xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 8b8036a66e3d59ffa58afb8d96fa2c73262155a7)
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 		/* All features use the same REG_C. */
911 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
912 		return priv->mtr_color_reg;
913 	case MLX5_COPY_MARK:
914 		/*
915 		 * Metadata COPY_MARK register using is in meter suffix sub
916 		 * flow while with meter. It's safe to share the same register.
917 		 */
918 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
919 	case MLX5_APP_TAG:
920 		/*
921 		 * If meter is enable, it will engage the register for color
922 		 * match and flow match. If meter color match is not using the
923 		 * REG_C_2, need to skip the REG_C_x be used by meter color
924 		 * match.
925 		 * If meter is disable, free to use all available registers.
926 		 */
927 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
928 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
929 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
930 		if (id > (uint32_t)(REG_C_7 - start_reg))
931 			return rte_flow_error_set(error, EINVAL,
932 						  RTE_FLOW_ERROR_TYPE_ITEM,
933 						  NULL, "invalid tag id");
934 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
935 			return rte_flow_error_set(error, ENOTSUP,
936 						  RTE_FLOW_ERROR_TYPE_ITEM,
937 						  NULL, "unsupported tag id");
938 		/*
939 		 * This case means meter is using the REG_C_x great than 2.
940 		 * Take care not to conflict with meter color REG_C_x.
941 		 * If the available index REG_C_y >= REG_C_x, skip the
942 		 * color register.
943 		 */
944 		if (skip_mtr_reg && priv->sh->flow_mreg_c
945 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
946 			if (id >= (uint32_t)(REG_C_7 - start_reg))
947 				return rte_flow_error_set(error, EINVAL,
948 						       RTE_FLOW_ERROR_TYPE_ITEM,
949 							NULL, "invalid tag id");
950 			if (priv->sh->flow_mreg_c
951 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
952 				return priv->sh->flow_mreg_c
953 					       [id + 1 + start_reg - REG_C_0];
954 			return rte_flow_error_set(error, ENOTSUP,
955 						  RTE_FLOW_ERROR_TYPE_ITEM,
956 						  NULL, "unsupported tag id");
957 		}
958 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
959 	}
960 	MLX5_ASSERT(false);
961 	return rte_flow_error_set(error, EINVAL,
962 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
963 				  NULL, "invalid feature name");
964 }
965 
966 /**
967  * Check extensive flow metadata register support.
968  *
969  * @param dev
970  *   Pointer to rte_eth_dev structure.
971  *
972  * @return
973  *   True if device supports extensive flow metadata register, otherwise false.
974  */
975 bool
976 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
977 {
978 	struct mlx5_priv *priv = dev->data->dev_private;
979 
980 	/*
981 	 * Having available reg_c can be regarded inclusively as supporting
982 	 * extensive flow metadata register, which could mean,
983 	 * - metadata register copy action by modify header.
984 	 * - 16 modify header actions is supported.
985 	 * - reg_c's are preserved across different domain (FDB and NIC) on
986 	 *   packet loopback by flow lookup miss.
987 	 */
988 	return priv->sh->flow_mreg_c[2] != REG_NON;
989 }
990 
991 /**
992  * Get the lowest priority.
993  *
994  * @param[in] dev
995  *   Pointer to the Ethernet device structure.
996  * @param[in] attributes
997  *   Pointer to device flow rule attributes.
998  *
999  * @return
1000  *   The value of lowest priority of flow.
1001  */
1002 uint32_t
1003 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1004 			  const struct rte_flow_attr *attr)
1005 {
1006 	struct mlx5_priv *priv = dev->data->dev_private;
1007 
1008 	if (!attr->group && !attr->transfer)
1009 		return priv->sh->flow_max_priority - 2;
1010 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1011 }
1012 
1013 /**
1014  * Calculate matcher priority of the flow.
1015  *
1016  * @param[in] dev
1017  *   Pointer to the Ethernet device structure.
1018  * @param[in] attr
1019  *   Pointer to device flow rule attributes.
1020  * @param[in] subpriority
1021  *   The priority based on the items.
1022  * @param[in] external
1023  *   Flow is user flow.
1024  * @return
1025  *   The matcher priority of the flow.
1026  */
1027 uint16_t
1028 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1029 			  const struct rte_flow_attr *attr,
1030 			  uint32_t subpriority, bool external)
1031 {
1032 	uint16_t priority = (uint16_t)attr->priority;
1033 	struct mlx5_priv *priv = dev->data->dev_private;
1034 
1035 	if (!attr->group && !attr->transfer) {
1036 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1037 			priority = priv->sh->flow_max_priority - 1;
1038 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1039 	} else if (!external && attr->transfer && attr->group == 0 &&
1040 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1041 		return (priv->sh->flow_max_priority - 1) * 3;
1042 	}
1043 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1044 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1045 	return priority * 3 + subpriority;
1046 }
1047 
1048 /**
1049  * Verify the @p item specifications (spec, last, mask) are compatible with the
1050  * NIC capabilities.
1051  *
1052  * @param[in] item
1053  *   Item specification.
1054  * @param[in] mask
1055  *   @p item->mask or flow default bit-masks.
1056  * @param[in] nic_mask
1057  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1058  * @param[in] size
1059  *   Bit-masks size in bytes.
1060  * @param[in] range_accepted
1061  *   True if range of values is accepted for specific fields, false otherwise.
1062  * @param[out] error
1063  *   Pointer to error structure.
1064  *
1065  * @return
1066  *   0 on success, a negative errno value otherwise and rte_errno is set.
1067  */
1068 int
1069 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1070 			  const uint8_t *mask,
1071 			  const uint8_t *nic_mask,
1072 			  unsigned int size,
1073 			  bool range_accepted,
1074 			  struct rte_flow_error *error)
1075 {
1076 	unsigned int i;
1077 
1078 	MLX5_ASSERT(nic_mask);
1079 	for (i = 0; i < size; ++i)
1080 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1081 			return rte_flow_error_set(error, ENOTSUP,
1082 						  RTE_FLOW_ERROR_TYPE_ITEM,
1083 						  item,
1084 						  "mask enables non supported"
1085 						  " bits");
1086 	if (!item->spec && (item->mask || item->last))
1087 		return rte_flow_error_set(error, EINVAL,
1088 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1089 					  "mask/last without a spec is not"
1090 					  " supported");
1091 	if (item->spec && item->last && !range_accepted) {
1092 		uint8_t spec[size];
1093 		uint8_t last[size];
1094 		unsigned int i;
1095 		int ret;
1096 
1097 		for (i = 0; i < size; ++i) {
1098 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1099 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1100 		}
1101 		ret = memcmp(spec, last, size);
1102 		if (ret != 0)
1103 			return rte_flow_error_set(error, EINVAL,
1104 						  RTE_FLOW_ERROR_TYPE_ITEM,
1105 						  item,
1106 						  "range is not valid");
1107 	}
1108 	return 0;
1109 }
1110 
1111 /**
1112  * Adjust the hash fields according to the @p flow information.
1113  *
1114  * @param[in] dev_flow.
1115  *   Pointer to the mlx5_flow.
1116  * @param[in] tunnel
1117  *   1 when the hash field is for a tunnel item.
1118  * @param[in] layer_types
1119  *   RTE_ETH_RSS_* types.
1120  * @param[in] hash_fields
1121  *   Item hash fields.
1122  *
1123  * @return
1124  *   The hash fields that should be used.
1125  */
1126 uint64_t
1127 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1128 			    int tunnel __rte_unused, uint64_t layer_types,
1129 			    uint64_t hash_fields)
1130 {
1131 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1132 	int rss_request_inner = rss_desc->level >= 2;
1133 
1134 	/* Check RSS hash level for tunnel. */
1135 	if (tunnel && rss_request_inner)
1136 		hash_fields |= IBV_RX_HASH_INNER;
1137 	else if (tunnel || rss_request_inner)
1138 		return 0;
1139 #endif
1140 	/* Check if requested layer matches RSS hash fields. */
1141 	if (!(rss_desc->types & layer_types))
1142 		return 0;
1143 	return hash_fields;
1144 }
1145 
1146 /**
1147  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1148  * if several tunnel rules are used on this queue, the tunnel ptype will be
1149  * cleared.
1150  *
1151  * @param rxq_ctrl
1152  *   Rx queue to update.
1153  */
1154 static void
1155 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1156 {
1157 	unsigned int i;
1158 	uint32_t tunnel_ptype = 0;
1159 
1160 	/* Look up for the ptype to use. */
1161 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1162 		if (!rxq_ctrl->flow_tunnels_n[i])
1163 			continue;
1164 		if (!tunnel_ptype) {
1165 			tunnel_ptype = tunnels_info[i].ptype;
1166 		} else {
1167 			tunnel_ptype = 0;
1168 			break;
1169 		}
1170 	}
1171 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1172 }
1173 
1174 /**
1175  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1176  * flow.
1177  *
1178  * @param[in] dev
1179  *   Pointer to the Ethernet device structure.
1180  * @param[in] dev_handle
1181  *   Pointer to device flow handle structure.
1182  */
1183 void
1184 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1185 		       struct mlx5_flow_handle *dev_handle)
1186 {
1187 	struct mlx5_priv *priv = dev->data->dev_private;
1188 	const int mark = dev_handle->mark;
1189 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1190 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1191 	unsigned int i;
1192 
1193 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1194 		struct mlx5_hrxq *hrxq;
1195 
1196 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1197 			      dev_handle->rix_hrxq);
1198 		if (hrxq)
1199 			ind_tbl = hrxq->ind_table;
1200 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1201 		struct mlx5_shared_action_rss *shared_rss;
1202 
1203 		shared_rss = mlx5_ipool_get
1204 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1205 			 dev_handle->rix_srss);
1206 		if (shared_rss)
1207 			ind_tbl = shared_rss->ind_tbl;
1208 	}
1209 	if (!ind_tbl)
1210 		return;
1211 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1212 		int idx = ind_tbl->queues[i];
1213 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1214 
1215 		MLX5_ASSERT(rxq_ctrl != NULL);
1216 		if (rxq_ctrl == NULL)
1217 			continue;
1218 		/*
1219 		 * To support metadata register copy on Tx loopback,
1220 		 * this must be always enabled (metadata may arive
1221 		 * from other port - not from local flows only.
1222 		 */
1223 		if (priv->config.dv_flow_en &&
1224 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1225 		    mlx5_flow_ext_mreg_supported(dev)) {
1226 			rxq_ctrl->rxq.mark = 1;
1227 			rxq_ctrl->flow_mark_n = 1;
1228 		} else if (mark) {
1229 			rxq_ctrl->rxq.mark = 1;
1230 			rxq_ctrl->flow_mark_n++;
1231 		}
1232 		if (tunnel) {
1233 			unsigned int j;
1234 
1235 			/* Increase the counter matching the flow. */
1236 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1237 				if ((tunnels_info[j].tunnel &
1238 				     dev_handle->layers) ==
1239 				    tunnels_info[j].tunnel) {
1240 					rxq_ctrl->flow_tunnels_n[j]++;
1241 					break;
1242 				}
1243 			}
1244 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1245 		}
1246 	}
1247 }
1248 
1249 /**
1250  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1251  *
1252  * @param[in] dev
1253  *   Pointer to the Ethernet device structure.
1254  * @param[in] flow
1255  *   Pointer to flow structure.
1256  */
1257 static void
1258 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1259 {
1260 	struct mlx5_priv *priv = dev->data->dev_private;
1261 	uint32_t handle_idx;
1262 	struct mlx5_flow_handle *dev_handle;
1263 
1264 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1265 		       handle_idx, dev_handle, next)
1266 		flow_drv_rxq_flags_set(dev, dev_handle);
1267 }
1268 
1269 /**
1270  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1271  * device flow if no other flow uses it with the same kind of request.
1272  *
1273  * @param dev
1274  *   Pointer to Ethernet device.
1275  * @param[in] dev_handle
1276  *   Pointer to the device flow handle structure.
1277  */
1278 static void
1279 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1280 			struct mlx5_flow_handle *dev_handle)
1281 {
1282 	struct mlx5_priv *priv = dev->data->dev_private;
1283 	const int mark = dev_handle->mark;
1284 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1285 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1286 	unsigned int i;
1287 
1288 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1289 		struct mlx5_hrxq *hrxq;
1290 
1291 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1292 			      dev_handle->rix_hrxq);
1293 		if (hrxq)
1294 			ind_tbl = hrxq->ind_table;
1295 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1296 		struct mlx5_shared_action_rss *shared_rss;
1297 
1298 		shared_rss = mlx5_ipool_get
1299 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1300 			 dev_handle->rix_srss);
1301 		if (shared_rss)
1302 			ind_tbl = shared_rss->ind_tbl;
1303 	}
1304 	if (!ind_tbl)
1305 		return;
1306 	MLX5_ASSERT(dev->data->dev_started);
1307 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1308 		int idx = ind_tbl->queues[i];
1309 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1310 
1311 		MLX5_ASSERT(rxq_ctrl != NULL);
1312 		if (rxq_ctrl == NULL)
1313 			continue;
1314 		if (priv->config.dv_flow_en &&
1315 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1316 		    mlx5_flow_ext_mreg_supported(dev)) {
1317 			rxq_ctrl->rxq.mark = 1;
1318 			rxq_ctrl->flow_mark_n = 1;
1319 		} else if (mark) {
1320 			rxq_ctrl->flow_mark_n--;
1321 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1322 		}
1323 		if (tunnel) {
1324 			unsigned int j;
1325 
1326 			/* Decrease the counter matching the flow. */
1327 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1328 				if ((tunnels_info[j].tunnel &
1329 				     dev_handle->layers) ==
1330 				    tunnels_info[j].tunnel) {
1331 					rxq_ctrl->flow_tunnels_n[j]--;
1332 					break;
1333 				}
1334 			}
1335 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1336 		}
1337 	}
1338 }
1339 
1340 /**
1341  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1342  * @p flow if no other flow uses it with the same kind of request.
1343  *
1344  * @param dev
1345  *   Pointer to Ethernet device.
1346  * @param[in] flow
1347  *   Pointer to the flow.
1348  */
1349 static void
1350 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1351 {
1352 	struct mlx5_priv *priv = dev->data->dev_private;
1353 	uint32_t handle_idx;
1354 	struct mlx5_flow_handle *dev_handle;
1355 
1356 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1357 		       handle_idx, dev_handle, next)
1358 		flow_drv_rxq_flags_trim(dev, dev_handle);
1359 }
1360 
1361 /**
1362  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1363  *
1364  * @param dev
1365  *   Pointer to Ethernet device.
1366  */
1367 static void
1368 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1369 {
1370 	struct mlx5_priv *priv = dev->data->dev_private;
1371 	unsigned int i;
1372 
1373 	for (i = 0; i != priv->rxqs_n; ++i) {
1374 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1375 		unsigned int j;
1376 
1377 		if (rxq == NULL || rxq->ctrl == NULL)
1378 			continue;
1379 		rxq->ctrl->flow_mark_n = 0;
1380 		rxq->ctrl->rxq.mark = 0;
1381 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1382 			rxq->ctrl->flow_tunnels_n[j] = 0;
1383 		rxq->ctrl->rxq.tunnel = 0;
1384 	}
1385 }
1386 
1387 /**
1388  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1389  *
1390  * @param[in] dev
1391  *   Pointer to the Ethernet device structure.
1392  */
1393 void
1394 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1395 {
1396 	struct mlx5_priv *priv = dev->data->dev_private;
1397 	unsigned int i;
1398 
1399 	for (i = 0; i != priv->rxqs_n; ++i) {
1400 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1401 		struct mlx5_rxq_data *data;
1402 
1403 		if (rxq == NULL || rxq->ctrl == NULL)
1404 			continue;
1405 		data = &rxq->ctrl->rxq;
1406 		if (!rte_flow_dynf_metadata_avail()) {
1407 			data->dynf_meta = 0;
1408 			data->flow_meta_mask = 0;
1409 			data->flow_meta_offset = -1;
1410 			data->flow_meta_port_mask = 0;
1411 		} else {
1412 			data->dynf_meta = 1;
1413 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1414 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1415 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1416 		}
1417 	}
1418 }
1419 
1420 /*
1421  * return a pointer to the desired action in the list of actions.
1422  *
1423  * @param[in] actions
1424  *   The list of actions to search the action in.
1425  * @param[in] action
1426  *   The action to find.
1427  *
1428  * @return
1429  *   Pointer to the action in the list, if found. NULL otherwise.
1430  */
1431 const struct rte_flow_action *
1432 mlx5_flow_find_action(const struct rte_flow_action *actions,
1433 		      enum rte_flow_action_type action)
1434 {
1435 	if (actions == NULL)
1436 		return NULL;
1437 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1438 		if (actions->type == action)
1439 			return actions;
1440 	return NULL;
1441 }
1442 
1443 /*
1444  * Validate the flag action.
1445  *
1446  * @param[in] action_flags
1447  *   Bit-fields that holds the actions detected until now.
1448  * @param[in] attr
1449  *   Attributes of flow that includes this action.
1450  * @param[out] error
1451  *   Pointer to error structure.
1452  *
1453  * @return
1454  *   0 on success, a negative errno value otherwise and rte_errno is set.
1455  */
1456 int
1457 mlx5_flow_validate_action_flag(uint64_t action_flags,
1458 			       const struct rte_flow_attr *attr,
1459 			       struct rte_flow_error *error)
1460 {
1461 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1462 		return rte_flow_error_set(error, EINVAL,
1463 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1464 					  "can't mark and flag in same flow");
1465 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1466 		return rte_flow_error_set(error, EINVAL,
1467 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1468 					  "can't have 2 flag"
1469 					  " actions in same flow");
1470 	if (attr->egress)
1471 		return rte_flow_error_set(error, ENOTSUP,
1472 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1473 					  "flag action not supported for "
1474 					  "egress");
1475 	return 0;
1476 }
1477 
1478 /*
1479  * Validate the mark action.
1480  *
1481  * @param[in] action
1482  *   Pointer to the queue action.
1483  * @param[in] action_flags
1484  *   Bit-fields that holds the actions detected until now.
1485  * @param[in] attr
1486  *   Attributes of flow that includes this action.
1487  * @param[out] error
1488  *   Pointer to error structure.
1489  *
1490  * @return
1491  *   0 on success, a negative errno value otherwise and rte_errno is set.
1492  */
1493 int
1494 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1495 			       uint64_t action_flags,
1496 			       const struct rte_flow_attr *attr,
1497 			       struct rte_flow_error *error)
1498 {
1499 	const struct rte_flow_action_mark *mark = action->conf;
1500 
1501 	if (!mark)
1502 		return rte_flow_error_set(error, EINVAL,
1503 					  RTE_FLOW_ERROR_TYPE_ACTION,
1504 					  action,
1505 					  "configuration cannot be null");
1506 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1507 		return rte_flow_error_set(error, EINVAL,
1508 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1509 					  &mark->id,
1510 					  "mark id must in 0 <= id < "
1511 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1512 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1513 		return rte_flow_error_set(error, EINVAL,
1514 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1515 					  "can't flag and mark in same flow");
1516 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1517 		return rte_flow_error_set(error, EINVAL,
1518 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1519 					  "can't have 2 mark actions in same"
1520 					  " flow");
1521 	if (attr->egress)
1522 		return rte_flow_error_set(error, ENOTSUP,
1523 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1524 					  "mark action not supported for "
1525 					  "egress");
1526 	return 0;
1527 }
1528 
1529 /*
1530  * Validate the drop action.
1531  *
1532  * @param[in] action_flags
1533  *   Bit-fields that holds the actions detected until now.
1534  * @param[in] attr
1535  *   Attributes of flow that includes this action.
1536  * @param[out] error
1537  *   Pointer to error structure.
1538  *
1539  * @return
1540  *   0 on success, a negative errno value otherwise and rte_errno is set.
1541  */
1542 int
1543 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1544 			       const struct rte_flow_attr *attr,
1545 			       struct rte_flow_error *error)
1546 {
1547 	if (attr->egress)
1548 		return rte_flow_error_set(error, ENOTSUP,
1549 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1550 					  "drop action not supported for "
1551 					  "egress");
1552 	return 0;
1553 }
1554 
1555 /*
1556  * Validate the queue action.
1557  *
1558  * @param[in] action
1559  *   Pointer to the queue action.
1560  * @param[in] action_flags
1561  *   Bit-fields that holds the actions detected until now.
1562  * @param[in] dev
1563  *   Pointer to the Ethernet device structure.
1564  * @param[in] attr
1565  *   Attributes of flow that includes this action.
1566  * @param[out] error
1567  *   Pointer to error structure.
1568  *
1569  * @return
1570  *   0 on success, a negative errno value otherwise and rte_errno is set.
1571  */
1572 int
1573 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1574 				uint64_t action_flags,
1575 				struct rte_eth_dev *dev,
1576 				const struct rte_flow_attr *attr,
1577 				struct rte_flow_error *error)
1578 {
1579 	struct mlx5_priv *priv = dev->data->dev_private;
1580 	const struct rte_flow_action_queue *queue = action->conf;
1581 
1582 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1583 		return rte_flow_error_set(error, EINVAL,
1584 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1585 					  "can't have 2 fate actions in"
1586 					  " same flow");
1587 	if (!priv->rxqs_n)
1588 		return rte_flow_error_set(error, EINVAL,
1589 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1590 					  NULL, "No Rx queues configured");
1591 	if (queue->index >= priv->rxqs_n)
1592 		return rte_flow_error_set(error, EINVAL,
1593 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1594 					  &queue->index,
1595 					  "queue index out of range");
1596 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1597 		return rte_flow_error_set(error, EINVAL,
1598 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1599 					  &queue->index,
1600 					  "queue is not configured");
1601 	if (attr->egress)
1602 		return rte_flow_error_set(error, ENOTSUP,
1603 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1604 					  "queue action not supported for "
1605 					  "egress");
1606 	return 0;
1607 }
1608 
1609 /**
1610  * Validate queue numbers for device RSS.
1611  *
1612  * @param[in] dev
1613  *   Configured device.
1614  * @param[in] queues
1615  *   Array of queue numbers.
1616  * @param[in] queues_n
1617  *   Size of the @p queues array.
1618  * @param[out] error
1619  *   On error, filled with a textual error description.
1620  * @param[out] queue
1621  *   On error, filled with an offending queue index in @p queues array.
1622  *
1623  * @return
1624  *   0 on success, a negative errno code on error.
1625  */
1626 static int
1627 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1628 			 const uint16_t *queues, uint32_t queues_n,
1629 			 const char **error, uint32_t *queue_idx)
1630 {
1631 	const struct mlx5_priv *priv = dev->data->dev_private;
1632 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1633 	uint32_t i;
1634 
1635 	for (i = 0; i != queues_n; ++i) {
1636 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1637 								   queues[i]);
1638 
1639 		if (queues[i] >= priv->rxqs_n) {
1640 			*error = "queue index out of range";
1641 			*queue_idx = i;
1642 			return -EINVAL;
1643 		}
1644 		if (rxq_ctrl == NULL) {
1645 			*error =  "queue is not configured";
1646 			*queue_idx = i;
1647 			return -EINVAL;
1648 		}
1649 		if (i == 0)
1650 			rxq_type = rxq_ctrl->type;
1651 		if (rxq_type != rxq_ctrl->type) {
1652 			*error = "combining hairpin and regular RSS queues is not supported";
1653 			*queue_idx = i;
1654 			return -ENOTSUP;
1655 		}
1656 	}
1657 	return 0;
1658 }
1659 
1660 /*
1661  * Validate the rss action.
1662  *
1663  * @param[in] dev
1664  *   Pointer to the Ethernet device structure.
1665  * @param[in] action
1666  *   Pointer to the queue action.
1667  * @param[out] error
1668  *   Pointer to error structure.
1669  *
1670  * @return
1671  *   0 on success, a negative errno value otherwise and rte_errno is set.
1672  */
1673 int
1674 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1675 			 const struct rte_flow_action *action,
1676 			 struct rte_flow_error *error)
1677 {
1678 	struct mlx5_priv *priv = dev->data->dev_private;
1679 	const struct rte_flow_action_rss *rss = action->conf;
1680 	int ret;
1681 	const char *message;
1682 	uint32_t queue_idx;
1683 
1684 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1685 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1686 		return rte_flow_error_set(error, ENOTSUP,
1687 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1688 					  &rss->func,
1689 					  "RSS hash function not supported");
1690 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1691 	if (rss->level > 2)
1692 #else
1693 	if (rss->level > 1)
1694 #endif
1695 		return rte_flow_error_set(error, ENOTSUP,
1696 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1697 					  &rss->level,
1698 					  "tunnel RSS is not supported");
1699 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1700 	if (rss->key_len == 0 && rss->key != NULL)
1701 		return rte_flow_error_set(error, ENOTSUP,
1702 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1703 					  &rss->key_len,
1704 					  "RSS hash key length 0");
1705 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1706 		return rte_flow_error_set(error, ENOTSUP,
1707 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1708 					  &rss->key_len,
1709 					  "RSS hash key too small");
1710 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1711 		return rte_flow_error_set(error, ENOTSUP,
1712 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1713 					  &rss->key_len,
1714 					  "RSS hash key too large");
1715 	if (rss->queue_num > priv->config.ind_table_max_size)
1716 		return rte_flow_error_set(error, ENOTSUP,
1717 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1718 					  &rss->queue_num,
1719 					  "number of queues too large");
1720 	if (rss->types & MLX5_RSS_HF_MASK)
1721 		return rte_flow_error_set(error, ENOTSUP,
1722 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1723 					  &rss->types,
1724 					  "some RSS protocols are not"
1725 					  " supported");
1726 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1727 	    !(rss->types & RTE_ETH_RSS_IP))
1728 		return rte_flow_error_set(error, EINVAL,
1729 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1730 					  "L3 partial RSS requested but L3 RSS"
1731 					  " type not specified");
1732 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1733 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1734 		return rte_flow_error_set(error, EINVAL,
1735 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1736 					  "L4 partial RSS requested but L4 RSS"
1737 					  " type not specified");
1738 	if (!priv->rxqs_n)
1739 		return rte_flow_error_set(error, EINVAL,
1740 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1741 					  NULL, "No Rx queues configured");
1742 	if (!rss->queue_num)
1743 		return rte_flow_error_set(error, EINVAL,
1744 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1745 					  NULL, "No queues configured");
1746 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1747 				       &message, &queue_idx);
1748 	if (ret != 0) {
1749 		return rte_flow_error_set(error, -ret,
1750 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1751 					  &rss->queue[queue_idx], message);
1752 	}
1753 	return 0;
1754 }
1755 
1756 /*
1757  * Validate the rss action.
1758  *
1759  * @param[in] action
1760  *   Pointer to the queue action.
1761  * @param[in] action_flags
1762  *   Bit-fields that holds the actions detected until now.
1763  * @param[in] dev
1764  *   Pointer to the Ethernet device structure.
1765  * @param[in] attr
1766  *   Attributes of flow that includes this action.
1767  * @param[in] item_flags
1768  *   Items that were detected.
1769  * @param[out] error
1770  *   Pointer to error structure.
1771  *
1772  * @return
1773  *   0 on success, a negative errno value otherwise and rte_errno is set.
1774  */
1775 int
1776 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1777 			      uint64_t action_flags,
1778 			      struct rte_eth_dev *dev,
1779 			      const struct rte_flow_attr *attr,
1780 			      uint64_t item_flags,
1781 			      struct rte_flow_error *error)
1782 {
1783 	const struct rte_flow_action_rss *rss = action->conf;
1784 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1785 	int ret;
1786 
1787 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1788 		return rte_flow_error_set(error, EINVAL,
1789 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1790 					  "can't have 2 fate actions"
1791 					  " in same flow");
1792 	ret = mlx5_validate_action_rss(dev, action, error);
1793 	if (ret)
1794 		return ret;
1795 	if (attr->egress)
1796 		return rte_flow_error_set(error, ENOTSUP,
1797 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1798 					  "rss action not supported for "
1799 					  "egress");
1800 	if (rss->level > 1 && !tunnel)
1801 		return rte_flow_error_set(error, EINVAL,
1802 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1803 					  "inner RSS is not supported for "
1804 					  "non-tunnel flows");
1805 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1806 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1807 		return rte_flow_error_set(error, EINVAL,
1808 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1809 					  "RSS on eCPRI is not supported now");
1810 	}
1811 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1812 	    !(item_flags &
1813 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1814 	    rss->level > 1)
1815 		return rte_flow_error_set(error, EINVAL,
1816 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1817 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1818 	return 0;
1819 }
1820 
1821 /*
1822  * Validate the default miss action.
1823  *
1824  * @param[in] action_flags
1825  *   Bit-fields that holds the actions detected until now.
1826  * @param[out] error
1827  *   Pointer to error structure.
1828  *
1829  * @return
1830  *   0 on success, a negative errno value otherwise and rte_errno is set.
1831  */
1832 int
1833 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1834 				const struct rte_flow_attr *attr,
1835 				struct rte_flow_error *error)
1836 {
1837 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1838 		return rte_flow_error_set(error, EINVAL,
1839 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1840 					  "can't have 2 fate actions in"
1841 					  " same flow");
1842 	if (attr->egress)
1843 		return rte_flow_error_set(error, ENOTSUP,
1844 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1845 					  "default miss action not supported "
1846 					  "for egress");
1847 	if (attr->group)
1848 		return rte_flow_error_set(error, ENOTSUP,
1849 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1850 					  "only group 0 is supported");
1851 	if (attr->transfer)
1852 		return rte_flow_error_set(error, ENOTSUP,
1853 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1854 					  NULL, "transfer is not supported");
1855 	return 0;
1856 }
1857 
1858 /*
1859  * Validate the count action.
1860  *
1861  * @param[in] dev
1862  *   Pointer to the Ethernet device structure.
1863  * @param[in] attr
1864  *   Attributes of flow that includes this action.
1865  * @param[out] error
1866  *   Pointer to error structure.
1867  *
1868  * @return
1869  *   0 on success, a negative errno value otherwise and rte_errno is set.
1870  */
1871 int
1872 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1873 				const struct rte_flow_attr *attr,
1874 				struct rte_flow_error *error)
1875 {
1876 	if (attr->egress)
1877 		return rte_flow_error_set(error, ENOTSUP,
1878 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1879 					  "count action not supported for "
1880 					  "egress");
1881 	return 0;
1882 }
1883 
1884 /*
1885  * Validate the ASO CT action.
1886  *
1887  * @param[in] dev
1888  *   Pointer to the Ethernet device structure.
1889  * @param[in] conntrack
1890  *   Pointer to the CT action profile.
1891  * @param[out] error
1892  *   Pointer to error structure.
1893  *
1894  * @return
1895  *   0 on success, a negative errno value otherwise and rte_errno is set.
1896  */
1897 int
1898 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1899 			const struct rte_flow_action_conntrack *conntrack,
1900 			struct rte_flow_error *error)
1901 {
1902 	RTE_SET_USED(dev);
1903 
1904 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1905 		return rte_flow_error_set(error, EINVAL,
1906 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1907 					  "Invalid CT state");
1908 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1909 		return rte_flow_error_set(error, EINVAL,
1910 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1911 					  "Invalid last TCP packet flag");
1912 	return 0;
1913 }
1914 
1915 /**
1916  * Verify the @p attributes will be correctly understood by the NIC and store
1917  * them in the @p flow if everything is correct.
1918  *
1919  * @param[in] dev
1920  *   Pointer to the Ethernet device structure.
1921  * @param[in] attributes
1922  *   Pointer to flow attributes
1923  * @param[out] error
1924  *   Pointer to error structure.
1925  *
1926  * @return
1927  *   0 on success, a negative errno value otherwise and rte_errno is set.
1928  */
1929 int
1930 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1931 			      const struct rte_flow_attr *attributes,
1932 			      struct rte_flow_error *error)
1933 {
1934 	struct mlx5_priv *priv = dev->data->dev_private;
1935 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1936 
1937 	if (attributes->group)
1938 		return rte_flow_error_set(error, ENOTSUP,
1939 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1940 					  NULL, "groups is not supported");
1941 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1942 	    attributes->priority >= priority_max)
1943 		return rte_flow_error_set(error, ENOTSUP,
1944 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1945 					  NULL, "priority out of range");
1946 	if (attributes->egress)
1947 		return rte_flow_error_set(error, ENOTSUP,
1948 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1949 					  "egress is not supported");
1950 	if (attributes->transfer && !priv->config.dv_esw_en)
1951 		return rte_flow_error_set(error, ENOTSUP,
1952 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1953 					  NULL, "transfer is not supported");
1954 	if (!attributes->ingress)
1955 		return rte_flow_error_set(error, EINVAL,
1956 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1957 					  NULL,
1958 					  "ingress attribute is mandatory");
1959 	return 0;
1960 }
1961 
1962 /**
1963  * Validate ICMP6 item.
1964  *
1965  * @param[in] item
1966  *   Item specification.
1967  * @param[in] item_flags
1968  *   Bit-fields that holds the items detected until now.
1969  * @param[in] ext_vlan_sup
1970  *   Whether extended VLAN features are supported or not.
1971  * @param[out] error
1972  *   Pointer to error structure.
1973  *
1974  * @return
1975  *   0 on success, a negative errno value otherwise and rte_errno is set.
1976  */
1977 int
1978 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1979 			       uint64_t item_flags,
1980 			       uint8_t target_protocol,
1981 			       struct rte_flow_error *error)
1982 {
1983 	const struct rte_flow_item_icmp6 *mask = item->mask;
1984 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1985 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1986 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1987 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1988 				      MLX5_FLOW_LAYER_OUTER_L4;
1989 	int ret;
1990 
1991 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1992 		return rte_flow_error_set(error, EINVAL,
1993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 					  "protocol filtering not compatible"
1995 					  " with ICMP6 layer");
1996 	if (!(item_flags & l3m))
1997 		return rte_flow_error_set(error, EINVAL,
1998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 					  "IPv6 is mandatory to filter on"
2000 					  " ICMP6");
2001 	if (item_flags & l4m)
2002 		return rte_flow_error_set(error, EINVAL,
2003 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2004 					  "multiple L4 layers not supported");
2005 	if (!mask)
2006 		mask = &rte_flow_item_icmp6_mask;
2007 	ret = mlx5_flow_item_acceptable
2008 		(item, (const uint8_t *)mask,
2009 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2010 		 sizeof(struct rte_flow_item_icmp6),
2011 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2012 	if (ret < 0)
2013 		return ret;
2014 	return 0;
2015 }
2016 
2017 /**
2018  * Validate ICMP item.
2019  *
2020  * @param[in] item
2021  *   Item specification.
2022  * @param[in] item_flags
2023  *   Bit-fields that holds the items detected until now.
2024  * @param[out] error
2025  *   Pointer to error structure.
2026  *
2027  * @return
2028  *   0 on success, a negative errno value otherwise and rte_errno is set.
2029  */
2030 int
2031 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2032 			     uint64_t item_flags,
2033 			     uint8_t target_protocol,
2034 			     struct rte_flow_error *error)
2035 {
2036 	const struct rte_flow_item_icmp *mask = item->mask;
2037 	const struct rte_flow_item_icmp nic_mask = {
2038 		.hdr.icmp_type = 0xff,
2039 		.hdr.icmp_code = 0xff,
2040 		.hdr.icmp_ident = RTE_BE16(0xffff),
2041 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2042 	};
2043 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2044 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2045 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2046 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2047 				      MLX5_FLOW_LAYER_OUTER_L4;
2048 	int ret;
2049 
2050 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2051 		return rte_flow_error_set(error, EINVAL,
2052 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2053 					  "protocol filtering not compatible"
2054 					  " with ICMP layer");
2055 	if (!(item_flags & l3m))
2056 		return rte_flow_error_set(error, EINVAL,
2057 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2058 					  "IPv4 is mandatory to filter"
2059 					  " on ICMP");
2060 	if (item_flags & l4m)
2061 		return rte_flow_error_set(error, EINVAL,
2062 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2063 					  "multiple L4 layers not supported");
2064 	if (!mask)
2065 		mask = &nic_mask;
2066 	ret = mlx5_flow_item_acceptable
2067 		(item, (const uint8_t *)mask,
2068 		 (const uint8_t *)&nic_mask,
2069 		 sizeof(struct rte_flow_item_icmp),
2070 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2071 	if (ret < 0)
2072 		return ret;
2073 	return 0;
2074 }
2075 
2076 /**
2077  * Validate Ethernet item.
2078  *
2079  * @param[in] item
2080  *   Item specification.
2081  * @param[in] item_flags
2082  *   Bit-fields that holds the items detected until now.
2083  * @param[out] error
2084  *   Pointer to error structure.
2085  *
2086  * @return
2087  *   0 on success, a negative errno value otherwise and rte_errno is set.
2088  */
2089 int
2090 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2091 			    uint64_t item_flags, bool ext_vlan_sup,
2092 			    struct rte_flow_error *error)
2093 {
2094 	const struct rte_flow_item_eth *mask = item->mask;
2095 	const struct rte_flow_item_eth nic_mask = {
2096 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2097 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2098 		.type = RTE_BE16(0xffff),
2099 		.has_vlan = ext_vlan_sup ? 1 : 0,
2100 	};
2101 	int ret;
2102 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2103 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2104 				       MLX5_FLOW_LAYER_OUTER_L2;
2105 
2106 	if (item_flags & ethm)
2107 		return rte_flow_error_set(error, ENOTSUP,
2108 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2109 					  "multiple L2 layers not supported");
2110 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2111 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2112 		return rte_flow_error_set(error, EINVAL,
2113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 					  "L2 layer should not follow "
2115 					  "L3 layers");
2116 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2117 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2118 		return rte_flow_error_set(error, EINVAL,
2119 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2120 					  "L2 layer should not follow VLAN");
2121 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2122 		return rte_flow_error_set(error, EINVAL,
2123 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2124 					  "L2 layer should not follow GTP");
2125 	if (!mask)
2126 		mask = &rte_flow_item_eth_mask;
2127 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2128 					(const uint8_t *)&nic_mask,
2129 					sizeof(struct rte_flow_item_eth),
2130 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2131 	return ret;
2132 }
2133 
2134 /**
2135  * Validate VLAN item.
2136  *
2137  * @param[in] item
2138  *   Item specification.
2139  * @param[in] item_flags
2140  *   Bit-fields that holds the items detected until now.
2141  * @param[in] dev
2142  *   Ethernet device flow is being created on.
2143  * @param[out] error
2144  *   Pointer to error structure.
2145  *
2146  * @return
2147  *   0 on success, a negative errno value otherwise and rte_errno is set.
2148  */
2149 int
2150 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2151 			     uint64_t item_flags,
2152 			     struct rte_eth_dev *dev,
2153 			     struct rte_flow_error *error)
2154 {
2155 	const struct rte_flow_item_vlan *spec = item->spec;
2156 	const struct rte_flow_item_vlan *mask = item->mask;
2157 	const struct rte_flow_item_vlan nic_mask = {
2158 		.tci = RTE_BE16(UINT16_MAX),
2159 		.inner_type = RTE_BE16(UINT16_MAX),
2160 	};
2161 	uint16_t vlan_tag = 0;
2162 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2163 	int ret;
2164 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2165 					MLX5_FLOW_LAYER_INNER_L4) :
2166 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2167 					MLX5_FLOW_LAYER_OUTER_L4);
2168 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2169 					MLX5_FLOW_LAYER_OUTER_VLAN;
2170 
2171 	if (item_flags & vlanm)
2172 		return rte_flow_error_set(error, EINVAL,
2173 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2174 					  "multiple VLAN layers not supported");
2175 	else if ((item_flags & l34m) != 0)
2176 		return rte_flow_error_set(error, EINVAL,
2177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2178 					  "VLAN cannot follow L3/L4 layer");
2179 	if (!mask)
2180 		mask = &rte_flow_item_vlan_mask;
2181 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2182 					(const uint8_t *)&nic_mask,
2183 					sizeof(struct rte_flow_item_vlan),
2184 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2185 	if (ret)
2186 		return ret;
2187 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2188 		struct mlx5_priv *priv = dev->data->dev_private;
2189 
2190 		if (priv->vmwa_context) {
2191 			/*
2192 			 * Non-NULL context means we have a virtual machine
2193 			 * and SR-IOV enabled, we have to create VLAN interface
2194 			 * to make hypervisor to setup E-Switch vport
2195 			 * context correctly. We avoid creating the multiple
2196 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2197 			 */
2198 			return rte_flow_error_set(error, EINVAL,
2199 						  RTE_FLOW_ERROR_TYPE_ITEM,
2200 						  item,
2201 						  "VLAN tag mask is not"
2202 						  " supported in virtual"
2203 						  " environment");
2204 		}
2205 	}
2206 	if (spec) {
2207 		vlan_tag = spec->tci;
2208 		vlan_tag &= mask->tci;
2209 	}
2210 	/*
2211 	 * From verbs perspective an empty VLAN is equivalent
2212 	 * to a packet without VLAN layer.
2213 	 */
2214 	if (!vlan_tag)
2215 		return rte_flow_error_set(error, EINVAL,
2216 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2217 					  item->spec,
2218 					  "VLAN cannot be empty");
2219 	return 0;
2220 }
2221 
2222 /**
2223  * Validate IPV4 item.
2224  *
2225  * @param[in] item
2226  *   Item specification.
2227  * @param[in] item_flags
2228  *   Bit-fields that holds the items detected until now.
2229  * @param[in] last_item
2230  *   Previous validated item in the pattern items.
2231  * @param[in] ether_type
2232  *   Type in the ethernet layer header (including dot1q).
2233  * @param[in] acc_mask
2234  *   Acceptable mask, if NULL default internal default mask
2235  *   will be used to check whether item fields are supported.
2236  * @param[in] range_accepted
2237  *   True if range of values is accepted for specific fields, false otherwise.
2238  * @param[out] error
2239  *   Pointer to error structure.
2240  *
2241  * @return
2242  *   0 on success, a negative errno value otherwise and rte_errno is set.
2243  */
2244 int
2245 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2246 			     uint64_t item_flags,
2247 			     uint64_t last_item,
2248 			     uint16_t ether_type,
2249 			     const struct rte_flow_item_ipv4 *acc_mask,
2250 			     bool range_accepted,
2251 			     struct rte_flow_error *error)
2252 {
2253 	const struct rte_flow_item_ipv4 *mask = item->mask;
2254 	const struct rte_flow_item_ipv4 *spec = item->spec;
2255 	const struct rte_flow_item_ipv4 nic_mask = {
2256 		.hdr = {
2257 			.src_addr = RTE_BE32(0xffffffff),
2258 			.dst_addr = RTE_BE32(0xffffffff),
2259 			.type_of_service = 0xff,
2260 			.next_proto_id = 0xff,
2261 		},
2262 	};
2263 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2264 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2265 				      MLX5_FLOW_LAYER_OUTER_L3;
2266 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2267 				      MLX5_FLOW_LAYER_OUTER_L4;
2268 	int ret;
2269 	uint8_t next_proto = 0xFF;
2270 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2271 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2272 				  MLX5_FLOW_LAYER_INNER_VLAN);
2273 
2274 	if ((last_item & l2_vlan) && ether_type &&
2275 	    ether_type != RTE_ETHER_TYPE_IPV4)
2276 		return rte_flow_error_set(error, EINVAL,
2277 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2278 					  "IPv4 cannot follow L2/VLAN layer "
2279 					  "which ether type is not IPv4");
2280 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2281 		if (mask && spec)
2282 			next_proto = mask->hdr.next_proto_id &
2283 				     spec->hdr.next_proto_id;
2284 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2285 			return rte_flow_error_set(error, EINVAL,
2286 						  RTE_FLOW_ERROR_TYPE_ITEM,
2287 						  item,
2288 						  "multiple tunnel "
2289 						  "not supported");
2290 	}
2291 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2292 		return rte_flow_error_set(error, EINVAL,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "wrong tunnel type - IPv6 specified "
2295 					  "but IPv4 item provided");
2296 	if (item_flags & l3m)
2297 		return rte_flow_error_set(error, ENOTSUP,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "multiple L3 layers not supported");
2300 	else if (item_flags & l4m)
2301 		return rte_flow_error_set(error, EINVAL,
2302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 					  "L3 cannot follow an L4 layer.");
2304 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2305 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2306 		return rte_flow_error_set(error, EINVAL,
2307 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2308 					  "L3 cannot follow an NVGRE layer.");
2309 	if (!mask)
2310 		mask = &rte_flow_item_ipv4_mask;
2311 	else if (mask->hdr.next_proto_id != 0 &&
2312 		 mask->hdr.next_proto_id != 0xff)
2313 		return rte_flow_error_set(error, EINVAL,
2314 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2315 					  "partial mask is not supported"
2316 					  " for protocol");
2317 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2318 					acc_mask ? (const uint8_t *)acc_mask
2319 						 : (const uint8_t *)&nic_mask,
2320 					sizeof(struct rte_flow_item_ipv4),
2321 					range_accepted, error);
2322 	if (ret < 0)
2323 		return ret;
2324 	return 0;
2325 }
2326 
2327 /**
2328  * Validate IPV6 item.
2329  *
2330  * @param[in] item
2331  *   Item specification.
2332  * @param[in] item_flags
2333  *   Bit-fields that holds the items detected until now.
2334  * @param[in] last_item
2335  *   Previous validated item in the pattern items.
2336  * @param[in] ether_type
2337  *   Type in the ethernet layer header (including dot1q).
2338  * @param[in] acc_mask
2339  *   Acceptable mask, if NULL default internal default mask
2340  *   will be used to check whether item fields are supported.
2341  * @param[out] error
2342  *   Pointer to error structure.
2343  *
2344  * @return
2345  *   0 on success, a negative errno value otherwise and rte_errno is set.
2346  */
2347 int
2348 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2349 			     uint64_t item_flags,
2350 			     uint64_t last_item,
2351 			     uint16_t ether_type,
2352 			     const struct rte_flow_item_ipv6 *acc_mask,
2353 			     struct rte_flow_error *error)
2354 {
2355 	const struct rte_flow_item_ipv6 *mask = item->mask;
2356 	const struct rte_flow_item_ipv6 *spec = item->spec;
2357 	const struct rte_flow_item_ipv6 nic_mask = {
2358 		.hdr = {
2359 			.src_addr =
2360 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2361 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2362 			.dst_addr =
2363 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2364 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2365 			.vtc_flow = RTE_BE32(0xffffffff),
2366 			.proto = 0xff,
2367 		},
2368 	};
2369 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2370 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2371 				      MLX5_FLOW_LAYER_OUTER_L3;
2372 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2373 				      MLX5_FLOW_LAYER_OUTER_L4;
2374 	int ret;
2375 	uint8_t next_proto = 0xFF;
2376 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2377 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2378 				  MLX5_FLOW_LAYER_INNER_VLAN);
2379 
2380 	if ((last_item & l2_vlan) && ether_type &&
2381 	    ether_type != RTE_ETHER_TYPE_IPV6)
2382 		return rte_flow_error_set(error, EINVAL,
2383 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2384 					  "IPv6 cannot follow L2/VLAN layer "
2385 					  "which ether type is not IPv6");
2386 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2387 		next_proto = spec->hdr.proto;
2388 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2389 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2390 			return rte_flow_error_set(error, EINVAL,
2391 						  RTE_FLOW_ERROR_TYPE_ITEM,
2392 						  item,
2393 						  "multiple tunnel "
2394 						  "not supported");
2395 	}
2396 	if (next_proto == IPPROTO_HOPOPTS  ||
2397 	    next_proto == IPPROTO_ROUTING  ||
2398 	    next_proto == IPPROTO_FRAGMENT ||
2399 	    next_proto == IPPROTO_ESP	   ||
2400 	    next_proto == IPPROTO_AH	   ||
2401 	    next_proto == IPPROTO_DSTOPTS)
2402 		return rte_flow_error_set(error, EINVAL,
2403 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2404 					  "IPv6 proto (next header) should "
2405 					  "not be set as extension header");
2406 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2407 		return rte_flow_error_set(error, EINVAL,
2408 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2409 					  "wrong tunnel type - IPv4 specified "
2410 					  "but IPv6 item provided");
2411 	if (item_flags & l3m)
2412 		return rte_flow_error_set(error, ENOTSUP,
2413 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2414 					  "multiple L3 layers not supported");
2415 	else if (item_flags & l4m)
2416 		return rte_flow_error_set(error, EINVAL,
2417 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2418 					  "L3 cannot follow an L4 layer.");
2419 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2420 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2421 		return rte_flow_error_set(error, EINVAL,
2422 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2423 					  "L3 cannot follow an NVGRE layer.");
2424 	if (!mask)
2425 		mask = &rte_flow_item_ipv6_mask;
2426 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2427 					acc_mask ? (const uint8_t *)acc_mask
2428 						 : (const uint8_t *)&nic_mask,
2429 					sizeof(struct rte_flow_item_ipv6),
2430 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2431 	if (ret < 0)
2432 		return ret;
2433 	return 0;
2434 }
2435 
2436 /**
2437  * Validate UDP item.
2438  *
2439  * @param[in] item
2440  *   Item specification.
2441  * @param[in] item_flags
2442  *   Bit-fields that holds the items detected until now.
2443  * @param[in] target_protocol
2444  *   The next protocol in the previous item.
2445  * @param[in] flow_mask
2446  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2447  * @param[out] error
2448  *   Pointer to error structure.
2449  *
2450  * @return
2451  *   0 on success, a negative errno value otherwise and rte_errno is set.
2452  */
2453 int
2454 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2455 			    uint64_t item_flags,
2456 			    uint8_t target_protocol,
2457 			    struct rte_flow_error *error)
2458 {
2459 	const struct rte_flow_item_udp *mask = item->mask;
2460 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2461 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2462 				      MLX5_FLOW_LAYER_OUTER_L3;
2463 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2464 				      MLX5_FLOW_LAYER_OUTER_L4;
2465 	int ret;
2466 
2467 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2468 		return rte_flow_error_set(error, EINVAL,
2469 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2470 					  "protocol filtering not compatible"
2471 					  " with UDP layer");
2472 	if (!(item_flags & l3m))
2473 		return rte_flow_error_set(error, EINVAL,
2474 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2475 					  "L3 is mandatory to filter on L4");
2476 	if (item_flags & l4m)
2477 		return rte_flow_error_set(error, EINVAL,
2478 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2479 					  "multiple L4 layers not supported");
2480 	if (!mask)
2481 		mask = &rte_flow_item_udp_mask;
2482 	ret = mlx5_flow_item_acceptable
2483 		(item, (const uint8_t *)mask,
2484 		 (const uint8_t *)&rte_flow_item_udp_mask,
2485 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2486 		 error);
2487 	if (ret < 0)
2488 		return ret;
2489 	return 0;
2490 }
2491 
2492 /**
2493  * Validate TCP item.
2494  *
2495  * @param[in] item
2496  *   Item specification.
2497  * @param[in] item_flags
2498  *   Bit-fields that holds the items detected until now.
2499  * @param[in] target_protocol
2500  *   The next protocol in the previous item.
2501  * @param[out] error
2502  *   Pointer to error structure.
2503  *
2504  * @return
2505  *   0 on success, a negative errno value otherwise and rte_errno is set.
2506  */
2507 int
2508 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2509 			    uint64_t item_flags,
2510 			    uint8_t target_protocol,
2511 			    const struct rte_flow_item_tcp *flow_mask,
2512 			    struct rte_flow_error *error)
2513 {
2514 	const struct rte_flow_item_tcp *mask = item->mask;
2515 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2516 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2517 				      MLX5_FLOW_LAYER_OUTER_L3;
2518 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2519 				      MLX5_FLOW_LAYER_OUTER_L4;
2520 	int ret;
2521 
2522 	MLX5_ASSERT(flow_mask);
2523 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2524 		return rte_flow_error_set(error, EINVAL,
2525 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2526 					  "protocol filtering not compatible"
2527 					  " with TCP layer");
2528 	if (!(item_flags & l3m))
2529 		return rte_flow_error_set(error, EINVAL,
2530 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2531 					  "L3 is mandatory to filter on L4");
2532 	if (item_flags & l4m)
2533 		return rte_flow_error_set(error, EINVAL,
2534 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2535 					  "multiple L4 layers not supported");
2536 	if (!mask)
2537 		mask = &rte_flow_item_tcp_mask;
2538 	ret = mlx5_flow_item_acceptable
2539 		(item, (const uint8_t *)mask,
2540 		 (const uint8_t *)flow_mask,
2541 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2542 		 error);
2543 	if (ret < 0)
2544 		return ret;
2545 	return 0;
2546 }
2547 
2548 /**
2549  * Validate VXLAN item.
2550  *
2551  * @param[in] dev
2552  *   Pointer to the Ethernet device structure.
2553  * @param[in] udp_dport
2554  *   UDP destination port
2555  * @param[in] item
2556  *   Item specification.
2557  * @param[in] item_flags
2558  *   Bit-fields that holds the items detected until now.
2559  * @param[in] attr
2560  *   Flow rule attributes.
2561  * @param[out] error
2562  *   Pointer to error structure.
2563  *
2564  * @return
2565  *   0 on success, a negative errno value otherwise and rte_errno is set.
2566  */
2567 int
2568 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2569 			      uint16_t udp_dport,
2570 			      const struct rte_flow_item *item,
2571 			      uint64_t item_flags,
2572 			      const struct rte_flow_attr *attr,
2573 			      struct rte_flow_error *error)
2574 {
2575 	const struct rte_flow_item_vxlan *spec = item->spec;
2576 	const struct rte_flow_item_vxlan *mask = item->mask;
2577 	int ret;
2578 	struct mlx5_priv *priv = dev->data->dev_private;
2579 	union vni {
2580 		uint32_t vlan_id;
2581 		uint8_t vni[4];
2582 	} id = { .vlan_id = 0, };
2583 	const struct rte_flow_item_vxlan nic_mask = {
2584 		.vni = "\xff\xff\xff",
2585 		.rsvd1 = 0xff,
2586 	};
2587 	const struct rte_flow_item_vxlan *valid_mask;
2588 
2589 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2590 		return rte_flow_error_set(error, ENOTSUP,
2591 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2592 					  "multiple tunnel layers not"
2593 					  " supported");
2594 	valid_mask = &rte_flow_item_vxlan_mask;
2595 	/*
2596 	 * Verify only UDPv4 is present as defined in
2597 	 * https://tools.ietf.org/html/rfc7348
2598 	 */
2599 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2600 		return rte_flow_error_set(error, EINVAL,
2601 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2602 					  "no outer UDP layer found");
2603 	if (!mask)
2604 		mask = &rte_flow_item_vxlan_mask;
2605 
2606 	if (priv->sh->steering_format_version !=
2607 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2608 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2609 		/* FDB domain & NIC domain non-zero group */
2610 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2611 			valid_mask = &nic_mask;
2612 		/* Group zero in NIC domain */
2613 		if (!attr->group && !attr->transfer &&
2614 		    priv->sh->tunnel_header_0_1)
2615 			valid_mask = &nic_mask;
2616 	}
2617 	ret = mlx5_flow_item_acceptable
2618 		(item, (const uint8_t *)mask,
2619 		 (const uint8_t *)valid_mask,
2620 		 sizeof(struct rte_flow_item_vxlan),
2621 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2622 	if (ret < 0)
2623 		return ret;
2624 	if (spec) {
2625 		memcpy(&id.vni[1], spec->vni, 3);
2626 		memcpy(&id.vni[1], mask->vni, 3);
2627 	}
2628 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2629 		return rte_flow_error_set(error, ENOTSUP,
2630 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2631 					  "VXLAN tunnel must be fully defined");
2632 	return 0;
2633 }
2634 
2635 /**
2636  * Validate VXLAN_GPE item.
2637  *
2638  * @param[in] item
2639  *   Item specification.
2640  * @param[in] item_flags
2641  *   Bit-fields that holds the items detected until now.
2642  * @param[in] priv
2643  *   Pointer to the private data structure.
2644  * @param[in] target_protocol
2645  *   The next protocol in the previous item.
2646  * @param[out] error
2647  *   Pointer to error structure.
2648  *
2649  * @return
2650  *   0 on success, a negative errno value otherwise and rte_errno is set.
2651  */
2652 int
2653 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2654 				  uint64_t item_flags,
2655 				  struct rte_eth_dev *dev,
2656 				  struct rte_flow_error *error)
2657 {
2658 	struct mlx5_priv *priv = dev->data->dev_private;
2659 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2660 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2661 	int ret;
2662 	union vni {
2663 		uint32_t vlan_id;
2664 		uint8_t vni[4];
2665 	} id = { .vlan_id = 0, };
2666 
2667 	if (!priv->config.l3_vxlan_en)
2668 		return rte_flow_error_set(error, ENOTSUP,
2669 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2670 					  "L3 VXLAN is not enabled by device"
2671 					  " parameter and/or not configured in"
2672 					  " firmware");
2673 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2674 		return rte_flow_error_set(error, ENOTSUP,
2675 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2676 					  "multiple tunnel layers not"
2677 					  " supported");
2678 	/*
2679 	 * Verify only UDPv4 is present as defined in
2680 	 * https://tools.ietf.org/html/rfc7348
2681 	 */
2682 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2683 		return rte_flow_error_set(error, EINVAL,
2684 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2685 					  "no outer UDP layer found");
2686 	if (!mask)
2687 		mask = &rte_flow_item_vxlan_gpe_mask;
2688 	ret = mlx5_flow_item_acceptable
2689 		(item, (const uint8_t *)mask,
2690 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2691 		 sizeof(struct rte_flow_item_vxlan_gpe),
2692 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2693 	if (ret < 0)
2694 		return ret;
2695 	if (spec) {
2696 		if (spec->protocol)
2697 			return rte_flow_error_set(error, ENOTSUP,
2698 						  RTE_FLOW_ERROR_TYPE_ITEM,
2699 						  item,
2700 						  "VxLAN-GPE protocol"
2701 						  " not supported");
2702 		memcpy(&id.vni[1], spec->vni, 3);
2703 		memcpy(&id.vni[1], mask->vni, 3);
2704 	}
2705 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2706 		return rte_flow_error_set(error, ENOTSUP,
2707 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2708 					  "VXLAN-GPE tunnel must be fully"
2709 					  " defined");
2710 	return 0;
2711 }
2712 /**
2713  * Validate GRE Key item.
2714  *
2715  * @param[in] item
2716  *   Item specification.
2717  * @param[in] item_flags
2718  *   Bit flags to mark detected items.
2719  * @param[in] gre_item
2720  *   Pointer to gre_item
2721  * @param[out] error
2722  *   Pointer to error structure.
2723  *
2724  * @return
2725  *   0 on success, a negative errno value otherwise and rte_errno is set.
2726  */
2727 int
2728 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2729 				uint64_t item_flags,
2730 				const struct rte_flow_item *gre_item,
2731 				struct rte_flow_error *error)
2732 {
2733 	const rte_be32_t *mask = item->mask;
2734 	int ret = 0;
2735 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2736 	const struct rte_flow_item_gre *gre_spec;
2737 	const struct rte_flow_item_gre *gre_mask;
2738 
2739 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2740 		return rte_flow_error_set(error, ENOTSUP,
2741 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 					  "Multiple GRE key not support");
2743 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2744 		return rte_flow_error_set(error, ENOTSUP,
2745 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2746 					  "No preceding GRE header");
2747 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2748 		return rte_flow_error_set(error, ENOTSUP,
2749 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2750 					  "GRE key following a wrong item");
2751 	gre_mask = gre_item->mask;
2752 	if (!gre_mask)
2753 		gre_mask = &rte_flow_item_gre_mask;
2754 	gre_spec = gre_item->spec;
2755 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2756 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2757 		return rte_flow_error_set(error, EINVAL,
2758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2759 					  "Key bit must be on");
2760 
2761 	if (!mask)
2762 		mask = &gre_key_default_mask;
2763 	ret = mlx5_flow_item_acceptable
2764 		(item, (const uint8_t *)mask,
2765 		 (const uint8_t *)&gre_key_default_mask,
2766 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2767 	return ret;
2768 }
2769 
2770 /**
2771  * Validate GRE item.
2772  *
2773  * @param[in] item
2774  *   Item specification.
2775  * @param[in] item_flags
2776  *   Bit flags to mark detected items.
2777  * @param[in] target_protocol
2778  *   The next protocol in the previous item.
2779  * @param[out] error
2780  *   Pointer to error structure.
2781  *
2782  * @return
2783  *   0 on success, a negative errno value otherwise and rte_errno is set.
2784  */
2785 int
2786 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2787 			    uint64_t item_flags,
2788 			    uint8_t target_protocol,
2789 			    struct rte_flow_error *error)
2790 {
2791 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2792 	const struct rte_flow_item_gre *mask = item->mask;
2793 	int ret;
2794 	const struct rte_flow_item_gre nic_mask = {
2795 		.c_rsvd0_ver = RTE_BE16(0xB000),
2796 		.protocol = RTE_BE16(UINT16_MAX),
2797 	};
2798 
2799 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2800 		return rte_flow_error_set(error, EINVAL,
2801 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2802 					  "protocol filtering not compatible"
2803 					  " with this GRE layer");
2804 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2805 		return rte_flow_error_set(error, ENOTSUP,
2806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 					  "multiple tunnel layers not"
2808 					  " supported");
2809 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2810 		return rte_flow_error_set(error, ENOTSUP,
2811 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2812 					  "L3 Layer is missing");
2813 	if (!mask)
2814 		mask = &rte_flow_item_gre_mask;
2815 	ret = mlx5_flow_item_acceptable
2816 		(item, (const uint8_t *)mask,
2817 		 (const uint8_t *)&nic_mask,
2818 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2819 		 error);
2820 	if (ret < 0)
2821 		return ret;
2822 #ifndef HAVE_MLX5DV_DR
2823 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2824 	if (spec && (spec->protocol & mask->protocol))
2825 		return rte_flow_error_set(error, ENOTSUP,
2826 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2827 					  "without MPLS support the"
2828 					  " specification cannot be used for"
2829 					  " filtering");
2830 #endif
2831 #endif
2832 	return 0;
2833 }
2834 
2835 /**
2836  * Validate Geneve item.
2837  *
2838  * @param[in] item
2839  *   Item specification.
2840  * @param[in] itemFlags
2841  *   Bit-fields that holds the items detected until now.
2842  * @param[in] enPriv
2843  *   Pointer to the private data structure.
2844  * @param[out] error
2845  *   Pointer to error structure.
2846  *
2847  * @return
2848  *   0 on success, a negative errno value otherwise and rte_errno is set.
2849  */
2850 
2851 int
2852 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2853 			       uint64_t item_flags,
2854 			       struct rte_eth_dev *dev,
2855 			       struct rte_flow_error *error)
2856 {
2857 	struct mlx5_priv *priv = dev->data->dev_private;
2858 	const struct rte_flow_item_geneve *spec = item->spec;
2859 	const struct rte_flow_item_geneve *mask = item->mask;
2860 	int ret;
2861 	uint16_t gbhdr;
2862 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2863 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2864 	const struct rte_flow_item_geneve nic_mask = {
2865 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2866 		.vni = "\xff\xff\xff",
2867 		.protocol = RTE_BE16(UINT16_MAX),
2868 	};
2869 
2870 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2871 		return rte_flow_error_set(error, ENOTSUP,
2872 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2873 					  "L3 Geneve is not enabled by device"
2874 					  " parameter and/or not configured in"
2875 					  " firmware");
2876 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2877 		return rte_flow_error_set(error, ENOTSUP,
2878 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2879 					  "multiple tunnel layers not"
2880 					  " supported");
2881 	/*
2882 	 * Verify only UDPv4 is present as defined in
2883 	 * https://tools.ietf.org/html/rfc7348
2884 	 */
2885 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2886 		return rte_flow_error_set(error, EINVAL,
2887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2888 					  "no outer UDP layer found");
2889 	if (!mask)
2890 		mask = &rte_flow_item_geneve_mask;
2891 	ret = mlx5_flow_item_acceptable
2892 				  (item, (const uint8_t *)mask,
2893 				   (const uint8_t *)&nic_mask,
2894 				   sizeof(struct rte_flow_item_geneve),
2895 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2896 	if (ret)
2897 		return ret;
2898 	if (spec) {
2899 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2900 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2901 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2902 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2903 			return rte_flow_error_set(error, ENOTSUP,
2904 						  RTE_FLOW_ERROR_TYPE_ITEM,
2905 						  item,
2906 						  "Geneve protocol unsupported"
2907 						  " fields are being used");
2908 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2909 			return rte_flow_error_set
2910 					(error, ENOTSUP,
2911 					 RTE_FLOW_ERROR_TYPE_ITEM,
2912 					 item,
2913 					 "Unsupported Geneve options length");
2914 	}
2915 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2916 		return rte_flow_error_set
2917 				    (error, ENOTSUP,
2918 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2919 				     "Geneve tunnel must be fully defined");
2920 	return 0;
2921 }
2922 
2923 /**
2924  * Validate Geneve TLV option item.
2925  *
2926  * @param[in] item
2927  *   Item specification.
2928  * @param[in] last_item
2929  *   Previous validated item in the pattern items.
2930  * @param[in] geneve_item
2931  *   Previous GENEVE item specification.
2932  * @param[in] dev
2933  *   Pointer to the rte_eth_dev structure.
2934  * @param[out] error
2935  *   Pointer to error structure.
2936  *
2937  * @return
2938  *   0 on success, a negative errno value otherwise and rte_errno is set.
2939  */
2940 int
2941 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2942 				   uint64_t last_item,
2943 				   const struct rte_flow_item *geneve_item,
2944 				   struct rte_eth_dev *dev,
2945 				   struct rte_flow_error *error)
2946 {
2947 	struct mlx5_priv *priv = dev->data->dev_private;
2948 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2949 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2950 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2951 	uint8_t data_max_supported =
2952 			hca_attr->max_geneve_tlv_option_data_len * 4;
2953 	struct mlx5_dev_config *config = &priv->config;
2954 	const struct rte_flow_item_geneve *geneve_spec;
2955 	const struct rte_flow_item_geneve *geneve_mask;
2956 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2957 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2958 	unsigned int i;
2959 	unsigned int data_len;
2960 	uint8_t tlv_option_len;
2961 	uint16_t optlen_m, optlen_v;
2962 	const struct rte_flow_item_geneve_opt full_mask = {
2963 		.option_class = RTE_BE16(0xffff),
2964 		.option_type = 0xff,
2965 		.option_len = 0x1f,
2966 	};
2967 
2968 	if (!mask)
2969 		mask = &rte_flow_item_geneve_opt_mask;
2970 	if (!spec)
2971 		return rte_flow_error_set
2972 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2973 			"Geneve TLV opt class/type/length must be specified");
2974 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2975 		return rte_flow_error_set
2976 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2977 			"Geneve TLV opt length exceeeds the limit (31)");
2978 	/* Check if class type and length masks are full. */
2979 	if (full_mask.option_class != mask->option_class ||
2980 	    full_mask.option_type != mask->option_type ||
2981 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2982 		return rte_flow_error_set
2983 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2984 			"Geneve TLV opt class/type/length masks must be full");
2985 	/* Check if length is supported */
2986 	if ((uint32_t)spec->option_len >
2987 			config->hca_attr.max_geneve_tlv_option_data_len)
2988 		return rte_flow_error_set
2989 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2990 			"Geneve TLV opt length not supported");
2991 	if (config->hca_attr.max_geneve_tlv_options > 1)
2992 		DRV_LOG(DEBUG,
2993 			"max_geneve_tlv_options supports more than 1 option");
2994 	/* Check GENEVE item preceding. */
2995 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2996 		return rte_flow_error_set
2997 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2998 			"Geneve opt item must be preceded with Geneve item");
2999 	geneve_spec = geneve_item->spec;
3000 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3001 					  &rte_flow_item_geneve_mask;
3002 	/* Check if GENEVE TLV option size doesn't exceed option length */
3003 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3004 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3005 		tlv_option_len = spec->option_len & mask->option_len;
3006 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3007 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3008 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3009 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3010 		if ((optlen_v & optlen_m) <= tlv_option_len)
3011 			return rte_flow_error_set
3012 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3013 				 "GENEVE TLV option length exceeds optlen");
3014 	}
3015 	/* Check if length is 0 or data is 0. */
3016 	if (spec->data == NULL || spec->option_len == 0)
3017 		return rte_flow_error_set
3018 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3019 			"Geneve TLV opt with zero data/length not supported");
3020 	/* Check not all data & mask are 0. */
3021 	data_len = spec->option_len * 4;
3022 	if (mask->data == NULL) {
3023 		for (i = 0; i < data_len; i++)
3024 			if (spec->data[i])
3025 				break;
3026 		if (i == data_len)
3027 			return rte_flow_error_set(error, ENOTSUP,
3028 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3029 				"Can't match on Geneve option data 0");
3030 	} else {
3031 		for (i = 0; i < data_len; i++)
3032 			if (spec->data[i] & mask->data[i])
3033 				break;
3034 		if (i == data_len)
3035 			return rte_flow_error_set(error, ENOTSUP,
3036 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 				"Can't match on Geneve option data and mask 0");
3038 		/* Check data mask supported. */
3039 		for (i = data_max_supported; i < data_len ; i++)
3040 			if (mask->data[i])
3041 				return rte_flow_error_set(error, ENOTSUP,
3042 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3043 					"Data mask is of unsupported size");
3044 	}
3045 	/* Check GENEVE option is supported in NIC. */
3046 	if (!config->hca_attr.geneve_tlv_opt)
3047 		return rte_flow_error_set
3048 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3049 			"Geneve TLV opt not supported");
3050 	/* Check if we already have geneve option with different type/class. */
3051 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3052 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3053 	if (geneve_opt_resource != NULL)
3054 		if (geneve_opt_resource->option_class != spec->option_class ||
3055 		    geneve_opt_resource->option_type != spec->option_type ||
3056 		    geneve_opt_resource->length != spec->option_len) {
3057 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3058 			return rte_flow_error_set(error, ENOTSUP,
3059 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3060 				"Only one Geneve TLV option supported");
3061 		}
3062 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3063 	return 0;
3064 }
3065 
3066 /**
3067  * Validate MPLS item.
3068  *
3069  * @param[in] dev
3070  *   Pointer to the rte_eth_dev structure.
3071  * @param[in] item
3072  *   Item specification.
3073  * @param[in] item_flags
3074  *   Bit-fields that holds the items detected until now.
3075  * @param[in] prev_layer
3076  *   The protocol layer indicated in previous item.
3077  * @param[out] error
3078  *   Pointer to error structure.
3079  *
3080  * @return
3081  *   0 on success, a negative errno value otherwise and rte_errno is set.
3082  */
3083 int
3084 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3085 			     const struct rte_flow_item *item __rte_unused,
3086 			     uint64_t item_flags __rte_unused,
3087 			     uint64_t prev_layer __rte_unused,
3088 			     struct rte_flow_error *error)
3089 {
3090 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3091 	const struct rte_flow_item_mpls *mask = item->mask;
3092 	struct mlx5_priv *priv = dev->data->dev_private;
3093 	int ret;
3094 
3095 	if (!priv->config.mpls_en)
3096 		return rte_flow_error_set(error, ENOTSUP,
3097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3098 					  "MPLS not supported or"
3099 					  " disabled in firmware"
3100 					  " configuration.");
3101 	/* MPLS over UDP, GRE is allowed */
3102 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3103 			    MLX5_FLOW_LAYER_GRE |
3104 			    MLX5_FLOW_LAYER_GRE_KEY)))
3105 		return rte_flow_error_set(error, EINVAL,
3106 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 					  "protocol filtering not compatible"
3108 					  " with MPLS layer");
3109 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3110 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3111 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3112 		return rte_flow_error_set(error, ENOTSUP,
3113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3114 					  "multiple tunnel layers not"
3115 					  " supported");
3116 	if (!mask)
3117 		mask = &rte_flow_item_mpls_mask;
3118 	ret = mlx5_flow_item_acceptable
3119 		(item, (const uint8_t *)mask,
3120 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3121 		 sizeof(struct rte_flow_item_mpls),
3122 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3123 	if (ret < 0)
3124 		return ret;
3125 	return 0;
3126 #else
3127 	return rte_flow_error_set(error, ENOTSUP,
3128 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3129 				  "MPLS is not supported by Verbs, please"
3130 				  " update.");
3131 #endif
3132 }
3133 
3134 /**
3135  * Validate NVGRE item.
3136  *
3137  * @param[in] item
3138  *   Item specification.
3139  * @param[in] item_flags
3140  *   Bit flags to mark detected items.
3141  * @param[in] target_protocol
3142  *   The next protocol in the previous item.
3143  * @param[out] error
3144  *   Pointer to error structure.
3145  *
3146  * @return
3147  *   0 on success, a negative errno value otherwise and rte_errno is set.
3148  */
3149 int
3150 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3151 			      uint64_t item_flags,
3152 			      uint8_t target_protocol,
3153 			      struct rte_flow_error *error)
3154 {
3155 	const struct rte_flow_item_nvgre *mask = item->mask;
3156 	int ret;
3157 
3158 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3159 		return rte_flow_error_set(error, EINVAL,
3160 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3161 					  "protocol filtering not compatible"
3162 					  " with this GRE layer");
3163 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3164 		return rte_flow_error_set(error, ENOTSUP,
3165 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3166 					  "multiple tunnel layers not"
3167 					  " supported");
3168 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3169 		return rte_flow_error_set(error, ENOTSUP,
3170 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3171 					  "L3 Layer is missing");
3172 	if (!mask)
3173 		mask = &rte_flow_item_nvgre_mask;
3174 	ret = mlx5_flow_item_acceptable
3175 		(item, (const uint8_t *)mask,
3176 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3177 		 sizeof(struct rte_flow_item_nvgre),
3178 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3179 	if (ret < 0)
3180 		return ret;
3181 	return 0;
3182 }
3183 
3184 /**
3185  * Validate eCPRI item.
3186  *
3187  * @param[in] item
3188  *   Item specification.
3189  * @param[in] item_flags
3190  *   Bit-fields that holds the items detected until now.
3191  * @param[in] last_item
3192  *   Previous validated item in the pattern items.
3193  * @param[in] ether_type
3194  *   Type in the ethernet layer header (including dot1q).
3195  * @param[in] acc_mask
3196  *   Acceptable mask, if NULL default internal default mask
3197  *   will be used to check whether item fields are supported.
3198  * @param[out] error
3199  *   Pointer to error structure.
3200  *
3201  * @return
3202  *   0 on success, a negative errno value otherwise and rte_errno is set.
3203  */
3204 int
3205 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3206 			      uint64_t item_flags,
3207 			      uint64_t last_item,
3208 			      uint16_t ether_type,
3209 			      const struct rte_flow_item_ecpri *acc_mask,
3210 			      struct rte_flow_error *error)
3211 {
3212 	const struct rte_flow_item_ecpri *mask = item->mask;
3213 	const struct rte_flow_item_ecpri nic_mask = {
3214 		.hdr = {
3215 			.common = {
3216 				.u32 =
3217 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3218 					.type = 0xFF,
3219 					}).u32),
3220 			},
3221 			.dummy[0] = 0xFFFFFFFF,
3222 		},
3223 	};
3224 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3225 					MLX5_FLOW_LAYER_OUTER_VLAN);
3226 	struct rte_flow_item_ecpri mask_lo;
3227 
3228 	if (!(last_item & outer_l2_vlan) &&
3229 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3230 		return rte_flow_error_set(error, EINVAL,
3231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3232 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3233 	if ((last_item & outer_l2_vlan) && ether_type &&
3234 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3235 		return rte_flow_error_set(error, EINVAL,
3236 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3237 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3238 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3239 		return rte_flow_error_set(error, EINVAL,
3240 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3241 					  "eCPRI with tunnel is not supported right now");
3242 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3243 		return rte_flow_error_set(error, ENOTSUP,
3244 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3245 					  "multiple L3 layers not supported");
3246 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3247 		return rte_flow_error_set(error, EINVAL,
3248 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3249 					  "eCPRI cannot coexist with a TCP layer");
3250 	/* In specification, eCPRI could be over UDP layer. */
3251 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3252 		return rte_flow_error_set(error, EINVAL,
3253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3254 					  "eCPRI over UDP layer is not yet supported right now");
3255 	/* Mask for type field in common header could be zero. */
3256 	if (!mask)
3257 		mask = &rte_flow_item_ecpri_mask;
3258 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3259 	/* Input mask is in big-endian format. */
3260 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3261 		return rte_flow_error_set(error, EINVAL,
3262 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3263 					  "partial mask is not supported for protocol");
3264 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3265 		return rte_flow_error_set(error, EINVAL,
3266 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3267 					  "message header mask must be after a type mask");
3268 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3269 					 acc_mask ? (const uint8_t *)acc_mask
3270 						  : (const uint8_t *)&nic_mask,
3271 					 sizeof(struct rte_flow_item_ecpri),
3272 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3273 }
3274 
3275 static int
3276 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3277 		   const struct rte_flow_attr *attr __rte_unused,
3278 		   const struct rte_flow_item items[] __rte_unused,
3279 		   const struct rte_flow_action actions[] __rte_unused,
3280 		   bool external __rte_unused,
3281 		   int hairpin __rte_unused,
3282 		   struct rte_flow_error *error)
3283 {
3284 	return rte_flow_error_set(error, ENOTSUP,
3285 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3286 }
3287 
3288 static struct mlx5_flow *
3289 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3290 		  const struct rte_flow_attr *attr __rte_unused,
3291 		  const struct rte_flow_item items[] __rte_unused,
3292 		  const struct rte_flow_action actions[] __rte_unused,
3293 		  struct rte_flow_error *error)
3294 {
3295 	rte_flow_error_set(error, ENOTSUP,
3296 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3297 	return NULL;
3298 }
3299 
3300 static int
3301 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3302 		    struct mlx5_flow *dev_flow __rte_unused,
3303 		    const struct rte_flow_attr *attr __rte_unused,
3304 		    const struct rte_flow_item items[] __rte_unused,
3305 		    const struct rte_flow_action actions[] __rte_unused,
3306 		    struct rte_flow_error *error)
3307 {
3308 	return rte_flow_error_set(error, ENOTSUP,
3309 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3310 }
3311 
3312 static int
3313 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3314 		struct rte_flow *flow __rte_unused,
3315 		struct rte_flow_error *error)
3316 {
3317 	return rte_flow_error_set(error, ENOTSUP,
3318 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3319 }
3320 
3321 static void
3322 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3323 		 struct rte_flow *flow __rte_unused)
3324 {
3325 }
3326 
3327 static void
3328 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3329 		  struct rte_flow *flow __rte_unused)
3330 {
3331 }
3332 
3333 static int
3334 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3335 		struct rte_flow *flow __rte_unused,
3336 		const struct rte_flow_action *actions __rte_unused,
3337 		void *data __rte_unused,
3338 		struct rte_flow_error *error)
3339 {
3340 	return rte_flow_error_set(error, ENOTSUP,
3341 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3342 }
3343 
3344 static int
3345 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3346 		      uint32_t domains __rte_unused,
3347 		      uint32_t flags __rte_unused)
3348 {
3349 	return 0;
3350 }
3351 
3352 /* Void driver to protect from null pointer reference. */
3353 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3354 	.validate = flow_null_validate,
3355 	.prepare = flow_null_prepare,
3356 	.translate = flow_null_translate,
3357 	.apply = flow_null_apply,
3358 	.remove = flow_null_remove,
3359 	.destroy = flow_null_destroy,
3360 	.query = flow_null_query,
3361 	.sync_domain = flow_null_sync_domain,
3362 };
3363 
3364 /**
3365  * Select flow driver type according to flow attributes and device
3366  * configuration.
3367  *
3368  * @param[in] dev
3369  *   Pointer to the dev structure.
3370  * @param[in] attr
3371  *   Pointer to the flow attributes.
3372  *
3373  * @return
3374  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3375  */
3376 static enum mlx5_flow_drv_type
3377 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3378 {
3379 	struct mlx5_priv *priv = dev->data->dev_private;
3380 	/* The OS can determine first a specific flow type (DV, VERBS) */
3381 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3382 
3383 	if (type != MLX5_FLOW_TYPE_MAX)
3384 		return type;
3385 	/* If no OS specific type - continue with DV/VERBS selection */
3386 	if (attr->transfer && priv->config.dv_esw_en)
3387 		type = MLX5_FLOW_TYPE_DV;
3388 	if (!attr->transfer)
3389 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3390 						 MLX5_FLOW_TYPE_VERBS;
3391 	return type;
3392 }
3393 
3394 #define flow_get_drv_ops(type) flow_drv_ops[type]
3395 
3396 /**
3397  * Flow driver validation API. This abstracts calling driver specific functions.
3398  * The type of flow driver is determined according to flow attributes.
3399  *
3400  * @param[in] dev
3401  *   Pointer to the dev structure.
3402  * @param[in] attr
3403  *   Pointer to the flow attributes.
3404  * @param[in] items
3405  *   Pointer to the list of items.
3406  * @param[in] actions
3407  *   Pointer to the list of actions.
3408  * @param[in] external
3409  *   This flow rule is created by request external to PMD.
3410  * @param[in] hairpin
3411  *   Number of hairpin TX actions, 0 means classic flow.
3412  * @param[out] error
3413  *   Pointer to the error structure.
3414  *
3415  * @return
3416  *   0 on success, a negative errno value otherwise and rte_errno is set.
3417  */
3418 static inline int
3419 flow_drv_validate(struct rte_eth_dev *dev,
3420 		  const struct rte_flow_attr *attr,
3421 		  const struct rte_flow_item items[],
3422 		  const struct rte_flow_action actions[],
3423 		  bool external, int hairpin, struct rte_flow_error *error)
3424 {
3425 	const struct mlx5_flow_driver_ops *fops;
3426 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3427 
3428 	fops = flow_get_drv_ops(type);
3429 	return fops->validate(dev, attr, items, actions, external,
3430 			      hairpin, error);
3431 }
3432 
3433 /**
3434  * Flow driver preparation API. This abstracts calling driver specific
3435  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3436  * calculates the size of memory required for device flow, allocates the memory,
3437  * initializes the device flow and returns the pointer.
3438  *
3439  * @note
3440  *   This function initializes device flow structure such as dv or verbs in
3441  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3442  *   rest. For example, adding returning device flow to flow->dev_flow list and
3443  *   setting backward reference to the flow should be done out of this function.
3444  *   layers field is not filled either.
3445  *
3446  * @param[in] dev
3447  *   Pointer to the dev structure.
3448  * @param[in] attr
3449  *   Pointer to the flow attributes.
3450  * @param[in] items
3451  *   Pointer to the list of items.
3452  * @param[in] actions
3453  *   Pointer to the list of actions.
3454  * @param[in] flow_idx
3455  *   This memory pool index to the flow.
3456  * @param[out] error
3457  *   Pointer to the error structure.
3458  *
3459  * @return
3460  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3461  */
3462 static inline struct mlx5_flow *
3463 flow_drv_prepare(struct rte_eth_dev *dev,
3464 		 const struct rte_flow *flow,
3465 		 const struct rte_flow_attr *attr,
3466 		 const struct rte_flow_item items[],
3467 		 const struct rte_flow_action actions[],
3468 		 uint32_t flow_idx,
3469 		 struct rte_flow_error *error)
3470 {
3471 	const struct mlx5_flow_driver_ops *fops;
3472 	enum mlx5_flow_drv_type type = flow->drv_type;
3473 	struct mlx5_flow *mlx5_flow = NULL;
3474 
3475 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3476 	fops = flow_get_drv_ops(type);
3477 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3478 	if (mlx5_flow)
3479 		mlx5_flow->flow_idx = flow_idx;
3480 	return mlx5_flow;
3481 }
3482 
3483 /**
3484  * Flow driver translation API. This abstracts calling driver specific
3485  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3486  * translates a generic flow into a driver flow. flow_drv_prepare() must
3487  * precede.
3488  *
3489  * @note
3490  *   dev_flow->layers could be filled as a result of parsing during translation
3491  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3492  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3493  *   flow->actions could be overwritten even though all the expanded dev_flows
3494  *   have the same actions.
3495  *
3496  * @param[in] dev
3497  *   Pointer to the rte dev structure.
3498  * @param[in, out] dev_flow
3499  *   Pointer to the mlx5 flow.
3500  * @param[in] attr
3501  *   Pointer to the flow attributes.
3502  * @param[in] items
3503  *   Pointer to the list of items.
3504  * @param[in] actions
3505  *   Pointer to the list of actions.
3506  * @param[out] error
3507  *   Pointer to the error structure.
3508  *
3509  * @return
3510  *   0 on success, a negative errno value otherwise and rte_errno is set.
3511  */
3512 static inline int
3513 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3514 		   const struct rte_flow_attr *attr,
3515 		   const struct rte_flow_item items[],
3516 		   const struct rte_flow_action actions[],
3517 		   struct rte_flow_error *error)
3518 {
3519 	const struct mlx5_flow_driver_ops *fops;
3520 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3521 
3522 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3523 	fops = flow_get_drv_ops(type);
3524 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3525 }
3526 
3527 /**
3528  * Flow driver apply API. This abstracts calling driver specific functions.
3529  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3530  * translated driver flows on to device. flow_drv_translate() must precede.
3531  *
3532  * @param[in] dev
3533  *   Pointer to Ethernet device structure.
3534  * @param[in, out] flow
3535  *   Pointer to flow structure.
3536  * @param[out] error
3537  *   Pointer to error structure.
3538  *
3539  * @return
3540  *   0 on success, a negative errno value otherwise and rte_errno is set.
3541  */
3542 static inline int
3543 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3544 	       struct rte_flow_error *error)
3545 {
3546 	const struct mlx5_flow_driver_ops *fops;
3547 	enum mlx5_flow_drv_type type = flow->drv_type;
3548 
3549 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3550 	fops = flow_get_drv_ops(type);
3551 	return fops->apply(dev, flow, error);
3552 }
3553 
3554 /**
3555  * Flow driver destroy API. This abstracts calling driver specific functions.
3556  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3557  * on device and releases resources of the flow.
3558  *
3559  * @param[in] dev
3560  *   Pointer to Ethernet device.
3561  * @param[in, out] flow
3562  *   Pointer to flow structure.
3563  */
3564 static inline void
3565 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3566 {
3567 	const struct mlx5_flow_driver_ops *fops;
3568 	enum mlx5_flow_drv_type type = flow->drv_type;
3569 
3570 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3571 	fops = flow_get_drv_ops(type);
3572 	fops->destroy(dev, flow);
3573 }
3574 
3575 /**
3576  * Flow driver find RSS policy tbl API. This abstracts calling driver
3577  * specific functions. Parent flow (rte_flow) should have driver
3578  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3579  *
3580  * @param[in] dev
3581  *   Pointer to Ethernet device.
3582  * @param[in, out] flow
3583  *   Pointer to flow structure.
3584  * @param[in] policy
3585  *   Pointer to meter policy table.
3586  * @param[in] rss_desc
3587  *   Pointer to rss_desc
3588  */
3589 static struct mlx5_flow_meter_sub_policy *
3590 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3591 		struct rte_flow *flow,
3592 		struct mlx5_flow_meter_policy *policy,
3593 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3594 {
3595 	const struct mlx5_flow_driver_ops *fops;
3596 	enum mlx5_flow_drv_type type = flow->drv_type;
3597 
3598 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3599 	fops = flow_get_drv_ops(type);
3600 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3601 }
3602 
3603 /**
3604  * Flow driver color tag rule API. This abstracts calling driver
3605  * specific functions. Parent flow (rte_flow) should have driver
3606  * type (drv_type). It will create the color tag rules in hierarchy meter.
3607  *
3608  * @param[in] dev
3609  *   Pointer to Ethernet device.
3610  * @param[in, out] flow
3611  *   Pointer to flow structure.
3612  * @param[in] fm
3613  *   Pointer to flow meter structure.
3614  * @param[in] src_port
3615  *   The src port this extra rule should use.
3616  * @param[in] item
3617  *   The src port id match item.
3618  * @param[out] error
3619  *   Pointer to error structure.
3620  */
3621 static int
3622 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3623 		struct rte_flow *flow,
3624 		struct mlx5_flow_meter_info *fm,
3625 		int32_t src_port,
3626 		const struct rte_flow_item *item,
3627 		struct rte_flow_error *error)
3628 {
3629 	const struct mlx5_flow_driver_ops *fops;
3630 	enum mlx5_flow_drv_type type = flow->drv_type;
3631 
3632 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3633 	fops = flow_get_drv_ops(type);
3634 	return fops->meter_hierarchy_rule_create(dev, fm,
3635 						src_port, item, error);
3636 }
3637 
3638 /**
3639  * Get RSS action from the action list.
3640  *
3641  * @param[in] dev
3642  *   Pointer to Ethernet device.
3643  * @param[in] actions
3644  *   Pointer to the list of actions.
3645  * @param[in] flow
3646  *   Parent flow structure pointer.
3647  *
3648  * @return
3649  *   Pointer to the RSS action if exist, else return NULL.
3650  */
3651 static const struct rte_flow_action_rss*
3652 flow_get_rss_action(struct rte_eth_dev *dev,
3653 		    const struct rte_flow_action actions[])
3654 {
3655 	struct mlx5_priv *priv = dev->data->dev_private;
3656 	const struct rte_flow_action_rss *rss = NULL;
3657 	struct mlx5_meter_policy_action_container *acg;
3658 	struct mlx5_meter_policy_action_container *acy;
3659 
3660 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3661 		switch (actions->type) {
3662 		case RTE_FLOW_ACTION_TYPE_RSS:
3663 			rss = actions->conf;
3664 			break;
3665 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3666 		{
3667 			const struct rte_flow_action_sample *sample =
3668 								actions->conf;
3669 			const struct rte_flow_action *act = sample->actions;
3670 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3671 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3672 					rss = act->conf;
3673 			break;
3674 		}
3675 		case RTE_FLOW_ACTION_TYPE_METER:
3676 		{
3677 			uint32_t mtr_idx;
3678 			struct mlx5_flow_meter_info *fm;
3679 			struct mlx5_flow_meter_policy *policy;
3680 			const struct rte_flow_action_meter *mtr = actions->conf;
3681 
3682 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3683 			if (fm && !fm->def_policy) {
3684 				policy = mlx5_flow_meter_policy_find(dev,
3685 						fm->policy_id, NULL);
3686 				MLX5_ASSERT(policy);
3687 				if (policy->is_hierarchy) {
3688 					policy =
3689 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3690 									policy);
3691 					if (!policy)
3692 						return NULL;
3693 				}
3694 				if (policy->is_rss) {
3695 					acg =
3696 					&policy->act_cnt[RTE_COLOR_GREEN];
3697 					acy =
3698 					&policy->act_cnt[RTE_COLOR_YELLOW];
3699 					if (acg->fate_action ==
3700 					    MLX5_FLOW_FATE_SHARED_RSS)
3701 						rss = acg->rss->conf;
3702 					else if (acy->fate_action ==
3703 						 MLX5_FLOW_FATE_SHARED_RSS)
3704 						rss = acy->rss->conf;
3705 				}
3706 			}
3707 			break;
3708 		}
3709 		default:
3710 			break;
3711 		}
3712 	}
3713 	return rss;
3714 }
3715 
3716 /**
3717  * Get ASO age action by index.
3718  *
3719  * @param[in] dev
3720  *   Pointer to the Ethernet device structure.
3721  * @param[in] age_idx
3722  *   Index to the ASO age action.
3723  *
3724  * @return
3725  *   The specified ASO age action.
3726  */
3727 struct mlx5_aso_age_action*
3728 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3729 {
3730 	uint16_t pool_idx = age_idx & UINT16_MAX;
3731 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3732 	struct mlx5_priv *priv = dev->data->dev_private;
3733 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3734 	struct mlx5_aso_age_pool *pool;
3735 
3736 	rte_rwlock_read_lock(&mng->resize_rwl);
3737 	pool = mng->pools[pool_idx];
3738 	rte_rwlock_read_unlock(&mng->resize_rwl);
3739 	return &pool->actions[offset - 1];
3740 }
3741 
3742 /* maps indirect action to translated direct in some actions array */
3743 struct mlx5_translated_action_handle {
3744 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3745 	int index; /**< Index in related array of rte_flow_action. */
3746 };
3747 
3748 /**
3749  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3750  * direct action if translation possible.
3751  * This functionality used to run same execution path for both direct and
3752  * indirect actions on flow create. All necessary preparations for indirect
3753  * action handling should be performed on *handle* actions list returned
3754  * from this call.
3755  *
3756  * @param[in] dev
3757  *   Pointer to Ethernet device.
3758  * @param[in] actions
3759  *   List of actions to translate.
3760  * @param[out] handle
3761  *   List to store translated indirect action object handles.
3762  * @param[in, out] indir_n
3763  *   Size of *handle* array. On return should be updated with number of
3764  *   indirect actions retrieved from the *actions* list.
3765  * @param[out] translated_actions
3766  *   List of actions where all indirect actions were translated to direct
3767  *   if possible. NULL if no translation took place.
3768  * @param[out] error
3769  *   Pointer to the error structure.
3770  *
3771  * @return
3772  *   0 on success, a negative errno value otherwise and rte_errno is set.
3773  */
3774 static int
3775 flow_action_handles_translate(struct rte_eth_dev *dev,
3776 			      const struct rte_flow_action actions[],
3777 			      struct mlx5_translated_action_handle *handle,
3778 			      int *indir_n,
3779 			      struct rte_flow_action **translated_actions,
3780 			      struct rte_flow_error *error)
3781 {
3782 	struct mlx5_priv *priv = dev->data->dev_private;
3783 	struct rte_flow_action *translated = NULL;
3784 	size_t actions_size;
3785 	int n;
3786 	int copied_n = 0;
3787 	struct mlx5_translated_action_handle *handle_end = NULL;
3788 
3789 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3790 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3791 			continue;
3792 		if (copied_n == *indir_n) {
3793 			return rte_flow_error_set
3794 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3795 				 NULL, "too many shared actions");
3796 		}
3797 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3798 			   sizeof(actions[n].conf));
3799 		handle[copied_n].index = n;
3800 		copied_n++;
3801 	}
3802 	n++;
3803 	*indir_n = copied_n;
3804 	if (!copied_n)
3805 		return 0;
3806 	actions_size = sizeof(struct rte_flow_action) * n;
3807 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3808 	if (!translated) {
3809 		rte_errno = ENOMEM;
3810 		return -ENOMEM;
3811 	}
3812 	memcpy(translated, actions, actions_size);
3813 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3814 		struct mlx5_shared_action_rss *shared_rss;
3815 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3816 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3817 		uint32_t idx = act_idx &
3818 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3819 
3820 		switch (type) {
3821 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3822 			shared_rss = mlx5_ipool_get
3823 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3824 			translated[handle->index].type =
3825 				RTE_FLOW_ACTION_TYPE_RSS;
3826 			translated[handle->index].conf =
3827 				&shared_rss->origin;
3828 			break;
3829 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3830 			translated[handle->index].type =
3831 						(enum rte_flow_action_type)
3832 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3833 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3834 			break;
3835 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3836 			if (priv->sh->flow_hit_aso_en) {
3837 				translated[handle->index].type =
3838 					(enum rte_flow_action_type)
3839 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3840 				translated[handle->index].conf =
3841 							 (void *)(uintptr_t)idx;
3842 				break;
3843 			}
3844 			/* Fall-through */
3845 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3846 			if (priv->sh->ct_aso_en) {
3847 				translated[handle->index].type =
3848 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3849 				translated[handle->index].conf =
3850 							 (void *)(uintptr_t)idx;
3851 				break;
3852 			}
3853 			/* Fall-through */
3854 		default:
3855 			mlx5_free(translated);
3856 			return rte_flow_error_set
3857 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3858 				 NULL, "invalid indirect action type");
3859 		}
3860 	}
3861 	*translated_actions = translated;
3862 	return 0;
3863 }
3864 
3865 /**
3866  * Get Shared RSS action from the action list.
3867  *
3868  * @param[in] dev
3869  *   Pointer to Ethernet device.
3870  * @param[in] shared
3871  *   Pointer to the list of actions.
3872  * @param[in] shared_n
3873  *   Actions list length.
3874  *
3875  * @return
3876  *   The MLX5 RSS action ID if exists, otherwise return 0.
3877  */
3878 static uint32_t
3879 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3880 			   struct mlx5_translated_action_handle *handle,
3881 			   int shared_n)
3882 {
3883 	struct mlx5_translated_action_handle *handle_end;
3884 	struct mlx5_priv *priv = dev->data->dev_private;
3885 	struct mlx5_shared_action_rss *shared_rss;
3886 
3887 
3888 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3889 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3890 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3891 		uint32_t idx = act_idx &
3892 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3893 		switch (type) {
3894 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3895 			shared_rss = mlx5_ipool_get
3896 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3897 									   idx);
3898 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3899 					   __ATOMIC_RELAXED);
3900 			return idx;
3901 		default:
3902 			break;
3903 		}
3904 	}
3905 	return 0;
3906 }
3907 
3908 static unsigned int
3909 find_graph_root(uint32_t rss_level)
3910 {
3911 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3912 			       MLX5_EXPANSION_ROOT_OUTER;
3913 }
3914 
3915 /**
3916  *  Get layer flags from the prefix flow.
3917  *
3918  *  Some flows may be split to several subflows, the prefix subflow gets the
3919  *  match items and the suffix sub flow gets the actions.
3920  *  Some actions need the user defined match item flags to get the detail for
3921  *  the action.
3922  *  This function helps the suffix flow to get the item layer flags from prefix
3923  *  subflow.
3924  *
3925  * @param[in] dev_flow
3926  *   Pointer the created preifx subflow.
3927  *
3928  * @return
3929  *   The layers get from prefix subflow.
3930  */
3931 static inline uint64_t
3932 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3933 {
3934 	uint64_t layers = 0;
3935 
3936 	/*
3937 	 * Layers bits could be localization, but usually the compiler will
3938 	 * help to do the optimization work for source code.
3939 	 * If no decap actions, use the layers directly.
3940 	 */
3941 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3942 		return dev_flow->handle->layers;
3943 	/* Convert L3 layers with decap action. */
3944 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3945 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3946 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3947 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3948 	/* Convert L4 layers with decap action.  */
3949 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3950 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3951 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3952 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3953 	return layers;
3954 }
3955 
3956 /**
3957  * Get metadata split action information.
3958  *
3959  * @param[in] actions
3960  *   Pointer to the list of actions.
3961  * @param[out] qrss
3962  *   Pointer to the return pointer.
3963  * @param[out] qrss_type
3964  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3965  *   if no QUEUE/RSS is found.
3966  * @param[out] encap_idx
3967  *   Pointer to the index of the encap action if exists, otherwise the last
3968  *   action index.
3969  *
3970  * @return
3971  *   Total number of actions.
3972  */
3973 static int
3974 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3975 				       const struct rte_flow_action **qrss,
3976 				       int *encap_idx)
3977 {
3978 	const struct rte_flow_action_raw_encap *raw_encap;
3979 	int actions_n = 0;
3980 	int raw_decap_idx = -1;
3981 
3982 	*encap_idx = -1;
3983 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3984 		switch (actions->type) {
3985 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3986 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3987 			*encap_idx = actions_n;
3988 			break;
3989 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3990 			raw_decap_idx = actions_n;
3991 			break;
3992 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3993 			raw_encap = actions->conf;
3994 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3995 				*encap_idx = raw_decap_idx != -1 ?
3996 						      raw_decap_idx : actions_n;
3997 			break;
3998 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3999 		case RTE_FLOW_ACTION_TYPE_RSS:
4000 			*qrss = actions;
4001 			break;
4002 		default:
4003 			break;
4004 		}
4005 		actions_n++;
4006 	}
4007 	if (*encap_idx == -1)
4008 		*encap_idx = actions_n;
4009 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4010 	return actions_n + 1;
4011 }
4012 
4013 /**
4014  * Check if the action will change packet.
4015  *
4016  * @param dev
4017  *   Pointer to Ethernet device.
4018  * @param[in] type
4019  *   action type.
4020  *
4021  * @return
4022  *   true if action will change packet, false otherwise.
4023  */
4024 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4025 					  enum rte_flow_action_type type)
4026 {
4027 	struct mlx5_priv *priv = dev->data->dev_private;
4028 
4029 	switch (type) {
4030 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4031 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4032 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4033 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4034 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4035 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4036 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4037 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4038 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4039 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4040 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4041 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4042 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4043 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4044 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4045 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4046 	case RTE_FLOW_ACTION_TYPE_SET_META:
4047 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4048 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4049 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4050 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4051 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4052 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4053 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4054 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4055 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4056 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4057 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4058 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4059 		return true;
4060 	case RTE_FLOW_ACTION_TYPE_FLAG:
4061 	case RTE_FLOW_ACTION_TYPE_MARK:
4062 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4063 			return true;
4064 		else
4065 			return false;
4066 	default:
4067 		return false;
4068 	}
4069 }
4070 
4071 /**
4072  * Check meter action from the action list.
4073  *
4074  * @param dev
4075  *   Pointer to Ethernet device.
4076  * @param[in] actions
4077  *   Pointer to the list of actions.
4078  * @param[out] has_mtr
4079  *   Pointer to the meter exist flag.
4080  * @param[out] has_modify
4081  *   Pointer to the flag showing there's packet change action.
4082  * @param[out] meter_id
4083  *   Pointer to the meter id.
4084  *
4085  * @return
4086  *   Total number of actions.
4087  */
4088 static int
4089 flow_check_meter_action(struct rte_eth_dev *dev,
4090 			const struct rte_flow_action actions[],
4091 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4092 {
4093 	const struct rte_flow_action_meter *mtr = NULL;
4094 	int actions_n = 0;
4095 
4096 	MLX5_ASSERT(has_mtr);
4097 	*has_mtr = false;
4098 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4099 		switch (actions->type) {
4100 		case RTE_FLOW_ACTION_TYPE_METER:
4101 			mtr = actions->conf;
4102 			*meter_id = mtr->mtr_id;
4103 			*has_mtr = true;
4104 			break;
4105 		default:
4106 			break;
4107 		}
4108 		if (!*has_mtr)
4109 			*has_modify |= flow_check_modify_action_type(dev,
4110 								actions->type);
4111 		actions_n++;
4112 	}
4113 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4114 	return actions_n + 1;
4115 }
4116 
4117 /**
4118  * Check if the flow should be split due to hairpin.
4119  * The reason for the split is that in current HW we can't
4120  * support encap and push-vlan on Rx, so if a flow contains
4121  * these actions we move it to Tx.
4122  *
4123  * @param dev
4124  *   Pointer to Ethernet device.
4125  * @param[in] attr
4126  *   Flow rule attributes.
4127  * @param[in] actions
4128  *   Associated actions (list terminated by the END action).
4129  *
4130  * @return
4131  *   > 0 the number of actions and the flow should be split,
4132  *   0 when no split required.
4133  */
4134 static int
4135 flow_check_hairpin_split(struct rte_eth_dev *dev,
4136 			 const struct rte_flow_attr *attr,
4137 			 const struct rte_flow_action actions[])
4138 {
4139 	int queue_action = 0;
4140 	int action_n = 0;
4141 	int split = 0;
4142 	const struct rte_flow_action_queue *queue;
4143 	const struct rte_flow_action_rss *rss;
4144 	const struct rte_flow_action_raw_encap *raw_encap;
4145 	const struct rte_eth_hairpin_conf *conf;
4146 
4147 	if (!attr->ingress)
4148 		return 0;
4149 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4150 		switch (actions->type) {
4151 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4152 			queue = actions->conf;
4153 			if (queue == NULL)
4154 				return 0;
4155 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4156 			if (conf == NULL || conf->tx_explicit != 0)
4157 				return 0;
4158 			queue_action = 1;
4159 			action_n++;
4160 			break;
4161 		case RTE_FLOW_ACTION_TYPE_RSS:
4162 			rss = actions->conf;
4163 			if (rss == NULL || rss->queue_num == 0)
4164 				return 0;
4165 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4166 			if (conf == NULL || conf->tx_explicit != 0)
4167 				return 0;
4168 			queue_action = 1;
4169 			action_n++;
4170 			break;
4171 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4172 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4173 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4174 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4175 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4176 			split++;
4177 			action_n++;
4178 			break;
4179 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4180 			raw_encap = actions->conf;
4181 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4182 				split++;
4183 			action_n++;
4184 			break;
4185 		default:
4186 			action_n++;
4187 			break;
4188 		}
4189 	}
4190 	if (split && queue_action)
4191 		return action_n;
4192 	return 0;
4193 }
4194 
4195 /* Declare flow create/destroy prototype in advance. */
4196 static uint32_t
4197 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4198 		 const struct rte_flow_attr *attr,
4199 		 const struct rte_flow_item items[],
4200 		 const struct rte_flow_action actions[],
4201 		 bool external, struct rte_flow_error *error);
4202 
4203 static void
4204 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4205 		  uint32_t flow_idx);
4206 
4207 int
4208 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4209 		      struct mlx5_list_entry *entry, void *cb_ctx)
4210 {
4211 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4212 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4213 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4214 
4215 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4216 }
4217 
4218 struct mlx5_list_entry *
4219 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4220 {
4221 	struct rte_eth_dev *dev = tool_ctx;
4222 	struct mlx5_priv *priv = dev->data->dev_private;
4223 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4224 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4225 	struct rte_flow_error *error = ctx->error;
4226 	uint32_t idx = 0;
4227 	int ret;
4228 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4229 	struct rte_flow_attr attr = {
4230 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4231 		.ingress = 1,
4232 	};
4233 	struct mlx5_rte_flow_item_tag tag_spec = {
4234 		.data = mark_id,
4235 	};
4236 	struct rte_flow_item items[] = {
4237 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4238 	};
4239 	struct rte_flow_action_mark ftag = {
4240 		.id = mark_id,
4241 	};
4242 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4243 		.dst = REG_B,
4244 		.src = REG_NON,
4245 	};
4246 	struct rte_flow_action_jump jump = {
4247 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4248 	};
4249 	struct rte_flow_action actions[] = {
4250 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4251 	};
4252 
4253 	/* Fill the register fileds in the flow. */
4254 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4255 	if (ret < 0)
4256 		return NULL;
4257 	tag_spec.id = ret;
4258 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4259 	if (ret < 0)
4260 		return NULL;
4261 	cp_mreg.src = ret;
4262 	/* Provide the full width of FLAG specific value. */
4263 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4264 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4265 	/* Build a new flow. */
4266 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4267 		items[0] = (struct rte_flow_item){
4268 			.type = (enum rte_flow_item_type)
4269 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4270 			.spec = &tag_spec,
4271 		};
4272 		items[1] = (struct rte_flow_item){
4273 			.type = RTE_FLOW_ITEM_TYPE_END,
4274 		};
4275 		actions[0] = (struct rte_flow_action){
4276 			.type = (enum rte_flow_action_type)
4277 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4278 			.conf = &ftag,
4279 		};
4280 		actions[1] = (struct rte_flow_action){
4281 			.type = (enum rte_flow_action_type)
4282 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4283 			.conf = &cp_mreg,
4284 		};
4285 		actions[2] = (struct rte_flow_action){
4286 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4287 			.conf = &jump,
4288 		};
4289 		actions[3] = (struct rte_flow_action){
4290 			.type = RTE_FLOW_ACTION_TYPE_END,
4291 		};
4292 	} else {
4293 		/* Default rule, wildcard match. */
4294 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4295 		items[0] = (struct rte_flow_item){
4296 			.type = RTE_FLOW_ITEM_TYPE_END,
4297 		};
4298 		actions[0] = (struct rte_flow_action){
4299 			.type = (enum rte_flow_action_type)
4300 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4301 			.conf = &cp_mreg,
4302 		};
4303 		actions[1] = (struct rte_flow_action){
4304 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4305 			.conf = &jump,
4306 		};
4307 		actions[2] = (struct rte_flow_action){
4308 			.type = RTE_FLOW_ACTION_TYPE_END,
4309 		};
4310 	}
4311 	/* Build a new entry. */
4312 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4313 	if (!mcp_res) {
4314 		rte_errno = ENOMEM;
4315 		return NULL;
4316 	}
4317 	mcp_res->idx = idx;
4318 	mcp_res->mark_id = mark_id;
4319 	/*
4320 	 * The copy Flows are not included in any list. There
4321 	 * ones are referenced from other Flows and can not
4322 	 * be applied, removed, deleted in ardbitrary order
4323 	 * by list traversing.
4324 	 */
4325 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4326 					&attr, items, actions, false, error);
4327 	if (!mcp_res->rix_flow) {
4328 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4329 		return NULL;
4330 	}
4331 	return &mcp_res->hlist_ent;
4332 }
4333 
4334 struct mlx5_list_entry *
4335 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4336 		      void *cb_ctx __rte_unused)
4337 {
4338 	struct rte_eth_dev *dev = tool_ctx;
4339 	struct mlx5_priv *priv = dev->data->dev_private;
4340 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4341 	uint32_t idx = 0;
4342 
4343 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4344 	if (!mcp_res) {
4345 		rte_errno = ENOMEM;
4346 		return NULL;
4347 	}
4348 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4349 	mcp_res->idx = idx;
4350 	return &mcp_res->hlist_ent;
4351 }
4352 
4353 void
4354 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4355 {
4356 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4357 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4358 	struct rte_eth_dev *dev = tool_ctx;
4359 	struct mlx5_priv *priv = dev->data->dev_private;
4360 
4361 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4362 }
4363 
4364 /**
4365  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4366  *
4367  * As mark_id is unique, if there's already a registered flow for the mark_id,
4368  * return by increasing the reference counter of the resource. Otherwise, create
4369  * the resource (mcp_res) and flow.
4370  *
4371  * Flow looks like,
4372  *   - If ingress port is ANY and reg_c[1] is mark_id,
4373  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4374  *
4375  * For default flow (zero mark_id), flow is like,
4376  *   - If ingress port is ANY,
4377  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4378  *
4379  * @param dev
4380  *   Pointer to Ethernet device.
4381  * @param mark_id
4382  *   ID of MARK action, zero means default flow for META.
4383  * @param[out] error
4384  *   Perform verbose error reporting if not NULL.
4385  *
4386  * @return
4387  *   Associated resource on success, NULL otherwise and rte_errno is set.
4388  */
4389 static struct mlx5_flow_mreg_copy_resource *
4390 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4391 			  struct rte_flow_error *error)
4392 {
4393 	struct mlx5_priv *priv = dev->data->dev_private;
4394 	struct mlx5_list_entry *entry;
4395 	struct mlx5_flow_cb_ctx ctx = {
4396 		.dev = dev,
4397 		.error = error,
4398 		.data = &mark_id,
4399 	};
4400 
4401 	/* Check if already registered. */
4402 	MLX5_ASSERT(priv->mreg_cp_tbl);
4403 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4404 	if (!entry)
4405 		return NULL;
4406 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4407 			    hlist_ent);
4408 }
4409 
4410 void
4411 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4412 {
4413 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4414 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4415 	struct rte_eth_dev *dev = tool_ctx;
4416 	struct mlx5_priv *priv = dev->data->dev_private;
4417 
4418 	MLX5_ASSERT(mcp_res->rix_flow);
4419 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4420 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4421 }
4422 
4423 /**
4424  * Release flow in RX_CP_TBL.
4425  *
4426  * @param dev
4427  *   Pointer to Ethernet device.
4428  * @flow
4429  *   Parent flow for wich copying is provided.
4430  */
4431 static void
4432 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4433 			  struct rte_flow *flow)
4434 {
4435 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4436 	struct mlx5_priv *priv = dev->data->dev_private;
4437 
4438 	if (!flow->rix_mreg_copy)
4439 		return;
4440 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4441 				 flow->rix_mreg_copy);
4442 	if (!mcp_res || !priv->mreg_cp_tbl)
4443 		return;
4444 	MLX5_ASSERT(mcp_res->rix_flow);
4445 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4446 	flow->rix_mreg_copy = 0;
4447 }
4448 
4449 /**
4450  * Remove the default copy action from RX_CP_TBL.
4451  *
4452  * This functions is called in the mlx5_dev_start(). No thread safe
4453  * is guaranteed.
4454  *
4455  * @param dev
4456  *   Pointer to Ethernet device.
4457  */
4458 static void
4459 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4460 {
4461 	struct mlx5_list_entry *entry;
4462 	struct mlx5_priv *priv = dev->data->dev_private;
4463 	struct mlx5_flow_cb_ctx ctx;
4464 	uint32_t mark_id;
4465 
4466 	/* Check if default flow is registered. */
4467 	if (!priv->mreg_cp_tbl)
4468 		return;
4469 	mark_id = MLX5_DEFAULT_COPY_ID;
4470 	ctx.data = &mark_id;
4471 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4472 	if (!entry)
4473 		return;
4474 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4475 }
4476 
4477 /**
4478  * Add the default copy action in in RX_CP_TBL.
4479  *
4480  * This functions is called in the mlx5_dev_start(). No thread safe
4481  * is guaranteed.
4482  *
4483  * @param dev
4484  *   Pointer to Ethernet device.
4485  * @param[out] error
4486  *   Perform verbose error reporting if not NULL.
4487  *
4488  * @return
4489  *   0 for success, negative value otherwise and rte_errno is set.
4490  */
4491 static int
4492 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4493 				  struct rte_flow_error *error)
4494 {
4495 	struct mlx5_priv *priv = dev->data->dev_private;
4496 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4497 	struct mlx5_flow_cb_ctx ctx;
4498 	uint32_t mark_id;
4499 
4500 	/* Check whether extensive metadata feature is engaged. */
4501 	if (!priv->config.dv_flow_en ||
4502 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4503 	    !mlx5_flow_ext_mreg_supported(dev) ||
4504 	    !priv->sh->dv_regc0_mask)
4505 		return 0;
4506 	/*
4507 	 * Add default mreg copy flow may be called multiple time, but
4508 	 * only be called once in stop. Avoid register it twice.
4509 	 */
4510 	mark_id = MLX5_DEFAULT_COPY_ID;
4511 	ctx.data = &mark_id;
4512 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4513 		return 0;
4514 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4515 	if (!mcp_res)
4516 		return -rte_errno;
4517 	return 0;
4518 }
4519 
4520 /**
4521  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4522  *
4523  * All the flow having Q/RSS action should be split by
4524  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4525  * performs the following,
4526  *   - CQE->flow_tag := reg_c[1] (MARK)
4527  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4528  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4529  * but there should be a flow per each MARK ID set by MARK action.
4530  *
4531  * For the aforementioned reason, if there's a MARK action in flow's action
4532  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4533  * the MARK ID to CQE's flow_tag like,
4534  *   - If reg_c[1] is mark_id,
4535  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4536  *
4537  * For SET_META action which stores value in reg_c[0], as the destination is
4538  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4539  * MARK ID means the default flow. The default flow looks like,
4540  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4541  *
4542  * @param dev
4543  *   Pointer to Ethernet device.
4544  * @param flow
4545  *   Pointer to flow structure.
4546  * @param[in] actions
4547  *   Pointer to the list of actions.
4548  * @param[out] error
4549  *   Perform verbose error reporting if not NULL.
4550  *
4551  * @return
4552  *   0 on success, negative value otherwise and rte_errno is set.
4553  */
4554 static int
4555 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4556 			    struct rte_flow *flow,
4557 			    const struct rte_flow_action *actions,
4558 			    struct rte_flow_error *error)
4559 {
4560 	struct mlx5_priv *priv = dev->data->dev_private;
4561 	struct mlx5_dev_config *config = &priv->config;
4562 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4563 	const struct rte_flow_action_mark *mark;
4564 
4565 	/* Check whether extensive metadata feature is engaged. */
4566 	if (!config->dv_flow_en ||
4567 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4568 	    !mlx5_flow_ext_mreg_supported(dev) ||
4569 	    !priv->sh->dv_regc0_mask)
4570 		return 0;
4571 	/* Find MARK action. */
4572 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4573 		switch (actions->type) {
4574 		case RTE_FLOW_ACTION_TYPE_FLAG:
4575 			mcp_res = flow_mreg_add_copy_action
4576 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4577 			if (!mcp_res)
4578 				return -rte_errno;
4579 			flow->rix_mreg_copy = mcp_res->idx;
4580 			return 0;
4581 		case RTE_FLOW_ACTION_TYPE_MARK:
4582 			mark = (const struct rte_flow_action_mark *)
4583 				actions->conf;
4584 			mcp_res =
4585 				flow_mreg_add_copy_action(dev, mark->id, error);
4586 			if (!mcp_res)
4587 				return -rte_errno;
4588 			flow->rix_mreg_copy = mcp_res->idx;
4589 			return 0;
4590 		default:
4591 			break;
4592 		}
4593 	}
4594 	return 0;
4595 }
4596 
4597 #define MLX5_MAX_SPLIT_ACTIONS 24
4598 #define MLX5_MAX_SPLIT_ITEMS 24
4599 
4600 /**
4601  * Split the hairpin flow.
4602  * Since HW can't support encap and push-vlan on Rx, we move these
4603  * actions to Tx.
4604  * If the count action is after the encap then we also
4605  * move the count action. in this case the count will also measure
4606  * the outer bytes.
4607  *
4608  * @param dev
4609  *   Pointer to Ethernet device.
4610  * @param[in] actions
4611  *   Associated actions (list terminated by the END action).
4612  * @param[out] actions_rx
4613  *   Rx flow actions.
4614  * @param[out] actions_tx
4615  *   Tx flow actions..
4616  * @param[out] pattern_tx
4617  *   The pattern items for the Tx flow.
4618  * @param[out] flow_id
4619  *   The flow ID connected to this flow.
4620  *
4621  * @return
4622  *   0 on success.
4623  */
4624 static int
4625 flow_hairpin_split(struct rte_eth_dev *dev,
4626 		   const struct rte_flow_action actions[],
4627 		   struct rte_flow_action actions_rx[],
4628 		   struct rte_flow_action actions_tx[],
4629 		   struct rte_flow_item pattern_tx[],
4630 		   uint32_t flow_id)
4631 {
4632 	const struct rte_flow_action_raw_encap *raw_encap;
4633 	const struct rte_flow_action_raw_decap *raw_decap;
4634 	struct mlx5_rte_flow_action_set_tag *set_tag;
4635 	struct rte_flow_action *tag_action;
4636 	struct mlx5_rte_flow_item_tag *tag_item;
4637 	struct rte_flow_item *item;
4638 	char *addr;
4639 	int encap = 0;
4640 
4641 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4642 		switch (actions->type) {
4643 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4644 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4645 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4646 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4647 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4648 			rte_memcpy(actions_tx, actions,
4649 			       sizeof(struct rte_flow_action));
4650 			actions_tx++;
4651 			break;
4652 		case RTE_FLOW_ACTION_TYPE_COUNT:
4653 			if (encap) {
4654 				rte_memcpy(actions_tx, actions,
4655 					   sizeof(struct rte_flow_action));
4656 				actions_tx++;
4657 			} else {
4658 				rte_memcpy(actions_rx, actions,
4659 					   sizeof(struct rte_flow_action));
4660 				actions_rx++;
4661 			}
4662 			break;
4663 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4664 			raw_encap = actions->conf;
4665 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4666 				memcpy(actions_tx, actions,
4667 				       sizeof(struct rte_flow_action));
4668 				actions_tx++;
4669 				encap = 1;
4670 			} else {
4671 				rte_memcpy(actions_rx, actions,
4672 					   sizeof(struct rte_flow_action));
4673 				actions_rx++;
4674 			}
4675 			break;
4676 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4677 			raw_decap = actions->conf;
4678 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4679 				memcpy(actions_tx, actions,
4680 				       sizeof(struct rte_flow_action));
4681 				actions_tx++;
4682 			} else {
4683 				rte_memcpy(actions_rx, actions,
4684 					   sizeof(struct rte_flow_action));
4685 				actions_rx++;
4686 			}
4687 			break;
4688 		default:
4689 			rte_memcpy(actions_rx, actions,
4690 				   sizeof(struct rte_flow_action));
4691 			actions_rx++;
4692 			break;
4693 		}
4694 	}
4695 	/* Add set meta action and end action for the Rx flow. */
4696 	tag_action = actions_rx;
4697 	tag_action->type = (enum rte_flow_action_type)
4698 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4699 	actions_rx++;
4700 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4701 	actions_rx++;
4702 	set_tag = (void *)actions_rx;
4703 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4704 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4705 		.data = flow_id,
4706 	};
4707 	MLX5_ASSERT(set_tag->id > REG_NON);
4708 	tag_action->conf = set_tag;
4709 	/* Create Tx item list. */
4710 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4711 	addr = (void *)&pattern_tx[2];
4712 	item = pattern_tx;
4713 	item->type = (enum rte_flow_item_type)
4714 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4715 	tag_item = (void *)addr;
4716 	tag_item->data = flow_id;
4717 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4718 	MLX5_ASSERT(set_tag->id > REG_NON);
4719 	item->spec = tag_item;
4720 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4721 	tag_item = (void *)addr;
4722 	tag_item->data = UINT32_MAX;
4723 	tag_item->id = UINT16_MAX;
4724 	item->mask = tag_item;
4725 	item->last = NULL;
4726 	item++;
4727 	item->type = RTE_FLOW_ITEM_TYPE_END;
4728 	return 0;
4729 }
4730 
4731 /**
4732  * The last stage of splitting chain, just creates the subflow
4733  * without any modification.
4734  *
4735  * @param[in] dev
4736  *   Pointer to Ethernet device.
4737  * @param[in] flow
4738  *   Parent flow structure pointer.
4739  * @param[in, out] sub_flow
4740  *   Pointer to return the created subflow, may be NULL.
4741  * @param[in] attr
4742  *   Flow rule attributes.
4743  * @param[in] items
4744  *   Pattern specification (list terminated by the END pattern item).
4745  * @param[in] actions
4746  *   Associated actions (list terminated by the END action).
4747  * @param[in] flow_split_info
4748  *   Pointer to flow split info structure.
4749  * @param[out] error
4750  *   Perform verbose error reporting if not NULL.
4751  * @return
4752  *   0 on success, negative value otherwise
4753  */
4754 static int
4755 flow_create_split_inner(struct rte_eth_dev *dev,
4756 			struct rte_flow *flow,
4757 			struct mlx5_flow **sub_flow,
4758 			const struct rte_flow_attr *attr,
4759 			const struct rte_flow_item items[],
4760 			const struct rte_flow_action actions[],
4761 			struct mlx5_flow_split_info *flow_split_info,
4762 			struct rte_flow_error *error)
4763 {
4764 	struct mlx5_flow *dev_flow;
4765 
4766 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4767 				    flow_split_info->flow_idx, error);
4768 	if (!dev_flow)
4769 		return -rte_errno;
4770 	dev_flow->flow = flow;
4771 	dev_flow->external = flow_split_info->external;
4772 	dev_flow->skip_scale = flow_split_info->skip_scale;
4773 	/* Subflow object was created, we must include one in the list. */
4774 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4775 		      dev_flow->handle, next);
4776 	/*
4777 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4778 	 * flow may need some user defined item layer flags, and pass the
4779 	 * Metadate rxq mark flag to suffix flow as well.
4780 	 */
4781 	if (flow_split_info->prefix_layers)
4782 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4783 	if (flow_split_info->prefix_mark)
4784 		dev_flow->handle->mark = 1;
4785 	if (sub_flow)
4786 		*sub_flow = dev_flow;
4787 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4788 	dev_flow->dv.table_id = flow_split_info->table_id;
4789 #endif
4790 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4791 }
4792 
4793 /**
4794  * Get the sub policy of a meter.
4795  *
4796  * @param[in] dev
4797  *   Pointer to Ethernet device.
4798  * @param[in] flow
4799  *   Parent flow structure pointer.
4800  * @param wks
4801  *   Pointer to thread flow work space.
4802  * @param[in] attr
4803  *   Flow rule attributes.
4804  * @param[in] items
4805  *   Pattern specification (list terminated by the END pattern item).
4806  * @param[out] error
4807  *   Perform verbose error reporting if not NULL.
4808  *
4809  * @return
4810  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4811  */
4812 static struct mlx5_flow_meter_sub_policy *
4813 get_meter_sub_policy(struct rte_eth_dev *dev,
4814 		     struct rte_flow *flow,
4815 		     struct mlx5_flow_workspace *wks,
4816 		     const struct rte_flow_attr *attr,
4817 		     const struct rte_flow_item items[],
4818 		     struct rte_flow_error *error)
4819 {
4820 	struct mlx5_flow_meter_policy *policy;
4821 	struct mlx5_flow_meter_policy *final_policy;
4822 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4823 
4824 	policy = wks->policy;
4825 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4826 	if (final_policy->is_rss || final_policy->is_queue) {
4827 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4828 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4829 		uint32_t i;
4830 
4831 		/*
4832 		 * This is a tmp dev_flow,
4833 		 * no need to register any matcher for it in translate.
4834 		 */
4835 		wks->skip_matcher_reg = 1;
4836 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4837 			struct mlx5_flow dev_flow = {0};
4838 			struct mlx5_flow_handle dev_handle = { {0} };
4839 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4840 
4841 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4842 				const struct rte_flow_action_rss *rss_act =
4843 					final_policy->act_cnt[i].rss->conf;
4844 				struct rte_flow_action rss_actions[2] = {
4845 					[0] = {
4846 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4847 					.conf = rss_act,
4848 					},
4849 					[1] = {
4850 					.type = RTE_FLOW_ACTION_TYPE_END,
4851 					.conf = NULL,
4852 					}
4853 				};
4854 
4855 				dev_flow.handle = &dev_handle;
4856 				dev_flow.ingress = attr->ingress;
4857 				dev_flow.flow = flow;
4858 				dev_flow.external = 0;
4859 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4860 				dev_flow.dv.transfer = attr->transfer;
4861 #endif
4862 				/**
4863 				 * Translate RSS action to get rss hash fields.
4864 				 */
4865 				if (flow_drv_translate(dev, &dev_flow, attr,
4866 						items, rss_actions, error))
4867 					goto exit;
4868 				rss_desc_v[i] = wks->rss_desc;
4869 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4870 				rss_desc_v[i].hash_fields =
4871 						dev_flow.hash_fields;
4872 				rss_desc_v[i].queue_num =
4873 						rss_desc_v[i].hash_fields ?
4874 						rss_desc_v[i].queue_num : 1;
4875 				rss_desc_v[i].tunnel =
4876 						!!(dev_flow.handle->layers &
4877 						   MLX5_FLOW_LAYER_TUNNEL);
4878 				/* Use the RSS queues in the containers. */
4879 				rss_desc_v[i].queue =
4880 					(uint16_t *)(uintptr_t)rss_act->queue;
4881 				rss_desc[i] = &rss_desc_v[i];
4882 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4883 				/* This is queue action. */
4884 				rss_desc_v[i] = wks->rss_desc;
4885 				rss_desc_v[i].key_len = 0;
4886 				rss_desc_v[i].hash_fields = 0;
4887 				rss_desc_v[i].queue =
4888 					&final_policy->act_cnt[i].queue;
4889 				rss_desc_v[i].queue_num = 1;
4890 				rss_desc[i] = &rss_desc_v[i];
4891 			} else {
4892 				rss_desc[i] = NULL;
4893 			}
4894 		}
4895 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4896 						flow, policy, rss_desc);
4897 	} else {
4898 		enum mlx5_meter_domain mtr_domain =
4899 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4900 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4901 						MLX5_MTR_DOMAIN_INGRESS);
4902 		sub_policy = policy->sub_policys[mtr_domain][0];
4903 	}
4904 	if (!sub_policy)
4905 		rte_flow_error_set(error, EINVAL,
4906 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4907 				   "Failed to get meter sub-policy.");
4908 exit:
4909 	return sub_policy;
4910 }
4911 
4912 /**
4913  * Split the meter flow.
4914  *
4915  * As meter flow will split to three sub flow, other than meter
4916  * action, the other actions make sense to only meter accepts
4917  * the packet. If it need to be dropped, no other additional
4918  * actions should be take.
4919  *
4920  * One kind of special action which decapsulates the L3 tunnel
4921  * header will be in the prefix sub flow, as not to take the
4922  * L3 tunnel header into account.
4923  *
4924  * @param[in] dev
4925  *   Pointer to Ethernet device.
4926  * @param[in] flow
4927  *   Parent flow structure pointer.
4928  * @param wks
4929  *   Pointer to thread flow work space.
4930  * @param[in] attr
4931  *   Flow rule attributes.
4932  * @param[in] items
4933  *   Pattern specification (list terminated by the END pattern item).
4934  * @param[out] sfx_items
4935  *   Suffix flow match items (list terminated by the END pattern item).
4936  * @param[in] actions
4937  *   Associated actions (list terminated by the END action).
4938  * @param[out] actions_sfx
4939  *   Suffix flow actions.
4940  * @param[out] actions_pre
4941  *   Prefix flow actions.
4942  * @param[out] mtr_flow_id
4943  *   Pointer to meter flow id.
4944  * @param[out] error
4945  *   Perform verbose error reporting if not NULL.
4946  *
4947  * @return
4948  *   0 on success, a negative errno value otherwise and rte_errno is set.
4949  */
4950 static int
4951 flow_meter_split_prep(struct rte_eth_dev *dev,
4952 		      struct rte_flow *flow,
4953 		      struct mlx5_flow_workspace *wks,
4954 		      const struct rte_flow_attr *attr,
4955 		      const struct rte_flow_item items[],
4956 		      struct rte_flow_item sfx_items[],
4957 		      const struct rte_flow_action actions[],
4958 		      struct rte_flow_action actions_sfx[],
4959 		      struct rte_flow_action actions_pre[],
4960 		      uint32_t *mtr_flow_id,
4961 		      struct rte_flow_error *error)
4962 {
4963 	struct mlx5_priv *priv = dev->data->dev_private;
4964 	struct mlx5_flow_meter_info *fm = wks->fm;
4965 	struct rte_flow_action *tag_action = NULL;
4966 	struct rte_flow_item *tag_item;
4967 	struct mlx5_rte_flow_action_set_tag *set_tag;
4968 	const struct rte_flow_action_raw_encap *raw_encap;
4969 	const struct rte_flow_action_raw_decap *raw_decap;
4970 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4971 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4972 	uint32_t tag_id = 0;
4973 	struct rte_flow_item *vlan_item_dst = NULL;
4974 	const struct rte_flow_item *vlan_item_src = NULL;
4975 	struct rte_flow_action *hw_mtr_action;
4976 	struct rte_flow_action *action_pre_head = NULL;
4977 	int32_t flow_src_port = priv->representor_id;
4978 	bool mtr_first;
4979 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4980 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4981 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4982 	uint32_t flow_id = 0;
4983 	uint32_t flow_id_reversed = 0;
4984 	uint8_t flow_id_bits = 0;
4985 	int shift;
4986 
4987 	/* Prepare the suffix subflow items. */
4988 	tag_item = sfx_items++;
4989 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4990 		struct mlx5_priv *port_priv;
4991 		const struct rte_flow_item_port_id *pid_v;
4992 		int item_type = items->type;
4993 
4994 		switch (item_type) {
4995 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4996 			pid_v = items->spec;
4997 			MLX5_ASSERT(pid_v);
4998 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4999 			if (!port_priv)
5000 				return rte_flow_error_set(error,
5001 						rte_errno,
5002 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5003 						pid_v,
5004 						"Failed to get port info.");
5005 			flow_src_port = port_priv->representor_id;
5006 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5007 			    flow_src_port != priv->representor_id) {
5008 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5009 								flow, fm,
5010 								flow_src_port,
5011 								items,
5012 								error))
5013 					return -rte_errno;
5014 			}
5015 			memcpy(sfx_items, items, sizeof(*sfx_items));
5016 			sfx_items++;
5017 			break;
5018 		case RTE_FLOW_ITEM_TYPE_VLAN:
5019 			/* Determine if copy vlan item below. */
5020 			vlan_item_src = items;
5021 			vlan_item_dst = sfx_items++;
5022 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5023 			break;
5024 		default:
5025 			break;
5026 		}
5027 	}
5028 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5029 	sfx_items++;
5030 	mtr_first = priv->sh->meter_aso_en &&
5031 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5032 	/* For ASO meter, meter must be before tag in TX direction. */
5033 	if (mtr_first) {
5034 		action_pre_head = actions_pre++;
5035 		/* Leave space for tag action. */
5036 		tag_action = actions_pre++;
5037 	}
5038 	/* Prepare the actions for prefix and suffix flow. */
5039 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5040 		struct rte_flow_action *action_cur = NULL;
5041 
5042 		switch (actions->type) {
5043 		case RTE_FLOW_ACTION_TYPE_METER:
5044 			if (mtr_first) {
5045 				action_cur = action_pre_head;
5046 			} else {
5047 				/* Leave space for tag action. */
5048 				tag_action = actions_pre++;
5049 				action_cur = actions_pre++;
5050 			}
5051 			break;
5052 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5053 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5054 			action_cur = actions_pre++;
5055 			break;
5056 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5057 			raw_encap = actions->conf;
5058 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5059 				action_cur = actions_pre++;
5060 			break;
5061 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5062 			raw_decap = actions->conf;
5063 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5064 				action_cur = actions_pre++;
5065 			break;
5066 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5067 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5068 			if (vlan_item_dst && vlan_item_src) {
5069 				memcpy(vlan_item_dst, vlan_item_src,
5070 					sizeof(*vlan_item_dst));
5071 				/*
5072 				 * Convert to internal match item, it is used
5073 				 * for vlan push and set vid.
5074 				 */
5075 				vlan_item_dst->type = (enum rte_flow_item_type)
5076 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5077 			}
5078 			break;
5079 		default:
5080 			break;
5081 		}
5082 		if (!action_cur)
5083 			action_cur = (fm->def_policy) ?
5084 					actions_sfx++ : actions_pre++;
5085 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5086 	}
5087 	/* Add end action to the actions. */
5088 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5089 	if (priv->sh->meter_aso_en) {
5090 		/**
5091 		 * For ASO meter, need to add an extra jump action explicitly,
5092 		 * to jump from meter to policer table.
5093 		 */
5094 		struct mlx5_flow_meter_sub_policy *sub_policy;
5095 		struct mlx5_flow_tbl_data_entry *tbl_data;
5096 
5097 		if (!fm->def_policy) {
5098 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5099 							  attr, items, error);
5100 			if (!sub_policy)
5101 				return -rte_errno;
5102 		} else {
5103 			enum mlx5_meter_domain mtr_domain =
5104 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5105 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5106 						MLX5_MTR_DOMAIN_INGRESS);
5107 
5108 			sub_policy =
5109 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5110 		}
5111 		tbl_data = container_of(sub_policy->tbl_rsc,
5112 					struct mlx5_flow_tbl_data_entry, tbl);
5113 		hw_mtr_action = actions_pre++;
5114 		hw_mtr_action->type = (enum rte_flow_action_type)
5115 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5116 		hw_mtr_action->conf = tbl_data->jump.action;
5117 	}
5118 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5119 	actions_pre++;
5120 	if (!tag_action)
5121 		return rte_flow_error_set(error, ENOMEM,
5122 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5123 					  NULL, "No tag action space.");
5124 	if (!mtr_flow_id) {
5125 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5126 		goto exit;
5127 	}
5128 	/* Only default-policy Meter creates mtr flow id. */
5129 	if (fm->def_policy) {
5130 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5131 		if (!tag_id)
5132 			return rte_flow_error_set(error, ENOMEM,
5133 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5134 					"Failed to allocate meter flow id.");
5135 		flow_id = tag_id - 1;
5136 		flow_id_bits = (!flow_id) ? 1 :
5137 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5138 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5139 		    mtr_reg_bits) {
5140 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5141 			return rte_flow_error_set(error, EINVAL,
5142 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5143 					"Meter flow id exceeds max limit.");
5144 		}
5145 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5146 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5147 	}
5148 	/* Build tag actions and items for meter_id/meter flow_id. */
5149 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5150 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5151 	tag_item_mask = tag_item_spec + 1;
5152 	/* Both flow_id and meter_id share the same register. */
5153 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5154 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5155 							    0, error),
5156 		.offset = mtr_id_offset,
5157 		.length = mtr_reg_bits,
5158 		.data = flow->meter,
5159 	};
5160 	/*
5161 	 * The color Reg bits used by flow_id are growing from
5162 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5163 	 */
5164 	for (shift = 0; shift < flow_id_bits; shift++)
5165 		flow_id_reversed = (flow_id_reversed << 1) |
5166 				((flow_id >> shift) & 0x1);
5167 	set_tag->data |=
5168 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5169 	tag_item_spec->id = set_tag->id;
5170 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5171 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5172 	tag_action->type = (enum rte_flow_action_type)
5173 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5174 	tag_action->conf = set_tag;
5175 	tag_item->type = (enum rte_flow_item_type)
5176 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5177 	tag_item->spec = tag_item_spec;
5178 	tag_item->last = NULL;
5179 	tag_item->mask = tag_item_mask;
5180 exit:
5181 	if (mtr_flow_id)
5182 		*mtr_flow_id = tag_id;
5183 	return 0;
5184 }
5185 
5186 /**
5187  * Split action list having QUEUE/RSS for metadata register copy.
5188  *
5189  * Once Q/RSS action is detected in user's action list, the flow action
5190  * should be split in order to copy metadata registers, which will happen in
5191  * RX_CP_TBL like,
5192  *   - CQE->flow_tag := reg_c[1] (MARK)
5193  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5194  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5195  * This is because the last action of each flow must be a terminal action
5196  * (QUEUE, RSS or DROP).
5197  *
5198  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5199  * stored and kept in the mlx5_flow structure per each sub_flow.
5200  *
5201  * The Q/RSS action is replaced with,
5202  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5203  * And the following JUMP action is added at the end,
5204  *   - JUMP, to RX_CP_TBL.
5205  *
5206  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5207  * flow_create_split_metadata() routine. The flow will look like,
5208  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5209  *
5210  * @param dev
5211  *   Pointer to Ethernet device.
5212  * @param[out] split_actions
5213  *   Pointer to store split actions to jump to CP_TBL.
5214  * @param[in] actions
5215  *   Pointer to the list of original flow actions.
5216  * @param[in] qrss
5217  *   Pointer to the Q/RSS action.
5218  * @param[in] actions_n
5219  *   Number of original actions.
5220  * @param[out] error
5221  *   Perform verbose error reporting if not NULL.
5222  *
5223  * @return
5224  *   non-zero unique flow_id on success, otherwise 0 and
5225  *   error/rte_error are set.
5226  */
5227 static uint32_t
5228 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5229 			  struct rte_flow_action *split_actions,
5230 			  const struct rte_flow_action *actions,
5231 			  const struct rte_flow_action *qrss,
5232 			  int actions_n, struct rte_flow_error *error)
5233 {
5234 	struct mlx5_priv *priv = dev->data->dev_private;
5235 	struct mlx5_rte_flow_action_set_tag *set_tag;
5236 	struct rte_flow_action_jump *jump;
5237 	const int qrss_idx = qrss - actions;
5238 	uint32_t flow_id = 0;
5239 	int ret = 0;
5240 
5241 	/*
5242 	 * Given actions will be split
5243 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5244 	 * - Add jump to mreg CP_TBL.
5245 	 * As a result, there will be one more action.
5246 	 */
5247 	++actions_n;
5248 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5249 	set_tag = (void *)(split_actions + actions_n);
5250 	/*
5251 	 * If tag action is not set to void(it means we are not the meter
5252 	 * suffix flow), add the tag action. Since meter suffix flow already
5253 	 * has the tag added.
5254 	 */
5255 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5256 		/*
5257 		 * Allocate the new subflow ID. This one is unique within
5258 		 * device and not shared with representors. Otherwise,
5259 		 * we would have to resolve multi-thread access synch
5260 		 * issue. Each flow on the shared device is appended
5261 		 * with source vport identifier, so the resulting
5262 		 * flows will be unique in the shared (by master and
5263 		 * representors) domain even if they have coinciding
5264 		 * IDs.
5265 		 */
5266 		mlx5_ipool_malloc(priv->sh->ipool
5267 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5268 		if (!flow_id)
5269 			return rte_flow_error_set(error, ENOMEM,
5270 						  RTE_FLOW_ERROR_TYPE_ACTION,
5271 						  NULL, "can't allocate id "
5272 						  "for split Q/RSS subflow");
5273 		/* Internal SET_TAG action to set flow ID. */
5274 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5275 			.data = flow_id,
5276 		};
5277 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5278 		if (ret < 0)
5279 			return ret;
5280 		set_tag->id = ret;
5281 		/* Construct new actions array. */
5282 		/* Replace QUEUE/RSS action. */
5283 		split_actions[qrss_idx] = (struct rte_flow_action){
5284 			.type = (enum rte_flow_action_type)
5285 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5286 			.conf = set_tag,
5287 		};
5288 	}
5289 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5290 	jump = (void *)(set_tag + 1);
5291 	*jump = (struct rte_flow_action_jump){
5292 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5293 	};
5294 	split_actions[actions_n - 2] = (struct rte_flow_action){
5295 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5296 		.conf = jump,
5297 	};
5298 	split_actions[actions_n - 1] = (struct rte_flow_action){
5299 		.type = RTE_FLOW_ACTION_TYPE_END,
5300 	};
5301 	return flow_id;
5302 }
5303 
5304 /**
5305  * Extend the given action list for Tx metadata copy.
5306  *
5307  * Copy the given action list to the ext_actions and add flow metadata register
5308  * copy action in order to copy reg_a set by WQE to reg_c[0].
5309  *
5310  * @param[out] ext_actions
5311  *   Pointer to the extended action list.
5312  * @param[in] actions
5313  *   Pointer to the list of actions.
5314  * @param[in] actions_n
5315  *   Number of actions in the list.
5316  * @param[out] error
5317  *   Perform verbose error reporting if not NULL.
5318  * @param[in] encap_idx
5319  *   The encap action inndex.
5320  *
5321  * @return
5322  *   0 on success, negative value otherwise
5323  */
5324 static int
5325 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5326 		       struct rte_flow_action *ext_actions,
5327 		       const struct rte_flow_action *actions,
5328 		       int actions_n, struct rte_flow_error *error,
5329 		       int encap_idx)
5330 {
5331 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5332 		(struct mlx5_flow_action_copy_mreg *)
5333 			(ext_actions + actions_n + 1);
5334 	int ret;
5335 
5336 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5337 	if (ret < 0)
5338 		return ret;
5339 	cp_mreg->dst = ret;
5340 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5341 	if (ret < 0)
5342 		return ret;
5343 	cp_mreg->src = ret;
5344 	if (encap_idx != 0)
5345 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5346 	if (encap_idx == actions_n - 1) {
5347 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5348 			.type = (enum rte_flow_action_type)
5349 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5350 			.conf = cp_mreg,
5351 		};
5352 		ext_actions[actions_n] = (struct rte_flow_action){
5353 			.type = RTE_FLOW_ACTION_TYPE_END,
5354 		};
5355 	} else {
5356 		ext_actions[encap_idx] = (struct rte_flow_action){
5357 			.type = (enum rte_flow_action_type)
5358 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5359 			.conf = cp_mreg,
5360 		};
5361 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5362 				sizeof(*ext_actions) * (actions_n - encap_idx));
5363 	}
5364 	return 0;
5365 }
5366 
5367 /**
5368  * Check the match action from the action list.
5369  *
5370  * @param[in] actions
5371  *   Pointer to the list of actions.
5372  * @param[in] attr
5373  *   Flow rule attributes.
5374  * @param[in] action
5375  *   The action to be check if exist.
5376  * @param[out] match_action_pos
5377  *   Pointer to the position of the matched action if exists, otherwise is -1.
5378  * @param[out] qrss_action_pos
5379  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5380  * @param[out] modify_after_mirror
5381  *   Pointer to the flag of modify action after FDB mirroring.
5382  *
5383  * @return
5384  *   > 0 the total number of actions.
5385  *   0 if not found match action in action list.
5386  */
5387 static int
5388 flow_check_match_action(const struct rte_flow_action actions[],
5389 			const struct rte_flow_attr *attr,
5390 			enum rte_flow_action_type action,
5391 			int *match_action_pos, int *qrss_action_pos,
5392 			int *modify_after_mirror)
5393 {
5394 	const struct rte_flow_action_sample *sample;
5395 	const struct rte_flow_action_raw_decap *decap;
5396 	int actions_n = 0;
5397 	uint32_t ratio = 0;
5398 	int sub_type = 0;
5399 	int flag = 0;
5400 	int fdb_mirror = 0;
5401 
5402 	*match_action_pos = -1;
5403 	*qrss_action_pos = -1;
5404 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5405 		if (actions->type == action) {
5406 			flag = 1;
5407 			*match_action_pos = actions_n;
5408 		}
5409 		switch (actions->type) {
5410 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5411 		case RTE_FLOW_ACTION_TYPE_RSS:
5412 			*qrss_action_pos = actions_n;
5413 			break;
5414 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5415 			sample = actions->conf;
5416 			ratio = sample->ratio;
5417 			sub_type = ((const struct rte_flow_action *)
5418 					(sample->actions))->type;
5419 			if (ratio == 1 && attr->transfer)
5420 				fdb_mirror = 1;
5421 			break;
5422 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5423 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5424 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5425 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5426 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5427 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5428 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5429 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5430 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5431 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5432 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5433 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5434 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5435 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5436 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5437 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5438 		case RTE_FLOW_ACTION_TYPE_FLAG:
5439 		case RTE_FLOW_ACTION_TYPE_MARK:
5440 		case RTE_FLOW_ACTION_TYPE_SET_META:
5441 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5442 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5443 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5444 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5445 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5446 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5447 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5448 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5449 		case RTE_FLOW_ACTION_TYPE_METER:
5450 			if (fdb_mirror)
5451 				*modify_after_mirror = 1;
5452 			break;
5453 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5454 			decap = actions->conf;
5455 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5456 				;
5457 			actions_n++;
5458 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5459 				const struct rte_flow_action_raw_encap *encap =
5460 								actions->conf;
5461 				if (decap->size <=
5462 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5463 				    encap->size >
5464 					MLX5_ENCAPSULATION_DECISION_SIZE)
5465 					/* L3 encap. */
5466 					break;
5467 			}
5468 			if (fdb_mirror)
5469 				*modify_after_mirror = 1;
5470 			break;
5471 		default:
5472 			break;
5473 		}
5474 		actions_n++;
5475 	}
5476 	if (flag && fdb_mirror && !*modify_after_mirror) {
5477 		/* FDB mirroring uses the destination array to implement
5478 		 * instead of FLOW_SAMPLER object.
5479 		 */
5480 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5481 			flag = 0;
5482 	}
5483 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5484 	return flag ? actions_n + 1 : 0;
5485 }
5486 
5487 #define SAMPLE_SUFFIX_ITEM 2
5488 
5489 /**
5490  * Split the sample flow.
5491  *
5492  * As sample flow will split to two sub flow, sample flow with
5493  * sample action, the other actions will move to new suffix flow.
5494  *
5495  * Also add unique tag id with tag action in the sample flow,
5496  * the same tag id will be as match in the suffix flow.
5497  *
5498  * @param dev
5499  *   Pointer to Ethernet device.
5500  * @param[in] add_tag
5501  *   Add extra tag action flag.
5502  * @param[out] sfx_items
5503  *   Suffix flow match items (list terminated by the END pattern item).
5504  * @param[in] actions
5505  *   Associated actions (list terminated by the END action).
5506  * @param[out] actions_sfx
5507  *   Suffix flow actions.
5508  * @param[out] actions_pre
5509  *   Prefix flow actions.
5510  * @param[in] actions_n
5511  *  The total number of actions.
5512  * @param[in] sample_action_pos
5513  *   The sample action position.
5514  * @param[in] qrss_action_pos
5515  *   The Queue/RSS action position.
5516  * @param[in] jump_table
5517  *   Add extra jump action flag.
5518  * @param[out] error
5519  *   Perform verbose error reporting if not NULL.
5520  *
5521  * @return
5522  *   0 on success, or unique flow_id, a negative errno value
5523  *   otherwise and rte_errno is set.
5524  */
5525 static int
5526 flow_sample_split_prep(struct rte_eth_dev *dev,
5527 		       int add_tag,
5528 		       struct rte_flow_item sfx_items[],
5529 		       const struct rte_flow_action actions[],
5530 		       struct rte_flow_action actions_sfx[],
5531 		       struct rte_flow_action actions_pre[],
5532 		       int actions_n,
5533 		       int sample_action_pos,
5534 		       int qrss_action_pos,
5535 		       int jump_table,
5536 		       struct rte_flow_error *error)
5537 {
5538 	struct mlx5_priv *priv = dev->data->dev_private;
5539 	struct mlx5_rte_flow_action_set_tag *set_tag;
5540 	struct mlx5_rte_flow_item_tag *tag_spec;
5541 	struct mlx5_rte_flow_item_tag *tag_mask;
5542 	struct rte_flow_action_jump *jump_action;
5543 	uint32_t tag_id = 0;
5544 	int index;
5545 	int append_index = 0;
5546 	int ret;
5547 
5548 	if (sample_action_pos < 0)
5549 		return rte_flow_error_set(error, EINVAL,
5550 					  RTE_FLOW_ERROR_TYPE_ACTION,
5551 					  NULL, "invalid position of sample "
5552 					  "action in list");
5553 	/* Prepare the actions for prefix and suffix flow. */
5554 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5555 		index = qrss_action_pos;
5556 		/* Put the preceding the Queue/RSS action into prefix flow. */
5557 		if (index != 0)
5558 			memcpy(actions_pre, actions,
5559 			       sizeof(struct rte_flow_action) * index);
5560 		/* Put others preceding the sample action into prefix flow. */
5561 		if (sample_action_pos > index + 1)
5562 			memcpy(actions_pre + index, actions + index + 1,
5563 			       sizeof(struct rte_flow_action) *
5564 			       (sample_action_pos - index - 1));
5565 		index = sample_action_pos - 1;
5566 		/* Put Queue/RSS action into Suffix flow. */
5567 		memcpy(actions_sfx, actions + qrss_action_pos,
5568 		       sizeof(struct rte_flow_action));
5569 		actions_sfx++;
5570 	} else {
5571 		index = sample_action_pos;
5572 		if (index != 0)
5573 			memcpy(actions_pre, actions,
5574 			       sizeof(struct rte_flow_action) * index);
5575 	}
5576 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5577 	 * For CX6DX and above, metadata registers Cx preserve their value,
5578 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5579 	 */
5580 	if (add_tag) {
5581 		/* Prepare the prefix tag action. */
5582 		append_index++;
5583 		set_tag = (void *)(actions_pre + actions_n + append_index);
5584 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5585 		if (ret < 0)
5586 			return ret;
5587 		mlx5_ipool_malloc(priv->sh->ipool
5588 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5589 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5590 			.id = ret,
5591 			.data = tag_id,
5592 		};
5593 		/* Prepare the suffix subflow items. */
5594 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5595 		tag_spec->data = tag_id;
5596 		tag_spec->id = set_tag->id;
5597 		tag_mask = tag_spec + 1;
5598 		tag_mask->data = UINT32_MAX;
5599 		sfx_items[0] = (struct rte_flow_item){
5600 			.type = (enum rte_flow_item_type)
5601 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5602 			.spec = tag_spec,
5603 			.last = NULL,
5604 			.mask = tag_mask,
5605 		};
5606 		sfx_items[1] = (struct rte_flow_item){
5607 			.type = (enum rte_flow_item_type)
5608 				RTE_FLOW_ITEM_TYPE_END,
5609 		};
5610 		/* Prepare the tag action in prefix subflow. */
5611 		actions_pre[index++] =
5612 			(struct rte_flow_action){
5613 			.type = (enum rte_flow_action_type)
5614 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5615 			.conf = set_tag,
5616 		};
5617 	}
5618 	memcpy(actions_pre + index, actions + sample_action_pos,
5619 	       sizeof(struct rte_flow_action));
5620 	index += 1;
5621 	/* For the modify action after the sample action in E-Switch mirroring,
5622 	 * Add the extra jump action in prefix subflow and jump into the next
5623 	 * table, then do the modify action in the new table.
5624 	 */
5625 	if (jump_table) {
5626 		/* Prepare the prefix jump action. */
5627 		append_index++;
5628 		jump_action = (void *)(actions_pre + actions_n + append_index);
5629 		jump_action->group = jump_table;
5630 		actions_pre[index++] =
5631 			(struct rte_flow_action){
5632 			.type = (enum rte_flow_action_type)
5633 				RTE_FLOW_ACTION_TYPE_JUMP,
5634 			.conf = jump_action,
5635 		};
5636 	}
5637 	actions_pre[index] = (struct rte_flow_action){
5638 		.type = (enum rte_flow_action_type)
5639 			RTE_FLOW_ACTION_TYPE_END,
5640 	};
5641 	/* Put the actions after sample into Suffix flow. */
5642 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5643 	       sizeof(struct rte_flow_action) *
5644 	       (actions_n - sample_action_pos - 1));
5645 	return tag_id;
5646 }
5647 
5648 /**
5649  * The splitting for metadata feature.
5650  *
5651  * - Q/RSS action on NIC Rx should be split in order to pass by
5652  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5653  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5654  *
5655  * - All the actions on NIC Tx should have a mreg copy action to
5656  *   copy reg_a from WQE to reg_c[0].
5657  *
5658  * @param dev
5659  *   Pointer to Ethernet device.
5660  * @param[in] flow
5661  *   Parent flow structure pointer.
5662  * @param[in] attr
5663  *   Flow rule attributes.
5664  * @param[in] items
5665  *   Pattern specification (list terminated by the END pattern item).
5666  * @param[in] actions
5667  *   Associated actions (list terminated by the END action).
5668  * @param[in] flow_split_info
5669  *   Pointer to flow split info structure.
5670  * @param[out] error
5671  *   Perform verbose error reporting if not NULL.
5672  * @return
5673  *   0 on success, negative value otherwise
5674  */
5675 static int
5676 flow_create_split_metadata(struct rte_eth_dev *dev,
5677 			   struct rte_flow *flow,
5678 			   const struct rte_flow_attr *attr,
5679 			   const struct rte_flow_item items[],
5680 			   const struct rte_flow_action actions[],
5681 			   struct mlx5_flow_split_info *flow_split_info,
5682 			   struct rte_flow_error *error)
5683 {
5684 	struct mlx5_priv *priv = dev->data->dev_private;
5685 	struct mlx5_dev_config *config = &priv->config;
5686 	const struct rte_flow_action *qrss = NULL;
5687 	struct rte_flow_action *ext_actions = NULL;
5688 	struct mlx5_flow *dev_flow = NULL;
5689 	uint32_t qrss_id = 0;
5690 	int mtr_sfx = 0;
5691 	size_t act_size;
5692 	int actions_n;
5693 	int encap_idx;
5694 	int ret;
5695 
5696 	/* Check whether extensive metadata feature is engaged. */
5697 	if (!config->dv_flow_en ||
5698 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5699 	    !mlx5_flow_ext_mreg_supported(dev))
5700 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5701 					       actions, flow_split_info, error);
5702 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5703 							   &encap_idx);
5704 	if (qrss) {
5705 		/* Exclude hairpin flows from splitting. */
5706 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5707 			const struct rte_flow_action_queue *queue;
5708 
5709 			queue = qrss->conf;
5710 			if (mlx5_rxq_get_type(dev, queue->index) ==
5711 			    MLX5_RXQ_TYPE_HAIRPIN)
5712 				qrss = NULL;
5713 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5714 			const struct rte_flow_action_rss *rss;
5715 
5716 			rss = qrss->conf;
5717 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5718 			    MLX5_RXQ_TYPE_HAIRPIN)
5719 				qrss = NULL;
5720 		}
5721 	}
5722 	if (qrss) {
5723 		/* Check if it is in meter suffix table. */
5724 		mtr_sfx = attr->group == (attr->transfer ?
5725 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5726 			  MLX5_FLOW_TABLE_LEVEL_METER);
5727 		/*
5728 		 * Q/RSS action on NIC Rx should be split in order to pass by
5729 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5730 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5731 		 */
5732 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5733 			   sizeof(struct rte_flow_action_set_tag) +
5734 			   sizeof(struct rte_flow_action_jump);
5735 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5736 					  SOCKET_ID_ANY);
5737 		if (!ext_actions)
5738 			return rte_flow_error_set(error, ENOMEM,
5739 						  RTE_FLOW_ERROR_TYPE_ACTION,
5740 						  NULL, "no memory to split "
5741 						  "metadata flow");
5742 		/*
5743 		 * If we are the suffix flow of meter, tag already exist.
5744 		 * Set the tag action to void.
5745 		 */
5746 		if (mtr_sfx)
5747 			ext_actions[qrss - actions].type =
5748 						RTE_FLOW_ACTION_TYPE_VOID;
5749 		else
5750 			ext_actions[qrss - actions].type =
5751 						(enum rte_flow_action_type)
5752 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5753 		/*
5754 		 * Create the new actions list with removed Q/RSS action
5755 		 * and appended set tag and jump to register copy table
5756 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5757 		 * in advance, because it is needed for set tag action.
5758 		 */
5759 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5760 						    qrss, actions_n, error);
5761 		if (!mtr_sfx && !qrss_id) {
5762 			ret = -rte_errno;
5763 			goto exit;
5764 		}
5765 	} else if (attr->egress && !attr->transfer) {
5766 		/*
5767 		 * All the actions on NIC Tx should have a metadata register
5768 		 * copy action to copy reg_a from WQE to reg_c[meta]
5769 		 */
5770 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5771 			   sizeof(struct mlx5_flow_action_copy_mreg);
5772 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5773 					  SOCKET_ID_ANY);
5774 		if (!ext_actions)
5775 			return rte_flow_error_set(error, ENOMEM,
5776 						  RTE_FLOW_ERROR_TYPE_ACTION,
5777 						  NULL, "no memory to split "
5778 						  "metadata flow");
5779 		/* Create the action list appended with copy register. */
5780 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5781 					     actions_n, error, encap_idx);
5782 		if (ret < 0)
5783 			goto exit;
5784 	}
5785 	/* Add the unmodified original or prefix subflow. */
5786 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5787 				      items, ext_actions ? ext_actions :
5788 				      actions, flow_split_info, error);
5789 	if (ret < 0)
5790 		goto exit;
5791 	MLX5_ASSERT(dev_flow);
5792 	if (qrss) {
5793 		const struct rte_flow_attr q_attr = {
5794 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5795 			.ingress = 1,
5796 		};
5797 		/* Internal PMD action to set register. */
5798 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5799 			.data = qrss_id,
5800 			.id = REG_NON,
5801 		};
5802 		struct rte_flow_item q_items[] = {
5803 			{
5804 				.type = (enum rte_flow_item_type)
5805 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5806 				.spec = &q_tag_spec,
5807 				.last = NULL,
5808 				.mask = NULL,
5809 			},
5810 			{
5811 				.type = RTE_FLOW_ITEM_TYPE_END,
5812 			},
5813 		};
5814 		struct rte_flow_action q_actions[] = {
5815 			{
5816 				.type = qrss->type,
5817 				.conf = qrss->conf,
5818 			},
5819 			{
5820 				.type = RTE_FLOW_ACTION_TYPE_END,
5821 			},
5822 		};
5823 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5824 
5825 		/*
5826 		 * Configure the tag item only if there is no meter subflow.
5827 		 * Since tag is already marked in the meter suffix subflow
5828 		 * we can just use the meter suffix items as is.
5829 		 */
5830 		if (qrss_id) {
5831 			/* Not meter subflow. */
5832 			MLX5_ASSERT(!mtr_sfx);
5833 			/*
5834 			 * Put unique id in prefix flow due to it is destroyed
5835 			 * after suffix flow and id will be freed after there
5836 			 * is no actual flows with this id and identifier
5837 			 * reallocation becomes possible (for example, for
5838 			 * other flows in other threads).
5839 			 */
5840 			dev_flow->handle->split_flow_id = qrss_id;
5841 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5842 						   error);
5843 			if (ret < 0)
5844 				goto exit;
5845 			q_tag_spec.id = ret;
5846 		}
5847 		dev_flow = NULL;
5848 		/* Add suffix subflow to execute Q/RSS. */
5849 		flow_split_info->prefix_layers = layers;
5850 		flow_split_info->prefix_mark = 0;
5851 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5852 					      &q_attr, mtr_sfx ? items :
5853 					      q_items, q_actions,
5854 					      flow_split_info, error);
5855 		if (ret < 0)
5856 			goto exit;
5857 		/* qrss ID should be freed if failed. */
5858 		qrss_id = 0;
5859 		MLX5_ASSERT(dev_flow);
5860 	}
5861 
5862 exit:
5863 	/*
5864 	 * We do not destroy the partially created sub_flows in case of error.
5865 	 * These ones are included into parent flow list and will be destroyed
5866 	 * by flow_drv_destroy.
5867 	 */
5868 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5869 			qrss_id);
5870 	mlx5_free(ext_actions);
5871 	return ret;
5872 }
5873 
5874 /**
5875  * Create meter internal drop flow with the original pattern.
5876  *
5877  * @param dev
5878  *   Pointer to Ethernet device.
5879  * @param[in] flow
5880  *   Parent flow structure pointer.
5881  * @param[in] attr
5882  *   Flow rule attributes.
5883  * @param[in] items
5884  *   Pattern specification (list terminated by the END pattern item).
5885  * @param[in] flow_split_info
5886  *   Pointer to flow split info structure.
5887  * @param[in] fm
5888  *   Pointer to flow meter structure.
5889  * @param[out] error
5890  *   Perform verbose error reporting if not NULL.
5891  * @return
5892  *   0 on success, negative value otherwise
5893  */
5894 static uint32_t
5895 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5896 			struct rte_flow *flow,
5897 			const struct rte_flow_attr *attr,
5898 			const struct rte_flow_item items[],
5899 			struct mlx5_flow_split_info *flow_split_info,
5900 			struct mlx5_flow_meter_info *fm,
5901 			struct rte_flow_error *error)
5902 {
5903 	struct mlx5_flow *dev_flow = NULL;
5904 	struct rte_flow_attr drop_attr = *attr;
5905 	struct rte_flow_action drop_actions[3];
5906 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5907 
5908 	MLX5_ASSERT(fm->drop_cnt);
5909 	drop_actions[0].type =
5910 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5911 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5912 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5913 	drop_actions[1].conf = NULL;
5914 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5915 	drop_actions[2].conf = NULL;
5916 	drop_split_info.external = false;
5917 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5918 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5919 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5920 	return flow_create_split_inner(dev, flow, &dev_flow,
5921 				&drop_attr, items, drop_actions,
5922 				&drop_split_info, error);
5923 }
5924 
5925 /**
5926  * The splitting for meter feature.
5927  *
5928  * - The meter flow will be split to two flows as prefix and
5929  *   suffix flow. The packets make sense only it pass the prefix
5930  *   meter action.
5931  *
5932  * - Reg_C_5 is used for the packet to match betweend prefix and
5933  *   suffix flow.
5934  *
5935  * @param dev
5936  *   Pointer to Ethernet device.
5937  * @param[in] flow
5938  *   Parent flow structure pointer.
5939  * @param[in] attr
5940  *   Flow rule attributes.
5941  * @param[in] items
5942  *   Pattern specification (list terminated by the END pattern item).
5943  * @param[in] actions
5944  *   Associated actions (list terminated by the END action).
5945  * @param[in] flow_split_info
5946  *   Pointer to flow split info structure.
5947  * @param[out] error
5948  *   Perform verbose error reporting if not NULL.
5949  * @return
5950  *   0 on success, negative value otherwise
5951  */
5952 static int
5953 flow_create_split_meter(struct rte_eth_dev *dev,
5954 			struct rte_flow *flow,
5955 			const struct rte_flow_attr *attr,
5956 			const struct rte_flow_item items[],
5957 			const struct rte_flow_action actions[],
5958 			struct mlx5_flow_split_info *flow_split_info,
5959 			struct rte_flow_error *error)
5960 {
5961 	struct mlx5_priv *priv = dev->data->dev_private;
5962 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5963 	struct rte_flow_action *sfx_actions = NULL;
5964 	struct rte_flow_action *pre_actions = NULL;
5965 	struct rte_flow_item *sfx_items = NULL;
5966 	struct mlx5_flow *dev_flow = NULL;
5967 	struct rte_flow_attr sfx_attr = *attr;
5968 	struct mlx5_flow_meter_info *fm = NULL;
5969 	uint8_t skip_scale_restore;
5970 	bool has_mtr = false;
5971 	bool has_modify = false;
5972 	bool set_mtr_reg = true;
5973 	bool is_mtr_hierarchy = false;
5974 	uint32_t meter_id = 0;
5975 	uint32_t mtr_idx = 0;
5976 	uint32_t mtr_flow_id = 0;
5977 	size_t act_size;
5978 	size_t item_size;
5979 	int actions_n = 0;
5980 	int ret = 0;
5981 
5982 	if (priv->mtr_en)
5983 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5984 						    &has_modify, &meter_id);
5985 	if (has_mtr) {
5986 		if (flow->meter) {
5987 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5988 			if (!fm)
5989 				return rte_flow_error_set(error, EINVAL,
5990 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5991 						NULL, "Meter not found.");
5992 		} else {
5993 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5994 			if (!fm)
5995 				return rte_flow_error_set(error, EINVAL,
5996 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5997 						NULL, "Meter not found.");
5998 			ret = mlx5_flow_meter_attach(priv, fm,
5999 						     &sfx_attr, error);
6000 			if (ret)
6001 				return -rte_errno;
6002 			flow->meter = mtr_idx;
6003 		}
6004 		MLX5_ASSERT(wks);
6005 		wks->fm = fm;
6006 		if (!fm->def_policy) {
6007 			wks->policy = mlx5_flow_meter_policy_find(dev,
6008 								  fm->policy_id,
6009 								  NULL);
6010 			MLX5_ASSERT(wks->policy);
6011 			if (wks->policy->is_hierarchy) {
6012 				wks->final_policy =
6013 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6014 								wks->policy);
6015 				if (!wks->final_policy)
6016 					return rte_flow_error_set(error,
6017 					EINVAL,
6018 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6019 				"Failed to find terminal policy of hierarchy.");
6020 				is_mtr_hierarchy = true;
6021 			}
6022 		}
6023 		/*
6024 		 * If it isn't default-policy Meter, and
6025 		 * 1. There's no action in flow to change
6026 		 *    packet (modify/encap/decap etc.), OR
6027 		 * 2. No drop count needed for this meter.
6028 		 * 3. It's not meter hierarchy.
6029 		 * Then no need to use regC to save meter id anymore.
6030 		 */
6031 		if (!fm->def_policy && !is_mtr_hierarchy &&
6032 		    (!has_modify || !fm->drop_cnt))
6033 			set_mtr_reg = false;
6034 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6035 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6036 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6037 		/* Suffix items: tag, vlan, port id, end. */
6038 #define METER_SUFFIX_ITEM 4
6039 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6040 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6041 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6042 					  0, SOCKET_ID_ANY);
6043 		if (!sfx_actions)
6044 			return rte_flow_error_set(error, ENOMEM,
6045 						  RTE_FLOW_ERROR_TYPE_ACTION,
6046 						  NULL, "no memory to split "
6047 						  "meter flow");
6048 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6049 			     act_size);
6050 		/* There's no suffix flow for meter of non-default policy. */
6051 		if (!fm->def_policy)
6052 			pre_actions = sfx_actions + 1;
6053 		else
6054 			pre_actions = sfx_actions + actions_n;
6055 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6056 					    items, sfx_items, actions,
6057 					    sfx_actions, pre_actions,
6058 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6059 					    error);
6060 		if (ret) {
6061 			ret = -rte_errno;
6062 			goto exit;
6063 		}
6064 		/* Add the prefix subflow. */
6065 		flow_split_info->prefix_mark = 0;
6066 		skip_scale_restore = flow_split_info->skip_scale;
6067 		flow_split_info->skip_scale |=
6068 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6069 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6070 					      attr, items, pre_actions,
6071 					      flow_split_info, error);
6072 		flow_split_info->skip_scale = skip_scale_restore;
6073 		if (ret) {
6074 			if (mtr_flow_id)
6075 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6076 			ret = -rte_errno;
6077 			goto exit;
6078 		}
6079 		if (mtr_flow_id) {
6080 			dev_flow->handle->split_flow_id = mtr_flow_id;
6081 			dev_flow->handle->is_meter_flow_id = 1;
6082 		}
6083 		if (!fm->def_policy) {
6084 			if (!set_mtr_reg && fm->drop_cnt)
6085 				ret =
6086 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6087 							&sfx_attr, items,
6088 							flow_split_info,
6089 							fm, error);
6090 			goto exit;
6091 		}
6092 		/* Setting the sfx group atrr. */
6093 		sfx_attr.group = sfx_attr.transfer ?
6094 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6095 				 MLX5_FLOW_TABLE_LEVEL_METER;
6096 		flow_split_info->prefix_layers =
6097 				flow_get_prefix_layer_flags(dev_flow);
6098 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6099 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6100 	}
6101 	/* Add the prefix subflow. */
6102 	ret = flow_create_split_metadata(dev, flow,
6103 					 &sfx_attr, sfx_items ?
6104 					 sfx_items : items,
6105 					 sfx_actions ? sfx_actions : actions,
6106 					 flow_split_info, error);
6107 exit:
6108 	if (sfx_actions)
6109 		mlx5_free(sfx_actions);
6110 	return ret;
6111 }
6112 
6113 /**
6114  * The splitting for sample feature.
6115  *
6116  * Once Sample action is detected in the action list, the flow actions should
6117  * be split into prefix sub flow and suffix sub flow.
6118  *
6119  * The original items remain in the prefix sub flow, all actions preceding the
6120  * sample action and the sample action itself will be copied to the prefix
6121  * sub flow, the actions following the sample action will be copied to the
6122  * suffix sub flow, Queue action always be located in the suffix sub flow.
6123  *
6124  * In order to make the packet from prefix sub flow matches with suffix sub
6125  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6126  * flow uses tag item with the unique flow id.
6127  *
6128  * @param dev
6129  *   Pointer to Ethernet device.
6130  * @param[in] flow
6131  *   Parent flow structure pointer.
6132  * @param[in] attr
6133  *   Flow rule attributes.
6134  * @param[in] items
6135  *   Pattern specification (list terminated by the END pattern item).
6136  * @param[in] actions
6137  *   Associated actions (list terminated by the END action).
6138  * @param[in] flow_split_info
6139  *   Pointer to flow split info structure.
6140  * @param[out] error
6141  *   Perform verbose error reporting if not NULL.
6142  * @return
6143  *   0 on success, negative value otherwise
6144  */
6145 static int
6146 flow_create_split_sample(struct rte_eth_dev *dev,
6147 			 struct rte_flow *flow,
6148 			 const struct rte_flow_attr *attr,
6149 			 const struct rte_flow_item items[],
6150 			 const struct rte_flow_action actions[],
6151 			 struct mlx5_flow_split_info *flow_split_info,
6152 			 struct rte_flow_error *error)
6153 {
6154 	struct mlx5_priv *priv = dev->data->dev_private;
6155 	struct rte_flow_action *sfx_actions = NULL;
6156 	struct rte_flow_action *pre_actions = NULL;
6157 	struct rte_flow_item *sfx_items = NULL;
6158 	struct mlx5_flow *dev_flow = NULL;
6159 	struct rte_flow_attr sfx_attr = *attr;
6160 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6161 	struct mlx5_flow_dv_sample_resource *sample_res;
6162 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6163 	struct mlx5_flow_tbl_resource *sfx_tbl;
6164 #endif
6165 	size_t act_size;
6166 	size_t item_size;
6167 	uint32_t fdb_tx = 0;
6168 	int32_t tag_id = 0;
6169 	int actions_n = 0;
6170 	int sample_action_pos;
6171 	int qrss_action_pos;
6172 	int add_tag = 0;
6173 	int modify_after_mirror = 0;
6174 	uint16_t jump_table = 0;
6175 	const uint32_t next_ft_step = 1;
6176 	int ret = 0;
6177 
6178 	if (priv->sampler_en)
6179 		actions_n = flow_check_match_action(actions, attr,
6180 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6181 					&sample_action_pos, &qrss_action_pos,
6182 					&modify_after_mirror);
6183 	if (actions_n) {
6184 		/* The prefix actions must includes sample, tag, end. */
6185 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6186 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6187 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6188 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6189 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6190 					  item_size), 0, SOCKET_ID_ANY);
6191 		if (!sfx_actions)
6192 			return rte_flow_error_set(error, ENOMEM,
6193 						  RTE_FLOW_ERROR_TYPE_ACTION,
6194 						  NULL, "no memory to split "
6195 						  "sample flow");
6196 		/* The representor_id is UINT16_MAX for uplink. */
6197 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6198 		/*
6199 		 * When reg_c_preserve is set, metadata registers Cx preserve
6200 		 * their value even through packet duplication.
6201 		 */
6202 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6203 		if (add_tag)
6204 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6205 					+ act_size);
6206 		if (modify_after_mirror)
6207 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6208 				     next_ft_step;
6209 		pre_actions = sfx_actions + actions_n;
6210 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6211 						actions, sfx_actions,
6212 						pre_actions, actions_n,
6213 						sample_action_pos,
6214 						qrss_action_pos, jump_table,
6215 						error);
6216 		if (tag_id < 0 || (add_tag && !tag_id)) {
6217 			ret = -rte_errno;
6218 			goto exit;
6219 		}
6220 		if (modify_after_mirror)
6221 			flow_split_info->skip_scale =
6222 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6223 		/* Add the prefix subflow. */
6224 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6225 					      items, pre_actions,
6226 					      flow_split_info, error);
6227 		if (ret) {
6228 			ret = -rte_errno;
6229 			goto exit;
6230 		}
6231 		dev_flow->handle->split_flow_id = tag_id;
6232 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6233 		if (!modify_after_mirror) {
6234 			/* Set the sfx group attr. */
6235 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6236 						dev_flow->dv.sample_res;
6237 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6238 						sample_res->normal_path_tbl;
6239 			sfx_tbl_data = container_of(sfx_tbl,
6240 						struct mlx5_flow_tbl_data_entry,
6241 						tbl);
6242 			sfx_attr.group = sfx_attr.transfer ?
6243 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6244 		} else {
6245 			MLX5_ASSERT(attr->transfer);
6246 			sfx_attr.group = jump_table;
6247 		}
6248 		flow_split_info->prefix_layers =
6249 				flow_get_prefix_layer_flags(dev_flow);
6250 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6251 		/* Suffix group level already be scaled with factor, set
6252 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6253 		 * again in translation.
6254 		 */
6255 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6256 #endif
6257 	}
6258 	/* Add the suffix subflow. */
6259 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6260 				      sfx_items ? sfx_items : items,
6261 				      sfx_actions ? sfx_actions : actions,
6262 				      flow_split_info, error);
6263 exit:
6264 	if (sfx_actions)
6265 		mlx5_free(sfx_actions);
6266 	return ret;
6267 }
6268 
6269 /**
6270  * Split the flow to subflow set. The splitters might be linked
6271  * in the chain, like this:
6272  * flow_create_split_outer() calls:
6273  *   flow_create_split_meter() calls:
6274  *     flow_create_split_metadata(meter_subflow_0) calls:
6275  *       flow_create_split_inner(metadata_subflow_0)
6276  *       flow_create_split_inner(metadata_subflow_1)
6277  *       flow_create_split_inner(metadata_subflow_2)
6278  *     flow_create_split_metadata(meter_subflow_1) calls:
6279  *       flow_create_split_inner(metadata_subflow_0)
6280  *       flow_create_split_inner(metadata_subflow_1)
6281  *       flow_create_split_inner(metadata_subflow_2)
6282  *
6283  * This provide flexible way to add new levels of flow splitting.
6284  * The all of successfully created subflows are included to the
6285  * parent flow dev_flow list.
6286  *
6287  * @param dev
6288  *   Pointer to Ethernet device.
6289  * @param[in] flow
6290  *   Parent flow structure pointer.
6291  * @param[in] attr
6292  *   Flow rule attributes.
6293  * @param[in] items
6294  *   Pattern specification (list terminated by the END pattern item).
6295  * @param[in] actions
6296  *   Associated actions (list terminated by the END action).
6297  * @param[in] flow_split_info
6298  *   Pointer to flow split info structure.
6299  * @param[out] error
6300  *   Perform verbose error reporting if not NULL.
6301  * @return
6302  *   0 on success, negative value otherwise
6303  */
6304 static int
6305 flow_create_split_outer(struct rte_eth_dev *dev,
6306 			struct rte_flow *flow,
6307 			const struct rte_flow_attr *attr,
6308 			const struct rte_flow_item items[],
6309 			const struct rte_flow_action actions[],
6310 			struct mlx5_flow_split_info *flow_split_info,
6311 			struct rte_flow_error *error)
6312 {
6313 	int ret;
6314 
6315 	ret = flow_create_split_sample(dev, flow, attr, items,
6316 				       actions, flow_split_info, error);
6317 	MLX5_ASSERT(ret <= 0);
6318 	return ret;
6319 }
6320 
6321 static inline struct mlx5_flow_tunnel *
6322 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6323 {
6324 	struct mlx5_flow_tunnel *tunnel;
6325 
6326 #pragma GCC diagnostic push
6327 #pragma GCC diagnostic ignored "-Wcast-qual"
6328 	tunnel = (typeof(tunnel))flow->tunnel;
6329 #pragma GCC diagnostic pop
6330 
6331 	return tunnel;
6332 }
6333 
6334 /**
6335  * Adjust flow RSS workspace if needed.
6336  *
6337  * @param wks
6338  *   Pointer to thread flow work space.
6339  * @param rss_desc
6340  *   Pointer to RSS descriptor.
6341  * @param[in] nrssq_num
6342  *   New RSS queue number.
6343  *
6344  * @return
6345  *   0 on success, -1 otherwise and rte_errno is set.
6346  */
6347 static int
6348 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6349 			  struct mlx5_flow_rss_desc *rss_desc,
6350 			  uint32_t nrssq_num)
6351 {
6352 	if (likely(nrssq_num <= wks->rssq_num))
6353 		return 0;
6354 	rss_desc->queue = realloc(rss_desc->queue,
6355 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6356 	if (!rss_desc->queue) {
6357 		rte_errno = ENOMEM;
6358 		return -1;
6359 	}
6360 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6361 	return 0;
6362 }
6363 
6364 /**
6365  * Create a flow and add it to @p list.
6366  *
6367  * @param dev
6368  *   Pointer to Ethernet device.
6369  * @param list
6370  *   Pointer to a TAILQ flow list. If this parameter NULL,
6371  *   no list insertion occurred, flow is just created,
6372  *   this is caller's responsibility to track the
6373  *   created flow.
6374  * @param[in] attr
6375  *   Flow rule attributes.
6376  * @param[in] items
6377  *   Pattern specification (list terminated by the END pattern item).
6378  * @param[in] actions
6379  *   Associated actions (list terminated by the END action).
6380  * @param[in] external
6381  *   This flow rule is created by request external to PMD.
6382  * @param[out] error
6383  *   Perform verbose error reporting if not NULL.
6384  *
6385  * @return
6386  *   A flow index on success, 0 otherwise and rte_errno is set.
6387  */
6388 static uint32_t
6389 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6390 		 const struct rte_flow_attr *attr,
6391 		 const struct rte_flow_item items[],
6392 		 const struct rte_flow_action original_actions[],
6393 		 bool external, struct rte_flow_error *error)
6394 {
6395 	struct mlx5_priv *priv = dev->data->dev_private;
6396 	struct rte_flow *flow = NULL;
6397 	struct mlx5_flow *dev_flow;
6398 	const struct rte_flow_action_rss *rss = NULL;
6399 	struct mlx5_translated_action_handle
6400 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6401 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6402 	union {
6403 		struct mlx5_flow_expand_rss buf;
6404 		uint8_t buffer[4096];
6405 	} expand_buffer;
6406 	union {
6407 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6408 		uint8_t buffer[2048];
6409 	} actions_rx;
6410 	union {
6411 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6412 		uint8_t buffer[2048];
6413 	} actions_hairpin_tx;
6414 	union {
6415 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6416 		uint8_t buffer[2048];
6417 	} items_tx;
6418 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6419 	struct mlx5_flow_rss_desc *rss_desc;
6420 	const struct rte_flow_action *p_actions_rx;
6421 	uint32_t i;
6422 	uint32_t idx = 0;
6423 	int hairpin_flow;
6424 	struct rte_flow_attr attr_tx = { .priority = 0 };
6425 	const struct rte_flow_action *actions;
6426 	struct rte_flow_action *translated_actions = NULL;
6427 	struct mlx5_flow_tunnel *tunnel;
6428 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6429 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6430 	struct mlx5_flow_split_info flow_split_info = {
6431 		.external = !!external,
6432 		.skip_scale = 0,
6433 		.flow_idx = 0,
6434 		.prefix_mark = 0,
6435 		.prefix_layers = 0,
6436 		.table_id = 0
6437 	};
6438 	int ret;
6439 
6440 	MLX5_ASSERT(wks);
6441 	rss_desc = &wks->rss_desc;
6442 	ret = flow_action_handles_translate(dev, original_actions,
6443 					    indir_actions,
6444 					    &indir_actions_n,
6445 					    &translated_actions, error);
6446 	if (ret < 0) {
6447 		MLX5_ASSERT(translated_actions == NULL);
6448 		return 0;
6449 	}
6450 	actions = translated_actions ? translated_actions : original_actions;
6451 	p_actions_rx = actions;
6452 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6453 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6454 				external, hairpin_flow, error);
6455 	if (ret < 0)
6456 		goto error_before_hairpin_split;
6457 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6458 	if (!flow) {
6459 		rte_errno = ENOMEM;
6460 		goto error_before_hairpin_split;
6461 	}
6462 	if (hairpin_flow > 0) {
6463 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6464 			rte_errno = EINVAL;
6465 			goto error_before_hairpin_split;
6466 		}
6467 		flow_hairpin_split(dev, actions, actions_rx.actions,
6468 				   actions_hairpin_tx.actions, items_tx.items,
6469 				   idx);
6470 		p_actions_rx = actions_rx.actions;
6471 	}
6472 	flow_split_info.flow_idx = idx;
6473 	flow->drv_type = flow_get_drv_type(dev, attr);
6474 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6475 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6476 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6477 	/* RSS Action only works on NIC RX domain */
6478 	if (attr->ingress && !attr->transfer)
6479 		rss = flow_get_rss_action(dev, p_actions_rx);
6480 	if (rss) {
6481 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6482 			return 0;
6483 		/*
6484 		 * The following information is required by
6485 		 * mlx5_flow_hashfields_adjust() in advance.
6486 		 */
6487 		rss_desc->level = rss->level;
6488 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6489 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6490 	}
6491 	flow->dev_handles = 0;
6492 	if (rss && rss->types) {
6493 		unsigned int graph_root;
6494 
6495 		graph_root = find_graph_root(rss->level);
6496 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6497 					   items, rss->types,
6498 					   mlx5_support_expansion, graph_root);
6499 		MLX5_ASSERT(ret > 0 &&
6500 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6501 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6502 			for (i = 0; i < buf->entries; ++i)
6503 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6504 		}
6505 	} else {
6506 		buf->entries = 1;
6507 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6508 	}
6509 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6510 						      indir_actions_n);
6511 	for (i = 0; i < buf->entries; ++i) {
6512 		/* Initialize flow split data. */
6513 		flow_split_info.prefix_layers = 0;
6514 		flow_split_info.prefix_mark = 0;
6515 		flow_split_info.skip_scale = 0;
6516 		/*
6517 		 * The splitter may create multiple dev_flows,
6518 		 * depending on configuration. In the simplest
6519 		 * case it just creates unmodified original flow.
6520 		 */
6521 		ret = flow_create_split_outer(dev, flow, attr,
6522 					      buf->entry[i].pattern,
6523 					      p_actions_rx, &flow_split_info,
6524 					      error);
6525 		if (ret < 0)
6526 			goto error;
6527 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6528 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6529 							   p_actions_rx,
6530 							   idx,
6531 							   wks->flows[0].tunnel,
6532 							   &default_miss_ctx,
6533 							   error);
6534 			if (ret < 0) {
6535 				mlx5_free(default_miss_ctx.queue);
6536 				goto error;
6537 			}
6538 		}
6539 	}
6540 	/* Create the tx flow. */
6541 	if (hairpin_flow) {
6542 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6543 		attr_tx.ingress = 0;
6544 		attr_tx.egress = 1;
6545 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6546 					 actions_hairpin_tx.actions,
6547 					 idx, error);
6548 		if (!dev_flow)
6549 			goto error;
6550 		dev_flow->flow = flow;
6551 		dev_flow->external = 0;
6552 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6553 			      dev_flow->handle, next);
6554 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6555 					 items_tx.items,
6556 					 actions_hairpin_tx.actions, error);
6557 		if (ret < 0)
6558 			goto error;
6559 	}
6560 	/*
6561 	 * Update the metadata register copy table. If extensive
6562 	 * metadata feature is enabled and registers are supported
6563 	 * we might create the extra rte_flow for each unique
6564 	 * MARK/FLAG action ID.
6565 	 *
6566 	 * The table is updated for ingress Flows only, because
6567 	 * the egress Flows belong to the different device and
6568 	 * copy table should be updated in peer NIC Rx domain.
6569 	 */
6570 	if (attr->ingress &&
6571 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6572 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6573 		if (ret)
6574 			goto error;
6575 	}
6576 	/*
6577 	 * If the flow is external (from application) OR device is started,
6578 	 * OR mreg discover, then apply immediately.
6579 	 */
6580 	if (external || dev->data->dev_started ||
6581 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6582 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6583 		ret = flow_drv_apply(dev, flow, error);
6584 		if (ret < 0)
6585 			goto error;
6586 	}
6587 	flow->type = type;
6588 	flow_rxq_flags_set(dev, flow);
6589 	rte_free(translated_actions);
6590 	tunnel = flow_tunnel_from_rule(wks->flows);
6591 	if (tunnel) {
6592 		flow->tunnel = 1;
6593 		flow->tunnel_id = tunnel->tunnel_id;
6594 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6595 		mlx5_free(default_miss_ctx.queue);
6596 	}
6597 	mlx5_flow_pop_thread_workspace();
6598 	return idx;
6599 error:
6600 	MLX5_ASSERT(flow);
6601 	ret = rte_errno; /* Save rte_errno before cleanup. */
6602 	flow_mreg_del_copy_action(dev, flow);
6603 	flow_drv_destroy(dev, flow);
6604 	if (rss_desc->shared_rss)
6605 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6606 			mlx5_ipool_get
6607 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6608 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6609 	mlx5_ipool_free(priv->flows[type], idx);
6610 	rte_errno = ret; /* Restore rte_errno. */
6611 	ret = rte_errno;
6612 	rte_errno = ret;
6613 	mlx5_flow_pop_thread_workspace();
6614 error_before_hairpin_split:
6615 	rte_free(translated_actions);
6616 	return 0;
6617 }
6618 
6619 /**
6620  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6621  * incoming packets to table 1.
6622  *
6623  * Other flow rules, requested for group n, will be created in
6624  * e-switch table n+1.
6625  * Jump action to e-switch group n will be created to group n+1.
6626  *
6627  * Used when working in switchdev mode, to utilise advantages of table 1
6628  * and above.
6629  *
6630  * @param dev
6631  *   Pointer to Ethernet device.
6632  *
6633  * @return
6634  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6635  */
6636 struct rte_flow *
6637 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6638 {
6639 	const struct rte_flow_attr attr = {
6640 		.group = 0,
6641 		.priority = 0,
6642 		.ingress = 1,
6643 		.egress = 0,
6644 		.transfer = 1,
6645 	};
6646 	const struct rte_flow_item pattern = {
6647 		.type = RTE_FLOW_ITEM_TYPE_END,
6648 	};
6649 	struct rte_flow_action_jump jump = {
6650 		.group = 1,
6651 	};
6652 	const struct rte_flow_action actions[] = {
6653 		{
6654 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6655 			.conf = &jump,
6656 		},
6657 		{
6658 			.type = RTE_FLOW_ACTION_TYPE_END,
6659 		},
6660 	};
6661 	struct rte_flow_error error;
6662 
6663 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6664 						   &attr, &pattern,
6665 						   actions, false, &error);
6666 }
6667 
6668 /**
6669  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6670  * and sq number, directs all packets to peer vport.
6671  *
6672  * @param dev
6673  *   Pointer to Ethernet device.
6674  * @param txq
6675  *   Txq index.
6676  *
6677  * @return
6678  *   Flow ID on success, 0 otherwise and rte_errno is set.
6679  */
6680 uint32_t
6681 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6682 {
6683 	struct rte_flow_attr attr = {
6684 		.group = 0,
6685 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6686 		.ingress = 1,
6687 		.egress = 0,
6688 		.transfer = 1,
6689 	};
6690 	struct rte_flow_item_port_id port_spec = {
6691 		.id = MLX5_PORT_ESW_MGR,
6692 	};
6693 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6694 		.queue = txq,
6695 	};
6696 	struct rte_flow_item pattern[] = {
6697 		{
6698 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6699 			.spec = &port_spec,
6700 		},
6701 		{
6702 			.type = (enum rte_flow_item_type)
6703 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6704 			.spec = &txq_spec,
6705 		},
6706 		{
6707 			.type = RTE_FLOW_ITEM_TYPE_END,
6708 		},
6709 	};
6710 	struct rte_flow_action_jump jump = {
6711 		.group = 1,
6712 	};
6713 	struct rte_flow_action_port_id port = {
6714 		.id = dev->data->port_id,
6715 	};
6716 	struct rte_flow_action actions[] = {
6717 		{
6718 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6719 			.conf = &jump,
6720 		},
6721 		{
6722 			.type = RTE_FLOW_ACTION_TYPE_END,
6723 		},
6724 	};
6725 	struct rte_flow_error error;
6726 
6727 	/*
6728 	 * Creates group 0, highest priority jump flow.
6729 	 * Matches txq to bypass kernel packets.
6730 	 */
6731 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6732 			     false, &error) == 0)
6733 		return 0;
6734 	/* Create group 1, lowest priority redirect flow for txq. */
6735 	attr.group = 1;
6736 	actions[0].conf = &port;
6737 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6738 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6739 				actions, false, &error);
6740 }
6741 
6742 /**
6743  * Validate a flow supported by the NIC.
6744  *
6745  * @see rte_flow_validate()
6746  * @see rte_flow_ops
6747  */
6748 int
6749 mlx5_flow_validate(struct rte_eth_dev *dev,
6750 		   const struct rte_flow_attr *attr,
6751 		   const struct rte_flow_item items[],
6752 		   const struct rte_flow_action original_actions[],
6753 		   struct rte_flow_error *error)
6754 {
6755 	int hairpin_flow;
6756 	struct mlx5_translated_action_handle
6757 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6758 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6759 	const struct rte_flow_action *actions;
6760 	struct rte_flow_action *translated_actions = NULL;
6761 	int ret = flow_action_handles_translate(dev, original_actions,
6762 						indir_actions,
6763 						&indir_actions_n,
6764 						&translated_actions, error);
6765 
6766 	if (ret)
6767 		return ret;
6768 	actions = translated_actions ? translated_actions : original_actions;
6769 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6770 	ret = flow_drv_validate(dev, attr, items, actions,
6771 				true, hairpin_flow, error);
6772 	rte_free(translated_actions);
6773 	return ret;
6774 }
6775 
6776 /**
6777  * Create a flow.
6778  *
6779  * @see rte_flow_create()
6780  * @see rte_flow_ops
6781  */
6782 struct rte_flow *
6783 mlx5_flow_create(struct rte_eth_dev *dev,
6784 		 const struct rte_flow_attr *attr,
6785 		 const struct rte_flow_item items[],
6786 		 const struct rte_flow_action actions[],
6787 		 struct rte_flow_error *error)
6788 {
6789 	/*
6790 	 * If the device is not started yet, it is not allowed to created a
6791 	 * flow from application. PMD default flows and traffic control flows
6792 	 * are not affected.
6793 	 */
6794 	if (unlikely(!dev->data->dev_started)) {
6795 		DRV_LOG(DEBUG, "port %u is not started when "
6796 			"inserting a flow", dev->data->port_id);
6797 		rte_flow_error_set(error, ENODEV,
6798 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6799 				   NULL,
6800 				   "port not started");
6801 		return NULL;
6802 	}
6803 
6804 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6805 						   attr, items, actions,
6806 						   true, error);
6807 }
6808 
6809 /**
6810  * Destroy a flow in a list.
6811  *
6812  * @param dev
6813  *   Pointer to Ethernet device.
6814  * @param[in] flow_idx
6815  *   Index of flow to destroy.
6816  */
6817 static void
6818 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6819 		  uint32_t flow_idx)
6820 {
6821 	struct mlx5_priv *priv = dev->data->dev_private;
6822 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6823 
6824 	if (!flow)
6825 		return;
6826 	MLX5_ASSERT(flow->type == type);
6827 	/*
6828 	 * Update RX queue flags only if port is started, otherwise it is
6829 	 * already clean.
6830 	 */
6831 	if (dev->data->dev_started)
6832 		flow_rxq_flags_trim(dev, flow);
6833 	flow_drv_destroy(dev, flow);
6834 	if (flow->tunnel) {
6835 		struct mlx5_flow_tunnel *tunnel;
6836 
6837 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6838 		RTE_VERIFY(tunnel);
6839 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6840 			mlx5_flow_tunnel_free(dev, tunnel);
6841 	}
6842 	flow_mreg_del_copy_action(dev, flow);
6843 	mlx5_ipool_free(priv->flows[type], flow_idx);
6844 }
6845 
6846 /**
6847  * Destroy all flows.
6848  *
6849  * @param dev
6850  *   Pointer to Ethernet device.
6851  * @param type
6852  *   Flow type to be flushed.
6853  * @param active
6854  *   If flushing is called avtively.
6855  */
6856 void
6857 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6858 		     bool active)
6859 {
6860 	struct mlx5_priv *priv = dev->data->dev_private;
6861 	uint32_t num_flushed = 0, fidx = 1;
6862 	struct rte_flow *flow;
6863 
6864 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6865 		flow_list_destroy(dev, type, fidx);
6866 		num_flushed++;
6867 	}
6868 	if (active) {
6869 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6870 			dev->data->port_id, num_flushed);
6871 	}
6872 }
6873 
6874 /**
6875  * Stop all default actions for flows.
6876  *
6877  * @param dev
6878  *   Pointer to Ethernet device.
6879  */
6880 void
6881 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6882 {
6883 	flow_mreg_del_default_copy_action(dev);
6884 	flow_rxq_flags_clear(dev);
6885 }
6886 
6887 /**
6888  * Start all default actions for flows.
6889  *
6890  * @param dev
6891  *   Pointer to Ethernet device.
6892  * @return
6893  *   0 on success, a negative errno value otherwise and rte_errno is set.
6894  */
6895 int
6896 mlx5_flow_start_default(struct rte_eth_dev *dev)
6897 {
6898 	struct rte_flow_error error;
6899 
6900 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6901 	return flow_mreg_add_default_copy_action(dev, &error);
6902 }
6903 
6904 /**
6905  * Release key of thread specific flow workspace data.
6906  */
6907 void
6908 flow_release_workspace(void *data)
6909 {
6910 	struct mlx5_flow_workspace *wks = data;
6911 	struct mlx5_flow_workspace *next;
6912 
6913 	while (wks) {
6914 		next = wks->next;
6915 		free(wks->rss_desc.queue);
6916 		free(wks);
6917 		wks = next;
6918 	}
6919 }
6920 
6921 /**
6922  * Get thread specific current flow workspace.
6923  *
6924  * @return pointer to thread specific flow workspace data, NULL on error.
6925  */
6926 struct mlx5_flow_workspace*
6927 mlx5_flow_get_thread_workspace(void)
6928 {
6929 	struct mlx5_flow_workspace *data;
6930 
6931 	data = mlx5_flow_os_get_specific_workspace();
6932 	MLX5_ASSERT(data && data->inuse);
6933 	if (!data || !data->inuse)
6934 		DRV_LOG(ERR, "flow workspace not initialized.");
6935 	return data;
6936 }
6937 
6938 /**
6939  * Allocate and init new flow workspace.
6940  *
6941  * @return pointer to flow workspace data, NULL on error.
6942  */
6943 static struct mlx5_flow_workspace*
6944 flow_alloc_thread_workspace(void)
6945 {
6946 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6947 
6948 	if (!data) {
6949 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6950 			"memory.");
6951 		return NULL;
6952 	}
6953 	data->rss_desc.queue = calloc(1,
6954 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6955 	if (!data->rss_desc.queue)
6956 		goto err;
6957 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6958 	return data;
6959 err:
6960 	if (data->rss_desc.queue)
6961 		free(data->rss_desc.queue);
6962 	free(data);
6963 	return NULL;
6964 }
6965 
6966 /**
6967  * Get new thread specific flow workspace.
6968  *
6969  * If current workspace inuse, create new one and set as current.
6970  *
6971  * @return pointer to thread specific flow workspace data, NULL on error.
6972  */
6973 static struct mlx5_flow_workspace*
6974 mlx5_flow_push_thread_workspace(void)
6975 {
6976 	struct mlx5_flow_workspace *curr;
6977 	struct mlx5_flow_workspace *data;
6978 
6979 	curr = mlx5_flow_os_get_specific_workspace();
6980 	if (!curr) {
6981 		data = flow_alloc_thread_workspace();
6982 		if (!data)
6983 			return NULL;
6984 	} else if (!curr->inuse) {
6985 		data = curr;
6986 	} else if (curr->next) {
6987 		data = curr->next;
6988 	} else {
6989 		data = flow_alloc_thread_workspace();
6990 		if (!data)
6991 			return NULL;
6992 		curr->next = data;
6993 		data->prev = curr;
6994 	}
6995 	data->inuse = 1;
6996 	data->flow_idx = 0;
6997 	/* Set as current workspace */
6998 	if (mlx5_flow_os_set_specific_workspace(data))
6999 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7000 	return data;
7001 }
7002 
7003 /**
7004  * Close current thread specific flow workspace.
7005  *
7006  * If previous workspace available, set it as current.
7007  *
7008  * @return pointer to thread specific flow workspace data, NULL on error.
7009  */
7010 static void
7011 mlx5_flow_pop_thread_workspace(void)
7012 {
7013 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7014 
7015 	if (!data)
7016 		return;
7017 	if (!data->inuse) {
7018 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7019 		return;
7020 	}
7021 	data->inuse = 0;
7022 	if (!data->prev)
7023 		return;
7024 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7025 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7026 }
7027 
7028 /**
7029  * Verify the flow list is empty
7030  *
7031  * @param dev
7032  *  Pointer to Ethernet device.
7033  *
7034  * @return the number of flows not released.
7035  */
7036 int
7037 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7038 {
7039 	struct mlx5_priv *priv = dev->data->dev_private;
7040 	struct rte_flow *flow;
7041 	uint32_t idx = 0;
7042 	int ret = 0, i;
7043 
7044 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7045 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7046 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7047 				dev->data->port_id, (void *)flow);
7048 			ret++;
7049 		}
7050 	}
7051 	return ret;
7052 }
7053 
7054 /**
7055  * Enable default hairpin egress flow.
7056  *
7057  * @param dev
7058  *   Pointer to Ethernet device.
7059  * @param queue
7060  *   The queue index.
7061  *
7062  * @return
7063  *   0 on success, a negative errno value otherwise and rte_errno is set.
7064  */
7065 int
7066 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7067 			    uint32_t queue)
7068 {
7069 	const struct rte_flow_attr attr = {
7070 		.egress = 1,
7071 		.priority = 0,
7072 	};
7073 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7074 		.queue = queue,
7075 	};
7076 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7077 		.queue = UINT32_MAX,
7078 	};
7079 	struct rte_flow_item items[] = {
7080 		{
7081 			.type = (enum rte_flow_item_type)
7082 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7083 			.spec = &queue_spec,
7084 			.last = NULL,
7085 			.mask = &queue_mask,
7086 		},
7087 		{
7088 			.type = RTE_FLOW_ITEM_TYPE_END,
7089 		},
7090 	};
7091 	struct rte_flow_action_jump jump = {
7092 		.group = MLX5_HAIRPIN_TX_TABLE,
7093 	};
7094 	struct rte_flow_action actions[2];
7095 	uint32_t flow_idx;
7096 	struct rte_flow_error error;
7097 
7098 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7099 	actions[0].conf = &jump;
7100 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7101 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7102 				    &attr, items, actions, false, &error);
7103 	if (!flow_idx) {
7104 		DRV_LOG(DEBUG,
7105 			"Failed to create ctrl flow: rte_errno(%d),"
7106 			" type(%d), message(%s)",
7107 			rte_errno, error.type,
7108 			error.message ? error.message : " (no stated reason)");
7109 		return -rte_errno;
7110 	}
7111 	return 0;
7112 }
7113 
7114 /**
7115  * Enable a control flow configured from the control plane.
7116  *
7117  * @param dev
7118  *   Pointer to Ethernet device.
7119  * @param eth_spec
7120  *   An Ethernet flow spec to apply.
7121  * @param eth_mask
7122  *   An Ethernet flow mask to apply.
7123  * @param vlan_spec
7124  *   A VLAN flow spec to apply.
7125  * @param vlan_mask
7126  *   A VLAN flow mask to apply.
7127  *
7128  * @return
7129  *   0 on success, a negative errno value otherwise and rte_errno is set.
7130  */
7131 int
7132 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7133 		    struct rte_flow_item_eth *eth_spec,
7134 		    struct rte_flow_item_eth *eth_mask,
7135 		    struct rte_flow_item_vlan *vlan_spec,
7136 		    struct rte_flow_item_vlan *vlan_mask)
7137 {
7138 	struct mlx5_priv *priv = dev->data->dev_private;
7139 	const struct rte_flow_attr attr = {
7140 		.ingress = 1,
7141 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7142 	};
7143 	struct rte_flow_item items[] = {
7144 		{
7145 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7146 			.spec = eth_spec,
7147 			.last = NULL,
7148 			.mask = eth_mask,
7149 		},
7150 		{
7151 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7152 					      RTE_FLOW_ITEM_TYPE_END,
7153 			.spec = vlan_spec,
7154 			.last = NULL,
7155 			.mask = vlan_mask,
7156 		},
7157 		{
7158 			.type = RTE_FLOW_ITEM_TYPE_END,
7159 		},
7160 	};
7161 	uint16_t queue[priv->reta_idx_n];
7162 	struct rte_flow_action_rss action_rss = {
7163 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7164 		.level = 0,
7165 		.types = priv->rss_conf.rss_hf,
7166 		.key_len = priv->rss_conf.rss_key_len,
7167 		.queue_num = priv->reta_idx_n,
7168 		.key = priv->rss_conf.rss_key,
7169 		.queue = queue,
7170 	};
7171 	struct rte_flow_action actions[] = {
7172 		{
7173 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7174 			.conf = &action_rss,
7175 		},
7176 		{
7177 			.type = RTE_FLOW_ACTION_TYPE_END,
7178 		},
7179 	};
7180 	uint32_t flow_idx;
7181 	struct rte_flow_error error;
7182 	unsigned int i;
7183 
7184 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7185 		return 0;
7186 	}
7187 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7188 		action_rss.types = 0;
7189 	for (i = 0; i != priv->reta_idx_n; ++i)
7190 		queue[i] = (*priv->reta_idx)[i];
7191 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7192 				    &attr, items, actions, false, &error);
7193 	if (!flow_idx)
7194 		return -rte_errno;
7195 	return 0;
7196 }
7197 
7198 /**
7199  * Enable a flow control configured from the control plane.
7200  *
7201  * @param dev
7202  *   Pointer to Ethernet device.
7203  * @param eth_spec
7204  *   An Ethernet flow spec to apply.
7205  * @param eth_mask
7206  *   An Ethernet flow mask to apply.
7207  *
7208  * @return
7209  *   0 on success, a negative errno value otherwise and rte_errno is set.
7210  */
7211 int
7212 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7213 	       struct rte_flow_item_eth *eth_spec,
7214 	       struct rte_flow_item_eth *eth_mask)
7215 {
7216 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7217 }
7218 
7219 /**
7220  * Create default miss flow rule matching lacp traffic
7221  *
7222  * @param dev
7223  *   Pointer to Ethernet device.
7224  * @param eth_spec
7225  *   An Ethernet flow spec to apply.
7226  *
7227  * @return
7228  *   0 on success, a negative errno value otherwise and rte_errno is set.
7229  */
7230 int
7231 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7232 {
7233 	/*
7234 	 * The LACP matching is done by only using ether type since using
7235 	 * a multicast dst mac causes kernel to give low priority to this flow.
7236 	 */
7237 	static const struct rte_flow_item_eth lacp_spec = {
7238 		.type = RTE_BE16(0x8809),
7239 	};
7240 	static const struct rte_flow_item_eth lacp_mask = {
7241 		.type = 0xffff,
7242 	};
7243 	const struct rte_flow_attr attr = {
7244 		.ingress = 1,
7245 	};
7246 	struct rte_flow_item items[] = {
7247 		{
7248 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7249 			.spec = &lacp_spec,
7250 			.mask = &lacp_mask,
7251 		},
7252 		{
7253 			.type = RTE_FLOW_ITEM_TYPE_END,
7254 		},
7255 	};
7256 	struct rte_flow_action actions[] = {
7257 		{
7258 			.type = (enum rte_flow_action_type)
7259 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7260 		},
7261 		{
7262 			.type = RTE_FLOW_ACTION_TYPE_END,
7263 		},
7264 	};
7265 	struct rte_flow_error error;
7266 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7267 					&attr, items, actions,
7268 					false, &error);
7269 
7270 	if (!flow_idx)
7271 		return -rte_errno;
7272 	return 0;
7273 }
7274 
7275 /**
7276  * Destroy a flow.
7277  *
7278  * @see rte_flow_destroy()
7279  * @see rte_flow_ops
7280  */
7281 int
7282 mlx5_flow_destroy(struct rte_eth_dev *dev,
7283 		  struct rte_flow *flow,
7284 		  struct rte_flow_error *error __rte_unused)
7285 {
7286 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7287 				(uintptr_t)(void *)flow);
7288 	return 0;
7289 }
7290 
7291 /**
7292  * Destroy all flows.
7293  *
7294  * @see rte_flow_flush()
7295  * @see rte_flow_ops
7296  */
7297 int
7298 mlx5_flow_flush(struct rte_eth_dev *dev,
7299 		struct rte_flow_error *error __rte_unused)
7300 {
7301 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7302 	return 0;
7303 }
7304 
7305 /**
7306  * Isolated mode.
7307  *
7308  * @see rte_flow_isolate()
7309  * @see rte_flow_ops
7310  */
7311 int
7312 mlx5_flow_isolate(struct rte_eth_dev *dev,
7313 		  int enable,
7314 		  struct rte_flow_error *error)
7315 {
7316 	struct mlx5_priv *priv = dev->data->dev_private;
7317 
7318 	if (dev->data->dev_started) {
7319 		rte_flow_error_set(error, EBUSY,
7320 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7321 				   NULL,
7322 				   "port must be stopped first");
7323 		return -rte_errno;
7324 	}
7325 	priv->isolated = !!enable;
7326 	if (enable)
7327 		dev->dev_ops = &mlx5_dev_ops_isolate;
7328 	else
7329 		dev->dev_ops = &mlx5_dev_ops;
7330 
7331 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7332 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7333 
7334 	return 0;
7335 }
7336 
7337 /**
7338  * Query a flow.
7339  *
7340  * @see rte_flow_query()
7341  * @see rte_flow_ops
7342  */
7343 static int
7344 flow_drv_query(struct rte_eth_dev *dev,
7345 	       uint32_t flow_idx,
7346 	       const struct rte_flow_action *actions,
7347 	       void *data,
7348 	       struct rte_flow_error *error)
7349 {
7350 	struct mlx5_priv *priv = dev->data->dev_private;
7351 	const struct mlx5_flow_driver_ops *fops;
7352 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7353 					       flow_idx);
7354 	enum mlx5_flow_drv_type ftype;
7355 
7356 	if (!flow) {
7357 		return rte_flow_error_set(error, ENOENT,
7358 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7359 			  NULL,
7360 			  "invalid flow handle");
7361 	}
7362 	ftype = flow->drv_type;
7363 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7364 	fops = flow_get_drv_ops(ftype);
7365 
7366 	return fops->query(dev, flow, actions, data, error);
7367 }
7368 
7369 /**
7370  * Query a flow.
7371  *
7372  * @see rte_flow_query()
7373  * @see rte_flow_ops
7374  */
7375 int
7376 mlx5_flow_query(struct rte_eth_dev *dev,
7377 		struct rte_flow *flow,
7378 		const struct rte_flow_action *actions,
7379 		void *data,
7380 		struct rte_flow_error *error)
7381 {
7382 	int ret;
7383 
7384 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7385 			     error);
7386 	if (ret < 0)
7387 		return ret;
7388 	return 0;
7389 }
7390 
7391 /**
7392  * Get rte_flow callbacks.
7393  *
7394  * @param dev
7395  *   Pointer to Ethernet device structure.
7396  * @param ops
7397  *   Pointer to operation-specific structure.
7398  *
7399  * @return 0
7400  */
7401 int
7402 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7403 		  const struct rte_flow_ops **ops)
7404 {
7405 	*ops = &mlx5_flow_ops;
7406 	return 0;
7407 }
7408 
7409 /**
7410  * Validate meter policy actions.
7411  * Dispatcher for action type specific validation.
7412  *
7413  * @param[in] dev
7414  *   Pointer to the Ethernet device structure.
7415  * @param[in] action
7416  *   The meter policy action object to validate.
7417  * @param[in] attr
7418  *   Attributes of flow to determine steering domain.
7419  * @param[out] is_rss
7420  *   Is RSS or not.
7421  * @param[out] domain_bitmap
7422  *   Domain bitmap.
7423  * @param[out] is_def_policy
7424  *   Is default policy or not.
7425  * @param[out] error
7426  *   Perform verbose error reporting if not NULL. Initialized in case of
7427  *   error only.
7428  *
7429  * @return
7430  *   0 on success, otherwise negative errno value.
7431  */
7432 int
7433 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7434 			const struct rte_flow_action *actions[RTE_COLORS],
7435 			struct rte_flow_attr *attr,
7436 			bool *is_rss,
7437 			uint8_t *domain_bitmap,
7438 			uint8_t *policy_mode,
7439 			struct rte_mtr_error *error)
7440 {
7441 	const struct mlx5_flow_driver_ops *fops;
7442 
7443 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7444 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7445 				       domain_bitmap, policy_mode, error);
7446 }
7447 
7448 /**
7449  * Destroy the meter table set.
7450  *
7451  * @param[in] dev
7452  *   Pointer to Ethernet device.
7453  * @param[in] mtr_policy
7454  *   Meter policy struct.
7455  */
7456 void
7457 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7458 		      struct mlx5_flow_meter_policy *mtr_policy)
7459 {
7460 	const struct mlx5_flow_driver_ops *fops;
7461 
7462 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7463 	fops->destroy_mtr_acts(dev, mtr_policy);
7464 }
7465 
7466 /**
7467  * Create policy action, lock free,
7468  * (mutex should be acquired by caller).
7469  * Dispatcher for action type specific call.
7470  *
7471  * @param[in] dev
7472  *   Pointer to the Ethernet device structure.
7473  * @param[in] mtr_policy
7474  *   Meter policy struct.
7475  * @param[in] action
7476  *   Action specification used to create meter actions.
7477  * @param[out] error
7478  *   Perform verbose error reporting if not NULL. Initialized in case of
7479  *   error only.
7480  *
7481  * @return
7482  *   0 on success, otherwise negative errno value.
7483  */
7484 int
7485 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7486 		      struct mlx5_flow_meter_policy *mtr_policy,
7487 		      const struct rte_flow_action *actions[RTE_COLORS],
7488 		      struct rte_mtr_error *error)
7489 {
7490 	const struct mlx5_flow_driver_ops *fops;
7491 
7492 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7493 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7494 }
7495 
7496 /**
7497  * Create policy rules, lock free,
7498  * (mutex should be acquired by caller).
7499  * Dispatcher for action type specific call.
7500  *
7501  * @param[in] dev
7502  *   Pointer to the Ethernet device structure.
7503  * @param[in] mtr_policy
7504  *   Meter policy struct.
7505  *
7506  * @return
7507  *   0 on success, -1 otherwise.
7508  */
7509 int
7510 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7511 			     struct mlx5_flow_meter_policy *mtr_policy)
7512 {
7513 	const struct mlx5_flow_driver_ops *fops;
7514 
7515 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7516 	return fops->create_policy_rules(dev, mtr_policy);
7517 }
7518 
7519 /**
7520  * Destroy policy rules, lock free,
7521  * (mutex should be acquired by caller).
7522  * Dispatcher for action type specific call.
7523  *
7524  * @param[in] dev
7525  *   Pointer to the Ethernet device structure.
7526  * @param[in] mtr_policy
7527  *   Meter policy struct.
7528  */
7529 void
7530 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7531 			     struct mlx5_flow_meter_policy *mtr_policy)
7532 {
7533 	const struct mlx5_flow_driver_ops *fops;
7534 
7535 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7536 	fops->destroy_policy_rules(dev, mtr_policy);
7537 }
7538 
7539 /**
7540  * Destroy the default policy table set.
7541  *
7542  * @param[in] dev
7543  *   Pointer to Ethernet device.
7544  */
7545 void
7546 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7547 {
7548 	const struct mlx5_flow_driver_ops *fops;
7549 
7550 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7551 	fops->destroy_def_policy(dev);
7552 }
7553 
7554 /**
7555  * Destroy the default policy table set.
7556  *
7557  * @param[in] dev
7558  *   Pointer to Ethernet device.
7559  *
7560  * @return
7561  *   0 on success, -1 otherwise.
7562  */
7563 int
7564 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7565 {
7566 	const struct mlx5_flow_driver_ops *fops;
7567 
7568 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7569 	return fops->create_def_policy(dev);
7570 }
7571 
7572 /**
7573  * Create the needed meter and suffix tables.
7574  *
7575  * @param[in] dev
7576  *   Pointer to Ethernet device.
7577  *
7578  * @return
7579  *   0 on success, -1 otherwise.
7580  */
7581 int
7582 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7583 			struct mlx5_flow_meter_info *fm,
7584 			uint32_t mtr_idx,
7585 			uint8_t domain_bitmap)
7586 {
7587 	const struct mlx5_flow_driver_ops *fops;
7588 
7589 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7590 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7591 }
7592 
7593 /**
7594  * Destroy the meter table set.
7595  *
7596  * @param[in] dev
7597  *   Pointer to Ethernet device.
7598  * @param[in] tbl
7599  *   Pointer to the meter table set.
7600  */
7601 void
7602 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7603 			   struct mlx5_flow_meter_info *fm)
7604 {
7605 	const struct mlx5_flow_driver_ops *fops;
7606 
7607 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7608 	fops->destroy_mtr_tbls(dev, fm);
7609 }
7610 
7611 /**
7612  * Destroy the global meter drop table.
7613  *
7614  * @param[in] dev
7615  *   Pointer to Ethernet device.
7616  */
7617 void
7618 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7619 {
7620 	const struct mlx5_flow_driver_ops *fops;
7621 
7622 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7623 	fops->destroy_mtr_drop_tbls(dev);
7624 }
7625 
7626 /**
7627  * Destroy the sub policy table with RX queue.
7628  *
7629  * @param[in] dev
7630  *   Pointer to Ethernet device.
7631  * @param[in] mtr_policy
7632  *   Pointer to meter policy table.
7633  */
7634 void
7635 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7636 		struct mlx5_flow_meter_policy *mtr_policy)
7637 {
7638 	const struct mlx5_flow_driver_ops *fops;
7639 
7640 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7641 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7642 }
7643 
7644 /**
7645  * Allocate the needed aso flow meter id.
7646  *
7647  * @param[in] dev
7648  *   Pointer to Ethernet device.
7649  *
7650  * @return
7651  *   Index to aso flow meter on success, NULL otherwise.
7652  */
7653 uint32_t
7654 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7655 {
7656 	const struct mlx5_flow_driver_ops *fops;
7657 
7658 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7659 	return fops->create_meter(dev);
7660 }
7661 
7662 /**
7663  * Free the aso flow meter id.
7664  *
7665  * @param[in] dev
7666  *   Pointer to Ethernet device.
7667  * @param[in] mtr_idx
7668  *  Index to aso flow meter to be free.
7669  *
7670  * @return
7671  *   0 on success.
7672  */
7673 void
7674 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7675 {
7676 	const struct mlx5_flow_driver_ops *fops;
7677 
7678 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7679 	fops->free_meter(dev, mtr_idx);
7680 }
7681 
7682 /**
7683  * Allocate a counter.
7684  *
7685  * @param[in] dev
7686  *   Pointer to Ethernet device structure.
7687  *
7688  * @return
7689  *   Index to allocated counter  on success, 0 otherwise.
7690  */
7691 uint32_t
7692 mlx5_counter_alloc(struct rte_eth_dev *dev)
7693 {
7694 	const struct mlx5_flow_driver_ops *fops;
7695 	struct rte_flow_attr attr = { .transfer = 0 };
7696 
7697 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7698 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7699 		return fops->counter_alloc(dev);
7700 	}
7701 	DRV_LOG(ERR,
7702 		"port %u counter allocate is not supported.",
7703 		 dev->data->port_id);
7704 	return 0;
7705 }
7706 
7707 /**
7708  * Free a counter.
7709  *
7710  * @param[in] dev
7711  *   Pointer to Ethernet device structure.
7712  * @param[in] cnt
7713  *   Index to counter to be free.
7714  */
7715 void
7716 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7717 {
7718 	const struct mlx5_flow_driver_ops *fops;
7719 	struct rte_flow_attr attr = { .transfer = 0 };
7720 
7721 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7722 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7723 		fops->counter_free(dev, cnt);
7724 		return;
7725 	}
7726 	DRV_LOG(ERR,
7727 		"port %u counter free is not supported.",
7728 		 dev->data->port_id);
7729 }
7730 
7731 /**
7732  * Query counter statistics.
7733  *
7734  * @param[in] dev
7735  *   Pointer to Ethernet device structure.
7736  * @param[in] cnt
7737  *   Index to counter to query.
7738  * @param[in] clear
7739  *   Set to clear counter statistics.
7740  * @param[out] pkts
7741  *   The counter hits packets number to save.
7742  * @param[out] bytes
7743  *   The counter hits bytes number to save.
7744  *
7745  * @return
7746  *   0 on success, a negative errno value otherwise.
7747  */
7748 int
7749 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7750 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7751 {
7752 	const struct mlx5_flow_driver_ops *fops;
7753 	struct rte_flow_attr attr = { .transfer = 0 };
7754 
7755 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7756 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7757 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7758 	}
7759 	DRV_LOG(ERR,
7760 		"port %u counter query is not supported.",
7761 		 dev->data->port_id);
7762 	return -ENOTSUP;
7763 }
7764 
7765 /**
7766  * Allocate a new memory for the counter values wrapped by all the needed
7767  * management.
7768  *
7769  * @param[in] sh
7770  *   Pointer to mlx5_dev_ctx_shared object.
7771  *
7772  * @return
7773  *   0 on success, a negative errno value otherwise.
7774  */
7775 static int
7776 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7777 {
7778 	struct mlx5_devx_mkey_attr mkey_attr;
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 i;
7789 
7790 	if (pgsize == (size_t)-1) {
7791 		DRV_LOG(ERR, "Failed to get mem page size");
7792 		rte_errno = ENOMEM;
7793 		return -ENOMEM;
7794 	}
7795 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7796 	if (!mem) {
7797 		rte_errno = ENOMEM;
7798 		return -ENOMEM;
7799 	}
7800 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7801 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7802 	mem_mng->umem = mlx5_os_umem_reg(sh->cdev->ctx, mem, size,
7803 						 IBV_ACCESS_LOCAL_WRITE);
7804 	if (!mem_mng->umem) {
7805 		rte_errno = errno;
7806 		mlx5_free(mem);
7807 		return -rte_errno;
7808 	}
7809 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7810 	mkey_attr.addr = (uintptr_t)mem;
7811 	mkey_attr.size = size;
7812 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7813 	mkey_attr.pd = sh->cdev->pdn;
7814 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7815 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7816 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->cdev->ctx, &mkey_attr);
7817 	if (!mem_mng->dm) {
7818 		mlx5_os_umem_dereg(mem_mng->umem);
7819 		rte_errno = errno;
7820 		mlx5_free(mem);
7821 		return -rte_errno;
7822 	}
7823 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7824 	raw_data = (volatile struct flow_counter_stats *)mem;
7825 	for (i = 0; i < raws_n; ++i) {
7826 		mem_mng->raws[i].mem_mng = mem_mng;
7827 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7828 	}
7829 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7830 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7831 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7832 				 next);
7833 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7834 	sh->cmng.mem_mng = mem_mng;
7835 	return 0;
7836 }
7837 
7838 /**
7839  * Set the statistic memory to the new counter pool.
7840  *
7841  * @param[in] sh
7842  *   Pointer to mlx5_dev_ctx_shared object.
7843  * @param[in] pool
7844  *   Pointer to the pool to set the statistic memory.
7845  *
7846  * @return
7847  *   0 on success, a negative errno value otherwise.
7848  */
7849 static int
7850 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7851 			       struct mlx5_flow_counter_pool *pool)
7852 {
7853 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7854 	/* Resize statistic memory once used out. */
7855 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7856 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7857 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7858 		return -1;
7859 	}
7860 	rte_spinlock_lock(&pool->sl);
7861 	pool->raw = cmng->mem_mng->raws + pool->index %
7862 		    MLX5_CNT_CONTAINER_RESIZE;
7863 	rte_spinlock_unlock(&pool->sl);
7864 	pool->raw_hw = NULL;
7865 	return 0;
7866 }
7867 
7868 #define MLX5_POOL_QUERY_FREQ_US 1000000
7869 
7870 /**
7871  * Set the periodic procedure for triggering asynchronous batch queries for all
7872  * the counter pools.
7873  *
7874  * @param[in] sh
7875  *   Pointer to mlx5_dev_ctx_shared object.
7876  */
7877 void
7878 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7879 {
7880 	uint32_t pools_n, us;
7881 
7882 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7883 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7884 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7885 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7886 		sh->cmng.query_thread_on = 0;
7887 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7888 	} else {
7889 		sh->cmng.query_thread_on = 1;
7890 	}
7891 }
7892 
7893 /**
7894  * The periodic procedure for triggering asynchronous batch queries for all the
7895  * counter pools. This function is probably called by the host thread.
7896  *
7897  * @param[in] arg
7898  *   The parameter for the alarm process.
7899  */
7900 void
7901 mlx5_flow_query_alarm(void *arg)
7902 {
7903 	struct mlx5_dev_ctx_shared *sh = arg;
7904 	int ret;
7905 	uint16_t pool_index = sh->cmng.pool_index;
7906 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7907 	struct mlx5_flow_counter_pool *pool;
7908 	uint16_t n_valid;
7909 
7910 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7911 		goto set_alarm;
7912 	rte_spinlock_lock(&cmng->pool_update_sl);
7913 	pool = cmng->pools[pool_index];
7914 	n_valid = cmng->n_valid;
7915 	rte_spinlock_unlock(&cmng->pool_update_sl);
7916 	/* Set the statistic memory to the new created pool. */
7917 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7918 		goto set_alarm;
7919 	if (pool->raw_hw)
7920 		/* There is a pool query in progress. */
7921 		goto set_alarm;
7922 	pool->raw_hw =
7923 		LIST_FIRST(&sh->cmng.free_stat_raws);
7924 	if (!pool->raw_hw)
7925 		/* No free counter statistics raw memory. */
7926 		goto set_alarm;
7927 	/*
7928 	 * Identify the counters released between query trigger and query
7929 	 * handle more efficiently. The counter released in this gap period
7930 	 * should wait for a new round of query as the new arrived packets
7931 	 * will not be taken into account.
7932 	 */
7933 	pool->query_gen++;
7934 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7935 					       MLX5_COUNTERS_PER_POOL,
7936 					       NULL, NULL,
7937 					       pool->raw_hw->mem_mng->dm->id,
7938 					       (void *)(uintptr_t)
7939 					       pool->raw_hw->data,
7940 					       sh->devx_comp,
7941 					       (uint64_t)(uintptr_t)pool);
7942 	if (ret) {
7943 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7944 			" %d", pool->min_dcs->id);
7945 		pool->raw_hw = NULL;
7946 		goto set_alarm;
7947 	}
7948 	LIST_REMOVE(pool->raw_hw, next);
7949 	sh->cmng.pending_queries++;
7950 	pool_index++;
7951 	if (pool_index >= n_valid)
7952 		pool_index = 0;
7953 set_alarm:
7954 	sh->cmng.pool_index = pool_index;
7955 	mlx5_set_query_alarm(sh);
7956 }
7957 
7958 /**
7959  * Check and callback event for new aged flow in the counter pool
7960  *
7961  * @param[in] sh
7962  *   Pointer to mlx5_dev_ctx_shared object.
7963  * @param[in] pool
7964  *   Pointer to Current counter pool.
7965  */
7966 static void
7967 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7968 		   struct mlx5_flow_counter_pool *pool)
7969 {
7970 	struct mlx5_priv *priv;
7971 	struct mlx5_flow_counter *cnt;
7972 	struct mlx5_age_info *age_info;
7973 	struct mlx5_age_param *age_param;
7974 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7975 	struct mlx5_counter_stats_raw *prev = pool->raw;
7976 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7977 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7978 	uint16_t expected = AGE_CANDIDATE;
7979 	uint32_t i;
7980 
7981 	pool->time_of_last_age_check = curr_time;
7982 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7983 		cnt = MLX5_POOL_GET_CNT(pool, i);
7984 		age_param = MLX5_CNT_TO_AGE(cnt);
7985 		if (__atomic_load_n(&age_param->state,
7986 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7987 			continue;
7988 		if (cur->data[i].hits != prev->data[i].hits) {
7989 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7990 					 __ATOMIC_RELAXED);
7991 			continue;
7992 		}
7993 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7994 				       time_delta,
7995 				       __ATOMIC_RELAXED) <= age_param->timeout)
7996 			continue;
7997 		/**
7998 		 * Hold the lock first, or if between the
7999 		 * state AGE_TMOUT and tailq operation the
8000 		 * release happened, the release procedure
8001 		 * may delete a non-existent tailq node.
8002 		 */
8003 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8004 		age_info = GET_PORT_AGE_INFO(priv);
8005 		rte_spinlock_lock(&age_info->aged_sl);
8006 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8007 						AGE_TMOUT, false,
8008 						__ATOMIC_RELAXED,
8009 						__ATOMIC_RELAXED)) {
8010 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8011 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8012 		}
8013 		rte_spinlock_unlock(&age_info->aged_sl);
8014 	}
8015 	mlx5_age_event_prepare(sh);
8016 }
8017 
8018 /**
8019  * Handler for the HW respond about ready values from an asynchronous batch
8020  * query. This function is probably called by the host thread.
8021  *
8022  * @param[in] sh
8023  *   The pointer to the shared device context.
8024  * @param[in] async_id
8025  *   The Devx async ID.
8026  * @param[in] status
8027  *   The status of the completion.
8028  */
8029 void
8030 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8031 				  uint64_t async_id, int status)
8032 {
8033 	struct mlx5_flow_counter_pool *pool =
8034 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8035 	struct mlx5_counter_stats_raw *raw_to_free;
8036 	uint8_t query_gen = pool->query_gen ^ 1;
8037 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8038 	enum mlx5_counter_type cnt_type =
8039 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8040 				MLX5_COUNTER_TYPE_ORIGIN;
8041 
8042 	if (unlikely(status)) {
8043 		raw_to_free = pool->raw_hw;
8044 	} else {
8045 		raw_to_free = pool->raw;
8046 		if (pool->is_aged)
8047 			mlx5_flow_aging_check(sh, pool);
8048 		rte_spinlock_lock(&pool->sl);
8049 		pool->raw = pool->raw_hw;
8050 		rte_spinlock_unlock(&pool->sl);
8051 		/* Be sure the new raw counters data is updated in memory. */
8052 		rte_io_wmb();
8053 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8054 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8055 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8056 				     &pool->counters[query_gen], next);
8057 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8058 		}
8059 	}
8060 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8061 	pool->raw_hw = NULL;
8062 	sh->cmng.pending_queries--;
8063 }
8064 
8065 static int
8066 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8067 		    const struct flow_grp_info *grp_info,
8068 		    struct rte_flow_error *error)
8069 {
8070 	if (grp_info->transfer && grp_info->external &&
8071 	    grp_info->fdb_def_rule) {
8072 		if (group == UINT32_MAX)
8073 			return rte_flow_error_set
8074 						(error, EINVAL,
8075 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8076 						 NULL,
8077 						 "group index not supported");
8078 		*table = group + 1;
8079 	} else {
8080 		*table = group;
8081 	}
8082 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8083 	return 0;
8084 }
8085 
8086 /**
8087  * Translate the rte_flow group index to HW table value.
8088  *
8089  * If tunnel offload is disabled, all group ids converted to flow table
8090  * id using the standard method.
8091  * If tunnel offload is enabled, group id can be converted using the
8092  * standard or tunnel conversion method. Group conversion method
8093  * selection depends on flags in `grp_info` parameter:
8094  * - Internal (grp_info.external == 0) groups conversion uses the
8095  *   standard method.
8096  * - Group ids in JUMP action converted with the tunnel conversion.
8097  * - Group id in rule attribute conversion depends on a rule type and
8098  *   group id value:
8099  *   ** non zero group attributes converted with the tunnel method
8100  *   ** zero group attribute in non-tunnel rule is converted using the
8101  *      standard method - there's only one root table
8102  *   ** zero group attribute in steer tunnel rule is converted with the
8103  *      standard method - single root table
8104  *   ** zero group attribute in match tunnel rule is a special OvS
8105  *      case: that value is used for portability reasons. That group
8106  *      id is converted with the tunnel conversion method.
8107  *
8108  * @param[in] dev
8109  *   Port device
8110  * @param[in] tunnel
8111  *   PMD tunnel offload object
8112  * @param[in] group
8113  *   rte_flow group index value.
8114  * @param[out] table
8115  *   HW table value.
8116  * @param[in] grp_info
8117  *   flags used for conversion
8118  * @param[out] error
8119  *   Pointer to error structure.
8120  *
8121  * @return
8122  *   0 on success, a negative errno value otherwise and rte_errno is set.
8123  */
8124 int
8125 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8126 			 const struct mlx5_flow_tunnel *tunnel,
8127 			 uint32_t group, uint32_t *table,
8128 			 const struct flow_grp_info *grp_info,
8129 			 struct rte_flow_error *error)
8130 {
8131 	int ret;
8132 	bool standard_translation;
8133 
8134 	if (!grp_info->skip_scale && grp_info->external &&
8135 	    group < MLX5_MAX_TABLES_EXTERNAL)
8136 		group *= MLX5_FLOW_TABLE_FACTOR;
8137 	if (is_tunnel_offload_active(dev)) {
8138 		standard_translation = !grp_info->external ||
8139 					grp_info->std_tbl_fix;
8140 	} else {
8141 		standard_translation = true;
8142 	}
8143 	DRV_LOG(DEBUG,
8144 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8145 		dev->data->port_id, group, grp_info->transfer,
8146 		grp_info->external, grp_info->fdb_def_rule,
8147 		standard_translation ? "STANDARD" : "TUNNEL");
8148 	if (standard_translation)
8149 		ret = flow_group_to_table(dev->data->port_id, group, table,
8150 					  grp_info, error);
8151 	else
8152 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8153 						      table, error);
8154 
8155 	return ret;
8156 }
8157 
8158 /**
8159  * Discover availability of metadata reg_c's.
8160  *
8161  * Iteratively use test flows to check availability.
8162  *
8163  * @param[in] dev
8164  *   Pointer to the Ethernet device structure.
8165  *
8166  * @return
8167  *   0 on success, a negative errno value otherwise and rte_errno is set.
8168  */
8169 int
8170 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8171 {
8172 	struct mlx5_priv *priv = dev->data->dev_private;
8173 	enum modify_reg idx;
8174 	int n = 0;
8175 
8176 	/* reg_c[0] and reg_c[1] are reserved. */
8177 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8178 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8179 	/* Discover availability of other reg_c's. */
8180 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8181 		struct rte_flow_attr attr = {
8182 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8183 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8184 			.ingress = 1,
8185 		};
8186 		struct rte_flow_item items[] = {
8187 			[0] = {
8188 				.type = RTE_FLOW_ITEM_TYPE_END,
8189 			},
8190 		};
8191 		struct rte_flow_action actions[] = {
8192 			[0] = {
8193 				.type = (enum rte_flow_action_type)
8194 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8195 				.conf = &(struct mlx5_flow_action_copy_mreg){
8196 					.src = REG_C_1,
8197 					.dst = idx,
8198 				},
8199 			},
8200 			[1] = {
8201 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8202 				.conf = &(struct rte_flow_action_jump){
8203 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8204 				},
8205 			},
8206 			[2] = {
8207 				.type = RTE_FLOW_ACTION_TYPE_END,
8208 			},
8209 		};
8210 		uint32_t flow_idx;
8211 		struct rte_flow *flow;
8212 		struct rte_flow_error error;
8213 
8214 		if (!priv->config.dv_flow_en)
8215 			break;
8216 		/* Create internal flow, validation skips copy action. */
8217 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8218 					items, actions, false, &error);
8219 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8220 				      flow_idx);
8221 		if (!flow)
8222 			continue;
8223 		priv->sh->flow_mreg_c[n++] = idx;
8224 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8225 	}
8226 	for (; n < MLX5_MREG_C_NUM; ++n)
8227 		priv->sh->flow_mreg_c[n] = REG_NON;
8228 	priv->sh->metadata_regc_check_flag = 1;
8229 	return 0;
8230 }
8231 
8232 int
8233 save_dump_file(const uint8_t *data, uint32_t size,
8234 	uint32_t type, uint64_t id, void *arg, FILE *file)
8235 {
8236 	char line[BUF_SIZE];
8237 	uint32_t out = 0;
8238 	uint32_t k;
8239 	uint32_t actions_num;
8240 	struct rte_flow_query_count *count;
8241 
8242 	memset(line, 0, BUF_SIZE);
8243 	switch (type) {
8244 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8245 		actions_num = *(uint32_t *)(arg);
8246 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8247 				type, id, actions_num);
8248 		break;
8249 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8250 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8251 				type, id);
8252 		break;
8253 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8254 		count = (struct rte_flow_query_count *)arg;
8255 		fprintf(file,
8256 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8257 			type, id, count->hits, count->bytes);
8258 		return 0;
8259 	default:
8260 		return -1;
8261 	}
8262 
8263 	for (k = 0; k < size; k++) {
8264 		/* Make sure we do not overrun the line buffer length. */
8265 		if (out >= BUF_SIZE - 4) {
8266 			line[out] = '\0';
8267 			break;
8268 		}
8269 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8270 				(data[k]) & 0xff);
8271 	}
8272 	fprintf(file, "%s\n", line);
8273 	return 0;
8274 }
8275 
8276 int
8277 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8278 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8279 {
8280 	struct rte_flow_action action[2];
8281 	enum mlx5_flow_drv_type ftype;
8282 	const struct mlx5_flow_driver_ops *fops;
8283 
8284 	if (!flow) {
8285 		return rte_flow_error_set(error, ENOENT,
8286 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8287 				NULL,
8288 				"invalid flow handle");
8289 	}
8290 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8291 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8292 	if (flow->counter) {
8293 		memset(count, 0, sizeof(struct rte_flow_query_count));
8294 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8295 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8296 						ftype < MLX5_FLOW_TYPE_MAX);
8297 		fops = flow_get_drv_ops(ftype);
8298 		return fops->query(dev, flow, action, count, error);
8299 	}
8300 	return -1;
8301 }
8302 
8303 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8304 /**
8305  * Dump flow ipool data to file
8306  *
8307  * @param[in] dev
8308  *   The pointer to Ethernet device.
8309  * @param[in] file
8310  *   A pointer to a file for output.
8311  * @param[out] error
8312  *   Perform verbose error reporting if not NULL. PMDs initialize this
8313  *   structure in case of error only.
8314  * @return
8315  *   0 on success, a negative value otherwise.
8316  */
8317 int
8318 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8319 	struct rte_flow *flow, FILE *file,
8320 	struct rte_flow_error *error)
8321 {
8322 	struct mlx5_priv *priv = dev->data->dev_private;
8323 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8324 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8325 	uint32_t handle_idx;
8326 	struct mlx5_flow_handle *dh;
8327 	struct rte_flow_query_count count;
8328 	uint32_t actions_num;
8329 	const uint8_t *data;
8330 	size_t size;
8331 	uint64_t id;
8332 	uint32_t type;
8333 	void *action = NULL;
8334 
8335 	if (!flow) {
8336 		return rte_flow_error_set(error, ENOENT,
8337 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8338 				NULL,
8339 				"invalid flow handle");
8340 	}
8341 	handle_idx = flow->dev_handles;
8342 	while (handle_idx) {
8343 		dh = mlx5_ipool_get(priv->sh->ipool
8344 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8345 		if (!dh)
8346 			continue;
8347 		handle_idx = dh->next.next;
8348 
8349 		/* query counter */
8350 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8351 		flow_dv_query_count_ptr(dev, flow->counter,
8352 						&action, error);
8353 		if (action) {
8354 			id = (uint64_t)(uintptr_t)action;
8355 			if (!mlx5_flow_query_counter(dev, flow, &count, error))
8356 				save_dump_file(NULL, 0, type,
8357 						id, (void *)&count, file);
8358 		}
8359 		/* Get modify_hdr and encap_decap buf from ipools. */
8360 		encap_decap = NULL;
8361 		modify_hdr = dh->dvh.modify_hdr;
8362 
8363 		if (dh->dvh.rix_encap_decap) {
8364 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8365 						[MLX5_IPOOL_DECAP_ENCAP],
8366 						dh->dvh.rix_encap_decap);
8367 		}
8368 		if (modify_hdr) {
8369 			data = (const uint8_t *)modify_hdr->actions;
8370 			size = (size_t)(modify_hdr->actions_num) * 8;
8371 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8372 			actions_num = modify_hdr->actions_num;
8373 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8374 			save_dump_file(data, size, type, id,
8375 						(void *)(&actions_num), file);
8376 		}
8377 		if (encap_decap) {
8378 			data = encap_decap->buf;
8379 			size = encap_decap->size;
8380 			id = (uint64_t)(uintptr_t)encap_decap->action;
8381 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8382 			save_dump_file(data, size, type,
8383 						id, NULL, file);
8384 		}
8385 	}
8386 	return 0;
8387 }
8388 
8389 /**
8390  * Dump all flow's encap_decap/modify_hdr/counter data to file
8391  *
8392  * @param[in] dev
8393  *   The pointer to Ethernet device.
8394  * @param[in] file
8395  *   A pointer to a file for output.
8396  * @param[out] error
8397  *   Perform verbose error reporting if not NULL. PMDs initialize this
8398  *   structure in case of error only.
8399  * @return
8400  *   0 on success, a negative value otherwise.
8401  */
8402 static int
8403 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8404 	FILE *file, struct rte_flow_error *error)
8405 {
8406 	struct mlx5_priv *priv = dev->data->dev_private;
8407 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8408 	struct mlx5_hlist *h;
8409 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8410 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8411 	struct rte_flow_query_count count;
8412 	uint32_t actions_num;
8413 	const uint8_t *data;
8414 	size_t size;
8415 	uint64_t id;
8416 	uint32_t type;
8417 	uint32_t i;
8418 	uint32_t j;
8419 	struct mlx5_list_inconst *l_inconst;
8420 	struct mlx5_list_entry *e;
8421 	int lcore_index;
8422 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8423 	uint32_t max;
8424 	void *action;
8425 
8426 	/* encap_decap hlist is lcore_share, get global core cache. */
8427 	i = MLX5_LIST_GLOBAL;
8428 	h = sh->encaps_decaps;
8429 	if (h) {
8430 		for (j = 0; j <= h->mask; j++) {
8431 			l_inconst = &h->buckets[j].l;
8432 			if (!l_inconst || !l_inconst->cache[i])
8433 				continue;
8434 
8435 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8436 			while (e) {
8437 				encap_decap =
8438 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8439 				data = encap_decap->buf;
8440 				size = encap_decap->size;
8441 				id = (uint64_t)(uintptr_t)encap_decap->action;
8442 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8443 				save_dump_file(data, size, type,
8444 					id, NULL, file);
8445 				e = LIST_NEXT(e, next);
8446 			}
8447 		}
8448 	}
8449 
8450 	/* get modify_hdr */
8451 	h = sh->modify_cmds;
8452 	if (h) {
8453 		lcore_index = rte_lcore_index(rte_lcore_id());
8454 		if (unlikely(lcore_index == -1)) {
8455 			lcore_index = MLX5_LIST_NLCORE;
8456 			rte_spinlock_lock(&h->l_const.lcore_lock);
8457 		}
8458 		i = lcore_index;
8459 
8460 		for (j = 0; j <= h->mask; j++) {
8461 			l_inconst = &h->buckets[j].l;
8462 			if (!l_inconst || !l_inconst->cache[i])
8463 				continue;
8464 
8465 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8466 			while (e) {
8467 				modify_hdr =
8468 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8469 				data = (const uint8_t *)modify_hdr->actions;
8470 				size = (size_t)(modify_hdr->actions_num) * 8;
8471 				actions_num = modify_hdr->actions_num;
8472 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8473 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8474 				save_dump_file(data, size, type, id,
8475 						(void *)(&actions_num), file);
8476 				e = LIST_NEXT(e, next);
8477 			}
8478 		}
8479 
8480 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8481 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8482 	}
8483 
8484 	/* get counter */
8485 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8486 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8487 	for (j = 1; j <= max; j++) {
8488 		action = NULL;
8489 		flow_dv_query_count_ptr(dev, j, &action, error);
8490 		if (action) {
8491 			if (!flow_dv_query_count(dev, j, &count, error)) {
8492 				type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8493 				id = (uint64_t)(uintptr_t)action;
8494 				save_dump_file(NULL, 0, type,
8495 						id, (void *)&count, file);
8496 			}
8497 		}
8498 	}
8499 	return 0;
8500 }
8501 #endif
8502 
8503 /**
8504  * Dump flow raw hw data to file
8505  *
8506  * @param[in] dev
8507  *    The pointer to Ethernet device.
8508  * @param[in] file
8509  *   A pointer to a file for output.
8510  * @param[out] error
8511  *   Perform verbose error reporting if not NULL. PMDs initialize this
8512  *   structure in case of error only.
8513  * @return
8514  *   0 on success, a nagative value otherwise.
8515  */
8516 int
8517 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8518 		   FILE *file,
8519 		   struct rte_flow_error *error __rte_unused)
8520 {
8521 	struct mlx5_priv *priv = dev->data->dev_private;
8522 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8523 	uint32_t handle_idx;
8524 	int ret;
8525 	struct mlx5_flow_handle *dh;
8526 	struct rte_flow *flow;
8527 
8528 	if (!priv->config.dv_flow_en) {
8529 		if (fputs("device dv flow disabled\n", file) <= 0)
8530 			return -errno;
8531 		return -ENOTSUP;
8532 	}
8533 
8534 	/* dump all */
8535 	if (!flow_idx) {
8536 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8537 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8538 			return -EINVAL;
8539 #endif
8540 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8541 					sh->rx_domain,
8542 					sh->tx_domain, file);
8543 	}
8544 	/* dump one */
8545 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8546 			(uintptr_t)(void *)flow_idx);
8547 	if (!flow)
8548 		return -EINVAL;
8549 
8550 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8551 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8552 #endif
8553 	handle_idx = flow->dev_handles;
8554 	while (handle_idx) {
8555 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8556 				handle_idx);
8557 		if (!dh)
8558 			return -ENOENT;
8559 		if (dh->drv_flow) {
8560 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8561 					file);
8562 			if (ret)
8563 				return -ENOENT;
8564 		}
8565 		handle_idx = dh->next.next;
8566 	}
8567 	return 0;
8568 }
8569 
8570 /**
8571  * Get aged-out flows.
8572  *
8573  * @param[in] dev
8574  *   Pointer to the Ethernet device structure.
8575  * @param[in] context
8576  *   The address of an array of pointers to the aged-out flows contexts.
8577  * @param[in] nb_countexts
8578  *   The length of context array pointers.
8579  * @param[out] error
8580  *   Perform verbose error reporting if not NULL. Initialized in case of
8581  *   error only.
8582  *
8583  * @return
8584  *   how many contexts get in success, otherwise negative errno value.
8585  *   if nb_contexts is 0, return the amount of all aged contexts.
8586  *   if nb_contexts is not 0 , return the amount of aged flows reported
8587  *   in the context array.
8588  */
8589 int
8590 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8591 			uint32_t nb_contexts, struct rte_flow_error *error)
8592 {
8593 	const struct mlx5_flow_driver_ops *fops;
8594 	struct rte_flow_attr attr = { .transfer = 0 };
8595 
8596 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8597 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8598 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8599 						    error);
8600 	}
8601 	DRV_LOG(ERR,
8602 		"port %u get aged flows is not supported.",
8603 		 dev->data->port_id);
8604 	return -ENOTSUP;
8605 }
8606 
8607 /* Wrapper for driver action_validate op callback */
8608 static int
8609 flow_drv_action_validate(struct rte_eth_dev *dev,
8610 			 const struct rte_flow_indir_action_conf *conf,
8611 			 const struct rte_flow_action *action,
8612 			 const struct mlx5_flow_driver_ops *fops,
8613 			 struct rte_flow_error *error)
8614 {
8615 	static const char err_msg[] = "indirect action validation unsupported";
8616 
8617 	if (!fops->action_validate) {
8618 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8619 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8620 				   NULL, err_msg);
8621 		return -rte_errno;
8622 	}
8623 	return fops->action_validate(dev, conf, action, error);
8624 }
8625 
8626 /**
8627  * Destroys the shared action by handle.
8628  *
8629  * @param dev
8630  *   Pointer to Ethernet device structure.
8631  * @param[in] handle
8632  *   Handle for the indirect action object to be destroyed.
8633  * @param[out] error
8634  *   Perform verbose error reporting if not NULL. PMDs initialize this
8635  *   structure in case of error only.
8636  *
8637  * @return
8638  *   0 on success, a negative errno value otherwise and rte_errno is set.
8639  *
8640  * @note: wrapper for driver action_create op callback.
8641  */
8642 static int
8643 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8644 			   struct rte_flow_action_handle *handle,
8645 			   struct rte_flow_error *error)
8646 {
8647 	static const char err_msg[] = "indirect action destruction unsupported";
8648 	struct rte_flow_attr attr = { .transfer = 0 };
8649 	const struct mlx5_flow_driver_ops *fops =
8650 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8651 
8652 	if (!fops->action_destroy) {
8653 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8654 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8655 				   NULL, err_msg);
8656 		return -rte_errno;
8657 	}
8658 	return fops->action_destroy(dev, handle, error);
8659 }
8660 
8661 /* Wrapper for driver action_destroy op callback */
8662 static int
8663 flow_drv_action_update(struct rte_eth_dev *dev,
8664 		       struct rte_flow_action_handle *handle,
8665 		       const void *update,
8666 		       const struct mlx5_flow_driver_ops *fops,
8667 		       struct rte_flow_error *error)
8668 {
8669 	static const char err_msg[] = "indirect action update unsupported";
8670 
8671 	if (!fops->action_update) {
8672 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8673 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8674 				   NULL, err_msg);
8675 		return -rte_errno;
8676 	}
8677 	return fops->action_update(dev, handle, update, error);
8678 }
8679 
8680 /* Wrapper for driver action_destroy op callback */
8681 static int
8682 flow_drv_action_query(struct rte_eth_dev *dev,
8683 		      const struct rte_flow_action_handle *handle,
8684 		      void *data,
8685 		      const struct mlx5_flow_driver_ops *fops,
8686 		      struct rte_flow_error *error)
8687 {
8688 	static const char err_msg[] = "indirect action query unsupported";
8689 
8690 	if (!fops->action_query) {
8691 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8692 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8693 				   NULL, err_msg);
8694 		return -rte_errno;
8695 	}
8696 	return fops->action_query(dev, handle, data, error);
8697 }
8698 
8699 /**
8700  * Create indirect action for reuse in multiple flow rules.
8701  *
8702  * @param dev
8703  *   Pointer to Ethernet device structure.
8704  * @param conf
8705  *   Pointer to indirect action object configuration.
8706  * @param[in] action
8707  *   Action configuration for indirect action object creation.
8708  * @param[out] error
8709  *   Perform verbose error reporting if not NULL. PMDs initialize this
8710  *   structure in case of error only.
8711  * @return
8712  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8713  */
8714 static struct rte_flow_action_handle *
8715 mlx5_action_handle_create(struct rte_eth_dev *dev,
8716 			  const struct rte_flow_indir_action_conf *conf,
8717 			  const struct rte_flow_action *action,
8718 			  struct rte_flow_error *error)
8719 {
8720 	static const char err_msg[] = "indirect action creation unsupported";
8721 	struct rte_flow_attr attr = { .transfer = 0 };
8722 	const struct mlx5_flow_driver_ops *fops =
8723 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8724 
8725 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8726 		return NULL;
8727 	if (!fops->action_create) {
8728 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8729 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8730 				   NULL, err_msg);
8731 		return NULL;
8732 	}
8733 	return fops->action_create(dev, conf, action, error);
8734 }
8735 
8736 /**
8737  * Updates inplace the indirect action configuration pointed by *handle*
8738  * with the configuration provided as *update* argument.
8739  * The update of the indirect action configuration effects all flow rules
8740  * reusing the action via handle.
8741  *
8742  * @param dev
8743  *   Pointer to Ethernet device structure.
8744  * @param[in] handle
8745  *   Handle for the indirect action to be updated.
8746  * @param[in] update
8747  *   Action specification used to modify the action pointed by handle.
8748  *   *update* could be of same type with the action pointed by the *handle*
8749  *   handle argument, or some other structures like a wrapper, depending on
8750  *   the indirect action type.
8751  * @param[out] error
8752  *   Perform verbose error reporting if not NULL. PMDs initialize this
8753  *   structure in case of error only.
8754  *
8755  * @return
8756  *   0 on success, a negative errno value otherwise and rte_errno is set.
8757  */
8758 static int
8759 mlx5_action_handle_update(struct rte_eth_dev *dev,
8760 		struct rte_flow_action_handle *handle,
8761 		const void *update,
8762 		struct rte_flow_error *error)
8763 {
8764 	struct rte_flow_attr attr = { .transfer = 0 };
8765 	const struct mlx5_flow_driver_ops *fops =
8766 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8767 	int ret;
8768 
8769 	ret = flow_drv_action_validate(dev, NULL,
8770 			(const struct rte_flow_action *)update, fops, error);
8771 	if (ret)
8772 		return ret;
8773 	return flow_drv_action_update(dev, handle, update, fops,
8774 				      error);
8775 }
8776 
8777 /**
8778  * Query the indirect action by handle.
8779  *
8780  * This function allows retrieving action-specific data such as counters.
8781  * Data is gathered by special action which may be present/referenced in
8782  * more than one flow rule definition.
8783  *
8784  * see @RTE_FLOW_ACTION_TYPE_COUNT
8785  *
8786  * @param dev
8787  *   Pointer to Ethernet device structure.
8788  * @param[in] handle
8789  *   Handle for the indirect action to query.
8790  * @param[in, out] data
8791  *   Pointer to storage for the associated query data type.
8792  * @param[out] error
8793  *   Perform verbose error reporting if not NULL. PMDs initialize this
8794  *   structure in case of error only.
8795  *
8796  * @return
8797  *   0 on success, a negative errno value otherwise and rte_errno is set.
8798  */
8799 static int
8800 mlx5_action_handle_query(struct rte_eth_dev *dev,
8801 			 const struct rte_flow_action_handle *handle,
8802 			 void *data,
8803 			 struct rte_flow_error *error)
8804 {
8805 	struct rte_flow_attr attr = { .transfer = 0 };
8806 	const struct mlx5_flow_driver_ops *fops =
8807 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8808 
8809 	return flow_drv_action_query(dev, handle, data, fops, error);
8810 }
8811 
8812 /**
8813  * Destroy all indirect actions (shared RSS).
8814  *
8815  * @param dev
8816  *   Pointer to Ethernet device.
8817  *
8818  * @return
8819  *   0 on success, a negative errno value otherwise and rte_errno is set.
8820  */
8821 int
8822 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8823 {
8824 	struct rte_flow_error error;
8825 	struct mlx5_priv *priv = dev->data->dev_private;
8826 	struct mlx5_shared_action_rss *shared_rss;
8827 	int ret = 0;
8828 	uint32_t idx;
8829 
8830 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8831 		      priv->rss_shared_actions, idx, shared_rss, next) {
8832 		ret |= mlx5_action_handle_destroy(dev,
8833 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8834 	}
8835 	return ret;
8836 }
8837 
8838 /**
8839  * Validate existing indirect actions against current device configuration
8840  * and attach them to device resources.
8841  *
8842  * @param dev
8843  *   Pointer to Ethernet device.
8844  *
8845  * @return
8846  *   0 on success, a negative errno value otherwise and rte_errno is set.
8847  */
8848 int
8849 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8850 {
8851 	struct mlx5_priv *priv = dev->data->dev_private;
8852 	struct mlx5_indexed_pool *ipool =
8853 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8854 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8855 	int ret = 0;
8856 	uint32_t idx;
8857 
8858 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8859 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8860 		const char *message;
8861 		uint32_t queue_idx;
8862 
8863 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8864 					       ind_tbl->queues_n,
8865 					       &message, &queue_idx);
8866 		if (ret != 0) {
8867 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8868 				dev->data->port_id, ind_tbl->queues[queue_idx],
8869 				message);
8870 			break;
8871 		}
8872 	}
8873 	if (ret != 0)
8874 		return ret;
8875 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8876 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8877 
8878 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8879 		if (ret != 0) {
8880 			DRV_LOG(ERR, "Port %u could not attach "
8881 				"indirection table obj %p",
8882 				dev->data->port_id, (void *)ind_tbl);
8883 			goto error;
8884 		}
8885 	}
8886 	return 0;
8887 error:
8888 	shared_rss_last = shared_rss;
8889 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8890 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8891 
8892 		if (shared_rss == shared_rss_last)
8893 			break;
8894 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8895 			DRV_LOG(CRIT, "Port %u could not detach "
8896 				"indirection table obj %p on rollback",
8897 				dev->data->port_id, (void *)ind_tbl);
8898 	}
8899 	return ret;
8900 }
8901 
8902 /**
8903  * Detach indirect actions of the device from its resources.
8904  *
8905  * @param dev
8906  *   Pointer to Ethernet device.
8907  *
8908  * @return
8909  *   0 on success, a negative errno value otherwise and rte_errno is set.
8910  */
8911 int
8912 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8913 {
8914 	struct mlx5_priv *priv = dev->data->dev_private;
8915 	struct mlx5_indexed_pool *ipool =
8916 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8917 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8918 	int ret = 0;
8919 	uint32_t idx;
8920 
8921 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8922 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8923 
8924 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8925 		if (ret != 0) {
8926 			DRV_LOG(ERR, "Port %u could not detach "
8927 				"indirection table obj %p",
8928 				dev->data->port_id, (void *)ind_tbl);
8929 			goto error;
8930 		}
8931 	}
8932 	return 0;
8933 error:
8934 	shared_rss_last = shared_rss;
8935 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8936 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8937 
8938 		if (shared_rss == shared_rss_last)
8939 			break;
8940 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8941 			DRV_LOG(CRIT, "Port %u could not attach "
8942 				"indirection table obj %p on rollback",
8943 				dev->data->port_id, (void *)ind_tbl);
8944 	}
8945 	return ret;
8946 }
8947 
8948 #ifndef HAVE_MLX5DV_DR
8949 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8950 #else
8951 #define MLX5_DOMAIN_SYNC_FLOW \
8952 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8953 #endif
8954 
8955 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8956 {
8957 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8958 	const struct mlx5_flow_driver_ops *fops;
8959 	int ret;
8960 	struct rte_flow_attr attr = { .transfer = 0 };
8961 
8962 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8963 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8964 	if (ret > 0)
8965 		ret = -ret;
8966 	return ret;
8967 }
8968 
8969 const struct mlx5_flow_tunnel *
8970 mlx5_get_tof(const struct rte_flow_item *item,
8971 	     const struct rte_flow_action *action,
8972 	     enum mlx5_tof_rule_type *rule_type)
8973 {
8974 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8975 		if (item->type == (typeof(item->type))
8976 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8977 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8978 			return flow_items_to_tunnel(item);
8979 		}
8980 	}
8981 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8982 		if (action->type == (typeof(action->type))
8983 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8984 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8985 			return flow_actions_to_tunnel(action);
8986 		}
8987 	}
8988 	return NULL;
8989 }
8990 
8991 /**
8992  * tunnel offload functionalilty is defined for DV environment only
8993  */
8994 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8995 __extension__
8996 union tunnel_offload_mark {
8997 	uint32_t val;
8998 	struct {
8999 		uint32_t app_reserve:8;
9000 		uint32_t table_id:15;
9001 		uint32_t transfer:1;
9002 		uint32_t _unused_:8;
9003 	};
9004 };
9005 
9006 static bool
9007 mlx5_access_tunnel_offload_db
9008 	(struct rte_eth_dev *dev,
9009 	 bool (*match)(struct rte_eth_dev *,
9010 		       struct mlx5_flow_tunnel *, const void *),
9011 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9012 	 void (*miss)(struct rte_eth_dev *, void *),
9013 	 void *ctx, bool lock_op);
9014 
9015 static int
9016 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9017 			     struct rte_flow *flow,
9018 			     const struct rte_flow_attr *attr,
9019 			     const struct rte_flow_action *app_actions,
9020 			     uint32_t flow_idx,
9021 			     const struct mlx5_flow_tunnel *tunnel,
9022 			     struct tunnel_default_miss_ctx *ctx,
9023 			     struct rte_flow_error *error)
9024 {
9025 	struct mlx5_priv *priv = dev->data->dev_private;
9026 	struct mlx5_flow *dev_flow;
9027 	struct rte_flow_attr miss_attr = *attr;
9028 	const struct rte_flow_item miss_items[2] = {
9029 		{
9030 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9031 			.spec = NULL,
9032 			.last = NULL,
9033 			.mask = NULL
9034 		},
9035 		{
9036 			.type = RTE_FLOW_ITEM_TYPE_END,
9037 			.spec = NULL,
9038 			.last = NULL,
9039 			.mask = NULL
9040 		}
9041 	};
9042 	union tunnel_offload_mark mark_id;
9043 	struct rte_flow_action_mark miss_mark;
9044 	struct rte_flow_action miss_actions[3] = {
9045 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9046 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9047 	};
9048 	const struct rte_flow_action_jump *jump_data;
9049 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9050 	struct flow_grp_info grp_info = {
9051 		.external = 1,
9052 		.transfer = attr->transfer,
9053 		.fdb_def_rule = !!priv->fdb_def_rule,
9054 		.std_tbl_fix = 0,
9055 	};
9056 	int ret;
9057 
9058 	if (!attr->transfer) {
9059 		uint32_t q_size;
9060 
9061 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9062 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9063 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9064 					 0, SOCKET_ID_ANY);
9065 		if (!ctx->queue)
9066 			return rte_flow_error_set
9067 				(error, ENOMEM,
9068 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9069 				NULL, "invalid default miss RSS");
9070 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9071 		ctx->action_rss.level = 0,
9072 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9073 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9074 		ctx->action_rss.queue_num = priv->reta_idx_n,
9075 		ctx->action_rss.key = priv->rss_conf.rss_key,
9076 		ctx->action_rss.queue = ctx->queue;
9077 		if (!priv->reta_idx_n || !priv->rxqs_n)
9078 			return rte_flow_error_set
9079 				(error, EINVAL,
9080 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9081 				NULL, "invalid port configuration");
9082 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9083 			ctx->action_rss.types = 0;
9084 		for (i = 0; i != priv->reta_idx_n; ++i)
9085 			ctx->queue[i] = (*priv->reta_idx)[i];
9086 	} else {
9087 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9088 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9089 	}
9090 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9091 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9092 	jump_data = app_actions->conf;
9093 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9094 	miss_attr.group = jump_data->group;
9095 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9096 				       &flow_table, &grp_info, error);
9097 	if (ret)
9098 		return rte_flow_error_set(error, EINVAL,
9099 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9100 					  NULL, "invalid tunnel id");
9101 	mark_id.app_reserve = 0;
9102 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9103 	mark_id.transfer = !!attr->transfer;
9104 	mark_id._unused_ = 0;
9105 	miss_mark.id = mark_id.val;
9106 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9107 				    miss_items, miss_actions, flow_idx, error);
9108 	if (!dev_flow)
9109 		return -rte_errno;
9110 	dev_flow->flow = flow;
9111 	dev_flow->external = true;
9112 	dev_flow->tunnel = tunnel;
9113 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9114 	/* Subflow object was created, we must include one in the list. */
9115 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9116 		      dev_flow->handle, next);
9117 	DRV_LOG(DEBUG,
9118 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9119 		dev->data->port_id, tunnel->app_tunnel.type,
9120 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9121 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9122 				  miss_actions, error);
9123 	if (!ret)
9124 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9125 						  error);
9126 
9127 	return ret;
9128 }
9129 
9130 static const struct mlx5_flow_tbl_data_entry  *
9131 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9132 {
9133 	struct mlx5_priv *priv = dev->data->dev_private;
9134 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9135 	struct mlx5_list_entry *he;
9136 	union tunnel_offload_mark mbits = { .val = mark };
9137 	union mlx5_flow_tbl_key table_key = {
9138 		{
9139 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9140 			.id = 0,
9141 			.reserved = 0,
9142 			.dummy = 0,
9143 			.is_fdb = !!mbits.transfer,
9144 			.is_egress = 0,
9145 		}
9146 	};
9147 	struct mlx5_flow_cb_ctx ctx = {
9148 		.data = &table_key.v64,
9149 	};
9150 
9151 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9152 	return he ?
9153 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9154 }
9155 
9156 static void
9157 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9158 				   struct mlx5_list_entry *entry)
9159 {
9160 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9161 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9162 
9163 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9164 			tunnel_flow_tbl_to_id(tte->flow_table));
9165 	mlx5_free(tte);
9166 }
9167 
9168 static int
9169 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9170 				  struct mlx5_list_entry *entry, void *cb_ctx)
9171 {
9172 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9173 	union tunnel_tbl_key tbl = {
9174 		.val = *(uint64_t *)(ctx->data),
9175 	};
9176 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9177 
9178 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9179 }
9180 
9181 static struct mlx5_list_entry *
9182 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9183 {
9184 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9185 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9186 	struct tunnel_tbl_entry *tte;
9187 	union tunnel_tbl_key tbl = {
9188 		.val = *(uint64_t *)(ctx->data),
9189 	};
9190 
9191 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9192 			  sizeof(*tte), 0,
9193 			  SOCKET_ID_ANY);
9194 	if (!tte)
9195 		goto err;
9196 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9197 			  &tte->flow_table);
9198 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9199 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9200 			tte->flow_table);
9201 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9202 				tte->flow_table);
9203 		goto err;
9204 	} else if (!tte->flow_table) {
9205 		goto err;
9206 	}
9207 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9208 	tte->tunnel_id = tbl.tunnel_id;
9209 	tte->group = tbl.group;
9210 	return &tte->hash;
9211 err:
9212 	if (tte)
9213 		mlx5_free(tte);
9214 	return NULL;
9215 }
9216 
9217 static struct mlx5_list_entry *
9218 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9219 				  struct mlx5_list_entry *oentry,
9220 				  void *cb_ctx __rte_unused)
9221 {
9222 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9223 						   0, SOCKET_ID_ANY);
9224 
9225 	if (!tte)
9226 		return NULL;
9227 	memcpy(tte, oentry, sizeof(*tte));
9228 	return &tte->hash;
9229 }
9230 
9231 static void
9232 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9233 				       struct mlx5_list_entry *entry)
9234 {
9235 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9236 
9237 	mlx5_free(tte);
9238 }
9239 
9240 static uint32_t
9241 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9242 				const struct mlx5_flow_tunnel *tunnel,
9243 				uint32_t group, uint32_t *table,
9244 				struct rte_flow_error *error)
9245 {
9246 	struct mlx5_list_entry *he;
9247 	struct tunnel_tbl_entry *tte;
9248 	union tunnel_tbl_key key = {
9249 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9250 		.group = group
9251 	};
9252 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9253 	struct mlx5_hlist *group_hash;
9254 	struct mlx5_flow_cb_ctx ctx = {
9255 		.data = &key.val,
9256 	};
9257 
9258 	group_hash = tunnel ? tunnel->groups : thub->groups;
9259 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9260 	if (!he)
9261 		return rte_flow_error_set(error, EINVAL,
9262 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9263 					  NULL,
9264 					  "tunnel group index not supported");
9265 	tte = container_of(he, typeof(*tte), hash);
9266 	*table = tte->flow_table;
9267 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9268 		dev->data->port_id, key.tunnel_id, group, *table);
9269 	return 0;
9270 }
9271 
9272 static void
9273 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9274 		      struct mlx5_flow_tunnel *tunnel)
9275 {
9276 	struct mlx5_priv *priv = dev->data->dev_private;
9277 	struct mlx5_indexed_pool *ipool;
9278 
9279 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9280 		dev->data->port_id, tunnel->tunnel_id);
9281 	LIST_REMOVE(tunnel, chain);
9282 	mlx5_hlist_destroy(tunnel->groups);
9283 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9284 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9285 }
9286 
9287 static bool
9288 mlx5_access_tunnel_offload_db
9289 	(struct rte_eth_dev *dev,
9290 	 bool (*match)(struct rte_eth_dev *,
9291 		       struct mlx5_flow_tunnel *, const void *),
9292 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9293 	 void (*miss)(struct rte_eth_dev *, void *),
9294 	 void *ctx, bool lock_op)
9295 {
9296 	bool verdict = false;
9297 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9298 	struct mlx5_flow_tunnel *tunnel;
9299 
9300 	rte_spinlock_lock(&thub->sl);
9301 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9302 		verdict = match(dev, tunnel, (const void *)ctx);
9303 		if (verdict)
9304 			break;
9305 	}
9306 	if (!lock_op)
9307 		rte_spinlock_unlock(&thub->sl);
9308 	if (verdict && hit)
9309 		hit(dev, tunnel, ctx);
9310 	if (!verdict && miss)
9311 		miss(dev, ctx);
9312 	if (lock_op)
9313 		rte_spinlock_unlock(&thub->sl);
9314 
9315 	return verdict;
9316 }
9317 
9318 struct tunnel_db_find_tunnel_id_ctx {
9319 	uint32_t tunnel_id;
9320 	struct mlx5_flow_tunnel *tunnel;
9321 };
9322 
9323 static bool
9324 find_tunnel_id_match(struct rte_eth_dev *dev,
9325 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9326 {
9327 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9328 
9329 	RTE_SET_USED(dev);
9330 	return tunnel->tunnel_id == ctx->tunnel_id;
9331 }
9332 
9333 static void
9334 find_tunnel_id_hit(struct rte_eth_dev *dev,
9335 		   struct mlx5_flow_tunnel *tunnel, void *x)
9336 {
9337 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9338 	RTE_SET_USED(dev);
9339 	ctx->tunnel = tunnel;
9340 }
9341 
9342 static struct mlx5_flow_tunnel *
9343 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9344 {
9345 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9346 		.tunnel_id = id,
9347 	};
9348 
9349 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9350 				      find_tunnel_id_hit, NULL, &ctx, true);
9351 
9352 	return ctx.tunnel;
9353 }
9354 
9355 static struct mlx5_flow_tunnel *
9356 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9357 			  const struct rte_flow_tunnel *app_tunnel)
9358 {
9359 	struct mlx5_priv *priv = dev->data->dev_private;
9360 	struct mlx5_indexed_pool *ipool;
9361 	struct mlx5_flow_tunnel *tunnel;
9362 	uint32_t id;
9363 
9364 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9365 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9366 	if (!tunnel)
9367 		return NULL;
9368 	if (id >= MLX5_MAX_TUNNELS) {
9369 		mlx5_ipool_free(ipool, id);
9370 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9371 		return NULL;
9372 	}
9373 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9374 					   priv->sh,
9375 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9376 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9377 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9378 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9379 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9380 	if (!tunnel->groups) {
9381 		mlx5_ipool_free(ipool, id);
9382 		return NULL;
9383 	}
9384 	/* initiate new PMD tunnel */
9385 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9386 	tunnel->tunnel_id = id;
9387 	tunnel->action.type = (typeof(tunnel->action.type))
9388 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9389 	tunnel->action.conf = tunnel;
9390 	tunnel->item.type = (typeof(tunnel->item.type))
9391 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9392 	tunnel->item.spec = tunnel;
9393 	tunnel->item.last = NULL;
9394 	tunnel->item.mask = NULL;
9395 
9396 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9397 		dev->data->port_id, tunnel->tunnel_id);
9398 
9399 	return tunnel;
9400 }
9401 
9402 struct tunnel_db_get_tunnel_ctx {
9403 	const struct rte_flow_tunnel *app_tunnel;
9404 	struct mlx5_flow_tunnel *tunnel;
9405 };
9406 
9407 static bool get_tunnel_match(struct rte_eth_dev *dev,
9408 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9409 {
9410 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9411 
9412 	RTE_SET_USED(dev);
9413 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9414 		       sizeof(*ctx->app_tunnel));
9415 }
9416 
9417 static void get_tunnel_hit(struct rte_eth_dev *dev,
9418 			   struct mlx5_flow_tunnel *tunnel, void *x)
9419 {
9420 	/* called under tunnel spinlock protection */
9421 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9422 
9423 	RTE_SET_USED(dev);
9424 	tunnel->refctn++;
9425 	ctx->tunnel = tunnel;
9426 }
9427 
9428 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9429 {
9430 	/* called under tunnel spinlock protection */
9431 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9432 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9433 
9434 	rte_spinlock_unlock(&thub->sl);
9435 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9436 	rte_spinlock_lock(&thub->sl);
9437 	if (ctx->tunnel) {
9438 		ctx->tunnel->refctn = 1;
9439 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9440 	}
9441 }
9442 
9443 
9444 static int
9445 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9446 		     const struct rte_flow_tunnel *app_tunnel,
9447 		     struct mlx5_flow_tunnel **tunnel)
9448 {
9449 	struct tunnel_db_get_tunnel_ctx ctx = {
9450 		.app_tunnel = app_tunnel,
9451 	};
9452 
9453 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9454 				      get_tunnel_miss, &ctx, true);
9455 	*tunnel = ctx.tunnel;
9456 	return ctx.tunnel ? 0 : -ENOMEM;
9457 }
9458 
9459 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9460 {
9461 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9462 
9463 	if (!thub)
9464 		return;
9465 	if (!LIST_EMPTY(&thub->tunnels))
9466 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9467 	mlx5_hlist_destroy(thub->groups);
9468 	mlx5_free(thub);
9469 }
9470 
9471 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9472 {
9473 	int err;
9474 	struct mlx5_flow_tunnel_hub *thub;
9475 
9476 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9477 			   0, SOCKET_ID_ANY);
9478 	if (!thub)
9479 		return -ENOMEM;
9480 	LIST_INIT(&thub->tunnels);
9481 	rte_spinlock_init(&thub->sl);
9482 	thub->groups = mlx5_hlist_create("flow groups", 64,
9483 					 false, true, sh,
9484 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9485 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9486 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9487 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9488 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9489 	if (!thub->groups) {
9490 		err = -rte_errno;
9491 		goto err;
9492 	}
9493 	sh->tunnel_hub = thub;
9494 
9495 	return 0;
9496 
9497 err:
9498 	if (thub->groups)
9499 		mlx5_hlist_destroy(thub->groups);
9500 	if (thub)
9501 		mlx5_free(thub);
9502 	return err;
9503 }
9504 
9505 static inline bool
9506 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9507 			  struct rte_flow_tunnel *tunnel,
9508 			  const char *err_msg)
9509 {
9510 	err_msg = NULL;
9511 	if (!is_tunnel_offload_active(dev)) {
9512 		err_msg = "tunnel offload was not activated";
9513 		goto out;
9514 	} else if (!tunnel) {
9515 		err_msg = "no application tunnel";
9516 		goto out;
9517 	}
9518 
9519 	switch (tunnel->type) {
9520 	default:
9521 		err_msg = "unsupported tunnel type";
9522 		goto out;
9523 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9524 	case RTE_FLOW_ITEM_TYPE_GRE:
9525 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9526 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9527 		break;
9528 	}
9529 
9530 out:
9531 	return !err_msg;
9532 }
9533 
9534 static int
9535 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9536 		    struct rte_flow_tunnel *app_tunnel,
9537 		    struct rte_flow_action **actions,
9538 		    uint32_t *num_of_actions,
9539 		    struct rte_flow_error *error)
9540 {
9541 	int ret;
9542 	struct mlx5_flow_tunnel *tunnel;
9543 	const char *err_msg = NULL;
9544 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9545 
9546 	if (!verdict)
9547 		return rte_flow_error_set(error, EINVAL,
9548 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9549 					  err_msg);
9550 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9551 	if (ret < 0) {
9552 		return rte_flow_error_set(error, ret,
9553 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9554 					  "failed to initialize pmd tunnel");
9555 	}
9556 	*actions = &tunnel->action;
9557 	*num_of_actions = 1;
9558 	return 0;
9559 }
9560 
9561 static int
9562 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9563 		       struct rte_flow_tunnel *app_tunnel,
9564 		       struct rte_flow_item **items,
9565 		       uint32_t *num_of_items,
9566 		       struct rte_flow_error *error)
9567 {
9568 	int ret;
9569 	struct mlx5_flow_tunnel *tunnel;
9570 	const char *err_msg = NULL;
9571 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9572 
9573 	if (!verdict)
9574 		return rte_flow_error_set(error, EINVAL,
9575 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9576 					  err_msg);
9577 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9578 	if (ret < 0) {
9579 		return rte_flow_error_set(error, ret,
9580 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9581 					  "failed to initialize pmd tunnel");
9582 	}
9583 	*items = &tunnel->item;
9584 	*num_of_items = 1;
9585 	return 0;
9586 }
9587 
9588 struct tunnel_db_element_release_ctx {
9589 	struct rte_flow_item *items;
9590 	struct rte_flow_action *actions;
9591 	uint32_t num_elements;
9592 	struct rte_flow_error *error;
9593 	int ret;
9594 };
9595 
9596 static bool
9597 tunnel_element_release_match(struct rte_eth_dev *dev,
9598 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9599 {
9600 	const struct tunnel_db_element_release_ctx *ctx = x;
9601 
9602 	RTE_SET_USED(dev);
9603 	if (ctx->num_elements != 1)
9604 		return false;
9605 	else if (ctx->items)
9606 		return ctx->items == &tunnel->item;
9607 	else if (ctx->actions)
9608 		return ctx->actions == &tunnel->action;
9609 
9610 	return false;
9611 }
9612 
9613 static void
9614 tunnel_element_release_hit(struct rte_eth_dev *dev,
9615 			   struct mlx5_flow_tunnel *tunnel, void *x)
9616 {
9617 	struct tunnel_db_element_release_ctx *ctx = x;
9618 	ctx->ret = 0;
9619 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9620 		mlx5_flow_tunnel_free(dev, tunnel);
9621 }
9622 
9623 static void
9624 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9625 {
9626 	struct tunnel_db_element_release_ctx *ctx = x;
9627 	RTE_SET_USED(dev);
9628 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9629 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9630 				      "invalid argument");
9631 }
9632 
9633 static int
9634 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9635 		       struct rte_flow_item *pmd_items,
9636 		       uint32_t num_items, struct rte_flow_error *err)
9637 {
9638 	struct tunnel_db_element_release_ctx ctx = {
9639 		.items = pmd_items,
9640 		.actions = NULL,
9641 		.num_elements = num_items,
9642 		.error = err,
9643 	};
9644 
9645 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9646 				      tunnel_element_release_hit,
9647 				      tunnel_element_release_miss, &ctx, false);
9648 
9649 	return ctx.ret;
9650 }
9651 
9652 static int
9653 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9654 			 struct rte_flow_action *pmd_actions,
9655 			 uint32_t num_actions, struct rte_flow_error *err)
9656 {
9657 	struct tunnel_db_element_release_ctx ctx = {
9658 		.items = NULL,
9659 		.actions = pmd_actions,
9660 		.num_elements = num_actions,
9661 		.error = err,
9662 	};
9663 
9664 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9665 				      tunnel_element_release_hit,
9666 				      tunnel_element_release_miss, &ctx, false);
9667 
9668 	return ctx.ret;
9669 }
9670 
9671 static int
9672 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9673 				  struct rte_mbuf *m,
9674 				  struct rte_flow_restore_info *info,
9675 				  struct rte_flow_error *err)
9676 {
9677 	uint64_t ol_flags = m->ol_flags;
9678 	const struct mlx5_flow_tbl_data_entry *tble;
9679 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9680 
9681 	if (!is_tunnel_offload_active(dev)) {
9682 		info->flags = 0;
9683 		return 0;
9684 	}
9685 
9686 	if ((ol_flags & mask) != mask)
9687 		goto err;
9688 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9689 	if (!tble) {
9690 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9691 			dev->data->port_id, m->hash.fdir.hi);
9692 		goto err;
9693 	}
9694 	MLX5_ASSERT(tble->tunnel);
9695 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9696 	info->group_id = tble->group_id;
9697 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9698 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9699 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9700 
9701 	return 0;
9702 
9703 err:
9704 	return rte_flow_error_set(err, EINVAL,
9705 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9706 				  "failed to get restore info");
9707 }
9708 
9709 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9710 static int
9711 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9712 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9713 			   __rte_unused struct rte_flow_action **actions,
9714 			   __rte_unused uint32_t *num_of_actions,
9715 			   __rte_unused struct rte_flow_error *error)
9716 {
9717 	return -ENOTSUP;
9718 }
9719 
9720 static int
9721 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9722 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9723 		       __rte_unused struct rte_flow_item **items,
9724 		       __rte_unused uint32_t *num_of_items,
9725 		       __rte_unused struct rte_flow_error *error)
9726 {
9727 	return -ENOTSUP;
9728 }
9729 
9730 static int
9731 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9732 			      __rte_unused struct rte_flow_item *pmd_items,
9733 			      __rte_unused uint32_t num_items,
9734 			      __rte_unused struct rte_flow_error *err)
9735 {
9736 	return -ENOTSUP;
9737 }
9738 
9739 static int
9740 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9741 				__rte_unused struct rte_flow_action *pmd_action,
9742 				__rte_unused uint32_t num_actions,
9743 				__rte_unused struct rte_flow_error *err)
9744 {
9745 	return -ENOTSUP;
9746 }
9747 
9748 static int
9749 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9750 				  __rte_unused struct rte_mbuf *m,
9751 				  __rte_unused struct rte_flow_restore_info *i,
9752 				  __rte_unused struct rte_flow_error *err)
9753 {
9754 	return -ENOTSUP;
9755 }
9756 
9757 static int
9758 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9759 			     __rte_unused struct rte_flow *flow,
9760 			     __rte_unused const struct rte_flow_attr *attr,
9761 			     __rte_unused const struct rte_flow_action *actions,
9762 			     __rte_unused uint32_t flow_idx,
9763 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9764 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9765 			     __rte_unused struct rte_flow_error *error)
9766 {
9767 	return -ENOTSUP;
9768 }
9769 
9770 static struct mlx5_flow_tunnel *
9771 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9772 		    __rte_unused uint32_t id)
9773 {
9774 	return NULL;
9775 }
9776 
9777 static void
9778 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9779 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9780 {
9781 }
9782 
9783 static uint32_t
9784 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9785 				__rte_unused const struct mlx5_flow_tunnel *t,
9786 				__rte_unused uint32_t group,
9787 				__rte_unused uint32_t *table,
9788 				struct rte_flow_error *error)
9789 {
9790 	return rte_flow_error_set(error, ENOTSUP,
9791 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9792 				  "tunnel offload requires DV support");
9793 }
9794 
9795 void
9796 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9797 			__rte_unused  uint16_t port_id)
9798 {
9799 }
9800 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9801 
9802 /* Flex flow item API */
9803 static struct rte_flow_item_flex_handle *
9804 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9805 			   const struct rte_flow_item_flex_conf *conf,
9806 			   struct rte_flow_error *error)
9807 {
9808 	static const char err_msg[] = "flex item creation unsupported";
9809 	struct rte_flow_attr attr = { .transfer = 0 };
9810 	const struct mlx5_flow_driver_ops *fops =
9811 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9812 
9813 	if (!fops->item_create) {
9814 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9815 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9816 				   NULL, err_msg);
9817 		return NULL;
9818 	}
9819 	return fops->item_create(dev, conf, error);
9820 }
9821 
9822 static int
9823 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9824 			    const struct rte_flow_item_flex_handle *handle,
9825 			    struct rte_flow_error *error)
9826 {
9827 	static const char err_msg[] = "flex item release unsupported";
9828 	struct rte_flow_attr attr = { .transfer = 0 };
9829 	const struct mlx5_flow_driver_ops *fops =
9830 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9831 
9832 	if (!fops->item_release) {
9833 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9834 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9835 				   NULL, err_msg);
9836 		return -rte_errno;
9837 	}
9838 	return fops->item_release(dev, handle, error);
9839 }
9840 
9841 static void
9842 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9843 {
9844 	int ret;
9845 	struct rte_flow_error error;
9846 
9847 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9848 		char *item_name;
9849 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9850 				    sizeof(item_name),
9851 				    (void *)(uintptr_t)item->type, &error);
9852 		if (ret > 0)
9853 			printf("%s ", item_name);
9854 		else
9855 			printf("%d\n", (int)item->type);
9856 	}
9857 	printf("END\n");
9858 }
9859 
9860 static int
9861 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9862 {
9863 	const struct rte_flow_item_udp *spec = udp_item->spec;
9864 	const struct rte_flow_item_udp *mask = udp_item->mask;
9865 	uint16_t udp_dport = 0;
9866 
9867 	if (spec != NULL) {
9868 		if (!mask)
9869 			mask = &rte_flow_item_udp_mask;
9870 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9871 				mask->hdr.dst_port);
9872 	}
9873 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9874 }
9875 
9876 static const struct mlx5_flow_expand_node *
9877 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9878 		unsigned int item_idx,
9879 		const struct mlx5_flow_expand_node graph[],
9880 		const struct mlx5_flow_expand_node *node)
9881 {
9882 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9883 
9884 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9885 			node != NULL &&
9886 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9887 		/*
9888 		 * The expansion node is VXLAN and it is also the last
9889 		 * expandable item in the pattern, so need to continue
9890 		 * expansion of the inner tunnel.
9891 		 */
9892 		MLX5_ASSERT(item_idx > 0);
9893 		prev_item = pattern + item_idx - 1;
9894 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9895 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9896 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9897 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9898 	}
9899 	return node;
9900 }
9901 
9902 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9903 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9904 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9905 };
9906 
9907 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9908 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9909 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9910 	{ 9, 10, 11 }, { 12, 13, 14 },
9911 };
9912 
9913 /**
9914  * Discover the number of available flow priorities.
9915  *
9916  * @param dev
9917  *   Ethernet device.
9918  *
9919  * @return
9920  *   On success, number of available flow priorities.
9921  *   On failure, a negative errno-style code and rte_errno is set.
9922  */
9923 int
9924 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9925 {
9926 	static const uint16_t vprio[] = {8, 16};
9927 	const struct mlx5_priv *priv = dev->data->dev_private;
9928 	const struct mlx5_flow_driver_ops *fops;
9929 	enum mlx5_flow_drv_type type;
9930 	int ret;
9931 
9932 	type = mlx5_flow_os_get_type();
9933 	if (type == MLX5_FLOW_TYPE_MAX) {
9934 		type = MLX5_FLOW_TYPE_VERBS;
9935 		if (priv->sh->devx && priv->config.dv_flow_en)
9936 			type = MLX5_FLOW_TYPE_DV;
9937 	}
9938 	fops = flow_get_drv_ops(type);
9939 	if (fops->discover_priorities == NULL) {
9940 		DRV_LOG(ERR, "Priority discovery not supported");
9941 		rte_errno = ENOTSUP;
9942 		return -rte_errno;
9943 	}
9944 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9945 	if (ret < 0)
9946 		return ret;
9947 	switch (ret) {
9948 	case 8:
9949 		ret = RTE_DIM(priority_map_3);
9950 		break;
9951 	case 16:
9952 		ret = RTE_DIM(priority_map_5);
9953 		break;
9954 	default:
9955 		rte_errno = ENOTSUP;
9956 		DRV_LOG(ERR,
9957 			"port %u maximum priority: %d expected 8/16",
9958 			dev->data->port_id, ret);
9959 		return -rte_errno;
9960 	}
9961 	DRV_LOG(INFO, "port %u supported flow priorities:"
9962 		" 0-%d for ingress or egress root table,"
9963 		" 0-%d for non-root table or transfer root table.",
9964 		dev->data->port_id, ret - 2,
9965 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9966 	return ret;
9967 }
9968 
9969 /**
9970  * Adjust flow priority based on the highest layer and the request priority.
9971  *
9972  * @param[in] dev
9973  *   Pointer to the Ethernet device structure.
9974  * @param[in] priority
9975  *   The rule base priority.
9976  * @param[in] subpriority
9977  *   The priority based on the items.
9978  *
9979  * @return
9980  *   The new priority.
9981  */
9982 uint32_t
9983 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
9984 			  uint32_t subpriority)
9985 {
9986 	uint32_t res = 0;
9987 	struct mlx5_priv *priv = dev->data->dev_private;
9988 
9989 	switch (priv->sh->flow_max_priority) {
9990 	case RTE_DIM(priority_map_3):
9991 		res = priority_map_3[priority][subpriority];
9992 		break;
9993 	case RTE_DIM(priority_map_5):
9994 		res = priority_map_5[priority][subpriority];
9995 		break;
9996 	}
9997 	return  res;
9998 }
9999