizeof(*ext_actions) * (actions_n - encap_idx));
5364 	}
5365 	return 0;
5366 }
5367 
5368 /**
5369  * Check the match action from the action list.
5370  *
5371  * @param[in] actions
5372  *   Pointer to the list of actions.
5373  * @param[in] attr
5374  *   Flow rule attributes.
5375  * @param[in] action
5376  *   The action to be check if exist.
5377  * @param[out] match_action_pos
5378  *   Pointer to the position of the matched action if exists, otherwise is -1.
5379  * @param[out] qrss_action_pos
5380  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5381  * @param[out] modify_after_mirror
5382  *   Pointer to the flag of modify action after FDB mirroring.
5383  *
5384  * @return
5385  *   > 0 the total number of actions.
5386  *   0 if not found match action in action list.
5387  */
5388 static int
5389 flow_check_match_action(const struct rte_flow_action actions[],
5390 			const struct rte_flow_attr *attr,
5391 			enum rte_flow_action_type action,
5392 			int *match_action_pos, int *qrss_action_pos,
5393 			int *modify_after_mirror)
5394 {
5395 	const struct rte_flow_action_sample *sample;
5396 	const struct rte_flow_action_raw_decap *decap;
5397 	int actions_n = 0;
5398 	uint32_t ratio = 0;
5399 	int sub_type = 0;
5400 	int flag = 0;
5401 	int fdb_mirror = 0;
5402 
5403 	*match_action_pos = -1;
5404 	*qrss_action_pos = -1;
5405 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5406 		if (actions->type == action) {
5407 			flag = 1;
5408 			*match_action_pos = actions_n;
5409 		}
5410 		switch (actions->type) {
5411 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5412 		case RTE_FLOW_ACTION_TYPE_RSS:
5413 			*qrss_action_pos = actions_n;
5414 			break;
5415 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5416 			sample = actions->conf;
5417 			ratio = sample->ratio;
5418 			sub_type = ((const struct rte_flow_action *)
5419 					(sample->actions))->type;
5420 			if (ratio == 1 && attr->transfer)
5421 				fdb_mirror = 1;
5422 			break;
5423 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5424 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5425 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5426 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5427 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5428 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5429 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5430 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5431 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5432 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5433 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5434 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5435 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5436 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5437 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5438 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5439 		case RTE_FLOW_ACTION_TYPE_FLAG:
5440 		case RTE_FLOW_ACTION_TYPE_MARK:
5441 		case RTE_FLOW_ACTION_TYPE_SET_META:
5442 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5443 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5444 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5445 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5446 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5447 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5448 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5449 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5450 		case RTE_FLOW_ACTION_TYPE_METER:
5451 			if (fdb_mirror)
5452 				*modify_after_mirror = 1;
5453 			break;
5454 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5455 			decap = actions->conf;
5456 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5457 				;
5458 			actions_n++;
5459 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5460 				const struct rte_flow_action_raw_encap *encap =
5461 								actions->conf;
5462 				if (decap->size <=
5463 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5464 				    encap->size >
5465 					MLX5_ENCAPSULATION_DECISION_SIZE)
5466 					/* L3 encap. */
5467 					break;
5468 			}
5469 			if (fdb_mirror)
5470 				*modify_after_mirror = 1;
5471 			break;
5472 		default:
5473 			break;
5474 		}
5475 		actions_n++;
5476 	}
5477 	if (flag && fdb_mirror && !*modify_after_mirror) {
5478 		/* FDB mirroring uses the destination array to implement
5479 		 * instead of FLOW_SAMPLER object.
5480 		 */
5481 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5482 			flag = 0;
5483 	}
5484 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5485 	return flag ? actions_n + 1 : 0;
5486 }
5487 
5488 #define SAMPLE_SUFFIX_ITEM 2
5489 
5490 /**
5491  * Split the sample flow.
5492  *
5493  * As sample flow will split to two sub flow, sample flow with
5494  * sample action, the other actions will move to new suffix flow.
5495  *
5496  * Also add unique tag id with tag action in the sample flow,
5497  * the same tag id will be as match in the suffix flow.
5498  *
5499  * @param dev
5500  *   Pointer to Ethernet device.
5501  * @param[in] add_tag
5502  *   Add extra tag action flag.
5503  * @param[out] sfx_items
5504  *   Suffix flow match items (list terminated by the END pattern item).
5505  * @param[in] actions
5506  *   Associated actions (list terminated by the END action).
5507  * @param[out] actions_sfx
5508  *   Suffix flow actions.
5509  * @param[out] actions_pre
5510  *   Prefix flow actions.
5511  * @param[in] actions_n
5512  *  The total number of actions.
5513  * @param[in] sample_action_pos
5514  *   The sample action position.
5515  * @param[in] qrss_action_pos
5516  *   The Queue/RSS action position.
5517  * @param[in] jump_table
5518  *   Add extra jump action flag.
5519  * @param[out] error
5520  *   Perform verbose error reporting if not NULL.
5521  *
5522  * @return
5523  *   0 on success, or unique flow_id, a negative errno value
5524  *   otherwise and rte_errno is set.
5525  */
5526 static int
5527 flow_sample_split_prep(struct rte_eth_dev *dev,
5528 		       int add_tag,
5529 		       struct rte_flow_item sfx_items[],
5530 		       const struct rte_flow_action actions[],
5531 		       struct rte_flow_action actions_sfx[],
5532 		       struct rte_flow_action actions_pre[],
5533 		       int actions_n,
5534 		       int sample_action_pos,
5535 		       int qrss_action_pos,
5536 		       int jump_table,
5537 		       struct rte_flow_error *error)
5538 {
5539 	struct mlx5_priv *priv = dev->data->dev_private;
5540 	struct mlx5_rte_flow_action_set_tag *set_tag;
5541 	struct mlx5_rte_flow_item_tag *tag_spec;
5542 	struct mlx5_rte_flow_item_tag *tag_mask;
5543 	struct rte_flow_action_jump *jump_action;
5544 	uint32_t tag_id = 0;
5545 	int index;
5546 	int append_index = 0;
5547 	int ret;
5548 
5549 	if (sample_action_pos < 0)
5550 		return rte_flow_error_set(error, EINVAL,
5551 					  RTE_FLOW_ERROR_TYPE_ACTION,
5552 					  NULL, "invalid position of sample "
5553 					  "action in list");
5554 	/* Prepare the actions for prefix and suffix flow. */
5555 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5556 		index = qrss_action_pos;
5557 		/* Put the preceding the Queue/RSS action into prefix flow. */
5558 		if (index != 0)
5559 			memcpy(actions_pre, actions,
5560 			       sizeof(struct rte_flow_action) * index);
5561 		/* Put others preceding the sample action into prefix flow. */
5562 		if (sample_action_pos > index + 1)
5563 			memcpy(actions_pre + index, actions + index + 1,
5564 			       sizeof(struct rte_flow_action) *
5565 			       (sample_action_pos - index - 1));
5566 		index = sample_action_pos - 1;
5567 		/* Put Queue/RSS action into Suffix flow. */
5568 		memcpy(actions_sfx, actions + qrss_action_pos,
5569 		       sizeof(struct rte_flow_action));
5570 		actions_sfx++;
5571 	} else {
5572 		index = sample_action_pos;
5573 		if (index != 0)
5574 			memcpy(actions_pre, actions,
5575 			       sizeof(struct rte_flow_action) * index);
5576 	}
5577 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5578 	 * For CX6DX and above, metadata registers Cx preserve their value,
5579 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5580 	 */
5581 	if (add_tag) {
5582 		/* Prepare the prefix tag action. */
5583 		append_index++;
5584 		set_tag = (void *)(actions_pre + actions_n + append_index);
5585 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5586 		if (ret < 0)
5587 			return ret;
5588 		mlx5_ipool_malloc(priv->sh->ipool
5589 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5590 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5591 			.id = ret,
5592 			.data = tag_id,
5593 		};
5594 		/* Prepare the suffix subflow items. */
5595 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5596 		tag_spec->data = tag_id;
5597 		tag_spec->id = set_tag->id;
5598 		tag_mask = tag_spec + 1;
5599 		tag_mask->data = UINT32_MAX;
5600 		sfx_items[0] = (struct rte_flow_item){
5601 			.type = (enum rte_flow_item_type)
5602 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5603 			.spec = tag_spec,
5604 			.last = NULL,
5605 			.mask = tag_mask,
5606 		};
5607 		sfx_items[1] = (struct rte_flow_item){
5608 			.type = (enum rte_flow_item_type)
5609 				RTE_FLOW_ITEM_TYPE_END,
5610 		};
5611 		/* Prepare the tag action in prefix subflow. */
5612 		actions_pre[index++] =
5613 			(struct rte_flow_action){
5614 			.type = (enum rte_flow_action_type)
5615 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5616 			.conf = set_tag,
5617 		};
5618 	}
5619 	memcpy(actions_pre + index, actions + sample_action_pos,
5620 	       sizeof(struct rte_flow_action));
5621 	index += 1;
5622 	/* For the modify action after the sample action in E-Switch mirroring,
5623 	 * Add the extra jump action in prefix subflow and jump into the next
5624 	 * table, then do the modify action in the new table.
5625 	 */
5626 	if (jump_table) {
5627 		/* Prepare the prefix jump action. */
5628 		append_index++;
5629 		jump_action = (void *)(actions_pre + actions_n + append_index);
5630 		jump_action->group = jump_table;
5631 		actions_pre[index++] =
5632 			(struct rte_flow_action){
5633 			.type = (enum rte_flow_action_type)
5634 				RTE_FLOW_ACTION_TYPE_JUMP,
5635 			.conf = jump_action,
5636 		};
5637 	}
5638 	actions_pre[index] = (struct rte_flow_action){
5639 		.type = (enum rte_flow_action_type)
5640 			RTE_FLOW_ACTION_TYPE_END,
5641 	};
5642 	/* Put the actions after sample into Suffix flow. */
5643 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5644 	       sizeof(struct rte_flow_action) *
5645 	       (actions_n - sample_action_pos - 1));
5646 	return tag_id;
5647 }
5648 
5649 /**
5650  * The splitting for metadata feature.
5651  *
5652  * - Q/RSS action on NIC Rx should be split in order to pass by
5653  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5654  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5655  *
5656  * - All the actions on NIC Tx should have a mreg copy action to
5657  *   copy reg_a from WQE to reg_c[0].
5658  *
5659  * @param dev
5660  *   Pointer to Ethernet device.
5661  * @param[in] flow
5662  *   Parent flow structure pointer.
5663  * @param[in] attr
5664  *   Flow rule attributes.
5665  * @param[in] items
5666  *   Pattern specification (list terminated by the END pattern item).
5667  * @param[in] actions
5668  *   Associated actions (list terminated by the END action).
5669  * @param[in] flow_split_info
5670  *   Pointer to flow split info structure.
5671  * @param[out] error
5672  *   Perform verbose error reporting if not NULL.
5673  * @return
5674  *   0 on success, negative value otherwise
5675  */
5676 static int
5677 flow_create_split_metadata(struct rte_eth_dev *dev,
5678 			   struct rte_flow *flow,
5679 			   const struct rte_flow_attr *attr,
5680 			   const struct rte_flow_item items[],
5681 			   const struct rte_flow_action actions[],
5682 			   struct mlx5_flow_split_info *flow_split_info,
5683 			   struct rte_flow_error *error)
5684 {
5685 	struct mlx5_priv *priv = dev->data->dev_private;
5686 	struct mlx5_dev_config *config = &priv->config;
5687 	const struct rte_flow_action *qrss = NULL;
5688 	struct rte_flow_action *ext_actions = NULL;
5689 	struct mlx5_flow *dev_flow = NULL;
5690 	uint32_t qrss_id = 0;
5691 	int mtr_sfx = 0;
5692 	size_t act_size;
5693 	int actions_n;
5694 	int encap_idx;
5695 	int ret;
5696 
5697 	/* Check whether extensive metadata feature is engaged. */
5698 	if (!config->dv_flow_en ||
5699 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5700 	    !mlx5_flow_ext_mreg_supported(dev))
5701 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5702 					       actions, flow_split_info, error);
5703 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5704 							   &encap_idx);
5705 	if (qrss) {
5706 		/* Exclude hairpin flows from splitting. */
5707 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5708 			const struct rte_flow_action_queue *queue;
5709 
5710 			queue = qrss->conf;
5711 			if (mlx5_rxq_get_type(dev, queue->index) ==
5712 			    MLX5_RXQ_TYPE_HAIRPIN)
5713 				qrss = NULL;
5714 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5715 			const struct rte_flow_action_rss *rss;
5716 
5717 			rss = qrss->conf;
5718 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5719 			    MLX5_RXQ_TYPE_HAIRPIN)
5720 				qrss = NULL;
5721 		}
5722 	}
5723 	if (qrss) {
5724 		/* Check if it is in meter suffix table. */
5725 		mtr_sfx = attr->group == (attr->transfer ?
5726 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5727 			  MLX5_FLOW_TABLE_LEVEL_METER);
5728 		/*
5729 		 * Q/RSS action on NIC Rx should be split in order to pass by
5730 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5731 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5732 		 */
5733 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5734 			   sizeof(struct rte_flow_action_set_tag) +
5735 			   sizeof(struct rte_flow_action_jump);
5736 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5737 					  SOCKET_ID_ANY);
5738 		if (!ext_actions)
5739 			return rte_flow_error_set(error, ENOMEM,
5740 						  RTE_FLOW_ERROR_TYPE_ACTION,
5741 						  NULL, "no memory to split "
5742 						  "metadata flow");
5743 		/*
5744 		 * If we are the suffix flow of meter, tag already exist.
5745 		 * Set the tag action to void.
5746 		 */
5747 		if (mtr_sfx)
5748 			ext_actions[qrss - actions].type =
5749 						RTE_FLOW_ACTION_TYPE_VOID;
5750 		else
5751 			ext_actions[qrss - actions].type =
5752 						(enum rte_flow_action_type)
5753 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5754 		/*
5755 		 * Create the new actions list with removed Q/RSS action
5756 		 * and appended set tag and jump to register copy table
5757 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5758 		 * in advance, because it is needed for set tag action.
5759 		 */
5760 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5761 						    qrss, actions_n, error);
5762 		if (!mtr_sfx && !qrss_id) {
5763 			ret = -rte_errno;
5764 			goto exit;
5765 		}
5766 	} else if (attr->egress && !attr->transfer) {
5767 		/*
5768 		 * All the actions on NIC Tx should have a metadata register
5769 		 * copy action to copy reg_a from WQE to reg_c[meta]
5770 		 */
5771 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5772 			   sizeof(struct mlx5_flow_action_copy_mreg);
5773 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5774 					  SOCKET_ID_ANY);
5775 		if (!ext_actions)
5776 			return rte_flow_error_set(error, ENOMEM,
5777 						  RTE_FLOW_ERROR_TYPE_ACTION,
5778 						  NULL, "no memory to split "
5779 						  "metadata flow");
5780 		/* Create the action list appended with copy register. */
5781 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5782 					     actions_n, error, encap_idx);
5783 		if (ret < 0)
5784 			goto exit;
5785 	}
5786 	/* Add the unmodified original or prefix subflow. */
5787 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5788 				      items, ext_actions ? ext_actions :
5789 				      actions, flow_split_info, error);
5790 	if (ret < 0)
5791 		goto exit;
5792 	MLX5_ASSERT(dev_flow);
5793 	if (qrss) {
5794 		const struct rte_flow_attr q_attr = {
5795 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5796 			.ingress = 1,
5797 		};
5798 		/* Internal PMD action to set register. */
5799 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5800 			.data = qrss_id,
5801 			.id = REG_NON,
5802 		};
5803 		struct rte_flow_item q_items[] = {
5804 			{
5805 				.type = (enum rte_flow_item_type)
5806 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5807 				.spec = &q_tag_spec,
5808 				.last = NULL,
5809 				.mask = NULL,
5810 			},
5811 			{
5812 				.type = RTE_FLOW_ITEM_TYPE_END,
5813 			},
5814 		};
5815 		struct rte_flow_action q_actions[] = {
5816 			{
5817 				.type = qrss->type,
5818 				.conf = qrss->conf,
5819 			},
5820 			{
5821 				.type = RTE_FLOW_ACTION_TYPE_END,
5822 			},
5823 		};
5824 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5825 
5826 		/*
5827 		 * Configure the tag item only if there is no meter subflow.
5828 		 * Since tag is already marked in the meter suffix subflow
5829 		 * we can just use the meter suffix items as is.
5830 		 */
5831 		if (qrss_id) {
5832 			/* Not meter subflow. */
5833 			MLX5_ASSERT(!mtr_sfx);
5834 			/*
5835 			 * Put unique id in prefix flow due to it is destroyed
5836 			 * after suffix flow and id will be freed after there
5837 			 * is no actual flows with this id and identifier
5838 			 * reallocation becomes possible (for example, for
5839 			 * other flows in other threads).
5840 			 */
5841 			dev_flow->handle->split_flow_id = qrss_id;
5842 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5843 						   error);
5844 			if (ret < 0)
5845 				goto exit;
5846 			q_tag_spec.id = ret;
5847 		}
5848 		dev_flow = NULL;
5849 		/* Add suffix subflow to execute Q/RSS. */
5850 		flow_split_info->prefix_layers = layers;
5851 		flow_split_info->prefix_mark = 0;
5852 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5853 					      &q_attr, mtr_sfx ? items :
5854 					      q_items, q_actions,
5855 					      flow_split_info, error);
5856 		if (ret < 0)
5857 			goto exit;
5858 		/* qrss ID should be freed if failed. */
5859 		qrss_id = 0;
5860 		MLX5_ASSERT(dev_flow);
5861 	}
5862 
5863 exit:
5864 	/*
5865 	 * We do not destroy the partially created sub_flows in case of error.
5866 	 * These ones are included into parent flow list and will be destroyed
5867 	 * by flow_drv_destroy.
5868 	 */
5869 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5870 			qrss_id);
5871 	mlx5_free(ext_actions);
5872 	return ret;
5873 }
5874 
5875 /**
5876  * Create meter internal drop flow with the original pattern.
5877  *
5878  * @param dev
5879  *   Pointer to Ethernet device.
5880  * @param[in] flow
5881  *   Parent flow structure pointer.
5882  * @param[in] attr
5883  *   Flow rule attributes.
5884  * @param[in] items
5885  *   Pattern specification (list terminated by the END pattern item).
5886  * @param[in] flow_split_info
5887  *   Pointer to flow split info structure.
5888  * @param[in] fm
5889  *   Pointer to flow meter structure.
5890  * @param[out] error
5891  *   Perform verbose error reporting if not NULL.
5892  * @return
5893  *   0 on success, negative value otherwise
5894  */
5895 static uint32_t
5896 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5897 			struct rte_flow *flow,
5898 			const struct rte_flow_attr *attr,
5899 			const struct rte_flow_item items[],
5900 			struct mlx5_flow_split_info *flow_split_info,
5901 			struct mlx5_flow_meter_info *fm,
5902 			struct rte_flow_error *error)
5903 {
5904 	struct mlx5_flow *dev_flow = NULL;
5905 	struct rte_flow_attr drop_attr = *attr;
5906 	struct rte_flow_action drop_actions[3];
5907 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5908 
5909 	MLX5_ASSERT(fm->drop_cnt);
5910 	drop_actions[0].type =
5911 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5912 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5913 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5914 	drop_actions[1].conf = NULL;
5915 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5916 	drop_actions[2].conf = NULL;
5917 	drop_split_info.external = false;
5918 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5919 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5920 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5921 	return flow_create_split_inner(dev, flow, &dev_flow,
5922 				&drop_attr, items, drop_actions,
5923 				&drop_split_info, error);
5924 }
5925 
5926 /**
5927  * The splitting for meter feature.
5928  *
5929  * - The meter flow will be split to two flows as prefix and
5930  *   suffix flow. The packets make sense only it pass the prefix
5931  *   meter action.
5932  *
5933  * - Reg_C_5 is used for the packet to match betweend prefix and
5934  *   suffix flow.
5935  *
5936  * @param dev
5937  *   Pointer to Ethernet device.
5938  * @param[in] flow
5939  *   Parent flow structure pointer.
5940  * @param[in] attr
5941  *   Flow rule attributes.
5942  * @param[in] items
5943  *   Pattern specification (list terminated by the END pattern item).
5944  * @param[in] actions
5945  *   Associated actions (list terminated by the END action).
5946  * @param[in] flow_split_info
5947  *   Pointer to flow split info structure.
5948  * @param[out] error
5949  *   Perform verbose error reporting if not NULL.
5950  * @return
5951  *   0 on success, negative value otherwise
5952  */
5953 static int
5954 flow_create_split_meter(struct rte_eth_dev *dev,
5955 			struct rte_flow *flow,
5956 			const struct rte_flow_attr *attr,
5957 			const struct rte_flow_item items[],
5958 			const struct rte_flow_action actions[],
5959 			struct mlx5_flow_split_info *flow_split_info,
5960 			struct rte_flow_error *error)
5961 {
5962 	struct mlx5_priv *priv = dev->data->dev_private;
5963 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5964 	struct rte_flow_action *sfx_actions = NULL;
5965 	struct rte_flow_action *pre_actions = NULL;
5966 	struct rte_flow_item *sfx_items = NULL;
5967 	struct mlx5_flow *dev_flow = NULL;
5968 	struct rte_flow_attr sfx_attr = *attr;
5969 	struct mlx5_flow_meter_info *fm = NULL;
5970 	uint8_t skip_scale_restore;
5971 	bool has_mtr = false;
5972 	bool has_modify = false;
5973 	bool set_mtr_reg = true;
5974 	bool is_mtr_hierarchy = false;
5975 	uint32_t meter_id = 0;
5976 	uint32_t mtr_idx = 0;
5977 	uint32_t mtr_flow_id = 0;
5978 	size_t act_size;
5979 	size_t item_size;
5980 	int actions_n = 0;
5981 	int ret = 0;
5982 
5983 	if (priv->mtr_en)
5984 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5985 						    &has_modify, &meter_id);
5986 	if (has_mtr) {
5987 		if (flow->meter) {
5988 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5989 			if (!fm)
5990 				return rte_flow_error_set(error, EINVAL,
5991 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5992 						NULL, "Meter not found.");
5993 		} else {
5994 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5995 			if (!fm)
5996 				return rte_flow_error_set(error, EINVAL,
5997 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5998 						NULL, "Meter not found.");
5999 			ret = mlx5_flow_meter_attach(priv, fm,
6000 						     &sfx_attr, error);
6001 			if (ret)
6002 				return -rte_errno;
6003 			flow->meter = mtr_idx;
6004 		}
6005 		MLX5_ASSERT(wks);
6006 		wks->fm = fm;
6007 		if (!fm->def_policy) {
6008 			wks->policy = mlx5_flow_meter_policy_find(dev,
6009 								  fm->policy_id,
6010 								  NULL);
6011 			MLX5_ASSERT(wks->policy);
6012 			if (wks->policy->is_hierarchy) {
6013 				wks->final_policy =
6014 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6015 								wks->policy);
6016 				if (!wks->final_policy)
6017 					return rte_flow_error_set(error,
6018 					EINVAL,
6019 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6020 				"Failed to find terminal policy of hierarchy.");
6021 				is_mtr_hierarchy = true;
6022 			}
6023 		}
6024 		/*
6025 		 * If it isn't default-policy Meter, and
6026 		 * 1. There's no action in flow to change
6027 		 *    packet (modify/encap/decap etc.), OR
6028 		 * 2. No drop count needed for this meter.
6029 		 * 3. It's not meter hierarchy.
6030 		 * Then no need to use regC to save meter id anymore.
6031 		 */
6032 		if (!fm->def_policy && !is_mtr_hierarchy &&
6033 		    (!has_modify || !fm->drop_cnt))
6034 			set_mtr_reg = false;
6035 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6036 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6037 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6038 		/* Suffix items: tag, vlan, port id, end. */
6039 #define METER_SUFFIX_ITEM 4
6040 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6041 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6042 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6043 					  0, SOCKET_ID_ANY);
6044 		if (!sfx_actions)
6045 			return rte_flow_error_set(error, ENOMEM,
6046 						  RTE_FLOW_ERROR_TYPE_ACTION,
6047 						  NULL, "no memory to split "
6048 						  "meter flow");
6049 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6050 			     act_size);
6051 		/* There's no suffix flow for meter of non-default policy. */
6052 		if (!fm->def_policy)
6053 			pre_actions = sfx_actions + 1;
6054 		else
6055 			pre_actions = sfx_actions + actions_n;
6056 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6057 					    items, sfx_items, actions,
6058 					    sfx_actions, pre_actions,
6059 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6060 					    error);
6061 		if (ret) {
6062 			ret = -rte_errno;
6063 			goto exit;
6064 		}
6065 		/* Add the prefix subflow. */
6066 		flow_split_info->prefix_mark = 0;
6067 		skip_scale_restore = flow_split_info->skip_scale;
6068 		flow_split_info->skip_scale |=
6069 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6070 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6071 					      attr, items, pre_actions,
6072 					      flow_split_info, error);
6073 		flow_split_info->skip_scale = skip_scale_restore;
6074 		if (ret) {
6075 			if (mtr_flow_id)
6076 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6077 			ret = -rte_errno;
6078 			goto exit;
6079 		}
6080 		if (mtr_flow_id) {
6081 			dev_flow->handle->split_flow_id = mtr_flow_id;
6082 			dev_flow->handle->is_meter_flow_id = 1;
6083 		}
6084 		if (!fm->def_policy) {
6085 			if (!set_mtr_reg && fm->drop_cnt)
6086 				ret =
6087 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6088 							&sfx_attr, items,
6089 							flow_split_info,
6090 							fm, error);
6091 			goto exit;
6092 		}
6093 		/* Setting the sfx group atrr. */
6094 		sfx_attr.group = sfx_attr.transfer ?
6095 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6096 				 MLX5_FLOW_TABLE_LEVEL_METER;
6097 		flow_split_info->prefix_layers =
6098 				flow_get_prefix_layer_flags(dev_flow);
6099 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6100 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6101 	}
6102 	/* Add the prefix subflow. */
6103 	ret = flow_create_split_metadata(dev, flow,
6104 					 &sfx_attr, sfx_items ?
6105 					 sfx_items : items,
6106 					 sfx_actions ? sfx_actions : actions,
6107 					 flow_split_info, error);
6108 exit:
6109 	if (sfx_actions)
6110 		mlx5_free(sfx_actions);
6111 	return ret;
6112 }
6113 
6114 /**
6115  * The splitting for sample feature.
6116  *
6117  * Once Sample action is detected in the action list, the flow actions should
6118  * be split into prefix sub flow and suffix sub flow.
6119  *
6120  * The original items remain in the prefix sub flow, all actions preceding the
6121  * sample action and the sample action itself will be copied to the prefix
6122  * sub flow, the actions following the sample action will be copied to the
6123  * suffix sub flow, Queue action always be located in the suffix sub flow.
6124  *
6125  * In order to make the packet from prefix sub flow matches with suffix sub
6126  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6127  * flow uses tag item with the unique flow id.
6128  *
6129  * @param dev
6130  *   Pointer to Ethernet device.
6131  * @param[in] flow
6132  *   Parent flow structure pointer.
6133  * @param[in] attr
6134  *   Flow rule attributes.
6135  * @param[in] items
6136  *   Pattern specification (list terminated by the END pattern item).
6137  * @param[in] actions
6138  *   Associated actions (list terminated by the END action).
6139  * @param[in] flow_split_info
6140  *   Pointer to flow split info structure.
6141  * @param[out] error
6142  *   Perform verbose error reporting if not NULL.
6143  * @return
6144  *   0 on success, negative value otherwise
6145  */
6146 static int
6147 flow_create_split_sample(struct rte_eth_dev *dev,
6148 			 struct rte_flow *flow,
6149 			 const struct rte_flow_attr *attr,
6150 			 const struct rte_flow_item items[],
6151 			 const struct rte_flow_action actions[],
6152 			 struct mlx5_flow_split_info *flow_split_info,
6153 			 struct rte_flow_error *error)
6154 {
6155 	struct mlx5_priv *priv = dev->data->dev_private;
6156 	struct rte_flow_action *sfx_actions = NULL;
6157 	struct rte_flow_action *pre_actions = NULL;
6158 	struct rte_flow_item *sfx_items = NULL;
6159 	struct mlx5_flow *dev_flow = NULL;
6160 	struct rte_flow_attr sfx_attr = *attr;
6161 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6162 	struct mlx5_flow_dv_sample_resource *sample_res;
6163 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6164 	struct mlx5_flow_tbl_resource *sfx_tbl;
6165 #endif
6166 	size_t act_size;
6167 	size_t item_size;
6168 	uint32_t fdb_tx = 0;
6169 	int32_t tag_id = 0;
6170 	int actions_n = 0;
6171 	int sample_action_pos;
6172 	int qrss_action_pos;
6173 	int add_tag = 0;
6174 	int modify_after_mirror = 0;
6175 	uint16_t jump_table = 0;
6176 	const uint32_t next_ft_step = 1;
6177 	int ret = 0;
6178 
6179 	if (priv->sampler_en)
6180 		actions_n = flow_check_match_action(actions, attr,
6181 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6182 					&sample_action_pos, &qrss_action_pos,
6183 					&modify_after_mirror);
6184 	if (actions_n) {
6185 		/* The prefix actions must includes sample, tag, end. */
6186 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6187 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6188 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6189 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6190 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6191 					  item_size), 0, SOCKET_ID_ANY);
6192 		if (!sfx_actions)
6193 			return rte_flow_error_set(error, ENOMEM,
6194 						  RTE_FLOW_ERROR_TYPE_ACTION,
6195 						  NULL, "no memory to split "
6196 						  "sample flow");
6197 		/* The representor_id is UINT16_MAX for uplink. */
6198 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6199 		/*
6200 		 * When reg_c_preserve is set, metadata registers Cx preserve
6201 		 * their value even through packet duplication.
6202 		 */
6203 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6204 		if (add_tag)
6205 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6206 					+ act_size);
6207 		if (modify_after_mirror)
6208 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6209 				     next_ft_step;
6210 		pre_actions = sfx_actions + actions_n;
6211 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6212 						actions, sfx_actions,
6213 						pre_actions, actions_n,
6214 						sample_action_pos,
6215 						qrss_action_pos, jump_table,
6216 						error);
6217 		if (tag_id < 0 || (add_tag && !tag_id)) {
6218 			ret = -rte_errno;
6219 			goto exit;
6220 		}
6221 		if (modify_after_mirror)
6222 			flow_split_info->skip_scale =
6223 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6224 		/* Add the prefix subflow. */
6225 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6226 					      items, pre_actions,
6227 					      flow_split_info, error);
6228 		if (ret) {
6229 			ret = -rte_errno;
6230 			goto exit;
6231 		}
6232 		dev_flow->handle->split_flow_id = tag_id;
6233 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6234 		if (!modify_after_mirror) {
6235 			/* Set the sfx group attr. */
6236 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6237 						dev_flow->dv.sample_res;
6238 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6239 						sample_res->normal_path_tbl;
6240 			sfx_tbl_data = container_of(sfx_tbl,
6241 						struct mlx5_flow_tbl_data_entry,
6242 						tbl);
6243 			sfx_attr.group = sfx_attr.transfer ?
6244 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6245 		} else {
6246 			MLX5_ASSERT(attr->transfer);
6247 			sfx_attr.group = jump_table;
6248 		}
6249 		flow_split_info->prefix_layers =
6250 				flow_get_prefix_layer_flags(dev_flow);
6251 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6252 		/* Suffix group level already be scaled with factor, set
6253 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6254 		 * again in translation.
6255 		 */
6256 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6257 #endif
6258 	}
6259 	/* Add the suffix subflow. */
6260 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6261 				      sfx_items ? sfx_items : items,
6262 				      sfx_actions ? sfx_actions : actions,
6263 				      flow_split_info, error);
6264 exit:
6265 	if (sfx_actions)
6266 		mlx5_free(sfx_actions);
6267 	return ret;
6268 }
6269 
6270 /**
6271  * Split the flow to subflow set. The splitters might be linked
6272  * in the chain, like this:
6273  * flow_create_split_outer() calls:
6274  *   flow_create_split_meter() calls:
6275  *     flow_create_split_metadata(meter_subflow_0) calls:
6276  *       flow_create_split_inner(metadata_subflow_0)
6277  *       flow_create_split_inner(metadata_subflow_1)
6278  *       flow_create_split_inner(metadata_subflow_2)
6279  *     flow_create_split_metadata(meter_subflow_1) calls:
6280  *       flow_create_split_inner(metadata_subflow_0)
6281  *       flow_create_split_inner(metadata_subflow_1)
6282  *       flow_create_split_inner(metadata_subflow_2)
6283  *
6284  * This provide flexible way to add new levels of flow splitting.
6285  * The all of successfully created subflows are included to the
6286  * parent flow dev_flow list.
6287  *
6288  * @param dev
6289  *   Pointer to Ethernet device.
6290  * @param[in] flow
6291  *   Parent flow structure pointer.
6292  * @param[in] attr
6293  *   Flow rule attributes.
6294  * @param[in] items
6295  *   Pattern specification (list terminated by the END pattern item).
6296  * @param[in] actions
6297  *   Associated actions (list terminated by the END action).
6298  * @param[in] flow_split_info
6299  *   Pointer to flow split info structure.
6300  * @param[out] error
6301  *   Perform verbose error reporting if not NULL.
6302  * @return
6303  *   0 on success, negative value otherwise
6304  */
6305 static int
6306 flow_create_split_outer(struct rte_eth_dev *dev,
6307 			struct rte_flow *flow,
6308 			const struct rte_flow_attr *attr,
6309 			const struct rte_flow_item items[],
6310 			const struct rte_flow_action actions[],
6311 			struct mlx5_flow_split_info *flow_split_info,
6312 			struct rte_flow_error *error)
6313 {
6314 	int ret;
6315 
6316 	ret = flow_create_split_sample(dev, flow, attr, items,
6317 				       actions, flow_split_info, error);
6318 	MLX5_ASSERT(ret <= 0);
6319 	return ret;
6320 }
6321 
6322 static inline struct mlx5_flow_tunnel *
6323 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6324 {
6325 	struct mlx5_flow_tunnel *tunnel;
6326 
6327 #pragma GCC diagnostic push
6328 #pragma GCC diagnostic ignored "-Wcast-qual"
6329 	tunnel = (typeof(tunnel))flow->tunnel;
6330 #pragma GCC diagnostic pop
6331 
6332 	return tunnel;
6333 }
6334 
6335 /**
6336  * Adjust flow RSS workspace if needed.
6337  *
6338  * @param wks
6339  *   Pointer to thread flow work space.
6340  * @param rss_desc
6341  *   Pointer to RSS descriptor.
6342  * @param[in] nrssq_num
6343  *   New RSS queue number.
6344  *
6345  * @return
6346  *   0 on success, -1 otherwise and rte_errno is set.
6347  */
6348 static int
6349 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6350 			  struct mlx5_flow_rss_desc *rss_desc,
6351 			  uint32_t nrssq_num)
6352 {
6353 	if (likely(nrssq_num <= wks->rssq_num))
6354 		return 0;
6355 	rss_desc->queue = realloc(rss_desc->queue,
6356 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6357 	if (!rss_desc->queue) {
6358 		rte_errno = ENOMEM;
6359 		return -1;
6360 	}
6361 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6362 	return 0;
6363 }
6364 
6365 /**
6366  * Create a flow and add it to @p list.
6367  *
6368  * @param dev
6369  *   Pointer to Ethernet device.
6370  * @param list
6371  *   Pointer to a TAILQ flow list. If this parameter NULL,
6372  *   no list insertion occurred, flow is just created,
6373  *   this is caller's responsibility to track the
6374  *   created flow.
6375  * @param[in] attr
6376  *   Flow rule attributes.
6377  * @param[in] items
6378  *   Pattern specification (list terminated by the END pattern item).
6379  * @param[in] actions
6380  *   Associated actions (list terminated by the END action).
6381  * @param[in] external
6382  *   This flow rule is created by request external to PMD.
6383  * @param[out] error
6384  *   Perform verbose error reporting if not NULL.
6385  *
6386  * @return
6387  *   A flow index on success, 0 otherwise and rte_errno is set.
6388  */
6389 static uint32_t
6390 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6391 		 const struct rte_flow_attr *attr,
6392 		 const struct rte_flow_item items[],
6393 		 const struct rte_flow_action original_actions[],
6394 		 bool external, struct rte_flow_error *error)
6395 {
6396 	struct mlx5_priv *priv = dev->data->dev_private;
6397 	struct rte_flow *flow = NULL;
6398 	struct mlx5_flow *dev_flow;
6399 	const struct rte_flow_action_rss *rss = NULL;
6400 	struct mlx5_translated_action_handle
6401 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6402 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6403 	union {
6404 		struct mlx5_flow_expand_rss buf;
6405 		uint8_t buffer[4096];
6406 	} expand_buffer;
6407 	union {
6408 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6409 		uint8_t buffer[2048];
6410 	} actions_rx;
6411 	union {
6412 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6413 		uint8_t buffer[2048];
6414 	} actions_hairpin_tx;
6415 	union {
6416 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6417 		uint8_t buffer[2048];
6418 	} items_tx;
6419 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6420 	struct mlx5_flow_rss_desc *rss_desc;
6421 	const struct rte_flow_action *p_actions_rx;
6422 	uint32_t i;
6423 	uint32_t idx = 0;
6424 	int hairpin_flow;
6425 	struct rte_flow_attr attr_tx = { .priority = 0 };
6426 	const struct rte_flow_action *actions;
6427 	struct rte_flow_action *translated_actions = NULL;
6428 	struct mlx5_flow_tunnel *tunnel;
6429 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6430 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6431 	struct mlx5_flow_split_info flow_split_info = {
6432 		.external = !!external,
6433 		.skip_scale = 0,
6434 		.flow_idx = 0,
6435 		.prefix_mark = 0,
6436 		.prefix_layers = 0,
6437 		.table_id = 0
6438 	};
6439 	int ret;
6440 
6441 	MLX5_ASSERT(wks);
6442 	rss_desc = &wks->rss_desc;
6443 	ret = flow_action_handles_translate(dev, original_actions,
6444 					    indir_actions,
6445 					    &indir_actions_n,
6446 					    &translated_actions, error);
6447 	if (ret < 0) {
6448 		MLX5_ASSERT(translated_actions == NULL);
6449 		return 0;
6450 	}
6451 	actions = translated_actions ? translated_actions : original_actions;
6452 	p_actions_rx = actions;
6453 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6454 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6455 				external, hairpin_flow, error);
6456 	if (ret < 0)
6457 		goto error_before_hairpin_split;
6458 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6459 	if (!flow) {
6460 		rte_errno = ENOMEM;
6461 		goto error_before_hairpin_split;
6462 	}
6463 	if (hairpin_flow > 0) {
6464 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6465 			rte_errno = EINVAL;
6466 			goto error_before_hairpin_split;
6467 		}
6468 		flow_hairpin_split(dev, actions, actions_rx.actions,
6469 				   actions_hairpin_tx.actions, items_tx.items,
6470 				   idx);
6471 		p_actions_rx = actions_rx.actions;
6472 	}
6473 	flow_split_info.flow_idx = idx;
6474 	flow->drv_type = flow_get_drv_type(dev, attr);
6475 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6476 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6477 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6478 	/* RSS Action only works on NIC RX domain */
6479 	if (attr->ingress && !attr->transfer)
6480 		rss = flow_get_rss_action(dev, p_actions_rx);
6481 	if (rss) {
6482 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6483 			return 0;
6484 		/*
6485 		 * The following information is required by
6486 		 * mlx5_flow_hashfields_adjust() in advance.
6487 		 */
6488 		rss_desc->level = rss->level;
6489 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6490 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6491 	}
6492 	flow->dev_handles = 0;
6493 	if (rss && rss->types) {
6494 		unsigned int graph_root;
6495 
6496 		graph_root = find_graph_root(rss->level);
6497 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6498 					   items, rss->types,
6499 					   mlx5_support_expansion, graph_root);
6500 		MLX5_ASSERT(ret > 0 &&
6501 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6502 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6503 			for (i = 0; i < buf->entries; ++i)
6504 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6505 		}
6506 	} else {
6507 		buf->entries = 1;
6508 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6509 	}
6510 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6511 						      indir_actions_n);
6512 	for (i = 0; i < buf->entries; ++i) {
6513 		/* Initialize flow split data. */
6514 		flow_split_info.prefix_layers = 0;
6515 		flow_split_info.prefix_mark = 0;
6516 		flow_split_info.skip_scale = 0;
6517 		/*
6518 		 * The splitter may create multiple dev_flows,
6519 		 * depending on configuration. In the simplest
6520 		 * case it just creates unmodified original flow.
6521 		 */
6522 		ret = flow_create_split_outer(dev, flow, attr,
6523 					      buf->entry[i].pattern,
6524 					      p_actions_rx, &flow_split_info,
6525 					      error);
6526 		if (ret < 0)
6527 			goto error;
6528 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6529 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6530 							   p_actions_rx,
6531 							   idx,
6532 							   wks->flows[0].tunnel,
6533 							   &default_miss_ctx,
6534 							   error);
6535 			if (ret < 0) {
6536 				mlx5_free(default_miss_ctx.queue);
6537 				goto error;
6538 			}
6539 		}
6540 	}
6541 	/* Create the tx flow. */
6542 	if (hairpin_flow) {
6543 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6544 		attr_tx.ingress = 0;
6545 		attr_tx.egress = 1;
6546 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6547 					 actions_hairpin_tx.actions,
6548 					 idx, error);
6549 		if (!dev_flow)
6550 			goto error;
6551 		dev_flow->flow = flow;
6552 		dev_flow->external = 0;
6553 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6554 			      dev_flow->handle, next);
6555 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6556 					 items_tx.items,
6557 					 actions_hairpin_tx.actions, error);
6558 		if (ret < 0)
6559 			goto error;
6560 	}
6561 	/*
6562 	 * Update the metadata register copy table. If extensive
6563 	 * metadata feature is enabled and registers are supported
6564 	 * we might create the extra rte_flow for each unique
6565 	 * MARK/FLAG action ID.
6566 	 *
6567 	 * The table is updated for ingress Flows only, because
6568 	 * the egress Flows belong to the different device and
6569 	 * copy table should be updated in peer NIC Rx domain.
6570 	 */
6571 	if (attr->ingress &&
6572 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6573 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6574 		if (ret)
6575 			goto error;
6576 	}
6577 	/*
6578 	 * If the flow is external (from application) OR device is started,
6579 	 * OR mreg discover, then apply immediately.
6580 	 */
6581 	if (external || dev->data->dev_started ||
6582 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6583 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6584 		ret = flow_drv_apply(dev, flow, error);
6585 		if (ret < 0)
6586 			goto error;
6587 	}
6588 	flow->type = type;
6589 	flow_rxq_flags_set(dev, flow);
6590 	rte_free(translated_actions);
6591 	tunnel = flow_tunnel_from_rule(wks->flows);
6592 	if (tunnel) {
6593 		flow->tunnel = 1;
6594 		flow->tunnel_id = tunnel->tunnel_id;
6595 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6596 		mlx5_free(default_miss_ctx.queue);
6597 	}
6598 	mlx5_flow_pop_thread_workspace();
6599 	return idx;
6600 error:
6601 	MLX5_ASSERT(flow);
6602 	ret = rte_errno; /* Save rte_errno before cleanup. */
6603 	flow_mreg_del_copy_action(dev, flow);
6604 	flow_drv_destroy(dev, flow);
6605 	if (rss_desc->shared_rss)
6606 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6607 			mlx5_ipool_get
6608 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6609 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6610 	mlx5_ipool_free(priv->flows[type], idx);
6611 	rte_errno = ret; /* Restore rte_errno. */
6612 	ret = rte_errno;
6613 	rte_errno = ret;
6614 	mlx5_flow_pop_thread_workspace();
6615 error_before_hairpin_split:
6616 	rte_free(translated_actions);
6617 	return 0;
6618 }
6619 
6620 /**
6621  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6622  * incoming packets to table 1.
6623  *
6624  * Other flow rules, requested for group n, will be created in
6625  * e-switch table n+1.
6626  * Jump action to e-switch group n will be created to group n+1.
6627  *
6628  * Used when working in switchdev mode, to utilise advantages of table 1
6629  * and above.
6630  *
6631  * @param dev
6632  *   Pointer to Ethernet device.
6633  *
6634  * @return
6635  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6636  */
6637 struct rte_flow *
6638 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6639 {
6640 	const struct rte_flow_attr attr = {
6641 		.group = 0,
6642 		.priority = 0,
6643 		.ingress = 1,
6644 		.egress = 0,
6645 		.transfer = 1,
6646 	};
6647 	const struct rte_flow_item pattern = {
6648 		.type = RTE_FLOW_ITEM_TYPE_END,
6649 	};
6650 	struct rte_flow_action_jump jump = {
6651 		.group = 1,
6652 	};
6653 	const struct rte_flow_action actions[] = {
6654 		{
6655 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6656 			.conf = &jump,
6657 		},
6658 		{
6659 			.type = RTE_FLOW_ACTION_TYPE_END,
6660 		},
6661 	};
6662 	struct rte_flow_error error;
6663 
6664 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6665 						   &attr, &pattern,
6666 						   actions, false, &error);
6667 }
6668 
6669 /**
6670  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6671  * and sq number, directs all packets to peer vport.
6672  *
6673  * @param dev
6674  *   Pointer to Ethernet device.
6675  * @param txq
6676  *   Txq index.
6677  *
6678  * @return
6679  *   Flow ID on success, 0 otherwise and rte_errno is set.
6680  */
6681 uint32_t
6682 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6683 {
6684 	struct rte_flow_attr attr = {
6685 		.group = 0,
6686 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6687 		.ingress = 1,
6688 		.egress = 0,
6689 		.transfer = 1,
6690 	};
6691 	struct rte_flow_item_port_id port_spec = {
6692 		.id = MLX5_PORT_ESW_MGR,
6693 	};
6694 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6695 		.queue = txq,
6696 	};
6697 	struct rte_flow_item pattern[] = {
6698 		{
6699 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6700 			.spec = &port_spec,
6701 		},
6702 		{
6703 			.type = (enum rte_flow_item_type)
6704 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6705 			.spec = &txq_spec,
6706 		},
6707 		{
6708 			.type = RTE_FLOW_ITEM_TYPE_END,
6709 		},
6710 	};
6711 	struct rte_flow_action_jump jump = {
6712 		.group = 1,
6713 	};
6714 	struct rte_flow_action_port_id port = {
6715 		.id = dev->data->port_id,
6716 	};
6717 	struct rte_flow_action actions[] = {
6718 		{
6719 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6720 			.conf = &jump,
6721 		},
6722 		{
6723 			.type = RTE_FLOW_ACTION_TYPE_END,
6724 		},
6725 	};
6726 	struct rte_flow_error error;
6727 
6728 	/*
6729 	 * Creates group 0, highest priority jump flow.
6730 	 * Matches txq to bypass kernel packets.
6731 	 */
6732 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6733 			     false, &error) == 0)
6734 		return 0;
6735 	/* Create group 1, lowest priority redirect flow for txq. */
6736 	attr.group = 1;
6737 	actions[0].conf = &port;
6738 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6739 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6740 				actions, false, &error);
6741 }
6742 
6743 /**
6744  * Validate a flow supported by the NIC.
6745  *
6746  * @see rte_flow_validate()
6747  * @see rte_flow_ops
6748  */
6749 int
6750 mlx5_flow_validate(struct rte_eth_dev *dev,
6751 		   const struct rte_flow_attr *attr,
6752 		   const struct rte_flow_item items[],
6753 		   const struct rte_flow_action original_actions[],
6754 		   struct rte_flow_error *error)
6755 {
6756 	int hairpin_flow;
6757 	struct mlx5_translated_action_handle
6758 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6759 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6760 	const struct rte_flow_action *actions;
6761 	struct rte_flow_action *translated_actions = NULL;
6762 	int ret = flow_action_handles_translate(dev, original_actions,
6763 						indir_actions,
6764 						&indir_actions_n,
6765 						&translated_actions, error);
6766 
6767 	if (ret)
6768 		return ret;
6769 	actions = translated_actions ? translated_actions : original_actions;
6770 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6771 	ret = flow_drv_validate(dev, attr, items, actions,
6772 				true, hairpin_flow, error);
6773 	rte_free(translated_actions);
6774 	return ret;
6775 }
6776 
6777 /**
6778  * Create a flow.
6779  *
6780  * @see rte_flow_create()
6781  * @see rte_flow_ops
6782  */
6783 struct rte_flow *
6784 mlx5_flow_create(struct rte_eth_dev *dev,
6785 		 const struct rte_flow_attr *attr,
6786 		 const struct rte_flow_item items[],
6787 		 const struct rte_flow_action actions[],
6788 		 struct rte_flow_error *error)
6789 {
6790 	/*
6791 	 * If the device is not started yet, it is not allowed to created a
6792 	 * flow from application. PMD default flows and traffic control flows
6793 	 * are not affected.
6794 	 */
6795 	if (unlikely(!dev->data->dev_started)) {
6796 		DRV_LOG(DEBUG, "port %u is not started when "
6797 			"inserting a flow", dev->data->port_id);
6798 		rte_flow_error_set(error, ENODEV,
6799 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6800 				   NULL,
6801 				   "port not started");
6802 		return NULL;
6803 	}
6804 
6805 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6806 						   attr, items, actions,
6807 						   true, error);
6808 }
6809 
6810 /**
6811  * Destroy a flow in a list.
6812  *
6813  * @param dev
6814  *   Pointer to Ethernet device.
6815  * @param[in] flow_idx
6816  *   Index of flow to destroy.
6817  */
6818 static void
6819 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6820 		  uint32_t flow_idx)
6821 {
6822 	struct mlx5_priv *priv = dev->data->dev_private;
6823 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6824 
6825 	if (!flow)
6826 		return;
6827 	MLX5_ASSERT(flow->type == type);
6828 	/*
6829 	 * Update RX queue flags only if port is started, otherwise it is
6830 	 * already clean.
6831 	 */
6832 	if (dev->data->dev_started)
6833 		flow_rxq_flags_trim(dev, flow);
6834 	flow_drv_destroy(dev, flow);
6835 	if (flow->tunnel) {
6836 		struct mlx5_flow_tunnel *tunnel;
6837 
6838 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6839 		RTE_VERIFY(tunnel);
6840 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6841 			mlx5_flow_tunnel_free(dev, tunnel);
6842 	}
6843 	flow_mreg_del_copy_action(dev, flow);
6844 	mlx5_ipool_free(priv->flows[type], flow_idx);
6845 }
6846 
6847 /**
6848  * Destroy all flows.
6849  *
6850  * @param dev
6851  *   Pointer to Ethernet device.
6852  * @param type
6853  *   Flow type to be flushed.
6854  * @param active
6855  *   If flushing is called avtively.
6856  */
6857 void
6858 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6859 		     bool active)
6860 {
6861 	struct mlx5_priv *priv = dev->data->dev_private;
6862 	uint32_t num_flushed = 0, fidx = 1;
6863 	struct rte_flow *flow;
6864 
6865 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6866 		flow_list_destroy(dev, type, fidx);
6867 		num_flushed++;
6868 	}
6869 	if (active) {
6870 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6871 			dev->data->port_id, num_flushed);
6872 	}
6873 }
6874 
6875 /**
6876  * Stop all default actions for flows.
6877  *
6878  * @param dev
6879  *   Pointer to Ethernet device.
6880  */
6881 void
6882 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6883 {
6884 	flow_mreg_del_default_copy_action(dev);
6885 	flow_rxq_flags_clear(dev);
6886 }
6887 
6888 /**
6889  * Start all default actions for flows.
6890  *
6891  * @param dev
6892  *   Pointer to Ethernet device.
6893  * @return
6894  *   0 on success, a negative errno value otherwise and rte_errno is set.
6895  */
6896 int
6897 mlx5_flow_start_default(struct rte_eth_dev *dev)
6898 {
6899 	struct rte_flow_error error;
6900 
6901 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6902 	return flow_mreg_add_default_copy_action(dev, &error);
6903 }
6904 
6905 /**
6906  * Release key of thread specific flow workspace data.
6907  */
6908 void
6909 flow_release_workspace(void *data)
6910 {
6911 	struct mlx5_flow_workspace *wks = data;
6912 	struct mlx5_flow_workspace *next;
6913 
6914 	while (wks) {
6915 		next = wks->next;
6916 		free(wks->rss_desc.queue);
6917 		free(wks);
6918 		wks = next;
6919 	}
6920 }
6921 
6922 /**
6923  * Get thread specific current flow workspace.
6924  *
6925  * @return pointer to thread specific flow workspace data, NULL on error.
6926  */
6927 struct mlx5_flow_workspace*
6928 mlx5_flow_get_thread_workspace(void)
6929 {
6930 	struct mlx5_flow_workspace *data;
6931 
6932 	data = mlx5_flow_os_get_specific_workspace();
6933 	MLX5_ASSERT(data && data->inuse);
6934 	if (!data || !data->inuse)
6935 		DRV_LOG(ERR, "flow workspace not initialized.");
6936 	return data;
6937 }
6938 
6939 /**
6940  * Allocate and init new flow workspace.
6941  *
6942  * @return pointer to flow workspace data, NULL on error.
6943  */
6944 static struct mlx5_flow_workspace*
6945 flow_alloc_thread_workspace(void)
6946 {
6947 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6948 
6949 	if (!data) {
6950 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6951 			"memory.");
6952 		return NULL;
6953 	}
6954 	data->rss_desc.queue = calloc(1,
6955 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6956 	if (!data->rss_desc.queue)
6957 		goto err;
6958 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6959 	return data;
6960 err:
6961 	if (data->rss_desc.queue)
6962 		free(data->rss_desc.queue);
6963 	free(data);
6964 	return NULL;
6965 }
6966 
6967 /**
6968  * Get new thread specific flow workspace.
6969  *
6970  * If current workspace inuse, create new one and set as current.
6971  *
6972  * @return pointer to thread specific flow workspace data, NULL on error.
6973  */
6974 static struct mlx5_flow_workspace*
6975 mlx5_flow_push_thread_workspace(void)
6976 {
6977 	struct mlx5_flow_workspace *curr;
6978 	struct mlx5_flow_workspace *data;
6979 
6980 	curr = mlx5_flow_os_get_specific_workspace();
6981 	if (!curr) {
6982 		data = flow_alloc_thread_workspace();
6983 		if (!data)
6984 			return NULL;
6985 	} else if (!curr->inuse) {
6986 		data = curr;
6987 	} else if (curr->next) {
6988 		data = curr->next;
6989 	} else {
6990 		data = flow_alloc_thread_workspace();
6991 		if (!data)
6992 			return NULL;
6993 		curr->next = data;
6994 		data->prev = curr;
6995 	}
6996 	data->inuse = 1;
6997 	data->flow_idx = 0;
6998 	/* Set as current workspace */
6999 	if (mlx5_flow_os_set_specific_workspace(data))
7000 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7001 	return data;
7002 }
7003 
7004 /**
7005  * Close current thread specific flow workspace.
7006  *
7007  * If previous workspace available, set it as current.
7008  *
7009  * @return pointer to thread specific flow workspace data, NULL on error.
7010  */
7011 static void
7012 mlx5_flow_pop_thread_workspace(void)
7013 {
7014 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7015 
7016 	if (!data)
7017 		return;
7018 	if (!data->inuse) {
7019 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7020 		return;
7021 	}
7022 	data->inuse = 0;
7023 	if (!data->prev)
7024 		return;
7025 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7026 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7027 }
7028 
7029 /**
7030  * Verify the flow list is empty
7031  *
7032  * @param dev
7033  *  Pointer to Ethernet device.
7034  *
7035  * @return the number of flows not released.
7036  */
7037 int
7038 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7039 {
7040 	struct mlx5_priv *priv = dev->data->dev_private;
7041 	struct rte_flow *flow;
7042 	uint32_t idx = 0;
7043 	int ret = 0, i;
7044 
7045 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7046 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7047 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7048 				dev->data->port_id, (void *)flow);
7049 			ret++;
7050 		}
7051 	}
7052 	return ret;
7053 }
7054 
7055 /**
7056  * Enable default hairpin egress flow.
7057  *
7058  * @param dev
7059  *   Pointer to Ethernet device.
7060  * @param queue
7061  *   The queue index.
7062  *
7063  * @return
7064  *   0 on success, a negative errno value otherwise and rte_errno is set.
7065  */
7066 int
7067 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7068 			    uint32_t queue)
7069 {
7070 	const struct rte_flow_attr attr = {
7071 		.egress = 1,
7072 		.priority = 0,
7073 	};
7074 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7075 		.queue = queue,
7076 	};
7077 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7078 		.queue = UINT32_MAX,
7079 	};
7080 	struct rte_flow_item items[] = {
7081 		{
7082 			.type = (enum rte_flow_item_type)
7083 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7084 			.spec = &queue_spec,
7085 			.last = NULL,
7086 			.mask = &queue_mask,
7087 		},
7088 		{
7089 			.type = RTE_FLOW_ITEM_TYPE_END,
7090 		},
7091 	};
7092 	struct rte_flow_action_jump jump = {
7093 		.group = MLX5_HAIRPIN_TX_TABLE,
7094 	};
7095 	struct rte_flow_action actions[2];
7096 	uint32_t flow_idx;
7097 	struct rte_flow_error error;
7098 
7099 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7100 	actions[0].conf = &jump;
7101 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7102 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7103 				    &attr, items, actions, false, &error);
7104 	if (!flow_idx) {
7105 		DRV_LOG(DEBUG,
7106 			"Failed to create ctrl flow: rte_errno(%d),"
7107 			" type(%d), message(%s)",
7108 			rte_errno, error.type,
7109 			error.message ? error.message : " (no stated reason)");
7110 		return -rte_errno;
7111 	}
7112 	return 0;
7113 }
7114 
7115 /**
7116  * Enable a control flow configured from the control plane.
7117  *
7118  * @param dev
7119  *   Pointer to Ethernet device.
7120  * @param eth_spec
7121  *   An Ethernet flow spec to apply.
7122  * @param eth_mask
7123  *   An Ethernet flow mask to apply.
7124  * @param vlan_spec
7125  *   A VLAN flow spec to apply.
7126  * @param vlan_mask
7127  *   A VLAN flow mask to apply.
7128  *
7129  * @return
7130  *   0 on success, a negative errno value otherwise and rte_errno is set.
7131  */
7132 int
7133 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7134 		    struct rte_flow_item_eth *eth_spec,
7135 		    struct rte_flow_item_eth *eth_mask,
7136 		    struct rte_flow_item_vlan *vlan_spec,
7137 		    struct rte_flow_item_vlan *vlan_mask)
7138 {
7139 	struct mlx5_priv *priv = dev->data->dev_private;
7140 	const struct rte_flow_attr attr = {
7141 		.ingress = 1,
7142 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7143 	};
7144 	struct rte_flow_item items[] = {
7145 		{
7146 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7147 			.spec = eth_spec,
7148 			.last = NULL,
7149 			.mask = eth_mask,
7150 		},
7151 		{
7152 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7153 					      RTE_FLOW_ITEM_TYPE_END,
7154 			.spec = vlan_spec,
7155 			.last = NULL,
7156 			.mask = vlan_mask,
7157 		},
7158 		{
7159 			.type = RTE_FLOW_ITEM_TYPE_END,
7160 		},
7161 	};
7162 	uint16_t queue[priv->reta_idx_n];
7163 	struct rte_flow_action_rss action_rss = {
7164 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7165 		.level = 0,
7166 		.types = priv->rss_conf.rss_hf,
7167 		.key_len = priv->rss_conf.rss_key_len,
7168 		.queue_num = priv->reta_idx_n,
7169 		.key = priv->rss_conf.rss_key,
7170 		.queue = queue,
7171 	};
7172 	struct rte_flow_action actions[] = {
7173 		{
7174 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7175 			.conf = &action_rss,
7176 		},
7177 		{
7178 			.type = RTE_FLOW_ACTION_TYPE_END,
7179 		},
7180 	};
7181 	uint32_t flow_idx;
7182 	struct rte_flow_error error;
7183 	unsigned int i;
7184 
7185 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7186 		return 0;
7187 	}
7188 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7189 		action_rss.types = 0;
7190 	for (i = 0; i != priv->reta_idx_n; ++i)
7191 		queue[i] = (*priv->reta_idx)[i];
7192 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7193 				    &attr, items, actions, false, &error);
7194 	if (!flow_idx)
7195 		return -rte_errno;
7196 	return 0;
7197 }
7198 
7199 /**
7200  * Enable a flow control configured from the control plane.
7201  *
7202  * @param dev
7203  *   Pointer to Ethernet device.
7204  * @param eth_spec
7205  *   An Ethernet flow spec to apply.
7206  * @param eth_mask
7207  *   An Ethernet flow mask to apply.
7208  *
7209  * @return
7210  *   0 on success, a negative errno value otherwise and rte_errno is set.
7211  */
7212 int
7213 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7214 	       struct rte_flow_item_eth *eth_spec,
7215 	       struct rte_flow_item_eth *eth_mask)
7216 {
7217 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7218 }
7219 
7220 /**
7221  * Create default miss flow rule matching lacp traffic
7222  *
7223  * @param dev
7224  *   Pointer to Ethernet device.
7225  * @param eth_spec
7226  *   An Ethernet flow spec to apply.
7227  *
7228  * @return
7229  *   0 on success, a negative errno value otherwise and rte_errno is set.
7230  */
7231 int
7232 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7233 {
7234 	/*
7235 	 * The LACP matching is done by only using ether type since using
7236 	 * a multicast dst mac causes kernel to give low priority to this flow.
7237 	 */
7238 	static const struct rte_flow_item_eth lacp_spec = {
7239 		.type = RTE_BE16(0x8809),
7240 	};
7241 	static const struct rte_flow_item_eth lacp_mask = {
7242 		.type = 0xffff,
7243 	};
7244 	const struct rte_flow_attr attr = {
7245 		.ingress = 1,
7246 	};
7247 	struct rte_flow_item items[] = {
7248 		{
7249 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7250 			.spec = &lacp_spec,
7251 			.mask = &lacp_mask,
7252 		},
7253 		{
7254 			.type = RTE_FLOW_ITEM_TYPE_END,
7255 		},
7256 	};
7257 	struct rte_flow_action actions[] = {
7258 		{
7259 			.type = (enum rte_flow_action_type)
7260 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7261 		},
7262 		{
7263 			.type = RTE_FLOW_ACTION_TYPE_END,
7264 		},
7265 	};
7266 	struct rte_flow_error error;
7267 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7268 					&attr, items, actions,
7269 					false, &error);
7270 
7271 	if (!flow_idx)
7272 		return -rte_errno;
7273 	return 0;
7274 }
7275 
7276 /**
7277  * Destroy a flow.
7278  *
7279  * @see rte_flow_destroy()
7280  * @see rte_flow_ops
7281  */
7282 int
7283 mlx5_flow_destroy(struct rte_eth_dev *dev,
7284 		  struct rte_flow *flow,
7285 		  struct rte_flow_error *error __rte_unused)
7286 {
7287 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7288 				(uintptr_t)(void *)flow);
7289 	return 0;
7290 }
7291 
7292 /**
7293  * Destroy all flows.
7294  *
7295  * @see rte_flow_flush()
7296  * @see rte_flow_ops
7297  */
7298 int
7299 mlx5_flow_flush(struct rte_eth_dev *dev,
7300 		struct rte_flow_error *error __rte_unused)
7301 {
7302 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7303 	return 0;
7304 }
7305 
7306 /**
7307  * Isolated mode.
7308  *
7309  * @see rte_flow_isolate()
7310  * @see rte_flow_ops
7311  */
7312 int
7313 mlx5_flow_isolate(struct rte_eth_dev *dev,
7314 		  int enable,
7315 		  struct rte_flow_error *error)
7316 {
7317 	struct mlx5_priv *priv = dev->data->dev_private;
7318 
7319 	if (dev->data->dev_started) {
7320 		rte_flow_error_set(error, EBUSY,
7321 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7322 				   NULL,
7323 				   "port must be stopped first");
7324 		return -rte_errno;
7325 	}
7326 	priv->isolated = !!enable;
7327 	if (enable)
7328 		dev->dev_ops = &mlx5_dev_ops_isolate;
7329 	else
7330 		dev->dev_ops = &mlx5_dev_ops;
7331 
7332 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7333 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7334 
7335 	return 0;
7336 }
7337 
7338 /**
7339  * Query a flow.
7340  *
7341  * @see rte_flow_query()
7342  * @see rte_flow_ops
7343  */
7344 static int
7345 flow_drv_query(struct rte_eth_dev *dev,
7346 	       uint32_t flow_idx,
7347 	       const struct rte_flow_action *actions,
7348 	       void *data,
7349 	       struct rte_flow_error *error)
7350 {
7351 	struct mlx5_priv *priv = dev->data->dev_private;
7352 	const struct mlx5_flow_driver_ops *fops;
7353 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7354 					       flow_idx);
7355 	enum mlx5_flow_drv_type ftype;
7356 
7357 	if (!flow) {
7358 		return rte_flow_error_set(error, ENOENT,
7359 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7360 			  NULL,
7361 			  "invalid flow handle");
7362 	}
7363 	ftype = flow->drv_type;
7364 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7365 	fops = flow_get_drv_ops(ftype);
7366 
7367 	return fops->query(dev, flow, actions, data, error);
7368 }
7369 
7370 /**
7371  * Query a flow.
7372  *
7373  * @see rte_flow_query()
7374  * @see rte_flow_ops
7375  */
7376 int
7377 mlx5_flow_query(struct rte_eth_dev *dev,
7378 		struct rte_flow *flow,
7379 		const struct rte_flow_action *actions,
7380 		void *data,
7381 		struct rte_flow_error *error)
7382 {
7383 	int ret;
7384 
7385 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7386 			     error);
7387 	if (ret < 0)
7388 		return ret;
7389 	return 0;
7390 }
7391 
7392 /**
7393  * Get rte_flow callbacks.
7394  *
7395  * @param dev
7396  *   Pointer to Ethernet device structure.
7397  * @param ops
7398  *   Pointer to operation-specific structure.
7399  *
7400  * @return 0
7401  */
7402 int
7403 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7404 		  const struct rte_flow_ops **ops)
7405 {
7406 	*ops = &mlx5_flow_ops;
7407 	return 0;
7408 }
7409 
7410 /**
7411  * Validate meter policy actions.
7412  * Dispatcher for action type specific validation.
7413  *
7414  * @param[in] dev
7415  *   Pointer to the Ethernet device structure.
7416  * @param[in] action
7417  *   The meter policy action object to validate.
7418  * @param[in] attr
7419  *   Attributes of flow to determine steering domain.
7420  * @param[out] is_rss
7421  *   Is RSS or not.
7422  * @param[out] domain_bitmap
7423  *   Domain bitmap.
7424  * @param[out] is_def_policy
7425  *   Is default policy or not.
7426  * @param[out] error
7427  *   Perform verbose error reporting if not NULL. Initialized in case of
7428  *   error only.
7429  *
7430  * @return
7431  *   0 on success, otherwise negative errno value.
7432  */
7433 int
7434 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7435 			const struct rte_flow_action *actions[RTE_COLORS],
7436 			struct rte_flow_attr *attr,
7437 			bool *is_rss,
7438 			uint8_t *domain_bitmap,
7439 			uint8_t *policy_mode,
7440 			struct rte_mtr_error *error)
7441 {
7442 	const struct mlx5_flow_driver_ops *fops;
7443 
7444 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7445 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7446 				       domain_bitmap, policy_mode, error);
7447 }
7448 
7449 /**
7450  * Destroy the meter table set.
7451  *
7452  * @param[in] dev
7453  *   Pointer to Ethernet device.
7454  * @param[in] mtr_policy
7455  *   Meter policy struct.
7456  */
7457 void
7458 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7459 		      struct mlx5_flow_meter_policy *mtr_policy)
7460 {
7461 	const struct mlx5_flow_driver_ops *fops;
7462 
7463 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7464 	fops->destroy_mtr_acts(dev, mtr_policy);
7465 }
7466 
7467 /**
7468  * Create policy action, lock free,
7469  * (mutex should be acquired by caller).
7470  * Dispatcher for action type specific call.
7471  *
7472  * @param[in] dev
7473  *   Pointer to the Ethernet device structure.
7474  * @param[in] mtr_policy
7475  *   Meter policy struct.
7476  * @param[in] action
7477  *   Action specification used to create meter actions.
7478  * @param[out] error
7479  *   Perform verbose error reporting if not NULL. Initialized in case of
7480  *   error only.
7481  *
7482  * @return
7483  *   0 on success, otherwise negative errno value.
7484  */
7485 int
7486 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7487 		      struct mlx5_flow_meter_policy *mtr_policy,
7488 		      const struct rte_flow_action *actions[RTE_COLORS],
7489 		      struct rte_mtr_error *error)
7490 {
7491 	const struct mlx5_flow_driver_ops *fops;
7492 
7493 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7494 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7495 }
7496 
7497 /**
7498  * Create policy rules, lock free,
7499  * (mutex should be acquired by caller).
7500  * Dispatcher for action type specific call.
7501  *
7502  * @param[in] dev
7503  *   Pointer to the Ethernet device structure.
7504  * @param[in] mtr_policy
7505  *   Meter policy struct.
7506  *
7507  * @return
7508  *   0 on success, -1 otherwise.
7509  */
7510 int
7511 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7512 			     struct mlx5_flow_meter_policy *mtr_policy)
7513 {
7514 	const struct mlx5_flow_driver_ops *fops;
7515 
7516 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7517 	return fops->create_policy_rules(dev, mtr_policy);
7518 }
7519 
7520 /**
7521  * Destroy policy rules, lock free,
7522  * (mutex should be acquired by caller).
7523  * Dispatcher for action type specific call.
7524  *
7525  * @param[in] dev
7526  *   Pointer to the Ethernet device structure.
7527  * @param[in] mtr_policy
7528  *   Meter policy struct.
7529  */
7530 void
7531 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7532 			     struct mlx5_flow_meter_policy *mtr_policy)
7533 {
7534 	const struct mlx5_flow_driver_ops *fops;
7535 
7536 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7537 	fops->destroy_policy_rules(dev, mtr_policy);
7538 }
7539 
7540 /**
7541  * Destroy the default policy table set.
7542  *
7543  * @param[in] dev
7544  *   Pointer to Ethernet device.
7545  */
7546 void
7547 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7548 {
7549 	const struct mlx5_flow_driver_ops *fops;
7550 
7551 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7552 	fops->destroy_def_policy(dev);
7553 }
7554 
7555 /**
7556  * Destroy the default policy table set.
7557  *
7558  * @param[in] dev
7559  *   Pointer to Ethernet device.
7560  *
7561  * @return
7562  *   0 on success, -1 otherwise.
7563  */
7564 int
7565 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7566 {
7567 	const struct mlx5_flow_driver_ops *fops;
7568 
7569 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7570 	return fops->create_def_policy(dev);
7571 }
7572 
7573 /**
7574  * Create the needed meter and suffix tables.
7575  *
7576  * @param[in] dev
7577  *   Pointer to Ethernet device.
7578  *
7579  * @return
7580  *   0 on success, -1 otherwise.
7581  */
7582 int
7583 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7584 			struct mlx5_flow_meter_info *fm,
7585 			uint32_t mtr_idx,
7586 			uint8_t domain_bitmap)
7587 {
7588 	const struct mlx5_flow_driver_ops *fops;
7589 
7590 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7591 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7592 }
7593 
7594 /**
7595  * Destroy the meter table set.
7596  *
7597  * @param[in] dev
7598  *   Pointer to Ethernet device.
7599  * @param[in] tbl
7600  *   Pointer to the meter table set.
7601  */
7602 void
7603 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7604 			   struct mlx5_flow_meter_info *fm)
7605 {
7606 	const struct mlx5_flow_driver_ops *fops;
7607 
7608 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7609 	fops->destroy_mtr_tbls(dev, fm);
7610 }
7611 
7612 /**
7613  * Destroy the global meter drop table.
7614  *
7615  * @param[in] dev
7616  *   Pointer to Ethernet device.
7617  */
7618 void
7619 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7620 {
7621 	const struct mlx5_flow_driver_ops *fops;
7622 
7623 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7624 	fops->destroy_mtr_drop_tbls(dev);
7625 }
7626 
7627 /**
7628  * Destroy the sub policy table with RX queue.
7629  *
7630  * @param[in] dev
7631  *   Pointer to Ethernet device.
7632  * @param[in] mtr_policy
7633  *   Pointer to meter policy table.
7634  */
7635 void
7636 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7637 		struct mlx5_flow_meter_policy *mtr_policy)
7638 {
7639 	const struct mlx5_flow_driver_ops *fops;
7640 
7641 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7642 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7643 }
7644 
7645 /**
7646  * Allocate the needed aso flow meter id.
7647  *
7648  * @param[in] dev
7649  *   Pointer to Ethernet device.
7650  *
7651  * @return
7652  *   Index to aso flow meter on success, NULL otherwise.
7653  */
7654 uint32_t
7655 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7656 {
7657 	const struct mlx5_flow_driver_ops *fops;
7658 
7659 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7660 	return fops->create_meter(dev);
7661 }
7662 
7663 /**
7664  * Free the aso flow meter id.
7665  *
7666  * @param[in] dev
7667  *   Pointer to Ethernet device.
7668  * @param[in] mtr_idx
7669  *  Index to aso flow meter to be free.
7670  *
7671  * @return
7672  *   0 on success.
7673  */
7674 void
7675 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7676 {
7677 	const struct mlx5_flow_driver_ops *fops;
7678 
7679 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7680 	fops->free_meter(dev, mtr_idx);
7681 }
7682 
7683 /**
7684  * Allocate a counter.
7685  *
7686  * @param[in] dev
7687  *   Pointer to Ethernet device structure.
7688  *
7689  * @return
7690  *   Index to allocated counter  on success, 0 otherwise.
7691  */
7692 uint32_t
7693 mlx5_counter_alloc(struct rte_eth_dev *dev)
7694 {
7695 	const struct mlx5_flow_driver_ops *fops;
7696 	struct rte_flow_attr attr = { .transfer = 0 };
7697 
7698 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7699 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7700 		return fops->counter_alloc(dev);
7701 	}
7702 	DRV_LOG(ERR,
7703 		"port %u counter allocate is not supported.",
7704 		 dev->data->port_id);
7705 	return 0;
7706 }
7707 
7708 /**
7709  * Free a counter.
7710  *
7711  * @param[in] dev
7712  *   Pointer to Ethernet device structure.
7713  * @param[in] cnt
7714  *   Index to counter to be free.
7715  */
7716 void
7717 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7718 {
7719 	const struct mlx5_flow_driver_ops *fops;
7720 	struct rte_flow_attr attr = { .transfer = 0 };
7721 
7722 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7723 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7724 		fops->counter_free(dev, cnt);
7725 		return;
7726 	}
7727 	DRV_LOG(ERR,
7728 		"port %u counter free is not supported.",
7729 		 dev->data->port_id);
7730 }
7731 
7732 /**
7733  * Query counter statistics.
7734  *
7735  * @param[in] dev
7736  *   Pointer to Ethernet device structure.
7737  * @param[in] cnt
7738  *   Index to counter to query.
7739  * @param[in] clear
7740  *   Set to clear counter statistics.
7741  * @param[out] pkts
7742  *   The counter hits packets number to save.
7743  * @param[out] bytes
7744  *   The counter hits bytes number to save.
7745  *
7746  * @return
7747  *   0 on success, a negative errno value otherwise.
7748  */
7749 int
7750 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7751 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7752 {
7753 	const struct mlx5_flow_driver_ops *fops;
7754 	struct rte_flow_attr attr = { .transfer = 0 };
7755 
7756 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7757 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7758 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7759 	}
7760 	DRV_LOG(ERR,
7761 		"port %u counter query is not supported.",
7762 		 dev->data->port_id);
7763 	return -ENOTSUP;
7764 }
7765 
7766 /**
7767  * Allocate a new memory for the counter values wrapped by all the needed
7768  * management.
7769  *
7770  * @param[in] sh
7771  *   Pointer to mlx5_dev_ctx_shared object.
7772  *
7773  * @return
7774  *   0 on success, a negative errno value otherwise.
7775  */
7776 static int
7777 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7778 {
7779 	struct mlx5_counter_stats_mem_mng *mem_mng;
7780 	volatile struct flow_counter_stats *raw_data;
7781 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7782 	int size = (sizeof(struct flow_counter_stats) *
7783 			MLX5_COUNTERS_PER_POOL +
7784 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7785 			sizeof(struct mlx5_counter_stats_mem_mng);
7786 	size_t pgsize = rte_mem_page_size();
7787 	uint8_t *mem;
7788 	int ret;
7789 	int i;
7790 
7791 	if (pgsize == (size_t)-1) {
7792 		DRV_LOG(ERR, "Failed to get mem page size");
7793 		rte_errno = ENOMEM;
7794 		return -ENOMEM;
7795 	}
7796 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7797 	if (!mem) {
7798 		rte_errno = ENOMEM;
7799 		return -ENOMEM;
7800 	}
7801 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7802 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7803 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7804 					  sh->cdev->pdn, mem, size,
7805 					  &mem_mng->wm);
7806 	if (ret) {
7807 		rte_errno = errno;
7808 		mlx5_free(mem);
7809 		return -rte_errno;
7810 	}
7811 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7812 	raw_data = (volatile struct flow_counter_stats *)mem;
7813 	for (i = 0; i < raws_n; ++i) {
7814 		mem_mng->raws[i].mem_mng = mem_mng;
7815 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7816 	}
7817 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7818 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7819 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7820 				 next);
7821 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7822 	sh->cmng.mem_mng = mem_mng;
7823 	return 0;
7824 }
7825 
7826 /**
7827  * Set the statistic memory to the new counter pool.
7828  *
7829  * @param[in] sh
7830  *   Pointer to mlx5_dev_ctx_shared object.
7831  * @param[in] pool
7832  *   Pointer to the pool to set the statistic memory.
7833  *
7834  * @return
7835  *   0 on success, a negative errno value otherwise.
7836  */
7837 static int
7838 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7839 			       struct mlx5_flow_counter_pool *pool)
7840 {
7841 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7842 	/* Resize statistic memory once used out. */
7843 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7844 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7845 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7846 		return -1;
7847 	}
7848 	rte_spinlock_lock(&pool->sl);
7849 	pool->raw = cmng->mem_mng->raws + pool->index %
7850 		    MLX5_CNT_CONTAINER_RESIZE;
7851 	rte_spinlock_unlock(&pool->sl);
7852 	pool->raw_hw = NULL;
7853 	return 0;
7854 }
7855 
7856 #define MLX5_POOL_QUERY_FREQ_US 1000000
7857 
7858 /**
7859  * Set the periodic procedure for triggering asynchronous batch queries for all
7860  * the counter pools.
7861  *
7862  * @param[in] sh
7863  *   Pointer to mlx5_dev_ctx_shared object.
7864  */
7865 void
7866 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7867 {
7868 	uint32_t pools_n, us;
7869 
7870 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7871 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7872 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7873 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7874 		sh->cmng.query_thread_on = 0;
7875 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7876 	} else {
7877 		sh->cmng.query_thread_on = 1;
7878 	}
7879 }
7880 
7881 /**
7882  * The periodic procedure for triggering asynchronous batch queries for all the
7883  * counter pools. This function is probably called by the host thread.
7884  *
7885  * @param[in] arg
7886  *   The parameter for the alarm process.
7887  */
7888 void
7889 mlx5_flow_query_alarm(void *arg)
7890 {
7891 	struct mlx5_dev_ctx_shared *sh = arg;
7892 	int ret;
7893 	uint16_t pool_index = sh->cmng.pool_index;
7894 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7895 	struct mlx5_flow_counter_pool *pool;
7896 	uint16_t n_valid;
7897 
7898 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7899 		goto set_alarm;
7900 	rte_spinlock_lock(&cmng->pool_update_sl);
7901 	pool = cmng->pools[pool_index];
7902 	n_valid = cmng->n_valid;
7903 	rte_spinlock_unlock(&cmng->pool_update_sl);
7904 	/* Set the statistic memory to the new created pool. */
7905 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7906 		goto set_alarm;
7907 	if (pool->raw_hw)
7908 		/* There is a pool query in progress. */
7909 		goto set_alarm;
7910 	pool->raw_hw =
7911 		LIST_FIRST(&sh->cmng.free_stat_raws);
7912 	if (!pool->raw_hw)
7913 		/* No free counter statistics raw memory. */
7914 		goto set_alarm;
7915 	/*
7916 	 * Identify the counters released between query trigger and query
7917 	 * handle more efficiently. The counter released in this gap period
7918 	 * should wait for a new round of query as the new arrived packets
7919 	 * will not be taken into account.
7920 	 */
7921 	pool->query_gen++;
7922 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7923 					       MLX5_COUNTERS_PER_POOL,
7924 					       NULL, NULL,
7925 					       pool->raw_hw->mem_mng->wm.lkey,
7926 					       (void *)(uintptr_t)
7927 					       pool->raw_hw->data,
7928 					       sh->devx_comp,
7929 					       (uint64_t)(uintptr_t)pool);
7930 	if (ret) {
7931 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7932 			" %d", pool->min_dcs->id);
7933 		pool->raw_hw = NULL;
7934 		goto set_alarm;
7935 	}
7936 	LIST_REMOVE(pool->raw_hw, next);
7937 	sh->cmng.pending_queries++;
7938 	pool_index++;
7939 	if (pool_index >= n_valid)
7940 		pool_index = 0;
7941 set_alarm:
7942 	sh->cmng.pool_index = pool_index;
7943 	mlx5_set_query_alarm(sh);
7944 }
7945 
7946 /**
7947  * Check and callback event for new aged flow in the counter pool
7948  *
7949  * @param[in] sh
7950  *   Pointer to mlx5_dev_ctx_shared object.
7951  * @param[in] pool
7952  *   Pointer to Current counter pool.
7953  */
7954 static void
7955 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7956 		   struct mlx5_flow_counter_pool *pool)
7957 {
7958 	struct mlx5_priv *priv;
7959 	struct mlx5_flow_counter *cnt;
7960 	struct mlx5_age_info *age_info;
7961 	struct mlx5_age_param *age_param;
7962 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7963 	struct mlx5_counter_stats_raw *prev = pool->raw;
7964 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7965 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7966 	uint16_t expected = AGE_CANDIDATE;
7967 	uint32_t i;
7968 
7969 	pool->time_of_last_age_check = curr_time;
7970 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7971 		cnt = MLX5_POOL_GET_CNT(pool, i);
7972 		age_param = MLX5_CNT_TO_AGE(cnt);
7973 		if (__atomic_load_n(&age_param->state,
7974 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7975 			continue;
7976 		if (cur->data[i].hits != prev->data[i].hits) {
7977 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7978 					 __ATOMIC_RELAXED);
7979 			continue;
7980 		}
7981 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7982 				       time_delta,
7983 				       __ATOMIC_RELAXED) <= age_param->timeout)
7984 			continue;
7985 		/**
7986 		 * Hold the lock first, or if between the
7987 		 * state AGE_TMOUT and tailq operation the
7988 		 * release happened, the release procedure
7989 		 * may delete a non-existent tailq node.
7990 		 */
7991 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7992 		age_info = GET_PORT_AGE_INFO(priv);
7993 		rte_spinlock_lock(&age_info->aged_sl);
7994 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7995 						AGE_TMOUT, false,
7996 						__ATOMIC_RELAXED,
7997 						__ATOMIC_RELAXED)) {
7998 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7999 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8000 		}
8001 		rte_spinlock_unlock(&age_info->aged_sl);
8002 	}
8003 	mlx5_age_event_prepare(sh);
8004 }
8005 
8006 /**
8007  * Handler for the HW respond about ready values from an asynchronous batch
8008  * query. This function is probably called by the host thread.
8009  *
8010  * @param[in] sh
8011  *   The pointer to the shared device context.
8012  * @param[in] async_id
8013  *   The Devx async ID.
8014  * @param[in] status
8015  *   The status of the completion.
8016  */
8017 void
8018 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8019 				  uint64_t async_id, int status)
8020 {
8021 	struct mlx5_flow_counter_pool *pool =
8022 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8023 	struct mlx5_counter_stats_raw *raw_to_free;
8024 	uint8_t query_gen = pool->query_gen ^ 1;
8025 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8026 	enum mlx5_counter_type cnt_type =
8027 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8028 				MLX5_COUNTER_TYPE_ORIGIN;
8029 
8030 	if (unlikely(status)) {
8031 		raw_to_free = pool->raw_hw;
8032 	} else {
8033 		raw_to_free = pool->raw;
8034 		if (pool->is_aged)
8035 			mlx5_flow_aging_check(sh, pool);
8036 		rte_spinlock_lock(&pool->sl);
8037 		pool->raw = pool->raw_hw;
8038 		rte_spinlock_unlock(&pool->sl);
8039 		/* Be sure the new raw counters data is updated in memory. */
8040 		rte_io_wmb();
8041 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8042 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8043 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8044 				     &pool->counters[query_gen], next);
8045 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8046 		}
8047 	}
8048 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8049 	pool->raw_hw = NULL;
8050 	sh->cmng.pending_queries--;
8051 }
8052 
8053 static int
8054 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8055 		    const struct flow_grp_info *grp_info,
8056 		    struct rte_flow_error *error)
8057 {
8058 	if (grp_info->transfer && grp_info->external &&
8059 	    grp_info->fdb_def_rule) {
8060 		if (group == UINT32_MAX)
8061 			return rte_flow_error_set
8062 						(error, EINVAL,
8063 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8064 						 NULL,
8065 						 "group index not supported");
8066 		*table = group + 1;
8067 	} else {
8068 		*table = group;
8069 	}
8070 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8071 	return 0;
8072 }
8073 
8074 /**
8075  * Translate the rte_flow group index to HW table value.
8076  *
8077  * If tunnel offload is disabled, all group ids converted to flow table
8078  * id using the standard method.
8079  * If tunnel offload is enabled, group id can be converted using the
8080  * standard or tunnel conversion method. Group conversion method
8081  * selection depends on flags in `grp_info` parameter:
8082  * - Internal (grp_info.external == 0) groups conversion uses the
8083  *   standard method.
8084  * - Group ids in JUMP action converted with the tunnel conversion.
8085  * - Group id in rule attribute conversion depends on a rule type and
8086  *   group id value:
8087  *   ** non zero group attributes converted with the tunnel method
8088  *   ** zero group attribute in non-tunnel rule is converted using the
8089  *      standard method - there's only one root table
8090  *   ** zero group attribute in steer tunnel rule is converted with the
8091  *      standard method - single root table
8092  *   ** zero group attribute in match tunnel rule is a special OvS
8093  *      case: that value is used for portability reasons. That group
8094  *      id is converted with the tunnel conversion method.
8095  *
8096  * @param[in] dev
8097  *   Port device
8098  * @param[in] tunnel
8099  *   PMD tunnel offload object
8100  * @param[in] group
8101  *   rte_flow group index value.
8102  * @param[out] table
8103  *   HW table value.
8104  * @param[in] grp_info
8105  *   flags used for conversion
8106  * @param[out] error
8107  *   Pointer to error structure.
8108  *
8109  * @return
8110  *   0 on success, a negative errno value otherwise and rte_errno is set.
8111  */
8112 int
8113 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8114 			 const struct mlx5_flow_tunnel *tunnel,
8115 			 uint32_t group, uint32_t *table,
8116 			 const struct flow_grp_info *grp_info,
8117 			 struct rte_flow_error *error)
8118 {
8119 	int ret;
8120 	bool standard_translation;
8121 
8122 	if (!grp_info->skip_scale && grp_info->external &&
8123 	    group < MLX5_MAX_TABLES_EXTERNAL)
8124 		group *= MLX5_FLOW_TABLE_FACTOR;
8125 	if (is_tunnel_offload_active(dev)) {
8126 		standard_translation = !grp_info->external ||
8127 					grp_info->std_tbl_fix;
8128 	} else {
8129 		standard_translation = true;
8130 	}
8131 	DRV_LOG(DEBUG,
8132 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8133 		dev->data->port_id, group, grp_info->transfer,
8134 		grp_info->external, grp_info->fdb_def_rule,
8135 		standard_translation ? "STANDARD" : "TUNNEL");
8136 	if (standard_translation)
8137 		ret = flow_group_to_table(dev->data->port_id, group, table,
8138 					  grp_info, error);
8139 	else
8140 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8141 						      table, error);
8142 
8143 	return ret;
8144 }
8145 
8146 /**
8147  * Discover availability of metadata reg_c's.
8148  *
8149  * Iteratively use test flows to check availability.
8150  *
8151  * @param[in] dev
8152  *   Pointer to the Ethernet device structure.
8153  *
8154  * @return
8155  *   0 on success, a negative errno value otherwise and rte_errno is set.
8156  */
8157 int
8158 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8159 {
8160 	struct mlx5_priv *priv = dev->data->dev_private;
8161 	enum modify_reg idx;
8162 	int n = 0;
8163 
8164 	/* reg_c[0] and reg_c[1] are reserved. */
8165 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8166 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8167 	/* Discover availability of other reg_c's. */
8168 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8169 		struct rte_flow_attr attr = {
8170 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8171 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8172 			.ingress = 1,
8173 		};
8174 		struct rte_flow_item items[] = {
8175 			[0] = {
8176 				.type = RTE_FLOW_ITEM_TYPE_END,
8177 			},
8178 		};
8179 		struct rte_flow_action actions[] = {
8180 			[0] = {
8181 				.type = (enum rte_flow_action_type)
8182 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8183 				.conf = &(struct mlx5_flow_action_copy_mreg){
8184 					.src = REG_C_1,
8185 					.dst = idx,
8186 				},
8187 			},
8188 			[1] = {
8189 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8190 				.conf = &(struct rte_flow_action_jump){
8191 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8192 				},
8193 			},
8194 			[2] = {
8195 				.type = RTE_FLOW_ACTION_TYPE_END,
8196 			},
8197 		};
8198 		uint32_t flow_idx;
8199 		struct rte_flow *flow;
8200 		struct rte_flow_error error;
8201 
8202 		if (!priv->config.dv_flow_en)
8203 			break;
8204 		/* Create internal flow, validation skips copy action. */
8205 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8206 					items, actions, false, &error);
8207 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8208 				      flow_idx);
8209 		if (!flow)
8210 			continue;
8211 		priv->sh->flow_mreg_c[n++] = idx;
8212 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8213 	}
8214 	for (; n < MLX5_MREG_C_NUM; ++n)
8215 		priv->sh->flow_mreg_c[n] = REG_NON;
8216 	priv->sh->metadata_regc_check_flag = 1;
8217 	return 0;
8218 }
8219 
8220 int
8221 save_dump_file(const uint8_t *data, uint32_t size,
8222 	uint32_t type, uint64_t id, void *arg, FILE *file)
8223 {
8224 	char line[BUF_SIZE];
8225 	uint32_t out = 0;
8226 	uint32_t k;
8227 	uint32_t actions_num;
8228 	struct rte_flow_query_count *count;
8229 
8230 	memset(line, 0, BUF_SIZE);
8231 	switch (type) {
8232 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8233 		actions_num = *(uint32_t *)(arg);
8234 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8235 				type, id, actions_num);
8236 		break;
8237 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8238 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8239 				type, id);
8240 		break;
8241 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8242 		count = (struct rte_flow_query_count *)arg;
8243 		fprintf(file,
8244 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8245 			type, id, count->hits, count->bytes);
8246 		return 0;
8247 	default:
8248 		return -1;
8249 	}
8250 
8251 	for (k = 0; k < size; k++) {
8252 		/* Make sure we do not overrun the line buffer length. */
8253 		if (out >= BUF_SIZE - 4) {
8254 			line[out] = '\0';
8255 			break;
8256 		}
8257 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8258 				(data[k]) & 0xff);
8259 	}
8260 	fprintf(file, "%s\n", line);
8261 	return 0;
8262 }
8263 
8264 int
8265 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8266 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8267 {
8268 	struct rte_flow_action action[2];
8269 	enum mlx5_flow_drv_type ftype;
8270 	const struct mlx5_flow_driver_ops *fops;
8271 
8272 	if (!flow) {
8273 		return rte_flow_error_set(error, ENOENT,
8274 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8275 				NULL,
8276 				"invalid flow handle");
8277 	}
8278 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8279 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8280 	if (flow->counter) {
8281 		memset(count, 0, sizeof(struct rte_flow_query_count));
8282 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8283 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8284 						ftype < MLX5_FLOW_TYPE_MAX);
8285 		fops = flow_get_drv_ops(ftype);
8286 		return fops->query(dev, flow, action, count, error);
8287 	}
8288 	return -1;
8289 }
8290 
8291 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8292 /**
8293  * Dump flow ipool data to file
8294  *
8295  * @param[in] dev
8296  *   The pointer to Ethernet device.
8297  * @param[in] file
8298  *   A pointer to a file for output.
8299  * @param[out] error
8300  *   Perform verbose error reporting if not NULL. PMDs initialize this
8301  *   structure in case of error only.
8302  * @return
8303  *   0 on success, a negative value otherwise.
8304  */
8305 int
8306 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8307 	struct rte_flow *flow, FILE *file,
8308 	struct rte_flow_error *error)
8309 {
8310 	struct mlx5_priv *priv = dev->data->dev_private;
8311 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8312 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8313 	uint32_t handle_idx;
8314 	struct mlx5_flow_handle *dh;
8315 	struct rte_flow_query_count count;
8316 	uint32_t actions_num;
8317 	const uint8_t *data;
8318 	size_t size;
8319 	uint64_t id;
8320 	uint32_t type;
8321 	void *action = NULL;
8322 
8323 	if (!flow) {
8324 		return rte_flow_error_set(error, ENOENT,
8325 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8326 				NULL,
8327 				"invalid flow handle");
8328 	}
8329 	handle_idx = flow->dev_handles;
8330 	while (handle_idx) {
8331 		dh = mlx5_ipool_get(priv->sh->ipool
8332 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8333 		if (!dh)
8334 			continue;
8335 		handle_idx = dh->next.next;
8336 
8337 		/* query counter */
8338 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8339 		flow_dv_query_count_ptr(dev, flow->counter,
8340 						&action, error);
8341 		if (action) {
8342 			id = (uint64_t)(uintptr_t)action;
8343 			if (!mlx5_flow_query_counter(dev, flow, &count, error))
8344 				save_dump_file(NULL, 0, type,
8345 						id, (void *)&count, file);
8346 		}
8347 		/* Get modify_hdr and encap_decap buf from ipools. */
8348 		encap_decap = NULL;
8349 		modify_hdr = dh->dvh.modify_hdr;
8350 
8351 		if (dh->dvh.rix_encap_decap) {
8352 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8353 						[MLX5_IPOOL_DECAP_ENCAP],
8354 						dh->dvh.rix_encap_decap);
8355 		}
8356 		if (modify_hdr) {
8357 			data = (const uint8_t *)modify_hdr->actions;
8358 			size = (size_t)(modify_hdr->actions_num) * 8;
8359 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8360 			actions_num = modify_hdr->actions_num;
8361 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8362 			save_dump_file(data, size, type, id,
8363 						(void *)(&actions_num), file);
8364 		}
8365 		if (encap_decap) {
8366 			data = encap_decap->buf;
8367 			size = encap_decap->size;
8368 			id = (uint64_t)(uintptr_t)encap_decap->action;
8369 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8370 			save_dump_file(data, size, type,
8371 						id, NULL, file);
8372 		}
8373 	}
8374 	return 0;
8375 }
8376 
8377 /**
8378  * Dump all flow's encap_decap/modify_hdr/counter data to file
8379  *
8380  * @param[in] dev
8381  *   The pointer to Ethernet device.
8382  * @param[in] file
8383  *   A pointer to a file for output.
8384  * @param[out] error
8385  *   Perform verbose error reporting if not NULL. PMDs initialize this
8386  *   structure in case of error only.
8387  * @return
8388  *   0 on success, a negative value otherwise.
8389  */
8390 static int
8391 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8392 	FILE *file, struct rte_flow_error *error)
8393 {
8394 	struct mlx5_priv *priv = dev->data->dev_private;
8395 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8396 	struct mlx5_hlist *h;
8397 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8398 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8399 	struct rte_flow_query_count count;
8400 	uint32_t actions_num;
8401 	const uint8_t *data;
8402 	size_t size;
8403 	uint64_t id;
8404 	uint32_t type;
8405 	uint32_t i;
8406 	uint32_t j;
8407 	struct mlx5_list_inconst *l_inconst;
8408 	struct mlx5_list_entry *e;
8409 	int lcore_index;
8410 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8411 	uint32_t max;
8412 	void *action;
8413 
8414 	/* encap_decap hlist is lcore_share, get global core cache. */
8415 	i = MLX5_LIST_GLOBAL;
8416 	h = sh->encaps_decaps;
8417 	if (h) {
8418 		for (j = 0; j <= h->mask; j++) {
8419 			l_inconst = &h->buckets[j].l;
8420 			if (!l_inconst || !l_inconst->cache[i])
8421 				continue;
8422 
8423 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8424 			while (e) {
8425 				encap_decap =
8426 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8427 				data = encap_decap->buf;
8428 				size = encap_decap->size;
8429 				id = (uint64_t)(uintptr_t)encap_decap->action;
8430 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8431 				save_dump_file(data, size, type,
8432 					id, NULL, file);
8433 				e = LIST_NEXT(e, next);
8434 			}
8435 		}
8436 	}
8437 
8438 	/* get modify_hdr */
8439 	h = sh->modify_cmds;
8440 	if (h) {
8441 		lcore_index = rte_lcore_index(rte_lcore_id());
8442 		if (unlikely(lcore_index == -1)) {
8443 			lcore_index = MLX5_LIST_NLCORE;
8444 			rte_spinlock_lock(&h->l_const.lcore_lock);
8445 		}
8446 		i = lcore_index;
8447 
8448 		for (j = 0; j <= h->mask; j++) {
8449 			l_inconst = &h->buckets[j].l;
8450 			if (!l_inconst || !l_inconst->cache[i])
8451 				continue;
8452 
8453 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8454 			while (e) {
8455 				modify_hdr =
8456 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8457 				data = (const uint8_t *)modify_hdr->actions;
8458 				size = (size_t)(modify_hdr->actions_num) * 8;
8459 				actions_num = modify_hdr->actions_num;
8460 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8461 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8462 				save_dump_file(data, size, type, id,
8463 						(void *)(&actions_num), file);
8464 				e = LIST_NEXT(e, next);
8465 			}
8466 		}
8467 
8468 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8469 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8470 	}
8471 
8472 	/* get counter */
8473 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8474 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8475 	for (j = 1; j <= max; j++) {
8476 		action = NULL;
8477 		flow_dv_query_count_ptr(dev, j, &action, error);
8478 		if (action) {
8479 			if (!flow_dv_query_count(dev, j, &count, error)) {
8480 				type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8481 				id = (uint64_t)(uintptr_t)action;
8482 				save_dump_file(NULL, 0, type,
8483 						id, (void *)&count, file);
8484 			}
8485 		}
8486 	}
8487 	return 0;
8488 }
8489 #endif
8490 
8491 /**
8492  * Dump flow raw hw data to file
8493  *
8494  * @param[in] dev
8495  *    The pointer to Ethernet device.
8496  * @param[in] file
8497  *   A pointer to a file for output.
8498  * @param[out] error
8499  *   Perform verbose error reporting if not NULL. PMDs initialize this
8500  *   structure in case of error only.
8501  * @return
8502  *   0 on success, a nagative value otherwise.
8503  */
8504 int
8505 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8506 		   FILE *file,
8507 		   struct rte_flow_error *error __rte_unused)
8508 {
8509 	struct mlx5_priv *priv = dev->data->dev_private;
8510 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8511 	uint32_t handle_idx;
8512 	int ret;
8513 	struct mlx5_flow_handle *dh;
8514 	struct rte_flow *flow;
8515 
8516 	if (!priv->config.dv_flow_en) {
8517 		if (fputs("device dv flow disabled\n", file) <= 0)
8518 			return -errno;
8519 		return -ENOTSUP;
8520 	}
8521 
8522 	/* dump all */
8523 	if (!flow_idx) {
8524 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8525 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8526 			return -EINVAL;
8527 #endif
8528 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8529 					sh->rx_domain,
8530 					sh->tx_domain, file);
8531 	}
8532 	/* dump one */
8533 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8534 			(uintptr_t)(void *)flow_idx);
8535 	if (!flow)
8536 		return -EINVAL;
8537 
8538 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8539 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8540 #endif
8541 	handle_idx = flow->dev_handles;
8542 	while (handle_idx) {
8543 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8544 				handle_idx);
8545 		if (!dh)
8546 			return -ENOENT;
8547 		if (dh->drv_flow) {
8548 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8549 					file);
8550 			if (ret)
8551 				return -ENOENT;
8552 		}
8553 		handle_idx = dh->next.next;
8554 	}
8555 	return 0;
8556 }
8557 
8558 /**
8559  * Get aged-out flows.
8560  *
8561  * @param[in] dev
8562  *   Pointer to the Ethernet device structure.
8563  * @param[in] context
8564  *   The address of an array of pointers to the aged-out flows contexts.
8565  * @param[in] nb_countexts
8566  *   The length of context array pointers.
8567  * @param[out] error
8568  *   Perform verbose error reporting if not NULL. Initialized in case of
8569  *   error only.
8570  *
8571  * @return
8572  *   how many contexts get in success, otherwise negative errno value.
8573  *   if nb_contexts is 0, return the amount of all aged contexts.
8574  *   if nb_contexts is not 0 , return the amount of aged flows reported
8575  *   in the context array.
8576  */
8577 int
8578 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8579 			uint32_t nb_contexts, struct rte_flow_error *error)
8580 {
8581 	const struct mlx5_flow_driver_ops *fops;
8582 	struct rte_flow_attr attr = { .transfer = 0 };
8583 
8584 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8585 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8586 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8587 						    error);
8588 	}
8589 	DRV_LOG(ERR,
8590 		"port %u get aged flows is not supported.",
8591 		 dev->data->port_id);
8592 	return -ENOTSUP;
8593 }
8594 
8595 /* Wrapper for driver action_validate op callback */
8596 static int
8597 flow_drv_action_validate(struct rte_eth_dev *dev,
8598 			 const struct rte_flow_indir_action_conf *conf,
8599 			 const struct rte_flow_action *action,
8600 			 const struct mlx5_flow_driver_ops *fops,
8601 			 struct rte_flow_error *error)
8602 {
8603 	static const char err_msg[] = "indirect action validation unsupported";
8604 
8605 	if (!fops->action_validate) {
8606 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8607 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8608 				   NULL, err_msg);
8609 		return -rte_errno;
8610 	}
8611 	return fops->action_validate(dev, conf, action, error);
8612 }
8613 
8614 /**
8615  * Destroys the shared action by handle.
8616  *
8617  * @param dev
8618  *   Pointer to Ethernet device structure.
8619  * @param[in] handle
8620  *   Handle for the indirect action object to be destroyed.
8621  * @param[out] error
8622  *   Perform verbose error reporting if not NULL. PMDs initialize this
8623  *   structure in case of error only.
8624  *
8625  * @return
8626  *   0 on success, a negative errno value otherwise and rte_errno is set.
8627  *
8628  * @note: wrapper for driver action_create op callback.
8629  */
8630 static int
8631 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8632 			   struct rte_flow_action_handle *handle,
8633 			   struct rte_flow_error *error)
8634 {
8635 	static const char err_msg[] = "indirect action destruction unsupported";
8636 	struct rte_flow_attr attr = { .transfer = 0 };
8637 	const struct mlx5_flow_driver_ops *fops =
8638 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8639 
8640 	if (!fops->action_destroy) {
8641 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8642 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8643 				   NULL, err_msg);
8644 		return -rte_errno;
8645 	}
8646 	return fops->action_destroy(dev, handle, error);
8647 }
8648 
8649 /* Wrapper for driver action_destroy op callback */
8650 static int
8651 flow_drv_action_update(struct rte_eth_dev *dev,
8652 		       struct rte_flow_action_handle *handle,
8653 		       const void *update,
8654 		       const struct mlx5_flow_driver_ops *fops,
8655 		       struct rte_flow_error *error)
8656 {
8657 	static const char err_msg[] = "indirect action update unsupported";
8658 
8659 	if (!fops->action_update) {
8660 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8661 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8662 				   NULL, err_msg);
8663 		return -rte_errno;
8664 	}
8665 	return fops->action_update(dev, handle, update, error);
8666 }
8667 
8668 /* Wrapper for driver action_destroy op callback */
8669 static int
8670 flow_drv_action_query(struct rte_eth_dev *dev,
8671 		      const struct rte_flow_action_handle *handle,
8672 		      void *data,
8673 		      const struct mlx5_flow_driver_ops *fops,
8674 		      struct rte_flow_error *error)
8675 {
8676 	static const char err_msg[] = "indirect action query unsupported";
8677 
8678 	if (!fops->action_query) {
8679 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8680 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8681 				   NULL, err_msg);
8682 		return -rte_errno;
8683 	}
8684 	return fops->action_query(dev, handle, data, error);
8685 }
8686 
8687 /**
8688  * Create indirect action for reuse in multiple flow rules.
8689  *
8690  * @param dev
8691  *   Pointer to Ethernet device structure.
8692  * @param conf
8693  *   Pointer to indirect action object configuration.
8694  * @param[in] action
8695  *   Action configuration for indirect action object creation.
8696  * @param[out] error
8697  *   Perform verbose error reporting if not NULL. PMDs initialize this
8698  *   structure in case of error only.
8699  * @return
8700  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8701  */
8702 static struct rte_flow_action_handle *
8703 mlx5_action_handle_create(struct rte_eth_dev *dev,
8704 			  const struct rte_flow_indir_action_conf *conf,
8705 			  const struct rte_flow_action *action,
8706 			  struct rte_flow_error *error)
8707 {
8708 	static const char err_msg[] = "indirect action creation unsupported";
8709 	struct rte_flow_attr attr = { .transfer = 0 };
8710 	const struct mlx5_flow_driver_ops *fops =
8711 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8712 
8713 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8714 		return NULL;
8715 	if (!fops->action_create) {
8716 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8717 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8718 				   NULL, err_msg);
8719 		return NULL;
8720 	}
8721 	return fops->action_create(dev, conf, action, error);
8722 }
8723 
8724 /**
8725  * Updates inplace the indirect action configuration pointed by *handle*
8726  * with the configuration provided as *update* argument.
8727  * The update of the indirect action configuration effects all flow rules
8728  * reusing the action via handle.
8729  *
8730  * @param dev
8731  *   Pointer to Ethernet device structure.
8732  * @param[in] handle
8733  *   Handle for the indirect action to be updated.
8734  * @param[in] update
8735  *   Action specification used to modify the action pointed by handle.
8736  *   *update* could be of same type with the action pointed by the *handle*
8737  *   handle argument, or some other structures like a wrapper, depending on
8738  *   the indirect action type.
8739  * @param[out] error
8740  *   Perform verbose error reporting if not NULL. PMDs initialize this
8741  *   structure in case of error only.
8742  *
8743  * @return
8744  *   0 on success, a negative errno value otherwise and rte_errno is set.
8745  */
8746 static int
8747 mlx5_action_handle_update(struct rte_eth_dev *dev,
8748 		struct rte_flow_action_handle *handle,
8749 		const void *update,
8750 		struct rte_flow_error *error)
8751 {
8752 	struct rte_flow_attr attr = { .transfer = 0 };
8753 	const struct mlx5_flow_driver_ops *fops =
8754 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8755 	int ret;
8756 
8757 	ret = flow_drv_action_validate(dev, NULL,
8758 			(const struct rte_flow_action *)update, fops, error);
8759 	if (ret)
8760 		return ret;
8761 	return flow_drv_action_update(dev, handle, update, fops,
8762 				      error);
8763 }
8764 
8765 /**
8766  * Query the indirect action by handle.
8767  *
8768  * This function allows retrieving action-specific data such as counters.
8769  * Data is gathered by special action which may be present/referenced in
8770  * more than one flow rule definition.
8771  *
8772  * see @RTE_FLOW_ACTION_TYPE_COUNT
8773  *
8774  * @param dev
8775  *   Pointer to Ethernet device structure.
8776  * @param[in] handle
8777  *   Handle for the indirect action to query.
8778  * @param[in, out] data
8779  *   Pointer to storage for the associated query data type.
8780  * @param[out] error
8781  *   Perform verbose error reporting if not NULL. PMDs initialize this
8782  *   structure in case of error only.
8783  *
8784  * @return
8785  *   0 on success, a negative errno value otherwise and rte_errno is set.
8786  */
8787 static int
8788 mlx5_action_handle_query(struct rte_eth_dev *dev,
8789 			 const struct rte_flow_action_handle *handle,
8790 			 void *data,
8791 			 struct rte_flow_error *error)
8792 {
8793 	struct rte_flow_attr attr = { .transfer = 0 };
8794 	const struct mlx5_flow_driver_ops *fops =
8795 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8796 
8797 	return flow_drv_action_query(dev, handle, data, fops, error);
8798 }
8799 
8800 /**
8801  * Destroy all indirect actions (shared RSS).
8802  *
8803  * @param dev
8804  *   Pointer to Ethernet device.
8805  *
8806  * @return
8807  *   0 on success, a negative errno value otherwise and rte_errno is set.
8808  */
8809 int
8810 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8811 {
8812 	struct rte_flow_error error;
8813 	struct mlx5_priv *priv = dev->data->dev_private;
8814 	struct mlx5_shared_action_rss *shared_rss;
8815 	int ret = 0;
8816 	uint32_t idx;
8817 
8818 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8819 		      priv->rss_shared_actions, idx, shared_rss, next) {
8820 		ret |= mlx5_action_handle_destroy(dev,
8821 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8822 	}
8823 	return ret;
8824 }
8825 
8826 /**
8827  * Validate existing indirect actions against current device configuration
8828  * and attach them to device resources.
8829  *
8830  * @param dev
8831  *   Pointer to Ethernet device.
8832  *
8833  * @return
8834  *   0 on success, a negative errno value otherwise and rte_errno is set.
8835  */
8836 int
8837 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8838 {
8839 	struct mlx5_priv *priv = dev->data->dev_private;
8840 	struct mlx5_indexed_pool *ipool =
8841 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8842 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8843 	int ret = 0;
8844 	uint32_t idx;
8845 
8846 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8847 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8848 		const char *message;
8849 		uint32_t queue_idx;
8850 
8851 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8852 					       ind_tbl->queues_n,
8853 					       &message, &queue_idx);
8854 		if (ret != 0) {
8855 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8856 				dev->data->port_id, ind_tbl->queues[queue_idx],
8857 				message);
8858 			break;
8859 		}
8860 	}
8861 	if (ret != 0)
8862 		return ret;
8863 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8864 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8865 
8866 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8867 		if (ret != 0) {
8868 			DRV_LOG(ERR, "Port %u could not attach "
8869 				"indirection table obj %p",
8870 				dev->data->port_id, (void *)ind_tbl);
8871 			goto error;
8872 		}
8873 	}
8874 	return 0;
8875 error:
8876 	shared_rss_last = shared_rss;
8877 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8878 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8879 
8880 		if (shared_rss == shared_rss_last)
8881 			break;
8882 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8883 			DRV_LOG(CRIT, "Port %u could not detach "
8884 				"indirection table obj %p on rollback",
8885 				dev->data->port_id, (void *)ind_tbl);
8886 	}
8887 	return ret;
8888 }
8889 
8890 /**
8891  * Detach indirect actions of the device from its resources.
8892  *
8893  * @param dev
8894  *   Pointer to Ethernet device.
8895  *
8896  * @return
8897  *   0 on success, a negative errno value otherwise and rte_errno is set.
8898  */
8899 int
8900 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8901 {
8902 	struct mlx5_priv *priv = dev->data->dev_private;
8903 	struct mlx5_indexed_pool *ipool =
8904 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8905 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8906 	int ret = 0;
8907 	uint32_t idx;
8908 
8909 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8910 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8911 
8912 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8913 		if (ret != 0) {
8914 			DRV_LOG(ERR, "Port %u could not detach "
8915 				"indirection table obj %p",
8916 				dev->data->port_id, (void *)ind_tbl);
8917 			goto error;
8918 		}
8919 	}
8920 	return 0;
8921 error:
8922 	shared_rss_last = shared_rss;
8923 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8924 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8925 
8926 		if (shared_rss == shared_rss_last)
8927 			break;
8928 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8929 			DRV_LOG(CRIT, "Port %u could not attach "
8930 				"indirection table obj %p on rollback",
8931 				dev->data->port_id, (void *)ind_tbl);
8932 	}
8933 	return ret;
8934 }
8935 
8936 #ifndef HAVE_MLX5DV_DR
8937 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8938 #else
8939 #define MLX5_DOMAIN_SYNC_FLOW \
8940 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8941 #endif
8942 
8943 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8944 {
8945 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8946 	const struct mlx5_flow_driver_ops *fops;
8947 	int ret;
8948 	struct rte_flow_attr attr = { .transfer = 0 };
8949 
8950 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8951 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8952 	if (ret > 0)
8953 		ret = -ret;
8954 	return ret;
8955 }
8956 
8957 const struct mlx5_flow_tunnel *
8958 mlx5_get_tof(const struct rte_flow_item *item,
8959 	     const struct rte_flow_action *action,
8960 	     enum mlx5_tof_rule_type *rule_type)
8961 {
8962 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8963 		if (item->type == (typeof(item->type))
8964 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8965 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8966 			return flow_items_to_tunnel(item);
8967 		}
8968 	}
8969 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8970 		if (action->type == (typeof(action->type))
8971 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8972 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8973 			return flow_actions_to_tunnel(action);
8974 		}
8975 	}
8976 	return NULL;
8977 }
8978 
8979 /**
8980  * tunnel offload functionalilty is defined for DV environment only
8981  */
8982 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8983 __extension__
8984 union tunnel_offload_mark {
8985 	uint32_t val;
8986 	struct {
8987 		uint32_t app_reserve:8;
8988 		uint32_t table_id:15;
8989 		uint32_t transfer:1;
8990 		uint32_t _unused_:8;
8991 	};
8992 };
8993 
8994 static bool
8995 mlx5_access_tunnel_offload_db
8996 	(struct rte_eth_dev *dev,
8997 	 bool (*match)(struct rte_eth_dev *,
8998 		       struct mlx5_flow_tunnel *, const void *),
8999 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9000 	 void (*miss)(struct rte_eth_dev *, void *),
9001 	 void *ctx, bool lock_op);
9002 
9003 static int
9004 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9005 			     struct rte_flow *flow,
9006 			     const struct rte_flow_attr *attr,
9007 			     const struct rte_flow_action *app_actions,
9008 			     uint32_t flow_idx,
9009 			     const struct mlx5_flow_tunnel *tunnel,
9010 			     struct tunnel_default_miss_ctx *ctx,
9011 			     struct rte_flow_error *error)
9012 {
9013 	struct mlx5_priv *priv = dev->data->dev_private;
9014 	struct mlx5_flow *dev_flow;
9015 	struct rte_flow_attr miss_attr = *attr;
9016 	const struct rte_flow_item miss_items[2] = {
9017 		{
9018 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9019 			.spec = NULL,
9020 			.last = NULL,
9021 			.mask = NULL
9022 		},
9023 		{
9024 			.type = RTE_FLOW_ITEM_TYPE_END,
9025 			.spec = NULL,
9026 			.last = NULL,
9027 			.mask = NULL
9028 		}
9029 	};
9030 	union tunnel_offload_mark mark_id;
9031 	struct rte_flow_action_mark miss_mark;
9032 	struct rte_flow_action miss_actions[3] = {
9033 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9034 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9035 	};
9036 	const struct rte_flow_action_jump *jump_data;
9037 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9038 	struct flow_grp_info grp_info = {
9039 		.external = 1,
9040 		.transfer = attr->transfer,
9041 		.fdb_def_rule = !!priv->fdb_def_rule,
9042 		.std_tbl_fix = 0,
9043 	};
9044 	int ret;
9045 
9046 	if (!attr->transfer) {
9047 		uint32_t q_size;
9048 
9049 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9050 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9051 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9052 					 0, SOCKET_ID_ANY);
9053 		if (!ctx->queue)
9054 			return rte_flow_error_set
9055 				(error, ENOMEM,
9056 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9057 				NULL, "invalid default miss RSS");
9058 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9059 		ctx->action_rss.level = 0,
9060 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9061 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9062 		ctx->action_rss.queue_num = priv->reta_idx_n,
9063 		ctx->action_rss.key = priv->rss_conf.rss_key,
9064 		ctx->action_rss.queue = ctx->queue;
9065 		if (!priv->reta_idx_n || !priv->rxqs_n)
9066 			return rte_flow_error_set
9067 				(error, EINVAL,
9068 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9069 				NULL, "invalid port configuration");
9070 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9071 			ctx->action_rss.types = 0;
9072 		for (i = 0; i != priv->reta_idx_n; ++i)
9073 			ctx->queue[i] = (*priv->reta_idx)[i];
9074 	} else {
9075 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9076 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9077 	}
9078 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9079 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9080 	jump_data = app_actions->conf;
9081 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9082 	miss_attr.group = jump_data->group;
9083 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9084 				       &flow_table, &grp_info, error);
9085 	if (ret)
9086 		return rte_flow_error_set(error, EINVAL,
9087 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9088 					  NULL, "invalid tunnel id");
9089 	mark_id.app_reserve = 0;
9090 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9091 	mark_id.transfer = !!attr->transfer;
9092 	mark_id._unused_ = 0;
9093 	miss_mark.id = mark_id.val;
9094 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9095 				    miss_items, miss_actions, flow_idx, error);
9096 	if (!dev_flow)
9097 		return -rte_errno;
9098 	dev_flow->flow = flow;
9099 	dev_flow->external = true;
9100 	dev_flow->tunnel = tunnel;
9101 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9102 	/* Subflow object was created, we must include one in the list. */
9103 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9104 		      dev_flow->handle, next);
9105 	DRV_LOG(DEBUG,
9106 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9107 		dev->data->port_id, tunnel->app_tunnel.type,
9108 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9109 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9110 				  miss_actions, error);
9111 	if (!ret)
9112 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9113 						  error);
9114 
9115 	return ret;
9116 }
9117 
9118 static const struct mlx5_flow_tbl_data_entry  *
9119 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9120 {
9121 	struct mlx5_priv *priv = dev->data->dev_private;
9122 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9123 	struct mlx5_list_entry *he;
9124 	union tunnel_offload_mark mbits = { .val = mark };
9125 	union mlx5_flow_tbl_key table_key = {
9126 		{
9127 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9128 			.id = 0,
9129 			.reserved = 0,
9130 			.dummy = 0,
9131 			.is_fdb = !!mbits.transfer,
9132 			.is_egress = 0,
9133 		}
9134 	};
9135 	struct mlx5_flow_cb_ctx ctx = {
9136 		.data = &table_key.v64,
9137 	};
9138 
9139 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9140 	return he ?
9141 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9142 }
9143 
9144 static void
9145 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9146 				   struct mlx5_list_entry *entry)
9147 {
9148 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9149 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9150 
9151 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9152 			tunnel_flow_tbl_to_id(tte->flow_table));
9153 	mlx5_free(tte);
9154 }
9155 
9156 static int
9157 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9158 				  struct mlx5_list_entry *entry, void *cb_ctx)
9159 {
9160 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9161 	union tunnel_tbl_key tbl = {
9162 		.val = *(uint64_t *)(ctx->data),
9163 	};
9164 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9165 
9166 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9167 }
9168 
9169 static struct mlx5_list_entry *
9170 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9171 {
9172 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9173 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9174 	struct tunnel_tbl_entry *tte;
9175 	union tunnel_tbl_key tbl = {
9176 		.val = *(uint64_t *)(ctx->data),
9177 	};
9178 
9179 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9180 			  sizeof(*tte), 0,
9181 			  SOCKET_ID_ANY);
9182 	if (!tte)
9183 		goto err;
9184 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9185 			  &tte->flow_table);
9186 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9187 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9188 			tte->flow_table);
9189 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9190 				tte->flow_table);
9191 		goto err;
9192 	} else if (!tte->flow_table) {
9193 		goto err;
9194 	}
9195 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9196 	tte->tunnel_id = tbl.tunnel_id;
9197 	tte->group = tbl.group;
9198 	return &tte->hash;
9199 err:
9200 	if (tte)
9201 		mlx5_free(tte);
9202 	return NULL;
9203 }
9204 
9205 static struct mlx5_list_entry *
9206 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9207 				  struct mlx5_list_entry *oentry,
9208 				  void *cb_ctx __rte_unused)
9209 {
9210 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9211 						   0, SOCKET_ID_ANY);
9212 
9213 	if (!tte)
9214 		return NULL;
9215 	memcpy(tte, oentry, sizeof(*tte));
9216 	return &tte->hash;
9217 }
9218 
9219 static void
9220 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9221 				       struct mlx5_list_entry *entry)
9222 {
9223 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9224 
9225 	mlx5_free(tte);
9226 }
9227 
9228 static uint32_t
9229 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9230 				const struct mlx5_flow_tunnel *tunnel,
9231 				uint32_t group, uint32_t *table,
9232 				struct rte_flow_error *error)
9233 {
9234 	struct mlx5_list_entry *he;
9235 	struct tunnel_tbl_entry *tte;
9236 	union tunnel_tbl_key key = {
9237 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9238 		.group = group
9239 	};
9240 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9241 	struct mlx5_hlist *group_hash;
9242 	struct mlx5_flow_cb_ctx ctx = {
9243 		.data = &key.val,
9244 	};
9245 
9246 	group_hash = tunnel ? tunnel->groups : thub->groups;
9247 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9248 	if (!he)
9249 		return rte_flow_error_set(error, EINVAL,
9250 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9251 					  NULL,
9252 					  "tunnel group index not supported");
9253 	tte = container_of(he, typeof(*tte), hash);
9254 	*table = tte->flow_table;
9255 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9256 		dev->data->port_id, key.tunnel_id, group, *table);
9257 	return 0;
9258 }
9259 
9260 static void
9261 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9262 		      struct mlx5_flow_tunnel *tunnel)
9263 {
9264 	struct mlx5_priv *priv = dev->data->dev_private;
9265 	struct mlx5_indexed_pool *ipool;
9266 
9267 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9268 		dev->data->port_id, tunnel->tunnel_id);
9269 	LIST_REMOVE(tunnel, chain);
9270 	mlx5_hlist_destroy(tunnel->groups);
9271 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9272 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9273 }
9274 
9275 static bool
9276 mlx5_access_tunnel_offload_db
9277 	(struct rte_eth_dev *dev,
9278 	 bool (*match)(struct rte_eth_dev *,
9279 		       struct mlx5_flow_tunnel *, const void *),
9280 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9281 	 void (*miss)(struct rte_eth_dev *, void *),
9282 	 void *ctx, bool lock_op)
9283 {
9284 	bool verdict = false;
9285 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9286 	struct mlx5_flow_tunnel *tunnel;
9287 
9288 	rte_spinlock_lock(&thub->sl);
9289 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9290 		verdict = match(dev, tunnel, (const void *)ctx);
9291 		if (verdict)
9292 			break;
9293 	}
9294 	if (!lock_op)
9295 		rte_spinlock_unlock(&thub->sl);
9296 	if (verdict && hit)
9297 		hit(dev, tunnel, ctx);
9298 	if (!verdict && miss)
9299 		miss(dev, ctx);
9300 	if (lock_op)
9301 		rte_spinlock_unlock(&thub->sl);
9302 
9303 	return verdict;
9304 }
9305 
9306 struct tunnel_db_find_tunnel_id_ctx {
9307 	uint32_t tunnel_id;
9308 	struct mlx5_flow_tunnel *tunnel;
9309 };
9310 
9311 static bool
9312 find_tunnel_id_match(struct rte_eth_dev *dev,
9313 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9314 {
9315 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9316 
9317 	RTE_SET_USED(dev);
9318 	return tunnel->tunnel_id == ctx->tunnel_id;
9319 }
9320 
9321 static void
9322 find_tunnel_id_hit(struct rte_eth_dev *dev,
9323 		   struct mlx5_flow_tunnel *tunnel, void *x)
9324 {
9325 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9326 	RTE_SET_USED(dev);
9327 	ctx->tunnel = tunnel;
9328 }
9329 
9330 static struct mlx5_flow_tunnel *
9331 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9332 {
9333 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9334 		.tunnel_id = id,
9335 	};
9336 
9337 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9338 				      find_tunnel_id_hit, NULL, &ctx, true);
9339 
9340 	return ctx.tunnel;
9341 }
9342 
9343 static struct mlx5_flow_tunnel *
9344 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9345 			  const struct rte_flow_tunnel *app_tunnel)
9346 {
9347 	struct mlx5_priv *priv = dev->data->dev_private;
9348 	struct mlx5_indexed_pool *ipool;
9349 	struct mlx5_flow_tunnel *tunnel;
9350 	uint32_t id;
9351 
9352 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9353 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9354 	if (!tunnel)
9355 		return NULL;
9356 	if (id >= MLX5_MAX_TUNNELS) {
9357 		mlx5_ipool_free(ipool, id);
9358 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9359 		return NULL;
9360 	}
9361 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9362 					   priv->sh,
9363 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9364 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9365 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9366 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9367 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9368 	if (!tunnel->groups) {
9369 		mlx5_ipool_free(ipool, id);
9370 		return NULL;
9371 	}
9372 	/* initiate new PMD tunnel */
9373 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9374 	tunnel->tunnel_id = id;
9375 	tunnel->action.type = (typeof(tunnel->action.type))
9376 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9377 	tunnel->action.conf = tunnel;
9378 	tunnel->item.type = (typeof(tunnel->item.type))
9379 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9380 	tunnel->item.spec = tunnel;
9381 	tunnel->item.last = NULL;
9382 	tunnel->item.mask = NULL;
9383 
9384 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9385 		dev->data->port_id, tunnel->tunnel_id);
9386 
9387 	return tunnel;
9388 }
9389 
9390 struct tunnel_db_get_tunnel_ctx {
9391 	const struct rte_flow_tunnel *app_tunnel;
9392 	struct mlx5_flow_tunnel *tunnel;
9393 };
9394 
9395 static bool get_tunnel_match(struct rte_eth_dev *dev,
9396 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9397 {
9398 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9399 
9400 	RTE_SET_USED(dev);
9401 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9402 		       sizeof(*ctx->app_tunnel));
9403 }
9404 
9405 static void get_tunnel_hit(struct rte_eth_dev *dev,
9406 			   struct mlx5_flow_tunnel *tunnel, void *x)
9407 {
9408 	/* called under tunnel spinlock protection */
9409 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9410 
9411 	RTE_SET_USED(dev);
9412 	tunnel->refctn++;
9413 	ctx->tunnel = tunnel;
9414 }
9415 
9416 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9417 {
9418 	/* called under tunnel spinlock protection */
9419 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9420 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9421 
9422 	rte_spinlock_unlock(&thub->sl);
9423 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9424 	rte_spinlock_lock(&thub->sl);
9425 	if (ctx->tunnel) {
9426 		ctx->tunnel->refctn = 1;
9427 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9428 	}
9429 }
9430 
9431 
9432 static int
9433 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9434 		     const struct rte_flow_tunnel *app_tunnel,
9435 		     struct mlx5_flow_tunnel **tunnel)
9436 {
9437 	struct tunnel_db_get_tunnel_ctx ctx = {
9438 		.app_tunnel = app_tunnel,
9439 	};
9440 
9441 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9442 				      get_tunnel_miss, &ctx, true);
9443 	*tunnel = ctx.tunnel;
9444 	return ctx.tunnel ? 0 : -ENOMEM;
9445 }
9446 
9447 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9448 {
9449 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9450 
9451 	if (!thub)
9452 		return;
9453 	if (!LIST_EMPTY(&thub->tunnels))
9454 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9455 	mlx5_hlist_destroy(thub->groups);
9456 	mlx5_free(thub);
9457 }
9458 
9459 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9460 {
9461 	int err;
9462 	struct mlx5_flow_tunnel_hub *thub;
9463 
9464 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9465 			   0, SOCKET_ID_ANY);
9466 	if (!thub)
9467 		return -ENOMEM;
9468 	LIST_INIT(&thub->tunnels);
9469 	rte_spinlock_init(&thub->sl);
9470 	thub->groups = mlx5_hlist_create("flow groups", 64,
9471 					 false, true, sh,
9472 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9473 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9474 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9475 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9476 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9477 	if (!thub->groups) {
9478 		err = -rte_errno;
9479 		goto err;
9480 	}
9481 	sh->tunnel_hub = thub;
9482 
9483 	return 0;
9484 
9485 err:
9486 	if (thub->groups)
9487 		mlx5_hlist_destroy(thub->groups);
9488 	if (thub)
9489 		mlx5_free(thub);
9490 	return err;
9491 }
9492 
9493 static inline int
9494 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9495 			  struct rte_flow_tunnel *tunnel,
9496 			  struct rte_flow_error *error)
9497 {
9498 	struct mlx5_priv *priv = dev->data->dev_private;
9499 
9500 	if (!priv->config.dv_flow_en)
9501 		return rte_flow_error_set(error, ENOTSUP,
9502 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9503 					  "flow DV interface is off");
9504 	if (!is_tunnel_offload_active(dev))
9505 		return rte_flow_error_set(error, ENOTSUP,
9506 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9507 					  "tunnel offload was not activated");
9508 	if (!tunnel)
9509 		return rte_flow_error_set(error, EINVAL,
9510 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9511 					  "no application tunnel");
9512 	switch (tunnel->type) {
9513 	default:
9514 		return rte_flow_error_set(error, EINVAL,
9515 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9516 					  "unsupported tunnel type");
9517 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9518 	case RTE_FLOW_ITEM_TYPE_GRE:
9519 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9520 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9521 		break;
9522 	}
9523 	return 0;
9524 }
9525 
9526 static int
9527 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9528 		    struct rte_flow_tunnel *app_tunnel,
9529 		    struct rte_flow_action **actions,
9530 		    uint32_t *num_of_actions,
9531 		    struct rte_flow_error *error)
9532 {
9533 	struct mlx5_flow_tunnel *tunnel;
9534 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9535 
9536 	if (ret)
9537 		return ret;
9538 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9539 	if (ret < 0) {
9540 		return rte_flow_error_set(error, ret,
9541 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9542 					  "failed to initialize pmd tunnel");
9543 	}
9544 	*actions = &tunnel->action;
9545 	*num_of_actions = 1;
9546 	return 0;
9547 }
9548 
9549 static int
9550 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9551 		       struct rte_flow_tunnel *app_tunnel,
9552 		       struct rte_flow_item **items,
9553 		       uint32_t *num_of_items,
9554 		       struct rte_flow_error *error)
9555 {
9556 	struct mlx5_flow_tunnel *tunnel;
9557 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9558 
9559 	if (ret)
9560 		return ret;
9561 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9562 	if (ret < 0) {
9563 		return rte_flow_error_set(error, ret,
9564 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9565 					  "failed to initialize pmd tunnel");
9566 	}
9567 	*items = &tunnel->item;
9568 	*num_of_items = 1;
9569 	return 0;
9570 }
9571 
9572 struct tunnel_db_element_release_ctx {
9573 	struct rte_flow_item *items;
9574 	struct rte_flow_action *actions;
9575 	uint32_t num_elements;
9576 	struct rte_flow_error *error;
9577 	int ret;
9578 };
9579 
9580 static bool
9581 tunnel_element_release_match(struct rte_eth_dev *dev,
9582 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9583 {
9584 	const struct tunnel_db_element_release_ctx *ctx = x;
9585 
9586 	RTE_SET_USED(dev);
9587 	if (ctx->num_elements != 1)
9588 		return false;
9589 	else if (ctx->items)
9590 		return ctx->items == &tunnel->item;
9591 	else if (ctx->actions)
9592 		return ctx->actions == &tunnel->action;
9593 
9594 	return false;
9595 }
9596 
9597 static void
9598 tunnel_element_release_hit(struct rte_eth_dev *dev,
9599 			   struct mlx5_flow_tunnel *tunnel, void *x)
9600 {
9601 	struct tunnel_db_element_release_ctx *ctx = x;
9602 	ctx->ret = 0;
9603 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9604 		mlx5_flow_tunnel_free(dev, tunnel);
9605 }
9606 
9607 static void
9608 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9609 {
9610 	struct tunnel_db_element_release_ctx *ctx = x;
9611 	RTE_SET_USED(dev);
9612 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9613 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9614 				      "invalid argument");
9615 }
9616 
9617 static int
9618 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9619 		       struct rte_flow_item *pmd_items,
9620 		       uint32_t num_items, struct rte_flow_error *err)
9621 {
9622 	struct tunnel_db_element_release_ctx ctx = {
9623 		.items = pmd_items,
9624 		.actions = NULL,
9625 		.num_elements = num_items,
9626 		.error = err,
9627 	};
9628 
9629 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9630 				      tunnel_element_release_hit,
9631 				      tunnel_element_release_miss, &ctx, false);
9632 
9633 	return ctx.ret;
9634 }
9635 
9636 static int
9637 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9638 			 struct rte_flow_action *pmd_actions,
9639 			 uint32_t num_actions, struct rte_flow_error *err)
9640 {
9641 	struct tunnel_db_element_release_ctx ctx = {
9642 		.items = NULL,
9643 		.actions = pmd_actions,
9644 		.num_elements = num_actions,
9645 		.error = err,
9646 	};
9647 
9648 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9649 				      tunnel_element_release_hit,
9650 				      tunnel_element_release_miss, &ctx, false);
9651 
9652 	return ctx.ret;
9653 }
9654 
9655 static int
9656 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9657 				  struct rte_mbuf *m,
9658 				  struct rte_flow_restore_info *info,
9659 				  struct rte_flow_error *err)
9660 {
9661 	uint64_t ol_flags = m->ol_flags;
9662 	const struct mlx5_flow_tbl_data_entry *tble;
9663 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9664 
9665 	if (!is_tunnel_offload_active(dev)) {
9666 		info->flags = 0;
9667 		return 0;
9668 	}
9669 
9670 	if ((ol_flags & mask) != mask)
9671 		goto err;
9672 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9673 	if (!tble) {
9674 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9675 			dev->data->port_id, m->hash.fdir.hi);
9676 		goto err;
9677 	}
9678 	MLX5_ASSERT(tble->tunnel);
9679 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9680 	info->group_id = tble->group_id;
9681 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9682 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9683 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9684 
9685 	return 0;
9686 
9687 err:
9688 	return rte_flow_error_set(err, EINVAL,
9689 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9690 				  "failed to get restore info");
9691 }
9692 
9693 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9694 static int
9695 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9696 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9697 			   __rte_unused struct rte_flow_action **actions,
9698 			   __rte_unused uint32_t *num_of_actions,
9699 			   __rte_unused struct rte_flow_error *error)
9700 {
9701 	return -ENOTSUP;
9702 }
9703 
9704 static int
9705 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9706 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9707 		       __rte_unused struct rte_flow_item **items,
9708 		       __rte_unused uint32_t *num_of_items,
9709 		       __rte_unused struct rte_flow_error *error)
9710 {
9711 	return -ENOTSUP;
9712 }
9713 
9714 static int
9715 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9716 			      __rte_unused struct rte_flow_item *pmd_items,
9717 			      __rte_unused uint32_t num_items,
9718 			      __rte_unused struct rte_flow_error *err)
9719 {
9720 	return -ENOTSUP;
9721 }
9722 
9723 static int
9724 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9725 				__rte_unused struct rte_flow_action *pmd_action,
9726 				__rte_unused uint32_t num_actions,
9727 				__rte_unused struct rte_flow_error *err)
9728 {
9729 	return -ENOTSUP;
9730 }
9731 
9732 static int
9733 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9734 				  __rte_unused struct rte_mbuf *m,
9735 				  __rte_unused struct rte_flow_restore_info *i,
9736 				  __rte_unused struct rte_flow_error *err)
9737 {
9738 	return -ENOTSUP;
9739 }
9740 
9741 static int
9742 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9743 			     __rte_unused struct rte_flow *flow,
9744 			     __rte_unused const struct rte_flow_attr *attr,
9745 			     __rte_unused const struct rte_flow_action *actions,
9746 			     __rte_unused uint32_t flow_idx,
9747 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9748 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9749 			     __rte_unused struct rte_flow_error *error)
9750 {
9751 	return -ENOTSUP;
9752 }
9753 
9754 static struct mlx5_flow_tunnel *
9755 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9756 		    __rte_unused uint32_t id)
9757 {
9758 	return NULL;
9759 }
9760 
9761 static void
9762 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9763 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9764 {
9765 }
9766 
9767 static uint32_t
9768 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9769 				__rte_unused const struct mlx5_flow_tunnel *t,
9770 				__rte_unused uint32_t group,
9771 				__rte_unused uint32_t *table,
9772 				struct rte_flow_error *error)
9773 {
9774 	return rte_flow_error_set(error, ENOTSUP,
9775 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9776 				  "tunnel offload requires DV support");
9777 }
9778 
9779 void
9780 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9781 			__rte_unused  uint16_t port_id)
9782 {
9783 }
9784 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9785 
9786 /* Flex flow item API */
9787 static struct rte_flow_item_flex_handle *
9788 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9789 			   const struct rte_flow_item_flex_conf *conf,
9790 			   struct rte_flow_error *error)
9791 {
9792 	static const char err_msg[] = "flex item creation unsupported";
9793 	struct rte_flow_attr attr = { .transfer = 0 };
9794 	const struct mlx5_flow_driver_ops *fops =
9795 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9796 
9797 	if (!fops->item_create) {
9798 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9799 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9800 				   NULL, err_msg);
9801 		return NULL;
9802 	}
9803 	return fops->item_create(dev, conf, error);
9804 }
9805 
9806 static int
9807 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9808 			    const struct rte_flow_item_flex_handle *handle,
9809 			    struct rte_flow_error *error)
9810 {
9811 	static const char err_msg[] = "flex item release unsupported";
9812 	struct rte_flow_attr attr = { .transfer = 0 };
9813 	const struct mlx5_flow_driver_ops *fops =
9814 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9815 
9816 	if (!fops->item_release) {
9817 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9818 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9819 				   NULL, err_msg);
9820 		return -rte_errno;
9821 	}
9822 	return fops->item_release(dev, handle, error);
9823 }
9824 
9825 static void
9826 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9827 {
9828 	int ret;
9829 	struct rte_flow_error error;
9830 
9831 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9832 		char *item_name;
9833 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9834 				    sizeof(item_name),
9835 				    (void *)(uintptr_t)item->type, &error);
9836 		if (ret > 0)
9837 			printf("%s ", item_name);
9838 		else
9839 			printf("%d\n", (int)item->type);
9840 	}
9841 	printf("END\n");
9842 }
9843 
9844 static int
9845 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9846 {
9847 	const struct rte_flow_item_udp *spec = udp_item->spec;
9848 	const struct rte_flow_item_udp *mask = udp_item->mask;
9849 	uint16_t udp_dport = 0;
9850 
9851 	if (spec != NULL) {
9852 		if (!mask)
9853 			mask = &rte_flow_item_udp_mask;
9854 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9855 				mask->hdr.dst_port);
9856 	}
9857 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9858 }
9859 
9860 static const struct mlx5_flow_expand_node *
9861 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9862 		unsigned int item_idx,
9863 		const struct mlx5_flow_expand_node graph[],
9864 		const struct mlx5_flow_expand_node *node)
9865 {
9866 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9867 
9868 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9869 			node != NULL &&
9870 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9871 		/*
9872 		 * The expansion node is VXLAN and it is also the last
9873 		 * expandable item in the pattern, so need to continue
9874 		 * expansion of the inner tunnel.
9875 		 */
9876 		MLX5_ASSERT(item_idx > 0);
9877 		prev_item = pattern + item_idx - 1;
9878 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9879 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9880 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9881 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9882 	}
9883 	return node;
9884 }
9885 
9886 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9887 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9888 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9889 };
9890 
9891 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9892 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9893 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9894 	{ 9, 10, 11 }, { 12, 13, 14 },
9895 };
9896 
9897 /**
9898  * Discover the number of available flow priorities.
9899  *
9900  * @param dev
9901  *   Ethernet device.
9902  *
9903  * @return
9904  *   On success, number of available flow priorities.
9905  *   On failure, a negative errno-style code and rte_errno is set.
9906  */
9907 int
9908 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9909 {
9910 	static const uint16_t vprio[] = {8, 16};
9911 	const struct mlx5_priv *priv = dev->data->dev_private;
9912 	const struct mlx5_flow_driver_ops *fops;
9913 	enum mlx5_flow_drv_type type;
9914 	int ret;
9915 
9916 	type = mlx5_flow_os_get_type();
9917 	if (type == MLX5_FLOW_TYPE_MAX) {
9918 		type = MLX5_FLOW_TYPE_VERBS;
9919 		if (priv->sh->devx && priv->config.dv_flow_en)
9920 			type = MLX5_FLOW_TYPE_DV;
9921 	}
9922 	fops = flow_get_drv_ops(type);
9923 	if (fops->discover_priorities == NULL) {
9924 		DRV_LOG(ERR, "Priority discovery not supported");
9925 		rte_errno = ENOTSUP;
9926 		return -rte_errno;
9927 	}
9928 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9929 	if (ret < 0)
9930 		return ret;
9931 	switch (ret) {
9932 	case 8:
9933 		ret = RTE_DIM(priority_map_3);
9934 		break;
9935 	case 16:
9936 		ret = RTE_DIM(priority_map_5);
9937 		break;
9938 	default:
9939 		rte_errno = ENOTSUP;
9940 		DRV_LOG(ERR,
9941 			"port %u maximum priority: %d expected 8/16",
9942 			dev->data->port_id, ret);
9943 		return -rte_errno;
9944 	}
9945 	DRV_LOG(INFO, "port %u supported flow priorities:"
9946 		" 0-%d for ingress or egress root table,"
9947 		" 0-%d for non-root table or transfer root table.",
9948 		dev->data->port_id, ret - 2,
9949 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9950 	return ret;
9951 }
9952 
9953 /**
9954  * Adjust flow priority based on the highest layer and the request priority.
9955  *
9956  * @param[in] dev
9957  *   Pointer to the Ethernet device structure.
9958  * @param[in] priority
9959  *   The rule base priority.
9960  * @param[in] subpriority
9961  *   The priority based on the items.
9962  *
9963  * @return
9964  *   The new priority.
9965  */
9966 uint32_t
9967 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
9968 			  uint32_t subpriority)
9969 {
9970 	uint32_t res = 0;
9971 	struct mlx5_priv *priv = dev->data->dev_private;
9972 
9973 	switch (priv->sh->flow_max_priority) {
9974 	case RTE_DIM(priority_map_3):
9975 		res = priority_map_3[priority][subpriority];
9976 		break;
9977 	case RTE_DIM(priority_map_5):
9978 		res = priority_map_5[priority][subpriority];
9979 		break;
9980 	}
9981 	return  res;
9982 }
9983