xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b53d106d34b5c638f5a2cbdfee0da5bd42d4383f)
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 	case RTE_FLOW_ITEM_TYPE_GENEVE:
262 		ether_type_m = item->mask ?
263 			       ((const struct rte_flow_item_geneve *)
264 			       (item->mask))->protocol :
265 			       rte_flow_item_geneve_mask.protocol;
266 		ether_type = ((const struct rte_flow_item_geneve *)
267 			     (item->spec))->protocol;
268 		ether_type_m = rte_be_to_cpu_16(ether_type_m);
269 		ether_type = rte_be_to_cpu_16(ether_type);
270 		switch (ether_type_m & ether_type) {
271 		case RTE_ETHER_TYPE_TEB:
272 			ret = RTE_FLOW_ITEM_TYPE_ETH;
273 			break;
274 		case RTE_ETHER_TYPE_IPV4:
275 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
276 			break;
277 		case RTE_ETHER_TYPE_IPV6:
278 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
279 			break;
280 		default:
281 			ret = RTE_FLOW_ITEM_TYPE_END;
282 		}
283 		break;
284 	default:
285 		ret = RTE_FLOW_ITEM_TYPE_VOID;
286 		break;
287 	}
288 	return ret;
289 }
290 
291 static const int *
292 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
293 		const int *next_node)
294 {
295 	const struct mlx5_flow_expand_node *node = NULL;
296 	const int *next = next_node;
297 
298 	while (next && *next) {
299 		/*
300 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
301 		 * flag set, because they were not found in the flow pattern.
302 		 */
303 		node = &graph[*next];
304 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
305 			break;
306 		next = node->next;
307 	}
308 	return next;
309 }
310 
311 #define MLX5_RSS_EXP_ELT_N 16
312 
313 /**
314  * Expand RSS flows into several possible flows according to the RSS hash
315  * fields requested and the driver capabilities.
316  *
317  * @param[out] buf
318  *   Buffer to store the result expansion.
319  * @param[in] size
320  *   Buffer size in bytes. If 0, @p buf can be NULL.
321  * @param[in] pattern
322  *   User flow pattern.
323  * @param[in] types
324  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
325  * @param[in] graph
326  *   Input graph to expand @p pattern according to @p types.
327  * @param[in] graph_root_index
328  *   Index of root node in @p graph, typically 0.
329  *
330  * @return
331  *   A positive value representing the size of @p buf in bytes regardless of
332  *   @p size on success, a negative errno value otherwise and rte_errno is
333  *   set, the following errors are defined:
334  *
335  *   -E2BIG: graph-depth @p graph is too deep.
336  *   -EINVAL: @p size has not enough space for expanded pattern.
337  */
338 static int
339 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
340 		     const struct rte_flow_item *pattern, uint64_t types,
341 		     const struct mlx5_flow_expand_node graph[],
342 		     int graph_root_index)
343 {
344 	const struct rte_flow_item *item;
345 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
346 	const int *next_node;
347 	const int *stack[MLX5_RSS_EXP_ELT_N];
348 	int stack_pos = 0;
349 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
350 	unsigned int i, item_idx, last_expand_item_idx = 0;
351 	size_t lsize;
352 	size_t user_pattern_size = 0;
353 	void *addr = NULL;
354 	const struct mlx5_flow_expand_node *next = NULL;
355 	struct rte_flow_item missed_item;
356 	int missed = 0;
357 	int elt = 0;
358 	const struct rte_flow_item *last_expand_item = NULL;
359 
360 	memset(&missed_item, 0, sizeof(missed_item));
361 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
362 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
363 	if (lsize > size)
364 		return -EINVAL;
365 	buf->entry[0].priority = 0;
366 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
367 	buf->entries = 0;
368 	addr = buf->entry[0].pattern;
369 	for (item = pattern, item_idx = 0;
370 			item->type != RTE_FLOW_ITEM_TYPE_END;
371 			item++, item_idx++) {
372 		if (!mlx5_flow_is_rss_expandable_item(item)) {
373 			user_pattern_size += sizeof(*item);
374 			continue;
375 		}
376 		last_expand_item = item;
377 		last_expand_item_idx = item_idx;
378 		i = 0;
379 		while (node->next && node->next[i]) {
380 			next = &graph[node->next[i]];
381 			if (next->type == item->type)
382 				break;
383 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
384 				node = next;
385 				i = 0;
386 			} else {
387 				++i;
388 			}
389 		}
390 		if (next)
391 			node = next;
392 		user_pattern_size += sizeof(*item);
393 	}
394 	user_pattern_size += sizeof(*item); /* Handle END item. */
395 	lsize += user_pattern_size;
396 	if (lsize > size)
397 		return -EINVAL;
398 	/* Copy the user pattern in the first entry of the buffer. */
399 	rte_memcpy(addr, pattern, user_pattern_size);
400 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
401 	buf->entries = 1;
402 	/* Start expanding. */
403 	memset(flow_items, 0, sizeof(flow_items));
404 	user_pattern_size -= sizeof(*item);
405 	/*
406 	 * Check if the last valid item has spec set, need complete pattern,
407 	 * and the pattern can be used for expansion.
408 	 */
409 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
410 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
411 		/* Item type END indicates expansion is not required. */
412 		return lsize;
413 	}
414 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
415 		next = NULL;
416 		missed = 1;
417 		i = 0;
418 		while (node->next && node->next[i]) {
419 			next = &graph[node->next[i]];
420 			if (next->type == missed_item.type) {
421 				flow_items[0].type = missed_item.type;
422 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
423 				break;
424 			}
425 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
426 				node = next;
427 				i = 0;
428 			} else {
429 				++i;
430 			}
431 			next = NULL;
432 		}
433 	}
434 	if (next && missed) {
435 		elt = 2; /* missed item + item end. */
436 		node = next;
437 		lsize += elt * sizeof(*item) + user_pattern_size;
438 		if (lsize > size)
439 			return -EINVAL;
440 		if (node->rss_types & types) {
441 			buf->entry[buf->entries].priority = 1;
442 			buf->entry[buf->entries].pattern = addr;
443 			buf->entries++;
444 			rte_memcpy(addr, buf->entry[0].pattern,
445 				   user_pattern_size);
446 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
447 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
448 			addr = (void *)(((uintptr_t)addr) +
449 					elt * sizeof(*item));
450 		}
451 	} else if (last_expand_item != NULL) {
452 		node = mlx5_flow_expand_rss_adjust_node(pattern,
453 				last_expand_item_idx, graph, node);
454 	}
455 	memset(flow_items, 0, sizeof(flow_items));
456 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
457 			node->next);
458 	stack[stack_pos] = next_node;
459 	node = next_node ? &graph[*next_node] : NULL;
460 	while (node) {
461 		flow_items[stack_pos].type = node->type;
462 		if (node->rss_types & types) {
463 			size_t n;
464 			/*
465 			 * compute the number of items to copy from the
466 			 * expansion and copy it.
467 			 * When the stack_pos is 0, there are 1 element in it,
468 			 * plus the addition END item.
469 			 */
470 			elt = stack_pos + 2;
471 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
472 			lsize += elt * sizeof(*item) + user_pattern_size;
473 			if (lsize > size)
474 				return -EINVAL;
475 			n = elt * sizeof(*item);
476 			buf->entry[buf->entries].priority =
477 				stack_pos + 1 + missed;
478 			buf->entry[buf->entries].pattern = addr;
479 			buf->entries++;
480 			rte_memcpy(addr, buf->entry[0].pattern,
481 				   user_pattern_size);
482 			addr = (void *)(((uintptr_t)addr) +
483 					user_pattern_size);
484 			rte_memcpy(addr, &missed_item,
485 				   missed * sizeof(*item));
486 			addr = (void *)(((uintptr_t)addr) +
487 				missed * sizeof(*item));
488 			rte_memcpy(addr, flow_items, n);
489 			addr = (void *)(((uintptr_t)addr) + n);
490 		}
491 		/* Go deeper. */
492 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
493 				node->next) {
494 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
495 					node->next);
496 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
497 				rte_errno = E2BIG;
498 				return -rte_errno;
499 			}
500 			stack[stack_pos] = next_node;
501 		} else if (*(next_node + 1)) {
502 			/* Follow up with the next possibility. */
503 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
504 					++next_node);
505 		} else if (!stack_pos) {
506 			/*
507 			 * Completing the traverse over the different paths.
508 			 * The next_node is advanced to the terminator.
509 			 */
510 			++next_node;
511 		} else {
512 			/* Move to the next path. */
513 			while (stack_pos) {
514 				next_node = stack[--stack_pos];
515 				next_node++;
516 				if (*next_node)
517 					break;
518 			}
519 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
520 					next_node);
521 			stack[stack_pos] = next_node;
522 		}
523 		node = next_node && *next_node ? &graph[*next_node] : NULL;
524 	};
525 	return lsize;
526 }
527 
528 enum mlx5_expansion {
529 	MLX5_EXPANSION_ROOT,
530 	MLX5_EXPANSION_ROOT_OUTER,
531 	MLX5_EXPANSION_OUTER_ETH,
532 	MLX5_EXPANSION_OUTER_VLAN,
533 	MLX5_EXPANSION_OUTER_IPV4,
534 	MLX5_EXPANSION_OUTER_IPV4_UDP,
535 	MLX5_EXPANSION_OUTER_IPV4_TCP,
536 	MLX5_EXPANSION_OUTER_IPV6,
537 	MLX5_EXPANSION_OUTER_IPV6_UDP,
538 	MLX5_EXPANSION_OUTER_IPV6_TCP,
539 	MLX5_EXPANSION_VXLAN,
540 	MLX5_EXPANSION_STD_VXLAN,
541 	MLX5_EXPANSION_L3_VXLAN,
542 	MLX5_EXPANSION_VXLAN_GPE,
543 	MLX5_EXPANSION_GRE,
544 	MLX5_EXPANSION_NVGRE,
545 	MLX5_EXPANSION_GRE_KEY,
546 	MLX5_EXPANSION_MPLS,
547 	MLX5_EXPANSION_ETH,
548 	MLX5_EXPANSION_VLAN,
549 	MLX5_EXPANSION_IPV4,
550 	MLX5_EXPANSION_IPV4_UDP,
551 	MLX5_EXPANSION_IPV4_TCP,
552 	MLX5_EXPANSION_IPV6,
553 	MLX5_EXPANSION_IPV6_UDP,
554 	MLX5_EXPANSION_IPV6_TCP,
555 	MLX5_EXPANSION_IPV6_FRAG_EXT,
556 	MLX5_EXPANSION_GTP,
557 	MLX5_EXPANSION_GENEVE,
558 };
559 
560 /** Supported expansion of items. */
561 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
562 	[MLX5_EXPANSION_ROOT] = {
563 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
564 						  MLX5_EXPANSION_IPV4,
565 						  MLX5_EXPANSION_IPV6),
566 		.type = RTE_FLOW_ITEM_TYPE_END,
567 	},
568 	[MLX5_EXPANSION_ROOT_OUTER] = {
569 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
570 						  MLX5_EXPANSION_OUTER_IPV4,
571 						  MLX5_EXPANSION_OUTER_IPV6),
572 		.type = RTE_FLOW_ITEM_TYPE_END,
573 	},
574 	[MLX5_EXPANSION_OUTER_ETH] = {
575 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
576 		.type = RTE_FLOW_ITEM_TYPE_ETH,
577 		.rss_types = 0,
578 	},
579 	[MLX5_EXPANSION_OUTER_VLAN] = {
580 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
581 						  MLX5_EXPANSION_OUTER_IPV6),
582 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
583 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
584 	},
585 	[MLX5_EXPANSION_OUTER_IPV4] = {
586 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
587 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
588 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
589 			 MLX5_EXPANSION_GRE,
590 			 MLX5_EXPANSION_NVGRE,
591 			 MLX5_EXPANSION_IPV4,
592 			 MLX5_EXPANSION_IPV6),
593 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
594 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
595 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
596 	},
597 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
598 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
599 						  MLX5_EXPANSION_VXLAN_GPE,
600 						  MLX5_EXPANSION_MPLS,
601 						  MLX5_EXPANSION_GENEVE,
602 						  MLX5_EXPANSION_GTP),
603 		.type = RTE_FLOW_ITEM_TYPE_UDP,
604 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
605 	},
606 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
607 		.type = RTE_FLOW_ITEM_TYPE_TCP,
608 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
609 	},
610 	[MLX5_EXPANSION_OUTER_IPV6] = {
611 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
612 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
613 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
614 			 MLX5_EXPANSION_IPV4,
615 			 MLX5_EXPANSION_IPV6,
616 			 MLX5_EXPANSION_GRE,
617 			 MLX5_EXPANSION_NVGRE),
618 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
619 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
620 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
621 	},
622 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
623 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
624 						  MLX5_EXPANSION_VXLAN_GPE,
625 						  MLX5_EXPANSION_MPLS,
626 						  MLX5_EXPANSION_GENEVE,
627 						  MLX5_EXPANSION_GTP),
628 		.type = RTE_FLOW_ITEM_TYPE_UDP,
629 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
630 	},
631 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
632 		.type = RTE_FLOW_ITEM_TYPE_TCP,
633 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
634 	},
635 	[MLX5_EXPANSION_VXLAN] = {
636 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
637 						  MLX5_EXPANSION_IPV4,
638 						  MLX5_EXPANSION_IPV6),
639 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
640 	},
641 	[MLX5_EXPANSION_STD_VXLAN] = {
642 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
643 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
644 	},
645 	[MLX5_EXPANSION_L3_VXLAN] = {
646 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
647 					MLX5_EXPANSION_IPV6),
648 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
649 	},
650 	[MLX5_EXPANSION_VXLAN_GPE] = {
651 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
652 						  MLX5_EXPANSION_IPV4,
653 						  MLX5_EXPANSION_IPV6),
654 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
655 	},
656 	[MLX5_EXPANSION_GRE] = {
657 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
658 						  MLX5_EXPANSION_IPV4,
659 						  MLX5_EXPANSION_IPV6,
660 						  MLX5_EXPANSION_GRE_KEY,
661 						  MLX5_EXPANSION_MPLS),
662 		.type = RTE_FLOW_ITEM_TYPE_GRE,
663 	},
664 	[MLX5_EXPANSION_GRE_KEY] = {
665 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
666 						  MLX5_EXPANSION_IPV6,
667 						  MLX5_EXPANSION_MPLS),
668 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
669 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
670 	},
671 	[MLX5_EXPANSION_NVGRE] = {
672 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
673 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
674 	},
675 	[MLX5_EXPANSION_MPLS] = {
676 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
677 						  MLX5_EXPANSION_IPV6,
678 						  MLX5_EXPANSION_ETH),
679 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
680 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
681 	},
682 	[MLX5_EXPANSION_ETH] = {
683 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
684 		.type = RTE_FLOW_ITEM_TYPE_ETH,
685 	},
686 	[MLX5_EXPANSION_VLAN] = {
687 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
688 						  MLX5_EXPANSION_IPV6),
689 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
690 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
691 	},
692 	[MLX5_EXPANSION_IPV4] = {
693 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
694 						  MLX5_EXPANSION_IPV4_TCP),
695 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
696 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
697 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
698 	},
699 	[MLX5_EXPANSION_IPV4_UDP] = {
700 		.type = RTE_FLOW_ITEM_TYPE_UDP,
701 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
702 	},
703 	[MLX5_EXPANSION_IPV4_TCP] = {
704 		.type = RTE_FLOW_ITEM_TYPE_TCP,
705 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
706 	},
707 	[MLX5_EXPANSION_IPV6] = {
708 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
709 						  MLX5_EXPANSION_IPV6_TCP,
710 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
711 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
712 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
713 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
714 	},
715 	[MLX5_EXPANSION_IPV6_UDP] = {
716 		.type = RTE_FLOW_ITEM_TYPE_UDP,
717 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
718 	},
719 	[MLX5_EXPANSION_IPV6_TCP] = {
720 		.type = RTE_FLOW_ITEM_TYPE_TCP,
721 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
722 	},
723 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
724 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
725 	},
726 	[MLX5_EXPANSION_GTP] = {
727 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
728 						  MLX5_EXPANSION_IPV6),
729 		.type = RTE_FLOW_ITEM_TYPE_GTP,
730 	},
731 	[MLX5_EXPANSION_GENEVE] = {
732 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
733 						  MLX5_EXPANSION_IPV4,
734 						  MLX5_EXPANSION_IPV6),
735 		.type = RTE_FLOW_ITEM_TYPE_GENEVE,
736 	},
737 };
738 
739 static struct rte_flow_action_handle *
740 mlx5_action_handle_create(struct rte_eth_dev *dev,
741 			  const struct rte_flow_indir_action_conf *conf,
742 			  const struct rte_flow_action *action,
743 			  struct rte_flow_error *error);
744 static int mlx5_action_handle_destroy
745 				(struct rte_eth_dev *dev,
746 				 struct rte_flow_action_handle *handle,
747 				 struct rte_flow_error *error);
748 static int mlx5_action_handle_update
749 				(struct rte_eth_dev *dev,
750 				 struct rte_flow_action_handle *handle,
751 				 const void *update,
752 				 struct rte_flow_error *error);
753 static int mlx5_action_handle_query
754 				(struct rte_eth_dev *dev,
755 				 const struct rte_flow_action_handle *handle,
756 				 void *data,
757 				 struct rte_flow_error *error);
758 static int
759 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
760 		    struct rte_flow_tunnel *app_tunnel,
761 		    struct rte_flow_action **actions,
762 		    uint32_t *num_of_actions,
763 		    struct rte_flow_error *error);
764 static int
765 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
766 		       struct rte_flow_tunnel *app_tunnel,
767 		       struct rte_flow_item **items,
768 		       uint32_t *num_of_items,
769 		       struct rte_flow_error *error);
770 static int
771 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
772 			      struct rte_flow_item *pmd_items,
773 			      uint32_t num_items, struct rte_flow_error *err);
774 static int
775 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
776 				struct rte_flow_action *pmd_actions,
777 				uint32_t num_actions,
778 				struct rte_flow_error *err);
779 static int
780 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
781 				  struct rte_mbuf *m,
782 				  struct rte_flow_restore_info *info,
783 				  struct rte_flow_error *err);
784 static struct rte_flow_item_flex_handle *
785 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
786 			   const struct rte_flow_item_flex_conf *conf,
787 			   struct rte_flow_error *error);
788 static int
789 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
790 			    const struct rte_flow_item_flex_handle *handle,
791 			    struct rte_flow_error *error);
792 
793 static const struct rte_flow_ops mlx5_flow_ops = {
794 	.validate = mlx5_flow_validate,
795 	.create = mlx5_flow_create,
796 	.destroy = mlx5_flow_destroy,
797 	.flush = mlx5_flow_flush,
798 	.isolate = mlx5_flow_isolate,
799 	.query = mlx5_flow_query,
800 	.dev_dump = mlx5_flow_dev_dump,
801 	.get_aged_flows = mlx5_flow_get_aged_flows,
802 	.action_handle_create = mlx5_action_handle_create,
803 	.action_handle_destroy = mlx5_action_handle_destroy,
804 	.action_handle_update = mlx5_action_handle_update,
805 	.action_handle_query = mlx5_action_handle_query,
806 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
807 	.tunnel_match = mlx5_flow_tunnel_match,
808 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
809 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
810 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
811 	.flex_item_create = mlx5_flow_flex_item_create,
812 	.flex_item_release = mlx5_flow_flex_item_release,
813 };
814 
815 /* Tunnel information. */
816 struct mlx5_flow_tunnel_info {
817 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
818 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
819 };
820 
821 static struct mlx5_flow_tunnel_info tunnels_info[] = {
822 	{
823 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
824 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
825 	},
826 	{
827 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
828 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
829 	},
830 	{
831 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
832 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
833 	},
834 	{
835 		.tunnel = MLX5_FLOW_LAYER_GRE,
836 		.ptype = RTE_PTYPE_TUNNEL_GRE,
837 	},
838 	{
839 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
840 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
841 	},
842 	{
843 		.tunnel = MLX5_FLOW_LAYER_MPLS,
844 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
845 	},
846 	{
847 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
848 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
849 	},
850 	{
851 		.tunnel = MLX5_FLOW_LAYER_IPIP,
852 		.ptype = RTE_PTYPE_TUNNEL_IP,
853 	},
854 	{
855 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
856 		.ptype = RTE_PTYPE_TUNNEL_IP,
857 	},
858 	{
859 		.tunnel = MLX5_FLOW_LAYER_GTP,
860 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
861 	},
862 };
863 
864 
865 
866 /**
867  * Translate tag ID to register.
868  *
869  * @param[in] dev
870  *   Pointer to the Ethernet device structure.
871  * @param[in] feature
872  *   The feature that request the register.
873  * @param[in] id
874  *   The request register ID.
875  * @param[out] error
876  *   Error description in case of any.
877  *
878  * @return
879  *   The request register on success, a negative errno
880  *   value otherwise and rte_errno is set.
881  */
882 int
883 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
884 		     enum mlx5_feature_name feature,
885 		     uint32_t id,
886 		     struct rte_flow_error *error)
887 {
888 	struct mlx5_priv *priv = dev->data->dev_private;
889 	struct mlx5_dev_config *config = &priv->config;
890 	enum modify_reg start_reg;
891 	bool skip_mtr_reg = false;
892 
893 	switch (feature) {
894 	case MLX5_HAIRPIN_RX:
895 		return REG_B;
896 	case MLX5_HAIRPIN_TX:
897 		return REG_A;
898 	case MLX5_METADATA_RX:
899 		switch (config->dv_xmeta_en) {
900 		case MLX5_XMETA_MODE_LEGACY:
901 			return REG_B;
902 		case MLX5_XMETA_MODE_META16:
903 			return REG_C_0;
904 		case MLX5_XMETA_MODE_META32:
905 			return REG_C_1;
906 		}
907 		break;
908 	case MLX5_METADATA_TX:
909 		return REG_A;
910 	case MLX5_METADATA_FDB:
911 		switch (config->dv_xmeta_en) {
912 		case MLX5_XMETA_MODE_LEGACY:
913 			return REG_NON;
914 		case MLX5_XMETA_MODE_META16:
915 			return REG_C_0;
916 		case MLX5_XMETA_MODE_META32:
917 			return REG_C_1;
918 		}
919 		break;
920 	case MLX5_FLOW_MARK:
921 		switch (config->dv_xmeta_en) {
922 		case MLX5_XMETA_MODE_LEGACY:
923 			return REG_NON;
924 		case MLX5_XMETA_MODE_META16:
925 			return REG_C_1;
926 		case MLX5_XMETA_MODE_META32:
927 			return REG_C_0;
928 		}
929 		break;
930 	case MLX5_MTR_ID:
931 		/*
932 		 * If meter color and meter id share one register, flow match
933 		 * should use the meter color register for match.
934 		 */
935 		if (priv->mtr_reg_share)
936 			return priv->mtr_color_reg;
937 		else
938 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
939 			       REG_C_3;
940 	case MLX5_MTR_COLOR:
941 	case MLX5_ASO_FLOW_HIT:
942 	case MLX5_ASO_CONNTRACK:
943 	case MLX5_SAMPLE_ID:
944 		/* All features use the same REG_C. */
945 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
946 		return priv->mtr_color_reg;
947 	case MLX5_COPY_MARK:
948 		/*
949 		 * Metadata COPY_MARK register using is in meter suffix sub
950 		 * flow while with meter. It's safe to share the same register.
951 		 */
952 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
953 	case MLX5_APP_TAG:
954 		/*
955 		 * If meter is enable, it will engage the register for color
956 		 * match and flow match. If meter color match is not using the
957 		 * REG_C_2, need to skip the REG_C_x be used by meter color
958 		 * match.
959 		 * If meter is disable, free to use all available registers.
960 		 */
961 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
962 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
963 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
964 		if (id > (uint32_t)(REG_C_7 - start_reg))
965 			return rte_flow_error_set(error, EINVAL,
966 						  RTE_FLOW_ERROR_TYPE_ITEM,
967 						  NULL, "invalid tag id");
968 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
969 			return rte_flow_error_set(error, ENOTSUP,
970 						  RTE_FLOW_ERROR_TYPE_ITEM,
971 						  NULL, "unsupported tag id");
972 		/*
973 		 * This case means meter is using the REG_C_x great than 2.
974 		 * Take care not to conflict with meter color REG_C_x.
975 		 * If the available index REG_C_y >= REG_C_x, skip the
976 		 * color register.
977 		 */
978 		if (skip_mtr_reg && priv->sh->flow_mreg_c
979 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
980 			if (id >= (uint32_t)(REG_C_7 - start_reg))
981 				return rte_flow_error_set(error, EINVAL,
982 						       RTE_FLOW_ERROR_TYPE_ITEM,
983 							NULL, "invalid tag id");
984 			if (priv->sh->flow_mreg_c
985 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
986 				return priv->sh->flow_mreg_c
987 					       [id + 1 + start_reg - REG_C_0];
988 			return rte_flow_error_set(error, ENOTSUP,
989 						  RTE_FLOW_ERROR_TYPE_ITEM,
990 						  NULL, "unsupported tag id");
991 		}
992 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
993 	}
994 	MLX5_ASSERT(false);
995 	return rte_flow_error_set(error, EINVAL,
996 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
997 				  NULL, "invalid feature name");
998 }
999 
1000 /**
1001  * Check extensive flow metadata register support.
1002  *
1003  * @param dev
1004  *   Pointer to rte_eth_dev structure.
1005  *
1006  * @return
1007  *   True if device supports extensive flow metadata register, otherwise false.
1008  */
1009 bool
1010 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1011 {
1012 	struct mlx5_priv *priv = dev->data->dev_private;
1013 
1014 	/*
1015 	 * Having available reg_c can be regarded inclusively as supporting
1016 	 * extensive flow metadata register, which could mean,
1017 	 * - metadata register copy action by modify header.
1018 	 * - 16 modify header actions is supported.
1019 	 * - reg_c's are preserved across different domain (FDB and NIC) on
1020 	 *   packet loopback by flow lookup miss.
1021 	 */
1022 	return priv->sh->flow_mreg_c[2] != REG_NON;
1023 }
1024 
1025 /**
1026  * Get the lowest priority.
1027  *
1028  * @param[in] dev
1029  *   Pointer to the Ethernet device structure.
1030  * @param[in] attributes
1031  *   Pointer to device flow rule attributes.
1032  *
1033  * @return
1034  *   The value of lowest priority of flow.
1035  */
1036 uint32_t
1037 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1038 			  const struct rte_flow_attr *attr)
1039 {
1040 	struct mlx5_priv *priv = dev->data->dev_private;
1041 
1042 	if (!attr->group && !attr->transfer)
1043 		return priv->sh->flow_max_priority - 2;
1044 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1045 }
1046 
1047 /**
1048  * Calculate matcher priority of the flow.
1049  *
1050  * @param[in] dev
1051  *   Pointer to the Ethernet device structure.
1052  * @param[in] attr
1053  *   Pointer to device flow rule attributes.
1054  * @param[in] subpriority
1055  *   The priority based on the items.
1056  * @param[in] external
1057  *   Flow is user flow.
1058  * @return
1059  *   The matcher priority of the flow.
1060  */
1061 uint16_t
1062 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1063 			  const struct rte_flow_attr *attr,
1064 			  uint32_t subpriority, bool external)
1065 {
1066 	uint16_t priority = (uint16_t)attr->priority;
1067 	struct mlx5_priv *priv = dev->data->dev_private;
1068 
1069 	if (!attr->group && !attr->transfer) {
1070 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1071 			priority = priv->sh->flow_max_priority - 1;
1072 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1073 	} else if (!external && attr->transfer && attr->group == 0 &&
1074 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1075 		return (priv->sh->flow_max_priority - 1) * 3;
1076 	}
1077 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1078 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1079 	return priority * 3 + subpriority;
1080 }
1081 
1082 /**
1083  * Verify the @p item specifications (spec, last, mask) are compatible with the
1084  * NIC capabilities.
1085  *
1086  * @param[in] item
1087  *   Item specification.
1088  * @param[in] mask
1089  *   @p item->mask or flow default bit-masks.
1090  * @param[in] nic_mask
1091  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1092  * @param[in] size
1093  *   Bit-masks size in bytes.
1094  * @param[in] range_accepted
1095  *   True if range of values is accepted for specific fields, false otherwise.
1096  * @param[out] error
1097  *   Pointer to error structure.
1098  *
1099  * @return
1100  *   0 on success, a negative errno value otherwise and rte_errno is set.
1101  */
1102 int
1103 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1104 			  const uint8_t *mask,
1105 			  const uint8_t *nic_mask,
1106 			  unsigned int size,
1107 			  bool range_accepted,
1108 			  struct rte_flow_error *error)
1109 {
1110 	unsigned int i;
1111 
1112 	MLX5_ASSERT(nic_mask);
1113 	for (i = 0; i < size; ++i)
1114 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1115 			return rte_flow_error_set(error, ENOTSUP,
1116 						  RTE_FLOW_ERROR_TYPE_ITEM,
1117 						  item,
1118 						  "mask enables non supported"
1119 						  " bits");
1120 	if (!item->spec && (item->mask || item->last))
1121 		return rte_flow_error_set(error, EINVAL,
1122 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1123 					  "mask/last without a spec is not"
1124 					  " supported");
1125 	if (item->spec && item->last && !range_accepted) {
1126 		uint8_t spec[size];
1127 		uint8_t last[size];
1128 		unsigned int i;
1129 		int ret;
1130 
1131 		for (i = 0; i < size; ++i) {
1132 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1133 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1134 		}
1135 		ret = memcmp(spec, last, size);
1136 		if (ret != 0)
1137 			return rte_flow_error_set(error, EINVAL,
1138 						  RTE_FLOW_ERROR_TYPE_ITEM,
1139 						  item,
1140 						  "range is not valid");
1141 	}
1142 	return 0;
1143 }
1144 
1145 /**
1146  * Adjust the hash fields according to the @p flow information.
1147  *
1148  * @param[in] dev_flow.
1149  *   Pointer to the mlx5_flow.
1150  * @param[in] tunnel
1151  *   1 when the hash field is for a tunnel item.
1152  * @param[in] layer_types
1153  *   RTE_ETH_RSS_* types.
1154  * @param[in] hash_fields
1155  *   Item hash fields.
1156  *
1157  * @return
1158  *   The hash fields that should be used.
1159  */
1160 uint64_t
1161 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1162 			    int tunnel __rte_unused, uint64_t layer_types,
1163 			    uint64_t hash_fields)
1164 {
1165 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1166 	int rss_request_inner = rss_desc->level >= 2;
1167 
1168 	/* Check RSS hash level for tunnel. */
1169 	if (tunnel && rss_request_inner)
1170 		hash_fields |= IBV_RX_HASH_INNER;
1171 	else if (tunnel || rss_request_inner)
1172 		return 0;
1173 #endif
1174 	/* Check if requested layer matches RSS hash fields. */
1175 	if (!(rss_desc->types & layer_types))
1176 		return 0;
1177 	return hash_fields;
1178 }
1179 
1180 /**
1181  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1182  * if several tunnel rules are used on this queue, the tunnel ptype will be
1183  * cleared.
1184  *
1185  * @param rxq_ctrl
1186  *   Rx queue to update.
1187  */
1188 static void
1189 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1190 {
1191 	unsigned int i;
1192 	uint32_t tunnel_ptype = 0;
1193 
1194 	/* Look up for the ptype to use. */
1195 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1196 		if (!rxq_ctrl->flow_tunnels_n[i])
1197 			continue;
1198 		if (!tunnel_ptype) {
1199 			tunnel_ptype = tunnels_info[i].ptype;
1200 		} else {
1201 			tunnel_ptype = 0;
1202 			break;
1203 		}
1204 	}
1205 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1206 }
1207 
1208 /**
1209  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1210  * flow.
1211  *
1212  * @param[in] dev
1213  *   Pointer to the Ethernet device structure.
1214  * @param[in] dev_handle
1215  *   Pointer to device flow handle structure.
1216  */
1217 void
1218 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1219 		       struct mlx5_flow_handle *dev_handle)
1220 {
1221 	struct mlx5_priv *priv = dev->data->dev_private;
1222 	const int mark = dev_handle->mark;
1223 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1224 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1225 	unsigned int i;
1226 
1227 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1228 		struct mlx5_hrxq *hrxq;
1229 
1230 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1231 			      dev_handle->rix_hrxq);
1232 		if (hrxq)
1233 			ind_tbl = hrxq->ind_table;
1234 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1235 		struct mlx5_shared_action_rss *shared_rss;
1236 
1237 		shared_rss = mlx5_ipool_get
1238 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1239 			 dev_handle->rix_srss);
1240 		if (shared_rss)
1241 			ind_tbl = shared_rss->ind_tbl;
1242 	}
1243 	if (!ind_tbl)
1244 		return;
1245 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1246 		int idx = ind_tbl->queues[i];
1247 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1248 
1249 		MLX5_ASSERT(rxq_ctrl != NULL);
1250 		if (rxq_ctrl == NULL)
1251 			continue;
1252 		/*
1253 		 * To support metadata register copy on Tx loopback,
1254 		 * this must be always enabled (metadata may arive
1255 		 * from other port - not from local flows only.
1256 		 */
1257 		if (priv->config.dv_flow_en &&
1258 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1259 		    mlx5_flow_ext_mreg_supported(dev)) {
1260 			rxq_ctrl->rxq.mark = 1;
1261 			rxq_ctrl->flow_mark_n = 1;
1262 		} else if (mark) {
1263 			rxq_ctrl->rxq.mark = 1;
1264 			rxq_ctrl->flow_mark_n++;
1265 		}
1266 		if (tunnel) {
1267 			unsigned int j;
1268 
1269 			/* Increase the counter matching the flow. */
1270 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1271 				if ((tunnels_info[j].tunnel &
1272 				     dev_handle->layers) ==
1273 				    tunnels_info[j].tunnel) {
1274 					rxq_ctrl->flow_tunnels_n[j]++;
1275 					break;
1276 				}
1277 			}
1278 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1279 		}
1280 	}
1281 }
1282 
1283 /**
1284  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1285  *
1286  * @param[in] dev
1287  *   Pointer to the Ethernet device structure.
1288  * @param[in] flow
1289  *   Pointer to flow structure.
1290  */
1291 static void
1292 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1293 {
1294 	struct mlx5_priv *priv = dev->data->dev_private;
1295 	uint32_t handle_idx;
1296 	struct mlx5_flow_handle *dev_handle;
1297 
1298 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1299 		       handle_idx, dev_handle, next)
1300 		flow_drv_rxq_flags_set(dev, dev_handle);
1301 }
1302 
1303 /**
1304  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1305  * device flow if no other flow uses it with the same kind of request.
1306  *
1307  * @param dev
1308  *   Pointer to Ethernet device.
1309  * @param[in] dev_handle
1310  *   Pointer to the device flow handle structure.
1311  */
1312 static void
1313 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1314 			struct mlx5_flow_handle *dev_handle)
1315 {
1316 	struct mlx5_priv *priv = dev->data->dev_private;
1317 	const int mark = dev_handle->mark;
1318 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1319 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1320 	unsigned int i;
1321 
1322 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1323 		struct mlx5_hrxq *hrxq;
1324 
1325 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1326 			      dev_handle->rix_hrxq);
1327 		if (hrxq)
1328 			ind_tbl = hrxq->ind_table;
1329 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1330 		struct mlx5_shared_action_rss *shared_rss;
1331 
1332 		shared_rss = mlx5_ipool_get
1333 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1334 			 dev_handle->rix_srss);
1335 		if (shared_rss)
1336 			ind_tbl = shared_rss->ind_tbl;
1337 	}
1338 	if (!ind_tbl)
1339 		return;
1340 	MLX5_ASSERT(dev->data->dev_started);
1341 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1342 		int idx = ind_tbl->queues[i];
1343 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1344 
1345 		MLX5_ASSERT(rxq_ctrl != NULL);
1346 		if (rxq_ctrl == NULL)
1347 			continue;
1348 		if (priv->config.dv_flow_en &&
1349 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1350 		    mlx5_flow_ext_mreg_supported(dev)) {
1351 			rxq_ctrl->rxq.mark = 1;
1352 			rxq_ctrl->flow_mark_n = 1;
1353 		} else if (mark) {
1354 			rxq_ctrl->flow_mark_n--;
1355 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1356 		}
1357 		if (tunnel) {
1358 			unsigned int j;
1359 
1360 			/* Decrease the counter matching the flow. */
1361 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1362 				if ((tunnels_info[j].tunnel &
1363 				     dev_handle->layers) ==
1364 				    tunnels_info[j].tunnel) {
1365 					rxq_ctrl->flow_tunnels_n[j]--;
1366 					break;
1367 				}
1368 			}
1369 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1370 		}
1371 	}
1372 }
1373 
1374 /**
1375  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1376  * @p flow if no other flow uses it with the same kind of request.
1377  *
1378  * @param dev
1379  *   Pointer to Ethernet device.
1380  * @param[in] flow
1381  *   Pointer to the flow.
1382  */
1383 static void
1384 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1385 {
1386 	struct mlx5_priv *priv = dev->data->dev_private;
1387 	uint32_t handle_idx;
1388 	struct mlx5_flow_handle *dev_handle;
1389 
1390 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1391 		       handle_idx, dev_handle, next)
1392 		flow_drv_rxq_flags_trim(dev, dev_handle);
1393 }
1394 
1395 /**
1396  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1397  *
1398  * @param dev
1399  *   Pointer to Ethernet device.
1400  */
1401 static void
1402 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1403 {
1404 	struct mlx5_priv *priv = dev->data->dev_private;
1405 	unsigned int i;
1406 
1407 	for (i = 0; i != priv->rxqs_n; ++i) {
1408 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1409 		unsigned int j;
1410 
1411 		if (rxq == NULL || rxq->ctrl == NULL)
1412 			continue;
1413 		rxq->ctrl->flow_mark_n = 0;
1414 		rxq->ctrl->rxq.mark = 0;
1415 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1416 			rxq->ctrl->flow_tunnels_n[j] = 0;
1417 		rxq->ctrl->rxq.tunnel = 0;
1418 	}
1419 }
1420 
1421 /**
1422  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1423  *
1424  * @param[in] dev
1425  *   Pointer to the Ethernet device structure.
1426  */
1427 void
1428 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1429 {
1430 	struct mlx5_priv *priv = dev->data->dev_private;
1431 	unsigned int i;
1432 
1433 	for (i = 0; i != priv->rxqs_n; ++i) {
1434 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1435 		struct mlx5_rxq_data *data;
1436 
1437 		if (rxq == NULL || rxq->ctrl == NULL)
1438 			continue;
1439 		data = &rxq->ctrl->rxq;
1440 		if (!rte_flow_dynf_metadata_avail()) {
1441 			data->dynf_meta = 0;
1442 			data->flow_meta_mask = 0;
1443 			data->flow_meta_offset = -1;
1444 			data->flow_meta_port_mask = 0;
1445 		} else {
1446 			data->dynf_meta = 1;
1447 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1448 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1449 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1450 		}
1451 	}
1452 }
1453 
1454 /*
1455  * return a pointer to the desired action in the list of actions.
1456  *
1457  * @param[in] actions
1458  *   The list of actions to search the action in.
1459  * @param[in] action
1460  *   The action to find.
1461  *
1462  * @return
1463  *   Pointer to the action in the list, if found. NULL otherwise.
1464  */
1465 const struct rte_flow_action *
1466 mlx5_flow_find_action(const struct rte_flow_action *actions,
1467 		      enum rte_flow_action_type action)
1468 {
1469 	if (actions == NULL)
1470 		return NULL;
1471 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1472 		if (actions->type == action)
1473 			return actions;
1474 	return NULL;
1475 }
1476 
1477 /*
1478  * Validate the flag action.
1479  *
1480  * @param[in] action_flags
1481  *   Bit-fields that holds the actions detected until now.
1482  * @param[in] attr
1483  *   Attributes of flow that includes this action.
1484  * @param[out] error
1485  *   Pointer to error structure.
1486  *
1487  * @return
1488  *   0 on success, a negative errno value otherwise and rte_errno is set.
1489  */
1490 int
1491 mlx5_flow_validate_action_flag(uint64_t action_flags,
1492 			       const struct rte_flow_attr *attr,
1493 			       struct rte_flow_error *error)
1494 {
1495 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1496 		return rte_flow_error_set(error, EINVAL,
1497 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1498 					  "can't mark and flag in same flow");
1499 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1502 					  "can't have 2 flag"
1503 					  " actions in same flow");
1504 	if (attr->egress)
1505 		return rte_flow_error_set(error, ENOTSUP,
1506 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1507 					  "flag action not supported for "
1508 					  "egress");
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Validate the mark action.
1514  *
1515  * @param[in] action
1516  *   Pointer to the queue action.
1517  * @param[in] action_flags
1518  *   Bit-fields that holds the actions detected until now.
1519  * @param[in] attr
1520  *   Attributes of flow that includes this action.
1521  * @param[out] error
1522  *   Pointer to error structure.
1523  *
1524  * @return
1525  *   0 on success, a negative errno value otherwise and rte_errno is set.
1526  */
1527 int
1528 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1529 			       uint64_t action_flags,
1530 			       const struct rte_flow_attr *attr,
1531 			       struct rte_flow_error *error)
1532 {
1533 	const struct rte_flow_action_mark *mark = action->conf;
1534 
1535 	if (!mark)
1536 		return rte_flow_error_set(error, EINVAL,
1537 					  RTE_FLOW_ERROR_TYPE_ACTION,
1538 					  action,
1539 					  "configuration cannot be null");
1540 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1541 		return rte_flow_error_set(error, EINVAL,
1542 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1543 					  &mark->id,
1544 					  "mark id must in 0 <= id < "
1545 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1546 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1547 		return rte_flow_error_set(error, EINVAL,
1548 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1549 					  "can't flag and mark in same flow");
1550 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1551 		return rte_flow_error_set(error, EINVAL,
1552 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1553 					  "can't have 2 mark actions in same"
1554 					  " flow");
1555 	if (attr->egress)
1556 		return rte_flow_error_set(error, ENOTSUP,
1557 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1558 					  "mark action not supported for "
1559 					  "egress");
1560 	return 0;
1561 }
1562 
1563 /*
1564  * Validate the drop action.
1565  *
1566  * @param[in] action_flags
1567  *   Bit-fields that holds the actions detected until now.
1568  * @param[in] attr
1569  *   Attributes of flow that includes this action.
1570  * @param[out] error
1571  *   Pointer to error structure.
1572  *
1573  * @return
1574  *   0 on success, a negative errno value otherwise and rte_errno is set.
1575  */
1576 int
1577 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1578 			       const struct rte_flow_attr *attr,
1579 			       struct rte_flow_error *error)
1580 {
1581 	if (attr->egress)
1582 		return rte_flow_error_set(error, ENOTSUP,
1583 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1584 					  "drop action not supported for "
1585 					  "egress");
1586 	return 0;
1587 }
1588 
1589 /*
1590  * Validate the queue action.
1591  *
1592  * @param[in] action
1593  *   Pointer to the queue action.
1594  * @param[in] action_flags
1595  *   Bit-fields that holds the actions detected until now.
1596  * @param[in] dev
1597  *   Pointer to the Ethernet device structure.
1598  * @param[in] attr
1599  *   Attributes of flow that includes this action.
1600  * @param[out] error
1601  *   Pointer to error structure.
1602  *
1603  * @return
1604  *   0 on success, a negative errno value otherwise and rte_errno is set.
1605  */
1606 int
1607 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1608 				uint64_t action_flags,
1609 				struct rte_eth_dev *dev,
1610 				const struct rte_flow_attr *attr,
1611 				struct rte_flow_error *error)
1612 {
1613 	struct mlx5_priv *priv = dev->data->dev_private;
1614 	const struct rte_flow_action_queue *queue = action->conf;
1615 
1616 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1617 		return rte_flow_error_set(error, EINVAL,
1618 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1619 					  "can't have 2 fate actions in"
1620 					  " same flow");
1621 	if (!priv->rxqs_n)
1622 		return rte_flow_error_set(error, EINVAL,
1623 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1624 					  NULL, "No Rx queues configured");
1625 	if (queue->index >= priv->rxqs_n)
1626 		return rte_flow_error_set(error, EINVAL,
1627 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1628 					  &queue->index,
1629 					  "queue index out of range");
1630 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1631 		return rte_flow_error_set(error, EINVAL,
1632 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1633 					  &queue->index,
1634 					  "queue is not configured");
1635 	if (attr->egress)
1636 		return rte_flow_error_set(error, ENOTSUP,
1637 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1638 					  "queue action not supported for "
1639 					  "egress");
1640 	return 0;
1641 }
1642 
1643 /**
1644  * Validate queue numbers for device RSS.
1645  *
1646  * @param[in] dev
1647  *   Configured device.
1648  * @param[in] queues
1649  *   Array of queue numbers.
1650  * @param[in] queues_n
1651  *   Size of the @p queues array.
1652  * @param[out] error
1653  *   On error, filled with a textual error description.
1654  * @param[out] queue
1655  *   On error, filled with an offending queue index in @p queues array.
1656  *
1657  * @return
1658  *   0 on success, a negative errno code on error.
1659  */
1660 static int
1661 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1662 			 const uint16_t *queues, uint32_t queues_n,
1663 			 const char **error, uint32_t *queue_idx)
1664 {
1665 	const struct mlx5_priv *priv = dev->data->dev_private;
1666 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1667 	uint32_t i;
1668 
1669 	for (i = 0; i != queues_n; ++i) {
1670 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1671 								   queues[i]);
1672 
1673 		if (queues[i] >= priv->rxqs_n) {
1674 			*error = "queue index out of range";
1675 			*queue_idx = i;
1676 			return -EINVAL;
1677 		}
1678 		if (rxq_ctrl == NULL) {
1679 			*error =  "queue is not configured";
1680 			*queue_idx = i;
1681 			return -EINVAL;
1682 		}
1683 		if (i == 0)
1684 			rxq_type = rxq_ctrl->type;
1685 		if (rxq_type != rxq_ctrl->type) {
1686 			*error = "combining hairpin and regular RSS queues is not supported";
1687 			*queue_idx = i;
1688 			return -ENOTSUP;
1689 		}
1690 	}
1691 	return 0;
1692 }
1693 
1694 /*
1695  * Validate the rss action.
1696  *
1697  * @param[in] dev
1698  *   Pointer to the Ethernet device structure.
1699  * @param[in] action
1700  *   Pointer to the queue action.
1701  * @param[out] error
1702  *   Pointer to error structure.
1703  *
1704  * @return
1705  *   0 on success, a negative errno value otherwise and rte_errno is set.
1706  */
1707 int
1708 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1709 			 const struct rte_flow_action *action,
1710 			 struct rte_flow_error *error)
1711 {
1712 	struct mlx5_priv *priv = dev->data->dev_private;
1713 	const struct rte_flow_action_rss *rss = action->conf;
1714 	int ret;
1715 	const char *message;
1716 	uint32_t queue_idx;
1717 
1718 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1719 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1720 		return rte_flow_error_set(error, ENOTSUP,
1721 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1722 					  &rss->func,
1723 					  "RSS hash function not supported");
1724 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1725 	if (rss->level > 2)
1726 #else
1727 	if (rss->level > 1)
1728 #endif
1729 		return rte_flow_error_set(error, ENOTSUP,
1730 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1731 					  &rss->level,
1732 					  "tunnel RSS is not supported");
1733 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1734 	if (rss->key_len == 0 && rss->key != NULL)
1735 		return rte_flow_error_set(error, ENOTSUP,
1736 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1737 					  &rss->key_len,
1738 					  "RSS hash key length 0");
1739 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1740 		return rte_flow_error_set(error, ENOTSUP,
1741 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1742 					  &rss->key_len,
1743 					  "RSS hash key too small");
1744 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1745 		return rte_flow_error_set(error, ENOTSUP,
1746 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1747 					  &rss->key_len,
1748 					  "RSS hash key too large");
1749 	if (rss->queue_num > priv->config.ind_table_max_size)
1750 		return rte_flow_error_set(error, ENOTSUP,
1751 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1752 					  &rss->queue_num,
1753 					  "number of queues too large");
1754 	if (rss->types & MLX5_RSS_HF_MASK)
1755 		return rte_flow_error_set(error, ENOTSUP,
1756 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1757 					  &rss->types,
1758 					  "some RSS protocols are not"
1759 					  " supported");
1760 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1761 	    !(rss->types & RTE_ETH_RSS_IP))
1762 		return rte_flow_error_set(error, EINVAL,
1763 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1764 					  "L3 partial RSS requested but L3 RSS"
1765 					  " type not specified");
1766 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1767 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1768 		return rte_flow_error_set(error, EINVAL,
1769 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1770 					  "L4 partial RSS requested but L4 RSS"
1771 					  " type not specified");
1772 	if (!priv->rxqs_n)
1773 		return rte_flow_error_set(error, EINVAL,
1774 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1775 					  NULL, "No Rx queues configured");
1776 	if (!rss->queue_num)
1777 		return rte_flow_error_set(error, EINVAL,
1778 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1779 					  NULL, "No queues configured");
1780 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1781 				       &message, &queue_idx);
1782 	if (ret != 0) {
1783 		return rte_flow_error_set(error, -ret,
1784 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1785 					  &rss->queue[queue_idx], message);
1786 	}
1787 	return 0;
1788 }
1789 
1790 /*
1791  * Validate the rss action.
1792  *
1793  * @param[in] action
1794  *   Pointer to the queue action.
1795  * @param[in] action_flags
1796  *   Bit-fields that holds the actions detected until now.
1797  * @param[in] dev
1798  *   Pointer to the Ethernet device structure.
1799  * @param[in] attr
1800  *   Attributes of flow that includes this action.
1801  * @param[in] item_flags
1802  *   Items that were detected.
1803  * @param[out] error
1804  *   Pointer to error structure.
1805  *
1806  * @return
1807  *   0 on success, a negative errno value otherwise and rte_errno is set.
1808  */
1809 int
1810 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1811 			      uint64_t action_flags,
1812 			      struct rte_eth_dev *dev,
1813 			      const struct rte_flow_attr *attr,
1814 			      uint64_t item_flags,
1815 			      struct rte_flow_error *error)
1816 {
1817 	const struct rte_flow_action_rss *rss = action->conf;
1818 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1819 	int ret;
1820 
1821 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1822 		return rte_flow_error_set(error, EINVAL,
1823 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1824 					  "can't have 2 fate actions"
1825 					  " in same flow");
1826 	ret = mlx5_validate_action_rss(dev, action, error);
1827 	if (ret)
1828 		return ret;
1829 	if (attr->egress)
1830 		return rte_flow_error_set(error, ENOTSUP,
1831 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1832 					  "rss action not supported for "
1833 					  "egress");
1834 	if (rss->level > 1 && !tunnel)
1835 		return rte_flow_error_set(error, EINVAL,
1836 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1837 					  "inner RSS is not supported for "
1838 					  "non-tunnel flows");
1839 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1840 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1841 		return rte_flow_error_set(error, EINVAL,
1842 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1843 					  "RSS on eCPRI is not supported now");
1844 	}
1845 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1846 	    !(item_flags &
1847 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1848 	    rss->level > 1)
1849 		return rte_flow_error_set(error, EINVAL,
1850 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1851 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1852 	return 0;
1853 }
1854 
1855 /*
1856  * Validate the default miss action.
1857  *
1858  * @param[in] action_flags
1859  *   Bit-fields that holds the actions detected until now.
1860  * @param[out] error
1861  *   Pointer to error structure.
1862  *
1863  * @return
1864  *   0 on success, a negative errno value otherwise and rte_errno is set.
1865  */
1866 int
1867 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1868 				const struct rte_flow_attr *attr,
1869 				struct rte_flow_error *error)
1870 {
1871 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1872 		return rte_flow_error_set(error, EINVAL,
1873 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1874 					  "can't have 2 fate actions in"
1875 					  " same flow");
1876 	if (attr->egress)
1877 		return rte_flow_error_set(error, ENOTSUP,
1878 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1879 					  "default miss action not supported "
1880 					  "for egress");
1881 	if (attr->group)
1882 		return rte_flow_error_set(error, ENOTSUP,
1883 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1884 					  "only group 0 is supported");
1885 	if (attr->transfer)
1886 		return rte_flow_error_set(error, ENOTSUP,
1887 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1888 					  NULL, "transfer is not supported");
1889 	return 0;
1890 }
1891 
1892 /*
1893  * Validate the count action.
1894  *
1895  * @param[in] dev
1896  *   Pointer to the Ethernet device structure.
1897  * @param[in] attr
1898  *   Attributes of flow that includes this action.
1899  * @param[out] error
1900  *   Pointer to error structure.
1901  *
1902  * @return
1903  *   0 on success, a negative errno value otherwise and rte_errno is set.
1904  */
1905 int
1906 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1907 				const struct rte_flow_attr *attr,
1908 				struct rte_flow_error *error)
1909 {
1910 	if (attr->egress)
1911 		return rte_flow_error_set(error, ENOTSUP,
1912 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1913 					  "count action not supported for "
1914 					  "egress");
1915 	return 0;
1916 }
1917 
1918 /*
1919  * Validate the ASO CT action.
1920  *
1921  * @param[in] dev
1922  *   Pointer to the Ethernet device structure.
1923  * @param[in] conntrack
1924  *   Pointer to the CT action profile.
1925  * @param[out] error
1926  *   Pointer to error structure.
1927  *
1928  * @return
1929  *   0 on success, a negative errno value otherwise and rte_errno is set.
1930  */
1931 int
1932 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1933 			const struct rte_flow_action_conntrack *conntrack,
1934 			struct rte_flow_error *error)
1935 {
1936 	RTE_SET_USED(dev);
1937 
1938 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1939 		return rte_flow_error_set(error, EINVAL,
1940 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1941 					  "Invalid CT state");
1942 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1943 		return rte_flow_error_set(error, EINVAL,
1944 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1945 					  "Invalid last TCP packet flag");
1946 	return 0;
1947 }
1948 
1949 /**
1950  * Verify the @p attributes will be correctly understood by the NIC and store
1951  * them in the @p flow if everything is correct.
1952  *
1953  * @param[in] dev
1954  *   Pointer to the Ethernet device structure.
1955  * @param[in] attributes
1956  *   Pointer to flow attributes
1957  * @param[out] error
1958  *   Pointer to error structure.
1959  *
1960  * @return
1961  *   0 on success, a negative errno value otherwise and rte_errno is set.
1962  */
1963 int
1964 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1965 			      const struct rte_flow_attr *attributes,
1966 			      struct rte_flow_error *error)
1967 {
1968 	struct mlx5_priv *priv = dev->data->dev_private;
1969 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1970 
1971 	if (attributes->group)
1972 		return rte_flow_error_set(error, ENOTSUP,
1973 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1974 					  NULL, "groups is not supported");
1975 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1976 	    attributes->priority >= priority_max)
1977 		return rte_flow_error_set(error, ENOTSUP,
1978 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1979 					  NULL, "priority out of range");
1980 	if (attributes->egress)
1981 		return rte_flow_error_set(error, ENOTSUP,
1982 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1983 					  "egress is not supported");
1984 	if (attributes->transfer && !priv->config.dv_esw_en)
1985 		return rte_flow_error_set(error, ENOTSUP,
1986 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1987 					  NULL, "transfer is not supported");
1988 	if (!attributes->ingress)
1989 		return rte_flow_error_set(error, EINVAL,
1990 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1991 					  NULL,
1992 					  "ingress attribute is mandatory");
1993 	return 0;
1994 }
1995 
1996 /**
1997  * Validate ICMP6 item.
1998  *
1999  * @param[in] item
2000  *   Item specification.
2001  * @param[in] item_flags
2002  *   Bit-fields that holds the items detected until now.
2003  * @param[in] ext_vlan_sup
2004  *   Whether extended VLAN features are supported or not.
2005  * @param[out] error
2006  *   Pointer to error structure.
2007  *
2008  * @return
2009  *   0 on success, a negative errno value otherwise and rte_errno is set.
2010  */
2011 int
2012 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2013 			       uint64_t item_flags,
2014 			       uint8_t target_protocol,
2015 			       struct rte_flow_error *error)
2016 {
2017 	const struct rte_flow_item_icmp6 *mask = item->mask;
2018 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2019 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2020 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2021 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2022 				      MLX5_FLOW_LAYER_OUTER_L4;
2023 	int ret;
2024 
2025 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2026 		return rte_flow_error_set(error, EINVAL,
2027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 					  "protocol filtering not compatible"
2029 					  " with ICMP6 layer");
2030 	if (!(item_flags & l3m))
2031 		return rte_flow_error_set(error, EINVAL,
2032 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2033 					  "IPv6 is mandatory to filter on"
2034 					  " ICMP6");
2035 	if (item_flags & l4m)
2036 		return rte_flow_error_set(error, EINVAL,
2037 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2038 					  "multiple L4 layers not supported");
2039 	if (!mask)
2040 		mask = &rte_flow_item_icmp6_mask;
2041 	ret = mlx5_flow_item_acceptable
2042 		(item, (const uint8_t *)mask,
2043 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2044 		 sizeof(struct rte_flow_item_icmp6),
2045 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2046 	if (ret < 0)
2047 		return ret;
2048 	return 0;
2049 }
2050 
2051 /**
2052  * Validate ICMP item.
2053  *
2054  * @param[in] item
2055  *   Item specification.
2056  * @param[in] item_flags
2057  *   Bit-fields that holds the items detected until now.
2058  * @param[out] error
2059  *   Pointer to error structure.
2060  *
2061  * @return
2062  *   0 on success, a negative errno value otherwise and rte_errno is set.
2063  */
2064 int
2065 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2066 			     uint64_t item_flags,
2067 			     uint8_t target_protocol,
2068 			     struct rte_flow_error *error)
2069 {
2070 	const struct rte_flow_item_icmp *mask = item->mask;
2071 	const struct rte_flow_item_icmp nic_mask = {
2072 		.hdr.icmp_type = 0xff,
2073 		.hdr.icmp_code = 0xff,
2074 		.hdr.icmp_ident = RTE_BE16(0xffff),
2075 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2076 	};
2077 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2078 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2079 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2080 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2081 				      MLX5_FLOW_LAYER_OUTER_L4;
2082 	int ret;
2083 
2084 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2085 		return rte_flow_error_set(error, EINVAL,
2086 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2087 					  "protocol filtering not compatible"
2088 					  " with ICMP layer");
2089 	if (!(item_flags & l3m))
2090 		return rte_flow_error_set(error, EINVAL,
2091 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2092 					  "IPv4 is mandatory to filter"
2093 					  " on ICMP");
2094 	if (item_flags & l4m)
2095 		return rte_flow_error_set(error, EINVAL,
2096 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2097 					  "multiple L4 layers not supported");
2098 	if (!mask)
2099 		mask = &nic_mask;
2100 	ret = mlx5_flow_item_acceptable
2101 		(item, (const uint8_t *)mask,
2102 		 (const uint8_t *)&nic_mask,
2103 		 sizeof(struct rte_flow_item_icmp),
2104 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2105 	if (ret < 0)
2106 		return ret;
2107 	return 0;
2108 }
2109 
2110 /**
2111  * Validate Ethernet item.
2112  *
2113  * @param[in] item
2114  *   Item specification.
2115  * @param[in] item_flags
2116  *   Bit-fields that holds the items detected until now.
2117  * @param[out] error
2118  *   Pointer to error structure.
2119  *
2120  * @return
2121  *   0 on success, a negative errno value otherwise and rte_errno is set.
2122  */
2123 int
2124 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2125 			    uint64_t item_flags, bool ext_vlan_sup,
2126 			    struct rte_flow_error *error)
2127 {
2128 	const struct rte_flow_item_eth *mask = item->mask;
2129 	const struct rte_flow_item_eth nic_mask = {
2130 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2131 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2132 		.type = RTE_BE16(0xffff),
2133 		.has_vlan = ext_vlan_sup ? 1 : 0,
2134 	};
2135 	int ret;
2136 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2137 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2138 				       MLX5_FLOW_LAYER_OUTER_L2;
2139 
2140 	if (item_flags & ethm)
2141 		return rte_flow_error_set(error, ENOTSUP,
2142 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2143 					  "multiple L2 layers not supported");
2144 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2145 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2146 		return rte_flow_error_set(error, EINVAL,
2147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2148 					  "L2 layer should not follow "
2149 					  "L3 layers");
2150 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2151 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2152 		return rte_flow_error_set(error, EINVAL,
2153 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2154 					  "L2 layer should not follow VLAN");
2155 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2156 		return rte_flow_error_set(error, EINVAL,
2157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 					  "L2 layer should not follow GTP");
2159 	if (!mask)
2160 		mask = &rte_flow_item_eth_mask;
2161 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2162 					(const uint8_t *)&nic_mask,
2163 					sizeof(struct rte_flow_item_eth),
2164 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2165 	return ret;
2166 }
2167 
2168 /**
2169  * Validate VLAN item.
2170  *
2171  * @param[in] item
2172  *   Item specification.
2173  * @param[in] item_flags
2174  *   Bit-fields that holds the items detected until now.
2175  * @param[in] dev
2176  *   Ethernet device flow is being created on.
2177  * @param[out] error
2178  *   Pointer to error structure.
2179  *
2180  * @return
2181  *   0 on success, a negative errno value otherwise and rte_errno is set.
2182  */
2183 int
2184 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2185 			     uint64_t item_flags,
2186 			     struct rte_eth_dev *dev,
2187 			     struct rte_flow_error *error)
2188 {
2189 	const struct rte_flow_item_vlan *spec = item->spec;
2190 	const struct rte_flow_item_vlan *mask = item->mask;
2191 	const struct rte_flow_item_vlan nic_mask = {
2192 		.tci = RTE_BE16(UINT16_MAX),
2193 		.inner_type = RTE_BE16(UINT16_MAX),
2194 	};
2195 	uint16_t vlan_tag = 0;
2196 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2197 	int ret;
2198 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2199 					MLX5_FLOW_LAYER_INNER_L4) :
2200 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2201 					MLX5_FLOW_LAYER_OUTER_L4);
2202 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2203 					MLX5_FLOW_LAYER_OUTER_VLAN;
2204 
2205 	if (item_flags & vlanm)
2206 		return rte_flow_error_set(error, EINVAL,
2207 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2208 					  "multiple VLAN layers not supported");
2209 	else if ((item_flags & l34m) != 0)
2210 		return rte_flow_error_set(error, EINVAL,
2211 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2212 					  "VLAN cannot follow L3/L4 layer");
2213 	if (!mask)
2214 		mask = &rte_flow_item_vlan_mask;
2215 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2216 					(const uint8_t *)&nic_mask,
2217 					sizeof(struct rte_flow_item_vlan),
2218 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2219 	if (ret)
2220 		return ret;
2221 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2222 		struct mlx5_priv *priv = dev->data->dev_private;
2223 
2224 		if (priv->vmwa_context) {
2225 			/*
2226 			 * Non-NULL context means we have a virtual machine
2227 			 * and SR-IOV enabled, we have to create VLAN interface
2228 			 * to make hypervisor to setup E-Switch vport
2229 			 * context correctly. We avoid creating the multiple
2230 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2231 			 */
2232 			return rte_flow_error_set(error, EINVAL,
2233 						  RTE_FLOW_ERROR_TYPE_ITEM,
2234 						  item,
2235 						  "VLAN tag mask is not"
2236 						  " supported in virtual"
2237 						  " environment");
2238 		}
2239 	}
2240 	if (spec) {
2241 		vlan_tag = spec->tci;
2242 		vlan_tag &= mask->tci;
2243 	}
2244 	/*
2245 	 * From verbs perspective an empty VLAN is equivalent
2246 	 * to a packet without VLAN layer.
2247 	 */
2248 	if (!vlan_tag)
2249 		return rte_flow_error_set(error, EINVAL,
2250 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2251 					  item->spec,
2252 					  "VLAN cannot be empty");
2253 	return 0;
2254 }
2255 
2256 /**
2257  * Validate IPV4 item.
2258  *
2259  * @param[in] item
2260  *   Item specification.
2261  * @param[in] item_flags
2262  *   Bit-fields that holds the items detected until now.
2263  * @param[in] last_item
2264  *   Previous validated item in the pattern items.
2265  * @param[in] ether_type
2266  *   Type in the ethernet layer header (including dot1q).
2267  * @param[in] acc_mask
2268  *   Acceptable mask, if NULL default internal default mask
2269  *   will be used to check whether item fields are supported.
2270  * @param[in] range_accepted
2271  *   True if range of values is accepted for specific fields, false otherwise.
2272  * @param[out] error
2273  *   Pointer to error structure.
2274  *
2275  * @return
2276  *   0 on success, a negative errno value otherwise and rte_errno is set.
2277  */
2278 int
2279 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2280 			     uint64_t item_flags,
2281 			     uint64_t last_item,
2282 			     uint16_t ether_type,
2283 			     const struct rte_flow_item_ipv4 *acc_mask,
2284 			     bool range_accepted,
2285 			     struct rte_flow_error *error)
2286 {
2287 	const struct rte_flow_item_ipv4 *mask = item->mask;
2288 	const struct rte_flow_item_ipv4 *spec = item->spec;
2289 	const struct rte_flow_item_ipv4 nic_mask = {
2290 		.hdr = {
2291 			.src_addr = RTE_BE32(0xffffffff),
2292 			.dst_addr = RTE_BE32(0xffffffff),
2293 			.type_of_service = 0xff,
2294 			.next_proto_id = 0xff,
2295 		},
2296 	};
2297 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2298 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2299 				      MLX5_FLOW_LAYER_OUTER_L3;
2300 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2301 				      MLX5_FLOW_LAYER_OUTER_L4;
2302 	int ret;
2303 	uint8_t next_proto = 0xFF;
2304 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2305 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2306 				  MLX5_FLOW_LAYER_INNER_VLAN);
2307 
2308 	if ((last_item & l2_vlan) && ether_type &&
2309 	    ether_type != RTE_ETHER_TYPE_IPV4)
2310 		return rte_flow_error_set(error, EINVAL,
2311 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2312 					  "IPv4 cannot follow L2/VLAN layer "
2313 					  "which ether type is not IPv4");
2314 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2315 		if (mask && spec)
2316 			next_proto = mask->hdr.next_proto_id &
2317 				     spec->hdr.next_proto_id;
2318 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2319 			return rte_flow_error_set(error, EINVAL,
2320 						  RTE_FLOW_ERROR_TYPE_ITEM,
2321 						  item,
2322 						  "multiple tunnel "
2323 						  "not supported");
2324 	}
2325 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2326 		return rte_flow_error_set(error, EINVAL,
2327 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2328 					  "wrong tunnel type - IPv6 specified "
2329 					  "but IPv4 item provided");
2330 	if (item_flags & l3m)
2331 		return rte_flow_error_set(error, ENOTSUP,
2332 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2333 					  "multiple L3 layers not supported");
2334 	else if (item_flags & l4m)
2335 		return rte_flow_error_set(error, EINVAL,
2336 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2337 					  "L3 cannot follow an L4 layer.");
2338 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2339 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2340 		return rte_flow_error_set(error, EINVAL,
2341 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2342 					  "L3 cannot follow an NVGRE layer.");
2343 	if (!mask)
2344 		mask = &rte_flow_item_ipv4_mask;
2345 	else if (mask->hdr.next_proto_id != 0 &&
2346 		 mask->hdr.next_proto_id != 0xff)
2347 		return rte_flow_error_set(error, EINVAL,
2348 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2349 					  "partial mask is not supported"
2350 					  " for protocol");
2351 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2352 					acc_mask ? (const uint8_t *)acc_mask
2353 						 : (const uint8_t *)&nic_mask,
2354 					sizeof(struct rte_flow_item_ipv4),
2355 					range_accepted, error);
2356 	if (ret < 0)
2357 		return ret;
2358 	return 0;
2359 }
2360 
2361 /**
2362  * Validate IPV6 item.
2363  *
2364  * @param[in] item
2365  *   Item specification.
2366  * @param[in] item_flags
2367  *   Bit-fields that holds the items detected until now.
2368  * @param[in] last_item
2369  *   Previous validated item in the pattern items.
2370  * @param[in] ether_type
2371  *   Type in the ethernet layer header (including dot1q).
2372  * @param[in] acc_mask
2373  *   Acceptable mask, if NULL default internal default mask
2374  *   will be used to check whether item fields are supported.
2375  * @param[out] error
2376  *   Pointer to error structure.
2377  *
2378  * @return
2379  *   0 on success, a negative errno value otherwise and rte_errno is set.
2380  */
2381 int
2382 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2383 			     uint64_t item_flags,
2384 			     uint64_t last_item,
2385 			     uint16_t ether_type,
2386 			     const struct rte_flow_item_ipv6 *acc_mask,
2387 			     struct rte_flow_error *error)
2388 {
2389 	const struct rte_flow_item_ipv6 *mask = item->mask;
2390 	const struct rte_flow_item_ipv6 *spec = item->spec;
2391 	const struct rte_flow_item_ipv6 nic_mask = {
2392 		.hdr = {
2393 			.src_addr =
2394 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2395 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2396 			.dst_addr =
2397 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2398 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2399 			.vtc_flow = RTE_BE32(0xffffffff),
2400 			.proto = 0xff,
2401 		},
2402 	};
2403 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2404 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2405 				      MLX5_FLOW_LAYER_OUTER_L3;
2406 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2407 				      MLX5_FLOW_LAYER_OUTER_L4;
2408 	int ret;
2409 	uint8_t next_proto = 0xFF;
2410 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2411 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2412 				  MLX5_FLOW_LAYER_INNER_VLAN);
2413 
2414 	if ((last_item & l2_vlan) && ether_type &&
2415 	    ether_type != RTE_ETHER_TYPE_IPV6)
2416 		return rte_flow_error_set(error, EINVAL,
2417 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2418 					  "IPv6 cannot follow L2/VLAN layer "
2419 					  "which ether type is not IPv6");
2420 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2421 		next_proto = spec->hdr.proto;
2422 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2423 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2424 			return rte_flow_error_set(error, EINVAL,
2425 						  RTE_FLOW_ERROR_TYPE_ITEM,
2426 						  item,
2427 						  "multiple tunnel "
2428 						  "not supported");
2429 	}
2430 	if (next_proto == IPPROTO_HOPOPTS  ||
2431 	    next_proto == IPPROTO_ROUTING  ||
2432 	    next_proto == IPPROTO_FRAGMENT ||
2433 	    next_proto == IPPROTO_ESP	   ||
2434 	    next_proto == IPPROTO_AH	   ||
2435 	    next_proto == IPPROTO_DSTOPTS)
2436 		return rte_flow_error_set(error, EINVAL,
2437 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2438 					  "IPv6 proto (next header) should "
2439 					  "not be set as extension header");
2440 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2441 		return rte_flow_error_set(error, EINVAL,
2442 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2443 					  "wrong tunnel type - IPv4 specified "
2444 					  "but IPv6 item provided");
2445 	if (item_flags & l3m)
2446 		return rte_flow_error_set(error, ENOTSUP,
2447 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2448 					  "multiple L3 layers not supported");
2449 	else if (item_flags & l4m)
2450 		return rte_flow_error_set(error, EINVAL,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 					  "L3 cannot follow an L4 layer.");
2453 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2454 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2455 		return rte_flow_error_set(error, EINVAL,
2456 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2457 					  "L3 cannot follow an NVGRE layer.");
2458 	if (!mask)
2459 		mask = &rte_flow_item_ipv6_mask;
2460 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2461 					acc_mask ? (const uint8_t *)acc_mask
2462 						 : (const uint8_t *)&nic_mask,
2463 					sizeof(struct rte_flow_item_ipv6),
2464 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2465 	if (ret < 0)
2466 		return ret;
2467 	return 0;
2468 }
2469 
2470 /**
2471  * Validate UDP item.
2472  *
2473  * @param[in] item
2474  *   Item specification.
2475  * @param[in] item_flags
2476  *   Bit-fields that holds the items detected until now.
2477  * @param[in] target_protocol
2478  *   The next protocol in the previous item.
2479  * @param[in] flow_mask
2480  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2481  * @param[out] error
2482  *   Pointer to error structure.
2483  *
2484  * @return
2485  *   0 on success, a negative errno value otherwise and rte_errno is set.
2486  */
2487 int
2488 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2489 			    uint64_t item_flags,
2490 			    uint8_t target_protocol,
2491 			    struct rte_flow_error *error)
2492 {
2493 	const struct rte_flow_item_udp *mask = item->mask;
2494 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2495 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2496 				      MLX5_FLOW_LAYER_OUTER_L3;
2497 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2498 				      MLX5_FLOW_LAYER_OUTER_L4;
2499 	int ret;
2500 
2501 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2502 		return rte_flow_error_set(error, EINVAL,
2503 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2504 					  "protocol filtering not compatible"
2505 					  " with UDP layer");
2506 	if (!(item_flags & l3m))
2507 		return rte_flow_error_set(error, EINVAL,
2508 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2509 					  "L3 is mandatory to filter on L4");
2510 	if (item_flags & l4m)
2511 		return rte_flow_error_set(error, EINVAL,
2512 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2513 					  "multiple L4 layers not supported");
2514 	if (!mask)
2515 		mask = &rte_flow_item_udp_mask;
2516 	ret = mlx5_flow_item_acceptable
2517 		(item, (const uint8_t *)mask,
2518 		 (const uint8_t *)&rte_flow_item_udp_mask,
2519 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2520 		 error);
2521 	if (ret < 0)
2522 		return ret;
2523 	return 0;
2524 }
2525 
2526 /**
2527  * Validate TCP item.
2528  *
2529  * @param[in] item
2530  *   Item specification.
2531  * @param[in] item_flags
2532  *   Bit-fields that holds the items detected until now.
2533  * @param[in] target_protocol
2534  *   The next protocol in the previous item.
2535  * @param[out] error
2536  *   Pointer to error structure.
2537  *
2538  * @return
2539  *   0 on success, a negative errno value otherwise and rte_errno is set.
2540  */
2541 int
2542 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2543 			    uint64_t item_flags,
2544 			    uint8_t target_protocol,
2545 			    const struct rte_flow_item_tcp *flow_mask,
2546 			    struct rte_flow_error *error)
2547 {
2548 	const struct rte_flow_item_tcp *mask = item->mask;
2549 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2550 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2551 				      MLX5_FLOW_LAYER_OUTER_L3;
2552 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2553 				      MLX5_FLOW_LAYER_OUTER_L4;
2554 	int ret;
2555 
2556 	MLX5_ASSERT(flow_mask);
2557 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2558 		return rte_flow_error_set(error, EINVAL,
2559 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2560 					  "protocol filtering not compatible"
2561 					  " with TCP layer");
2562 	if (!(item_flags & l3m))
2563 		return rte_flow_error_set(error, EINVAL,
2564 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2565 					  "L3 is mandatory to filter on L4");
2566 	if (item_flags & l4m)
2567 		return rte_flow_error_set(error, EINVAL,
2568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2569 					  "multiple L4 layers not supported");
2570 	if (!mask)
2571 		mask = &rte_flow_item_tcp_mask;
2572 	ret = mlx5_flow_item_acceptable
2573 		(item, (const uint8_t *)mask,
2574 		 (const uint8_t *)flow_mask,
2575 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2576 		 error);
2577 	if (ret < 0)
2578 		return ret;
2579 	return 0;
2580 }
2581 
2582 /**
2583  * Validate VXLAN item.
2584  *
2585  * @param[in] dev
2586  *   Pointer to the Ethernet device structure.
2587  * @param[in] udp_dport
2588  *   UDP destination port
2589  * @param[in] item
2590  *   Item specification.
2591  * @param[in] item_flags
2592  *   Bit-fields that holds the items detected until now.
2593  * @param[in] attr
2594  *   Flow rule attributes.
2595  * @param[out] error
2596  *   Pointer to error structure.
2597  *
2598  * @return
2599  *   0 on success, a negative errno value otherwise and rte_errno is set.
2600  */
2601 int
2602 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2603 			      uint16_t udp_dport,
2604 			      const struct rte_flow_item *item,
2605 			      uint64_t item_flags,
2606 			      const struct rte_flow_attr *attr,
2607 			      struct rte_flow_error *error)
2608 {
2609 	const struct rte_flow_item_vxlan *spec = item->spec;
2610 	const struct rte_flow_item_vxlan *mask = item->mask;
2611 	int ret;
2612 	struct mlx5_priv *priv = dev->data->dev_private;
2613 	union vni {
2614 		uint32_t vlan_id;
2615 		uint8_t vni[4];
2616 	} id = { .vlan_id = 0, };
2617 	const struct rte_flow_item_vxlan nic_mask = {
2618 		.vni = "\xff\xff\xff",
2619 		.rsvd1 = 0xff,
2620 	};
2621 	const struct rte_flow_item_vxlan *valid_mask;
2622 
2623 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2624 		return rte_flow_error_set(error, ENOTSUP,
2625 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2626 					  "multiple tunnel layers not"
2627 					  " supported");
2628 	valid_mask = &rte_flow_item_vxlan_mask;
2629 	/*
2630 	 * Verify only UDPv4 is present as defined in
2631 	 * https://tools.ietf.org/html/rfc7348
2632 	 */
2633 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2634 		return rte_flow_error_set(error, EINVAL,
2635 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2636 					  "no outer UDP layer found");
2637 	if (!mask)
2638 		mask = &rte_flow_item_vxlan_mask;
2639 
2640 	if (priv->sh->steering_format_version !=
2641 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2642 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2643 		/* FDB domain & NIC domain non-zero group */
2644 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2645 			valid_mask = &nic_mask;
2646 		/* Group zero in NIC domain */
2647 		if (!attr->group && !attr->transfer &&
2648 		    priv->sh->tunnel_header_0_1)
2649 			valid_mask = &nic_mask;
2650 	}
2651 	ret = mlx5_flow_item_acceptable
2652 		(item, (const uint8_t *)mask,
2653 		 (const uint8_t *)valid_mask,
2654 		 sizeof(struct rte_flow_item_vxlan),
2655 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2656 	if (ret < 0)
2657 		return ret;
2658 	if (spec) {
2659 		memcpy(&id.vni[1], spec->vni, 3);
2660 		memcpy(&id.vni[1], mask->vni, 3);
2661 	}
2662 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2663 		return rte_flow_error_set(error, ENOTSUP,
2664 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2665 					  "VXLAN tunnel must be fully defined");
2666 	return 0;
2667 }
2668 
2669 /**
2670  * Validate VXLAN_GPE item.
2671  *
2672  * @param[in] item
2673  *   Item specification.
2674  * @param[in] item_flags
2675  *   Bit-fields that holds the items detected until now.
2676  * @param[in] priv
2677  *   Pointer to the private data structure.
2678  * @param[in] target_protocol
2679  *   The next protocol in the previous item.
2680  * @param[out] error
2681  *   Pointer to error structure.
2682  *
2683  * @return
2684  *   0 on success, a negative errno value otherwise and rte_errno is set.
2685  */
2686 int
2687 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2688 				  uint64_t item_flags,
2689 				  struct rte_eth_dev *dev,
2690 				  struct rte_flow_error *error)
2691 {
2692 	struct mlx5_priv *priv = dev->data->dev_private;
2693 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2694 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2695 	int ret;
2696 	union vni {
2697 		uint32_t vlan_id;
2698 		uint8_t vni[4];
2699 	} id = { .vlan_id = 0, };
2700 
2701 	if (!priv->config.l3_vxlan_en)
2702 		return rte_flow_error_set(error, ENOTSUP,
2703 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2704 					  "L3 VXLAN is not enabled by device"
2705 					  " parameter and/or not configured in"
2706 					  " firmware");
2707 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2708 		return rte_flow_error_set(error, ENOTSUP,
2709 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2710 					  "multiple tunnel layers not"
2711 					  " supported");
2712 	/*
2713 	 * Verify only UDPv4 is present as defined in
2714 	 * https://tools.ietf.org/html/rfc7348
2715 	 */
2716 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2717 		return rte_flow_error_set(error, EINVAL,
2718 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2719 					  "no outer UDP layer found");
2720 	if (!mask)
2721 		mask = &rte_flow_item_vxlan_gpe_mask;
2722 	ret = mlx5_flow_item_acceptable
2723 		(item, (const uint8_t *)mask,
2724 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2725 		 sizeof(struct rte_flow_item_vxlan_gpe),
2726 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2727 	if (ret < 0)
2728 		return ret;
2729 	if (spec) {
2730 		if (spec->protocol)
2731 			return rte_flow_error_set(error, ENOTSUP,
2732 						  RTE_FLOW_ERROR_TYPE_ITEM,
2733 						  item,
2734 						  "VxLAN-GPE protocol"
2735 						  " not supported");
2736 		memcpy(&id.vni[1], spec->vni, 3);
2737 		memcpy(&id.vni[1], mask->vni, 3);
2738 	}
2739 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2740 		return rte_flow_error_set(error, ENOTSUP,
2741 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 					  "VXLAN-GPE tunnel must be fully"
2743 					  " defined");
2744 	return 0;
2745 }
2746 /**
2747  * Validate GRE Key item.
2748  *
2749  * @param[in] item
2750  *   Item specification.
2751  * @param[in] item_flags
2752  *   Bit flags to mark detected items.
2753  * @param[in] gre_item
2754  *   Pointer to gre_item
2755  * @param[out] error
2756  *   Pointer to error structure.
2757  *
2758  * @return
2759  *   0 on success, a negative errno value otherwise and rte_errno is set.
2760  */
2761 int
2762 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2763 				uint64_t item_flags,
2764 				const struct rte_flow_item *gre_item,
2765 				struct rte_flow_error *error)
2766 {
2767 	const rte_be32_t *mask = item->mask;
2768 	int ret = 0;
2769 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2770 	const struct rte_flow_item_gre *gre_spec;
2771 	const struct rte_flow_item_gre *gre_mask;
2772 
2773 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2774 		return rte_flow_error_set(error, ENOTSUP,
2775 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2776 					  "Multiple GRE key not support");
2777 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2778 		return rte_flow_error_set(error, ENOTSUP,
2779 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2780 					  "No preceding GRE header");
2781 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2782 		return rte_flow_error_set(error, ENOTSUP,
2783 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2784 					  "GRE key following a wrong item");
2785 	gre_mask = gre_item->mask;
2786 	if (!gre_mask)
2787 		gre_mask = &rte_flow_item_gre_mask;
2788 	gre_spec = gre_item->spec;
2789 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2790 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2791 		return rte_flow_error_set(error, EINVAL,
2792 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 					  "Key bit must be on");
2794 
2795 	if (!mask)
2796 		mask = &gre_key_default_mask;
2797 	ret = mlx5_flow_item_acceptable
2798 		(item, (const uint8_t *)mask,
2799 		 (const uint8_t *)&gre_key_default_mask,
2800 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2801 	return ret;
2802 }
2803 
2804 /**
2805  * Validate GRE item.
2806  *
2807  * @param[in] item
2808  *   Item specification.
2809  * @param[in] item_flags
2810  *   Bit flags to mark detected items.
2811  * @param[in] target_protocol
2812  *   The next protocol in the previous item.
2813  * @param[out] error
2814  *   Pointer to error structure.
2815  *
2816  * @return
2817  *   0 on success, a negative errno value otherwise and rte_errno is set.
2818  */
2819 int
2820 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2821 			    uint64_t item_flags,
2822 			    uint8_t target_protocol,
2823 			    struct rte_flow_error *error)
2824 {
2825 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2826 	const struct rte_flow_item_gre *mask = item->mask;
2827 	int ret;
2828 	const struct rte_flow_item_gre nic_mask = {
2829 		.c_rsvd0_ver = RTE_BE16(0xB000),
2830 		.protocol = RTE_BE16(UINT16_MAX),
2831 	};
2832 
2833 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2834 		return rte_flow_error_set(error, EINVAL,
2835 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2836 					  "protocol filtering not compatible"
2837 					  " with this GRE layer");
2838 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2839 		return rte_flow_error_set(error, ENOTSUP,
2840 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2841 					  "multiple tunnel layers not"
2842 					  " supported");
2843 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2844 		return rte_flow_error_set(error, ENOTSUP,
2845 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2846 					  "L3 Layer is missing");
2847 	if (!mask)
2848 		mask = &rte_flow_item_gre_mask;
2849 	ret = mlx5_flow_item_acceptable
2850 		(item, (const uint8_t *)mask,
2851 		 (const uint8_t *)&nic_mask,
2852 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2853 		 error);
2854 	if (ret < 0)
2855 		return ret;
2856 #ifndef HAVE_MLX5DV_DR
2857 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2858 	if (spec && (spec->protocol & mask->protocol))
2859 		return rte_flow_error_set(error, ENOTSUP,
2860 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2861 					  "without MPLS support the"
2862 					  " specification cannot be used for"
2863 					  " filtering");
2864 #endif
2865 #endif
2866 	return 0;
2867 }
2868 
2869 /**
2870  * Validate Geneve item.
2871  *
2872  * @param[in] item
2873  *   Item specification.
2874  * @param[in] itemFlags
2875  *   Bit-fields that holds the items detected until now.
2876  * @param[in] enPriv
2877  *   Pointer to the private data structure.
2878  * @param[out] error
2879  *   Pointer to error structure.
2880  *
2881  * @return
2882  *   0 on success, a negative errno value otherwise and rte_errno is set.
2883  */
2884 
2885 int
2886 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2887 			       uint64_t item_flags,
2888 			       struct rte_eth_dev *dev,
2889 			       struct rte_flow_error *error)
2890 {
2891 	struct mlx5_priv *priv = dev->data->dev_private;
2892 	const struct rte_flow_item_geneve *spec = item->spec;
2893 	const struct rte_flow_item_geneve *mask = item->mask;
2894 	int ret;
2895 	uint16_t gbhdr;
2896 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2897 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2898 	const struct rte_flow_item_geneve nic_mask = {
2899 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2900 		.vni = "\xff\xff\xff",
2901 		.protocol = RTE_BE16(UINT16_MAX),
2902 	};
2903 
2904 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2905 		return rte_flow_error_set(error, ENOTSUP,
2906 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2907 					  "L3 Geneve is not enabled by device"
2908 					  " parameter and/or not configured in"
2909 					  " firmware");
2910 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2911 		return rte_flow_error_set(error, ENOTSUP,
2912 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2913 					  "multiple tunnel layers not"
2914 					  " supported");
2915 	/*
2916 	 * Verify only UDPv4 is present as defined in
2917 	 * https://tools.ietf.org/html/rfc7348
2918 	 */
2919 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2920 		return rte_flow_error_set(error, EINVAL,
2921 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2922 					  "no outer UDP layer found");
2923 	if (!mask)
2924 		mask = &rte_flow_item_geneve_mask;
2925 	ret = mlx5_flow_item_acceptable
2926 				  (item, (const uint8_t *)mask,
2927 				   (const uint8_t *)&nic_mask,
2928 				   sizeof(struct rte_flow_item_geneve),
2929 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2930 	if (ret)
2931 		return ret;
2932 	if (spec) {
2933 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2934 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2935 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2936 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2937 			return rte_flow_error_set(error, ENOTSUP,
2938 						  RTE_FLOW_ERROR_TYPE_ITEM,
2939 						  item,
2940 						  "Geneve protocol unsupported"
2941 						  " fields are being used");
2942 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2943 			return rte_flow_error_set
2944 					(error, ENOTSUP,
2945 					 RTE_FLOW_ERROR_TYPE_ITEM,
2946 					 item,
2947 					 "Unsupported Geneve options length");
2948 	}
2949 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2950 		return rte_flow_error_set
2951 				    (error, ENOTSUP,
2952 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2953 				     "Geneve tunnel must be fully defined");
2954 	return 0;
2955 }
2956 
2957 /**
2958  * Validate Geneve TLV option item.
2959  *
2960  * @param[in] item
2961  *   Item specification.
2962  * @param[in] last_item
2963  *   Previous validated item in the pattern items.
2964  * @param[in] geneve_item
2965  *   Previous GENEVE item specification.
2966  * @param[in] dev
2967  *   Pointer to the rte_eth_dev structure.
2968  * @param[out] error
2969  *   Pointer to error structure.
2970  *
2971  * @return
2972  *   0 on success, a negative errno value otherwise and rte_errno is set.
2973  */
2974 int
2975 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2976 				   uint64_t last_item,
2977 				   const struct rte_flow_item *geneve_item,
2978 				   struct rte_eth_dev *dev,
2979 				   struct rte_flow_error *error)
2980 {
2981 	struct mlx5_priv *priv = dev->data->dev_private;
2982 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2983 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2984 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2985 	uint8_t data_max_supported =
2986 			hca_attr->max_geneve_tlv_option_data_len * 4;
2987 	struct mlx5_dev_config *config = &priv->config;
2988 	const struct rte_flow_item_geneve *geneve_spec;
2989 	const struct rte_flow_item_geneve *geneve_mask;
2990 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2991 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2992 	unsigned int i;
2993 	unsigned int data_len;
2994 	uint8_t tlv_option_len;
2995 	uint16_t optlen_m, optlen_v;
2996 	const struct rte_flow_item_geneve_opt full_mask = {
2997 		.option_class = RTE_BE16(0xffff),
2998 		.option_type = 0xff,
2999 		.option_len = 0x1f,
3000 	};
3001 
3002 	if (!mask)
3003 		mask = &rte_flow_item_geneve_opt_mask;
3004 	if (!spec)
3005 		return rte_flow_error_set
3006 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3007 			"Geneve TLV opt class/type/length must be specified");
3008 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3009 		return rte_flow_error_set
3010 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3011 			"Geneve TLV opt length exceeeds the limit (31)");
3012 	/* Check if class type and length masks are full. */
3013 	if (full_mask.option_class != mask->option_class ||
3014 	    full_mask.option_type != mask->option_type ||
3015 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
3016 		return rte_flow_error_set
3017 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3018 			"Geneve TLV opt class/type/length masks must be full");
3019 	/* Check if length is supported */
3020 	if ((uint32_t)spec->option_len >
3021 			config->hca_attr.max_geneve_tlv_option_data_len)
3022 		return rte_flow_error_set
3023 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 			"Geneve TLV opt length not supported");
3025 	if (config->hca_attr.max_geneve_tlv_options > 1)
3026 		DRV_LOG(DEBUG,
3027 			"max_geneve_tlv_options supports more than 1 option");
3028 	/* Check GENEVE item preceding. */
3029 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3030 		return rte_flow_error_set
3031 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3032 			"Geneve opt item must be preceded with Geneve item");
3033 	geneve_spec = geneve_item->spec;
3034 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3035 					  &rte_flow_item_geneve_mask;
3036 	/* Check if GENEVE TLV option size doesn't exceed option length */
3037 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3038 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3039 		tlv_option_len = spec->option_len & mask->option_len;
3040 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3041 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3042 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3043 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3044 		if ((optlen_v & optlen_m) <= tlv_option_len)
3045 			return rte_flow_error_set
3046 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3047 				 "GENEVE TLV option length exceeds optlen");
3048 	}
3049 	/* Check if length is 0 or data is 0. */
3050 	if (spec->data == NULL || spec->option_len == 0)
3051 		return rte_flow_error_set
3052 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3053 			"Geneve TLV opt with zero data/length not supported");
3054 	/* Check not all data & mask are 0. */
3055 	data_len = spec->option_len * 4;
3056 	if (mask->data == NULL) {
3057 		for (i = 0; i < data_len; i++)
3058 			if (spec->data[i])
3059 				break;
3060 		if (i == data_len)
3061 			return rte_flow_error_set(error, ENOTSUP,
3062 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3063 				"Can't match on Geneve option data 0");
3064 	} else {
3065 		for (i = 0; i < data_len; i++)
3066 			if (spec->data[i] & mask->data[i])
3067 				break;
3068 		if (i == data_len)
3069 			return rte_flow_error_set(error, ENOTSUP,
3070 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3071 				"Can't match on Geneve option data and mask 0");
3072 		/* Check data mask supported. */
3073 		for (i = data_max_supported; i < data_len ; i++)
3074 			if (mask->data[i])
3075 				return rte_flow_error_set(error, ENOTSUP,
3076 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3077 					"Data mask is of unsupported size");
3078 	}
3079 	/* Check GENEVE option is supported in NIC. */
3080 	if (!config->hca_attr.geneve_tlv_opt)
3081 		return rte_flow_error_set
3082 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3083 			"Geneve TLV opt not supported");
3084 	/* Check if we already have geneve option with different type/class. */
3085 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3086 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3087 	if (geneve_opt_resource != NULL)
3088 		if (geneve_opt_resource->option_class != spec->option_class ||
3089 		    geneve_opt_resource->option_type != spec->option_type ||
3090 		    geneve_opt_resource->length != spec->option_len) {
3091 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3092 			return rte_flow_error_set(error, ENOTSUP,
3093 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3094 				"Only one Geneve TLV option supported");
3095 		}
3096 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3097 	return 0;
3098 }
3099 
3100 /**
3101  * Validate MPLS item.
3102  *
3103  * @param[in] dev
3104  *   Pointer to the rte_eth_dev structure.
3105  * @param[in] item
3106  *   Item specification.
3107  * @param[in] item_flags
3108  *   Bit-fields that holds the items detected until now.
3109  * @param[in] prev_layer
3110  *   The protocol layer indicated in previous item.
3111  * @param[out] error
3112  *   Pointer to error structure.
3113  *
3114  * @return
3115  *   0 on success, a negative errno value otherwise and rte_errno is set.
3116  */
3117 int
3118 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3119 			     const struct rte_flow_item *item __rte_unused,
3120 			     uint64_t item_flags __rte_unused,
3121 			     uint64_t prev_layer __rte_unused,
3122 			     struct rte_flow_error *error)
3123 {
3124 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3125 	const struct rte_flow_item_mpls *mask = item->mask;
3126 	struct mlx5_priv *priv = dev->data->dev_private;
3127 	int ret;
3128 
3129 	if (!priv->config.mpls_en)
3130 		return rte_flow_error_set(error, ENOTSUP,
3131 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3132 					  "MPLS not supported or"
3133 					  " disabled in firmware"
3134 					  " configuration.");
3135 	/* MPLS over UDP, GRE is allowed */
3136 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3137 			    MLX5_FLOW_LAYER_GRE |
3138 			    MLX5_FLOW_LAYER_GRE_KEY)))
3139 		return rte_flow_error_set(error, EINVAL,
3140 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3141 					  "protocol filtering not compatible"
3142 					  " with MPLS layer");
3143 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3144 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3145 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3146 		return rte_flow_error_set(error, ENOTSUP,
3147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3148 					  "multiple tunnel layers not"
3149 					  " supported");
3150 	if (!mask)
3151 		mask = &rte_flow_item_mpls_mask;
3152 	ret = mlx5_flow_item_acceptable
3153 		(item, (const uint8_t *)mask,
3154 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3155 		 sizeof(struct rte_flow_item_mpls),
3156 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3157 	if (ret < 0)
3158 		return ret;
3159 	return 0;
3160 #else
3161 	return rte_flow_error_set(error, ENOTSUP,
3162 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3163 				  "MPLS is not supported by Verbs, please"
3164 				  " update.");
3165 #endif
3166 }
3167 
3168 /**
3169  * Validate NVGRE item.
3170  *
3171  * @param[in] item
3172  *   Item specification.
3173  * @param[in] item_flags
3174  *   Bit flags to mark detected items.
3175  * @param[in] target_protocol
3176  *   The next protocol in the previous item.
3177  * @param[out] error
3178  *   Pointer to error structure.
3179  *
3180  * @return
3181  *   0 on success, a negative errno value otherwise and rte_errno is set.
3182  */
3183 int
3184 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3185 			      uint64_t item_flags,
3186 			      uint8_t target_protocol,
3187 			      struct rte_flow_error *error)
3188 {
3189 	const struct rte_flow_item_nvgre *mask = item->mask;
3190 	int ret;
3191 
3192 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3193 		return rte_flow_error_set(error, EINVAL,
3194 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3195 					  "protocol filtering not compatible"
3196 					  " with this GRE layer");
3197 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3198 		return rte_flow_error_set(error, ENOTSUP,
3199 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3200 					  "multiple tunnel layers not"
3201 					  " supported");
3202 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3203 		return rte_flow_error_set(error, ENOTSUP,
3204 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3205 					  "L3 Layer is missing");
3206 	if (!mask)
3207 		mask = &rte_flow_item_nvgre_mask;
3208 	ret = mlx5_flow_item_acceptable
3209 		(item, (const uint8_t *)mask,
3210 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3211 		 sizeof(struct rte_flow_item_nvgre),
3212 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3213 	if (ret < 0)
3214 		return ret;
3215 	return 0;
3216 }
3217 
3218 /**
3219  * Validate eCPRI item.
3220  *
3221  * @param[in] item
3222  *   Item specification.
3223  * @param[in] item_flags
3224  *   Bit-fields that holds the items detected until now.
3225  * @param[in] last_item
3226  *   Previous validated item in the pattern items.
3227  * @param[in] ether_type
3228  *   Type in the ethernet layer header (including dot1q).
3229  * @param[in] acc_mask
3230  *   Acceptable mask, if NULL default internal default mask
3231  *   will be used to check whether item fields are supported.
3232  * @param[out] error
3233  *   Pointer to error structure.
3234  *
3235  * @return
3236  *   0 on success, a negative errno value otherwise and rte_errno is set.
3237  */
3238 int
3239 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3240 			      uint64_t item_flags,
3241 			      uint64_t last_item,
3242 			      uint16_t ether_type,
3243 			      const struct rte_flow_item_ecpri *acc_mask,
3244 			      struct rte_flow_error *error)
3245 {
3246 	const struct rte_flow_item_ecpri *mask = item->mask;
3247 	const struct rte_flow_item_ecpri nic_mask = {
3248 		.hdr = {
3249 			.common = {
3250 				.u32 =
3251 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3252 					.type = 0xFF,
3253 					}).u32),
3254 			},
3255 			.dummy[0] = 0xFFFFFFFF,
3256 		},
3257 	};
3258 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3259 					MLX5_FLOW_LAYER_OUTER_VLAN);
3260 	struct rte_flow_item_ecpri mask_lo;
3261 
3262 	if (!(last_item & outer_l2_vlan) &&
3263 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3264 		return rte_flow_error_set(error, EINVAL,
3265 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3266 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3267 	if ((last_item & outer_l2_vlan) && ether_type &&
3268 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3269 		return rte_flow_error_set(error, EINVAL,
3270 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3271 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3272 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3273 		return rte_flow_error_set(error, EINVAL,
3274 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3275 					  "eCPRI with tunnel is not supported right now");
3276 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3277 		return rte_flow_error_set(error, ENOTSUP,
3278 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3279 					  "multiple L3 layers not supported");
3280 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3281 		return rte_flow_error_set(error, EINVAL,
3282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3283 					  "eCPRI cannot coexist with a TCP layer");
3284 	/* In specification, eCPRI could be over UDP layer. */
3285 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3286 		return rte_flow_error_set(error, EINVAL,
3287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3288 					  "eCPRI over UDP layer is not yet supported right now");
3289 	/* Mask for type field in common header could be zero. */
3290 	if (!mask)
3291 		mask = &rte_flow_item_ecpri_mask;
3292 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3293 	/* Input mask is in big-endian format. */
3294 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3295 		return rte_flow_error_set(error, EINVAL,
3296 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3297 					  "partial mask is not supported for protocol");
3298 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3299 		return rte_flow_error_set(error, EINVAL,
3300 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3301 					  "message header mask must be after a type mask");
3302 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3303 					 acc_mask ? (const uint8_t *)acc_mask
3304 						  : (const uint8_t *)&nic_mask,
3305 					 sizeof(struct rte_flow_item_ecpri),
3306 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3307 }
3308 
3309 static int
3310 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3311 		   const struct rte_flow_attr *attr __rte_unused,
3312 		   const struct rte_flow_item items[] __rte_unused,
3313 		   const struct rte_flow_action actions[] __rte_unused,
3314 		   bool external __rte_unused,
3315 		   int hairpin __rte_unused,
3316 		   struct rte_flow_error *error)
3317 {
3318 	return rte_flow_error_set(error, ENOTSUP,
3319 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3320 }
3321 
3322 static struct mlx5_flow *
3323 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3324 		  const struct rte_flow_attr *attr __rte_unused,
3325 		  const struct rte_flow_item items[] __rte_unused,
3326 		  const struct rte_flow_action actions[] __rte_unused,
3327 		  struct rte_flow_error *error)
3328 {
3329 	rte_flow_error_set(error, ENOTSUP,
3330 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3331 	return NULL;
3332 }
3333 
3334 static int
3335 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3336 		    struct mlx5_flow *dev_flow __rte_unused,
3337 		    const struct rte_flow_attr *attr __rte_unused,
3338 		    const struct rte_flow_item items[] __rte_unused,
3339 		    const struct rte_flow_action actions[] __rte_unused,
3340 		    struct rte_flow_error *error)
3341 {
3342 	return rte_flow_error_set(error, ENOTSUP,
3343 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3344 }
3345 
3346 static int
3347 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3348 		struct rte_flow *flow __rte_unused,
3349 		struct rte_flow_error *error)
3350 {
3351 	return rte_flow_error_set(error, ENOTSUP,
3352 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3353 }
3354 
3355 static void
3356 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3357 		 struct rte_flow *flow __rte_unused)
3358 {
3359 }
3360 
3361 static void
3362 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3363 		  struct rte_flow *flow __rte_unused)
3364 {
3365 }
3366 
3367 static int
3368 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3369 		struct rte_flow *flow __rte_unused,
3370 		const struct rte_flow_action *actions __rte_unused,
3371 		void *data __rte_unused,
3372 		struct rte_flow_error *error)
3373 {
3374 	return rte_flow_error_set(error, ENOTSUP,
3375 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3376 }
3377 
3378 static int
3379 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3380 		      uint32_t domains __rte_unused,
3381 		      uint32_t flags __rte_unused)
3382 {
3383 	return 0;
3384 }
3385 
3386 /* Void driver to protect from null pointer reference. */
3387 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3388 	.validate = flow_null_validate,
3389 	.prepare = flow_null_prepare,
3390 	.translate = flow_null_translate,
3391 	.apply = flow_null_apply,
3392 	.remove = flow_null_remove,
3393 	.destroy = flow_null_destroy,
3394 	.query = flow_null_query,
3395 	.sync_domain = flow_null_sync_domain,
3396 };
3397 
3398 /**
3399  * Select flow driver type according to flow attributes and device
3400  * configuration.
3401  *
3402  * @param[in] dev
3403  *   Pointer to the dev structure.
3404  * @param[in] attr
3405  *   Pointer to the flow attributes.
3406  *
3407  * @return
3408  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3409  */
3410 static enum mlx5_flow_drv_type
3411 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3412 {
3413 	struct mlx5_priv *priv = dev->data->dev_private;
3414 	/* The OS can determine first a specific flow type (DV, VERBS) */
3415 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3416 
3417 	if (type != MLX5_FLOW_TYPE_MAX)
3418 		return type;
3419 	/* If no OS specific type - continue with DV/VERBS selection */
3420 	if (attr->transfer && priv->config.dv_esw_en)
3421 		type = MLX5_FLOW_TYPE_DV;
3422 	if (!attr->transfer)
3423 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3424 						 MLX5_FLOW_TYPE_VERBS;
3425 	return type;
3426 }
3427 
3428 #define flow_get_drv_ops(type) flow_drv_ops[type]
3429 
3430 /**
3431  * Flow driver validation API. This abstracts calling driver specific functions.
3432  * The type of flow driver is determined according to flow attributes.
3433  *
3434  * @param[in] dev
3435  *   Pointer to the dev structure.
3436  * @param[in] attr
3437  *   Pointer to the flow attributes.
3438  * @param[in] items
3439  *   Pointer to the list of items.
3440  * @param[in] actions
3441  *   Pointer to the list of actions.
3442  * @param[in] external
3443  *   This flow rule is created by request external to PMD.
3444  * @param[in] hairpin
3445  *   Number of hairpin TX actions, 0 means classic flow.
3446  * @param[out] error
3447  *   Pointer to the error structure.
3448  *
3449  * @return
3450  *   0 on success, a negative errno value otherwise and rte_errno is set.
3451  */
3452 static inline int
3453 flow_drv_validate(struct rte_eth_dev *dev,
3454 		  const struct rte_flow_attr *attr,
3455 		  const struct rte_flow_item items[],
3456 		  const struct rte_flow_action actions[],
3457 		  bool external, int hairpin, struct rte_flow_error *error)
3458 {
3459 	const struct mlx5_flow_driver_ops *fops;
3460 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3461 
3462 	fops = flow_get_drv_ops(type);
3463 	return fops->validate(dev, attr, items, actions, external,
3464 			      hairpin, error);
3465 }
3466 
3467 /**
3468  * Flow driver preparation API. This abstracts calling driver specific
3469  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3470  * calculates the size of memory required for device flow, allocates the memory,
3471  * initializes the device flow and returns the pointer.
3472  *
3473  * @note
3474  *   This function initializes device flow structure such as dv or verbs in
3475  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3476  *   rest. For example, adding returning device flow to flow->dev_flow list and
3477  *   setting backward reference to the flow should be done out of this function.
3478  *   layers field is not filled either.
3479  *
3480  * @param[in] dev
3481  *   Pointer to the dev structure.
3482  * @param[in] attr
3483  *   Pointer to the flow attributes.
3484  * @param[in] items
3485  *   Pointer to the list of items.
3486  * @param[in] actions
3487  *   Pointer to the list of actions.
3488  * @param[in] flow_idx
3489  *   This memory pool index to the flow.
3490  * @param[out] error
3491  *   Pointer to the error structure.
3492  *
3493  * @return
3494  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3495  */
3496 static inline struct mlx5_flow *
3497 flow_drv_prepare(struct rte_eth_dev *dev,
3498 		 const struct rte_flow *flow,
3499 		 const struct rte_flow_attr *attr,
3500 		 const struct rte_flow_item items[],
3501 		 const struct rte_flow_action actions[],
3502 		 uint32_t flow_idx,
3503 		 struct rte_flow_error *error)
3504 {
3505 	const struct mlx5_flow_driver_ops *fops;
3506 	enum mlx5_flow_drv_type type = flow->drv_type;
3507 	struct mlx5_flow *mlx5_flow = NULL;
3508 
3509 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3510 	fops = flow_get_drv_ops(type);
3511 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3512 	if (mlx5_flow)
3513 		mlx5_flow->flow_idx = flow_idx;
3514 	return mlx5_flow;
3515 }
3516 
3517 /**
3518  * Flow driver translation API. This abstracts calling driver specific
3519  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3520  * translates a generic flow into a driver flow. flow_drv_prepare() must
3521  * precede.
3522  *
3523  * @note
3524  *   dev_flow->layers could be filled as a result of parsing during translation
3525  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3526  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3527  *   flow->actions could be overwritten even though all the expanded dev_flows
3528  *   have the same actions.
3529  *
3530  * @param[in] dev
3531  *   Pointer to the rte dev structure.
3532  * @param[in, out] dev_flow
3533  *   Pointer to the mlx5 flow.
3534  * @param[in] attr
3535  *   Pointer to the flow attributes.
3536  * @param[in] items
3537  *   Pointer to the list of items.
3538  * @param[in] actions
3539  *   Pointer to the list of actions.
3540  * @param[out] error
3541  *   Pointer to the error structure.
3542  *
3543  * @return
3544  *   0 on success, a negative errno value otherwise and rte_errno is set.
3545  */
3546 static inline int
3547 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3548 		   const struct rte_flow_attr *attr,
3549 		   const struct rte_flow_item items[],
3550 		   const struct rte_flow_action actions[],
3551 		   struct rte_flow_error *error)
3552 {
3553 	const struct mlx5_flow_driver_ops *fops;
3554 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3555 
3556 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3557 	fops = flow_get_drv_ops(type);
3558 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3559 }
3560 
3561 /**
3562  * Flow driver apply API. This abstracts calling driver specific functions.
3563  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3564  * translated driver flows on to device. flow_drv_translate() must precede.
3565  *
3566  * @param[in] dev
3567  *   Pointer to Ethernet device structure.
3568  * @param[in, out] flow
3569  *   Pointer to flow structure.
3570  * @param[out] error
3571  *   Pointer to error structure.
3572  *
3573  * @return
3574  *   0 on success, a negative errno value otherwise and rte_errno is set.
3575  */
3576 static inline int
3577 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3578 	       struct rte_flow_error *error)
3579 {
3580 	const struct mlx5_flow_driver_ops *fops;
3581 	enum mlx5_flow_drv_type type = flow->drv_type;
3582 
3583 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3584 	fops = flow_get_drv_ops(type);
3585 	return fops->apply(dev, flow, error);
3586 }
3587 
3588 /**
3589  * Flow driver destroy API. This abstracts calling driver specific functions.
3590  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3591  * on device and releases resources of the flow.
3592  *
3593  * @param[in] dev
3594  *   Pointer to Ethernet device.
3595  * @param[in, out] flow
3596  *   Pointer to flow structure.
3597  */
3598 static inline void
3599 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3600 {
3601 	const struct mlx5_flow_driver_ops *fops;
3602 	enum mlx5_flow_drv_type type = flow->drv_type;
3603 
3604 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3605 	fops = flow_get_drv_ops(type);
3606 	fops->destroy(dev, flow);
3607 }
3608 
3609 /**
3610  * Flow driver find RSS policy tbl API. This abstracts calling driver
3611  * specific functions. Parent flow (rte_flow) should have driver
3612  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3613  *
3614  * @param[in] dev
3615  *   Pointer to Ethernet device.
3616  * @param[in, out] flow
3617  *   Pointer to flow structure.
3618  * @param[in] policy
3619  *   Pointer to meter policy table.
3620  * @param[in] rss_desc
3621  *   Pointer to rss_desc
3622  */
3623 static struct mlx5_flow_meter_sub_policy *
3624 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3625 		struct rte_flow *flow,
3626 		struct mlx5_flow_meter_policy *policy,
3627 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
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_sub_policy_rss_prepare(dev, policy, rss_desc);
3635 }
3636 
3637 /**
3638  * Flow driver color tag rule API. This abstracts calling driver
3639  * specific functions. Parent flow (rte_flow) should have driver
3640  * type (drv_type). It will create the color tag rules in hierarchy meter.
3641  *
3642  * @param[in] dev
3643  *   Pointer to Ethernet device.
3644  * @param[in, out] flow
3645  *   Pointer to flow structure.
3646  * @param[in] fm
3647  *   Pointer to flow meter structure.
3648  * @param[in] src_port
3649  *   The src port this extra rule should use.
3650  * @param[in] item
3651  *   The src port id match item.
3652  * @param[out] error
3653  *   Pointer to error structure.
3654  */
3655 static int
3656 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3657 		struct rte_flow *flow,
3658 		struct mlx5_flow_meter_info *fm,
3659 		int32_t src_port,
3660 		const struct rte_flow_item *item,
3661 		struct rte_flow_error *error)
3662 {
3663 	const struct mlx5_flow_driver_ops *fops;
3664 	enum mlx5_flow_drv_type type = flow->drv_type;
3665 
3666 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3667 	fops = flow_get_drv_ops(type);
3668 	return fops->meter_hierarchy_rule_create(dev, fm,
3669 						src_port, item, error);
3670 }
3671 
3672 /**
3673  * Get RSS action from the action list.
3674  *
3675  * @param[in] dev
3676  *   Pointer to Ethernet device.
3677  * @param[in] actions
3678  *   Pointer to the list of actions.
3679  * @param[in] flow
3680  *   Parent flow structure pointer.
3681  *
3682  * @return
3683  *   Pointer to the RSS action if exist, else return NULL.
3684  */
3685 static const struct rte_flow_action_rss*
3686 flow_get_rss_action(struct rte_eth_dev *dev,
3687 		    const struct rte_flow_action actions[])
3688 {
3689 	struct mlx5_priv *priv = dev->data->dev_private;
3690 	const struct rte_flow_action_rss *rss = NULL;
3691 	struct mlx5_meter_policy_action_container *acg;
3692 	struct mlx5_meter_policy_action_container *acy;
3693 
3694 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3695 		switch (actions->type) {
3696 		case RTE_FLOW_ACTION_TYPE_RSS:
3697 			rss = actions->conf;
3698 			break;
3699 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3700 		{
3701 			const struct rte_flow_action_sample *sample =
3702 								actions->conf;
3703 			const struct rte_flow_action *act = sample->actions;
3704 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3705 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3706 					rss = act->conf;
3707 			break;
3708 		}
3709 		case RTE_FLOW_ACTION_TYPE_METER:
3710 		{
3711 			uint32_t mtr_idx;
3712 			struct mlx5_flow_meter_info *fm;
3713 			struct mlx5_flow_meter_policy *policy;
3714 			const struct rte_flow_action_meter *mtr = actions->conf;
3715 
3716 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3717 			if (fm && !fm->def_policy) {
3718 				policy = mlx5_flow_meter_policy_find(dev,
3719 						fm->policy_id, NULL);
3720 				MLX5_ASSERT(policy);
3721 				if (policy->is_hierarchy) {
3722 					policy =
3723 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3724 									policy);
3725 					if (!policy)
3726 						return NULL;
3727 				}
3728 				if (policy->is_rss) {
3729 					acg =
3730 					&policy->act_cnt[RTE_COLOR_GREEN];
3731 					acy =
3732 					&policy->act_cnt[RTE_COLOR_YELLOW];
3733 					if (acg->fate_action ==
3734 					    MLX5_FLOW_FATE_SHARED_RSS)
3735 						rss = acg->rss->conf;
3736 					else if (acy->fate_action ==
3737 						 MLX5_FLOW_FATE_SHARED_RSS)
3738 						rss = acy->rss->conf;
3739 				}
3740 			}
3741 			break;
3742 		}
3743 		default:
3744 			break;
3745 		}
3746 	}
3747 	return rss;
3748 }
3749 
3750 /**
3751  * Get ASO age action by index.
3752  *
3753  * @param[in] dev
3754  *   Pointer to the Ethernet device structure.
3755  * @param[in] age_idx
3756  *   Index to the ASO age action.
3757  *
3758  * @return
3759  *   The specified ASO age action.
3760  */
3761 struct mlx5_aso_age_action*
3762 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3763 {
3764 	uint16_t pool_idx = age_idx & UINT16_MAX;
3765 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3766 	struct mlx5_priv *priv = dev->data->dev_private;
3767 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3768 	struct mlx5_aso_age_pool *pool;
3769 
3770 	rte_rwlock_read_lock(&mng->resize_rwl);
3771 	pool = mng->pools[pool_idx];
3772 	rte_rwlock_read_unlock(&mng->resize_rwl);
3773 	return &pool->actions[offset - 1];
3774 }
3775 
3776 /* maps indirect action to translated direct in some actions array */
3777 struct mlx5_translated_action_handle {
3778 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3779 	int index; /**< Index in related array of rte_flow_action. */
3780 };
3781 
3782 /**
3783  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3784  * direct action if translation possible.
3785  * This functionality used to run same execution path for both direct and
3786  * indirect actions on flow create. All necessary preparations for indirect
3787  * action handling should be performed on *handle* actions list returned
3788  * from this call.
3789  *
3790  * @param[in] dev
3791  *   Pointer to Ethernet device.
3792  * @param[in] actions
3793  *   List of actions to translate.
3794  * @param[out] handle
3795  *   List to store translated indirect action object handles.
3796  * @param[in, out] indir_n
3797  *   Size of *handle* array. On return should be updated with number of
3798  *   indirect actions retrieved from the *actions* list.
3799  * @param[out] translated_actions
3800  *   List of actions where all indirect actions were translated to direct
3801  *   if possible. NULL if no translation took place.
3802  * @param[out] error
3803  *   Pointer to the error structure.
3804  *
3805  * @return
3806  *   0 on success, a negative errno value otherwise and rte_errno is set.
3807  */
3808 static int
3809 flow_action_handles_translate(struct rte_eth_dev *dev,
3810 			      const struct rte_flow_action actions[],
3811 			      struct mlx5_translated_action_handle *handle,
3812 			      int *indir_n,
3813 			      struct rte_flow_action **translated_actions,
3814 			      struct rte_flow_error *error)
3815 {
3816 	struct mlx5_priv *priv = dev->data->dev_private;
3817 	struct rte_flow_action *translated = NULL;
3818 	size_t actions_size;
3819 	int n;
3820 	int copied_n = 0;
3821 	struct mlx5_translated_action_handle *handle_end = NULL;
3822 
3823 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3824 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3825 			continue;
3826 		if (copied_n == *indir_n) {
3827 			return rte_flow_error_set
3828 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3829 				 NULL, "too many shared actions");
3830 		}
3831 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3832 			   sizeof(actions[n].conf));
3833 		handle[copied_n].index = n;
3834 		copied_n++;
3835 	}
3836 	n++;
3837 	*indir_n = copied_n;
3838 	if (!copied_n)
3839 		return 0;
3840 	actions_size = sizeof(struct rte_flow_action) * n;
3841 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3842 	if (!translated) {
3843 		rte_errno = ENOMEM;
3844 		return -ENOMEM;
3845 	}
3846 	memcpy(translated, actions, actions_size);
3847 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3848 		struct mlx5_shared_action_rss *shared_rss;
3849 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3850 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3851 		uint32_t idx = act_idx &
3852 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3853 
3854 		switch (type) {
3855 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3856 			shared_rss = mlx5_ipool_get
3857 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3858 			translated[handle->index].type =
3859 				RTE_FLOW_ACTION_TYPE_RSS;
3860 			translated[handle->index].conf =
3861 				&shared_rss->origin;
3862 			break;
3863 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3864 			translated[handle->index].type =
3865 						(enum rte_flow_action_type)
3866 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3867 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3868 			break;
3869 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3870 			if (priv->sh->flow_hit_aso_en) {
3871 				translated[handle->index].type =
3872 					(enum rte_flow_action_type)
3873 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3874 				translated[handle->index].conf =
3875 							 (void *)(uintptr_t)idx;
3876 				break;
3877 			}
3878 			/* Fall-through */
3879 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3880 			if (priv->sh->ct_aso_en) {
3881 				translated[handle->index].type =
3882 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3883 				translated[handle->index].conf =
3884 							 (void *)(uintptr_t)idx;
3885 				break;
3886 			}
3887 			/* Fall-through */
3888 		default:
3889 			mlx5_free(translated);
3890 			return rte_flow_error_set
3891 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3892 				 NULL, "invalid indirect action type");
3893 		}
3894 	}
3895 	*translated_actions = translated;
3896 	return 0;
3897 }
3898 
3899 /**
3900  * Get Shared RSS action from the action list.
3901  *
3902  * @param[in] dev
3903  *   Pointer to Ethernet device.
3904  * @param[in] shared
3905  *   Pointer to the list of actions.
3906  * @param[in] shared_n
3907  *   Actions list length.
3908  *
3909  * @return
3910  *   The MLX5 RSS action ID if exists, otherwise return 0.
3911  */
3912 static uint32_t
3913 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3914 			   struct mlx5_translated_action_handle *handle,
3915 			   int shared_n)
3916 {
3917 	struct mlx5_translated_action_handle *handle_end;
3918 	struct mlx5_priv *priv = dev->data->dev_private;
3919 	struct mlx5_shared_action_rss *shared_rss;
3920 
3921 
3922 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3923 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3924 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3925 		uint32_t idx = act_idx &
3926 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3927 		switch (type) {
3928 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3929 			shared_rss = mlx5_ipool_get
3930 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3931 									   idx);
3932 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3933 					   __ATOMIC_RELAXED);
3934 			return idx;
3935 		default:
3936 			break;
3937 		}
3938 	}
3939 	return 0;
3940 }
3941 
3942 static unsigned int
3943 find_graph_root(uint32_t rss_level)
3944 {
3945 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3946 			       MLX5_EXPANSION_ROOT_OUTER;
3947 }
3948 
3949 /**
3950  *  Get layer flags from the prefix flow.
3951  *
3952  *  Some flows may be split to several subflows, the prefix subflow gets the
3953  *  match items and the suffix sub flow gets the actions.
3954  *  Some actions need the user defined match item flags to get the detail for
3955  *  the action.
3956  *  This function helps the suffix flow to get the item layer flags from prefix
3957  *  subflow.
3958  *
3959  * @param[in] dev_flow
3960  *   Pointer the created preifx subflow.
3961  *
3962  * @return
3963  *   The layers get from prefix subflow.
3964  */
3965 static inline uint64_t
3966 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3967 {
3968 	uint64_t layers = 0;
3969 
3970 	/*
3971 	 * Layers bits could be localization, but usually the compiler will
3972 	 * help to do the optimization work for source code.
3973 	 * If no decap actions, use the layers directly.
3974 	 */
3975 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3976 		return dev_flow->handle->layers;
3977 	/* Convert L3 layers with decap action. */
3978 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3979 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3980 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3981 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3982 	/* Convert L4 layers with decap action.  */
3983 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3984 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3985 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3986 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3987 	return layers;
3988 }
3989 
3990 /**
3991  * Get metadata split action information.
3992  *
3993  * @param[in] actions
3994  *   Pointer to the list of actions.
3995  * @param[out] qrss
3996  *   Pointer to the return pointer.
3997  * @param[out] qrss_type
3998  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3999  *   if no QUEUE/RSS is found.
4000  * @param[out] encap_idx
4001  *   Pointer to the index of the encap action if exists, otherwise the last
4002  *   action index.
4003  *
4004  * @return
4005  *   Total number of actions.
4006  */
4007 static int
4008 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4009 				       const struct rte_flow_action **qrss,
4010 				       int *encap_idx)
4011 {
4012 	const struct rte_flow_action_raw_encap *raw_encap;
4013 	int actions_n = 0;
4014 	int raw_decap_idx = -1;
4015 
4016 	*encap_idx = -1;
4017 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4018 		switch (actions->type) {
4019 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4020 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4021 			*encap_idx = actions_n;
4022 			break;
4023 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4024 			raw_decap_idx = actions_n;
4025 			break;
4026 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4027 			raw_encap = actions->conf;
4028 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4029 				*encap_idx = raw_decap_idx != -1 ?
4030 						      raw_decap_idx : actions_n;
4031 			break;
4032 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4033 		case RTE_FLOW_ACTION_TYPE_RSS:
4034 			*qrss = actions;
4035 			break;
4036 		default:
4037 			break;
4038 		}
4039 		actions_n++;
4040 	}
4041 	if (*encap_idx == -1)
4042 		*encap_idx = actions_n;
4043 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4044 	return actions_n + 1;
4045 }
4046 
4047 /**
4048  * Check if the action will change packet.
4049  *
4050  * @param dev
4051  *   Pointer to Ethernet device.
4052  * @param[in] type
4053  *   action type.
4054  *
4055  * @return
4056  *   true if action will change packet, false otherwise.
4057  */
4058 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4059 					  enum rte_flow_action_type type)
4060 {
4061 	struct mlx5_priv *priv = dev->data->dev_private;
4062 
4063 	switch (type) {
4064 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4065 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4066 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4067 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4068 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4069 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4070 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4071 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4072 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4073 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4074 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4075 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4076 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4077 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4078 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4079 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4080 	case RTE_FLOW_ACTION_TYPE_SET_META:
4081 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4082 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4083 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4084 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4085 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4086 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4087 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4088 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4089 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4090 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4091 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4092 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4093 		return true;
4094 	case RTE_FLOW_ACTION_TYPE_FLAG:
4095 	case RTE_FLOW_ACTION_TYPE_MARK:
4096 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4097 			return true;
4098 		else
4099 			return false;
4100 	default:
4101 		return false;
4102 	}
4103 }
4104 
4105 /**
4106  * Check meter action from the action list.
4107  *
4108  * @param dev
4109  *   Pointer to Ethernet device.
4110  * @param[in] actions
4111  *   Pointer to the list of actions.
4112  * @param[out] has_mtr
4113  *   Pointer to the meter exist flag.
4114  * @param[out] has_modify
4115  *   Pointer to the flag showing there's packet change action.
4116  * @param[out] meter_id
4117  *   Pointer to the meter id.
4118  *
4119  * @return
4120  *   Total number of actions.
4121  */
4122 static int
4123 flow_check_meter_action(struct rte_eth_dev *dev,
4124 			const struct rte_flow_action actions[],
4125 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4126 {
4127 	const struct rte_flow_action_meter *mtr = NULL;
4128 	int actions_n = 0;
4129 
4130 	MLX5_ASSERT(has_mtr);
4131 	*has_mtr = false;
4132 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4133 		switch (actions->type) {
4134 		case RTE_FLOW_ACTION_TYPE_METER:
4135 			mtr = actions->conf;
4136 			*meter_id = mtr->mtr_id;
4137 			*has_mtr = true;
4138 			break;
4139 		default:
4140 			break;
4141 		}
4142 		if (!*has_mtr)
4143 			*has_modify |= flow_check_modify_action_type(dev,
4144 								actions->type);
4145 		actions_n++;
4146 	}
4147 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4148 	return actions_n + 1;
4149 }
4150 
4151 /**
4152  * Check if the flow should be split due to hairpin.
4153  * The reason for the split is that in current HW we can't
4154  * support encap and push-vlan on Rx, so if a flow contains
4155  * these actions we move it to Tx.
4156  *
4157  * @param dev
4158  *   Pointer to Ethernet device.
4159  * @param[in] attr
4160  *   Flow rule attributes.
4161  * @param[in] actions
4162  *   Associated actions (list terminated by the END action).
4163  *
4164  * @return
4165  *   > 0 the number of actions and the flow should be split,
4166  *   0 when no split required.
4167  */
4168 static int
4169 flow_check_hairpin_split(struct rte_eth_dev *dev,
4170 			 const struct rte_flow_attr *attr,
4171 			 const struct rte_flow_action actions[])
4172 {
4173 	int queue_action = 0;
4174 	int action_n = 0;
4175 	int split = 0;
4176 	const struct rte_flow_action_queue *queue;
4177 	const struct rte_flow_action_rss *rss;
4178 	const struct rte_flow_action_raw_encap *raw_encap;
4179 	const struct rte_eth_hairpin_conf *conf;
4180 
4181 	if (!attr->ingress)
4182 		return 0;
4183 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4184 		switch (actions->type) {
4185 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4186 			queue = actions->conf;
4187 			if (queue == NULL)
4188 				return 0;
4189 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4190 			if (conf == NULL || conf->tx_explicit != 0)
4191 				return 0;
4192 			queue_action = 1;
4193 			action_n++;
4194 			break;
4195 		case RTE_FLOW_ACTION_TYPE_RSS:
4196 			rss = actions->conf;
4197 			if (rss == NULL || rss->queue_num == 0)
4198 				return 0;
4199 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4200 			if (conf == NULL || conf->tx_explicit != 0)
4201 				return 0;
4202 			queue_action = 1;
4203 			action_n++;
4204 			break;
4205 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4206 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4207 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4208 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4209 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4210 			split++;
4211 			action_n++;
4212 			break;
4213 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4214 			raw_encap = actions->conf;
4215 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4216 				split++;
4217 			action_n++;
4218 			break;
4219 		default:
4220 			action_n++;
4221 			break;
4222 		}
4223 	}
4224 	if (split && queue_action)
4225 		return action_n;
4226 	return 0;
4227 }
4228 
4229 /* Declare flow create/destroy prototype in advance. */
4230 static uint32_t
4231 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4232 		 const struct rte_flow_attr *attr,
4233 		 const struct rte_flow_item items[],
4234 		 const struct rte_flow_action actions[],
4235 		 bool external, struct rte_flow_error *error);
4236 
4237 static void
4238 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4239 		  uint32_t flow_idx);
4240 
4241 int
4242 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4243 		      struct mlx5_list_entry *entry, void *cb_ctx)
4244 {
4245 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4246 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4247 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4248 
4249 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4250 }
4251 
4252 struct mlx5_list_entry *
4253 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4254 {
4255 	struct rte_eth_dev *dev = tool_ctx;
4256 	struct mlx5_priv *priv = dev->data->dev_private;
4257 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4258 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4259 	struct rte_flow_error *error = ctx->error;
4260 	uint32_t idx = 0;
4261 	int ret;
4262 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4263 	struct rte_flow_attr attr = {
4264 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4265 		.ingress = 1,
4266 	};
4267 	struct mlx5_rte_flow_item_tag tag_spec = {
4268 		.data = mark_id,
4269 	};
4270 	struct rte_flow_item items[] = {
4271 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4272 	};
4273 	struct rte_flow_action_mark ftag = {
4274 		.id = mark_id,
4275 	};
4276 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4277 		.dst = REG_B,
4278 		.src = REG_NON,
4279 	};
4280 	struct rte_flow_action_jump jump = {
4281 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4282 	};
4283 	struct rte_flow_action actions[] = {
4284 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4285 	};
4286 
4287 	/* Fill the register fileds in the flow. */
4288 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4289 	if (ret < 0)
4290 		return NULL;
4291 	tag_spec.id = ret;
4292 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4293 	if (ret < 0)
4294 		return NULL;
4295 	cp_mreg.src = ret;
4296 	/* Provide the full width of FLAG specific value. */
4297 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4298 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4299 	/* Build a new flow. */
4300 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4301 		items[0] = (struct rte_flow_item){
4302 			.type = (enum rte_flow_item_type)
4303 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4304 			.spec = &tag_spec,
4305 		};
4306 		items[1] = (struct rte_flow_item){
4307 			.type = RTE_FLOW_ITEM_TYPE_END,
4308 		};
4309 		actions[0] = (struct rte_flow_action){
4310 			.type = (enum rte_flow_action_type)
4311 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4312 			.conf = &ftag,
4313 		};
4314 		actions[1] = (struct rte_flow_action){
4315 			.type = (enum rte_flow_action_type)
4316 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4317 			.conf = &cp_mreg,
4318 		};
4319 		actions[2] = (struct rte_flow_action){
4320 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4321 			.conf = &jump,
4322 		};
4323 		actions[3] = (struct rte_flow_action){
4324 			.type = RTE_FLOW_ACTION_TYPE_END,
4325 		};
4326 	} else {
4327 		/* Default rule, wildcard match. */
4328 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4329 		items[0] = (struct rte_flow_item){
4330 			.type = RTE_FLOW_ITEM_TYPE_END,
4331 		};
4332 		actions[0] = (struct rte_flow_action){
4333 			.type = (enum rte_flow_action_type)
4334 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4335 			.conf = &cp_mreg,
4336 		};
4337 		actions[1] = (struct rte_flow_action){
4338 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4339 			.conf = &jump,
4340 		};
4341 		actions[2] = (struct rte_flow_action){
4342 			.type = RTE_FLOW_ACTION_TYPE_END,
4343 		};
4344 	}
4345 	/* Build a new entry. */
4346 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4347 	if (!mcp_res) {
4348 		rte_errno = ENOMEM;
4349 		return NULL;
4350 	}
4351 	mcp_res->idx = idx;
4352 	mcp_res->mark_id = mark_id;
4353 	/*
4354 	 * The copy Flows are not included in any list. There
4355 	 * ones are referenced from other Flows and can not
4356 	 * be applied, removed, deleted in ardbitrary order
4357 	 * by list traversing.
4358 	 */
4359 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4360 					&attr, items, actions, false, error);
4361 	if (!mcp_res->rix_flow) {
4362 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4363 		return NULL;
4364 	}
4365 	return &mcp_res->hlist_ent;
4366 }
4367 
4368 struct mlx5_list_entry *
4369 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4370 		      void *cb_ctx __rte_unused)
4371 {
4372 	struct rte_eth_dev *dev = tool_ctx;
4373 	struct mlx5_priv *priv = dev->data->dev_private;
4374 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4375 	uint32_t idx = 0;
4376 
4377 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4378 	if (!mcp_res) {
4379 		rte_errno = ENOMEM;
4380 		return NULL;
4381 	}
4382 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4383 	mcp_res->idx = idx;
4384 	return &mcp_res->hlist_ent;
4385 }
4386 
4387 void
4388 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4389 {
4390 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4391 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4392 	struct rte_eth_dev *dev = tool_ctx;
4393 	struct mlx5_priv *priv = dev->data->dev_private;
4394 
4395 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4396 }
4397 
4398 /**
4399  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4400  *
4401  * As mark_id is unique, if there's already a registered flow for the mark_id,
4402  * return by increasing the reference counter of the resource. Otherwise, create
4403  * the resource (mcp_res) and flow.
4404  *
4405  * Flow looks like,
4406  *   - If ingress port is ANY and reg_c[1] is mark_id,
4407  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4408  *
4409  * For default flow (zero mark_id), flow is like,
4410  *   - If ingress port is ANY,
4411  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4412  *
4413  * @param dev
4414  *   Pointer to Ethernet device.
4415  * @param mark_id
4416  *   ID of MARK action, zero means default flow for META.
4417  * @param[out] error
4418  *   Perform verbose error reporting if not NULL.
4419  *
4420  * @return
4421  *   Associated resource on success, NULL otherwise and rte_errno is set.
4422  */
4423 static struct mlx5_flow_mreg_copy_resource *
4424 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4425 			  struct rte_flow_error *error)
4426 {
4427 	struct mlx5_priv *priv = dev->data->dev_private;
4428 	struct mlx5_list_entry *entry;
4429 	struct mlx5_flow_cb_ctx ctx = {
4430 		.dev = dev,
4431 		.error = error,
4432 		.data = &mark_id,
4433 	};
4434 
4435 	/* Check if already registered. */
4436 	MLX5_ASSERT(priv->mreg_cp_tbl);
4437 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4438 	if (!entry)
4439 		return NULL;
4440 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4441 			    hlist_ent);
4442 }
4443 
4444 void
4445 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4446 {
4447 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4448 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4449 	struct rte_eth_dev *dev = tool_ctx;
4450 	struct mlx5_priv *priv = dev->data->dev_private;
4451 
4452 	MLX5_ASSERT(mcp_res->rix_flow);
4453 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4454 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4455 }
4456 
4457 /**
4458  * Release flow in RX_CP_TBL.
4459  *
4460  * @param dev
4461  *   Pointer to Ethernet device.
4462  * @flow
4463  *   Parent flow for wich copying is provided.
4464  */
4465 static void
4466 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4467 			  struct rte_flow *flow)
4468 {
4469 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4470 	struct mlx5_priv *priv = dev->data->dev_private;
4471 
4472 	if (!flow->rix_mreg_copy)
4473 		return;
4474 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4475 				 flow->rix_mreg_copy);
4476 	if (!mcp_res || !priv->mreg_cp_tbl)
4477 		return;
4478 	MLX5_ASSERT(mcp_res->rix_flow);
4479 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4480 	flow->rix_mreg_copy = 0;
4481 }
4482 
4483 /**
4484  * Remove the default copy action from RX_CP_TBL.
4485  *
4486  * This functions is called in the mlx5_dev_start(). No thread safe
4487  * is guaranteed.
4488  *
4489  * @param dev
4490  *   Pointer to Ethernet device.
4491  */
4492 static void
4493 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4494 {
4495 	struct mlx5_list_entry *entry;
4496 	struct mlx5_priv *priv = dev->data->dev_private;
4497 	struct mlx5_flow_cb_ctx ctx;
4498 	uint32_t mark_id;
4499 
4500 	/* Check if default flow is registered. */
4501 	if (!priv->mreg_cp_tbl)
4502 		return;
4503 	mark_id = MLX5_DEFAULT_COPY_ID;
4504 	ctx.data = &mark_id;
4505 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4506 	if (!entry)
4507 		return;
4508 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4509 }
4510 
4511 /**
4512  * Add the default copy action in in RX_CP_TBL.
4513  *
4514  * This functions is called in the mlx5_dev_start(). No thread safe
4515  * is guaranteed.
4516  *
4517  * @param dev
4518  *   Pointer to Ethernet device.
4519  * @param[out] error
4520  *   Perform verbose error reporting if not NULL.
4521  *
4522  * @return
4523  *   0 for success, negative value otherwise and rte_errno is set.
4524  */
4525 static int
4526 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4527 				  struct rte_flow_error *error)
4528 {
4529 	struct mlx5_priv *priv = dev->data->dev_private;
4530 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4531 	struct mlx5_flow_cb_ctx ctx;
4532 	uint32_t mark_id;
4533 
4534 	/* Check whether extensive metadata feature is engaged. */
4535 	if (!priv->config.dv_flow_en ||
4536 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4537 	    !mlx5_flow_ext_mreg_supported(dev) ||
4538 	    !priv->sh->dv_regc0_mask)
4539 		return 0;
4540 	/*
4541 	 * Add default mreg copy flow may be called multiple time, but
4542 	 * only be called once in stop. Avoid register it twice.
4543 	 */
4544 	mark_id = MLX5_DEFAULT_COPY_ID;
4545 	ctx.data = &mark_id;
4546 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4547 		return 0;
4548 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4549 	if (!mcp_res)
4550 		return -rte_errno;
4551 	return 0;
4552 }
4553 
4554 /**
4555  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4556  *
4557  * All the flow having Q/RSS action should be split by
4558  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4559  * performs the following,
4560  *   - CQE->flow_tag := reg_c[1] (MARK)
4561  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4562  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4563  * but there should be a flow per each MARK ID set by MARK action.
4564  *
4565  * For the aforementioned reason, if there's a MARK action in flow's action
4566  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4567  * the MARK ID to CQE's flow_tag like,
4568  *   - If reg_c[1] is mark_id,
4569  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4570  *
4571  * For SET_META action which stores value in reg_c[0], as the destination is
4572  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4573  * MARK ID means the default flow. The default flow looks like,
4574  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4575  *
4576  * @param dev
4577  *   Pointer to Ethernet device.
4578  * @param flow
4579  *   Pointer to flow structure.
4580  * @param[in] actions
4581  *   Pointer to the list of actions.
4582  * @param[out] error
4583  *   Perform verbose error reporting if not NULL.
4584  *
4585  * @return
4586  *   0 on success, negative value otherwise and rte_errno is set.
4587  */
4588 static int
4589 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4590 			    struct rte_flow *flow,
4591 			    const struct rte_flow_action *actions,
4592 			    struct rte_flow_error *error)
4593 {
4594 	struct mlx5_priv *priv = dev->data->dev_private;
4595 	struct mlx5_dev_config *config = &priv->config;
4596 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4597 	const struct rte_flow_action_mark *mark;
4598 
4599 	/* Check whether extensive metadata feature is engaged. */
4600 	if (!config->dv_flow_en ||
4601 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4602 	    !mlx5_flow_ext_mreg_supported(dev) ||
4603 	    !priv->sh->dv_regc0_mask)
4604 		return 0;
4605 	/* Find MARK action. */
4606 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4607 		switch (actions->type) {
4608 		case RTE_FLOW_ACTION_TYPE_FLAG:
4609 			mcp_res = flow_mreg_add_copy_action
4610 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4611 			if (!mcp_res)
4612 				return -rte_errno;
4613 			flow->rix_mreg_copy = mcp_res->idx;
4614 			return 0;
4615 		case RTE_FLOW_ACTION_TYPE_MARK:
4616 			mark = (const struct rte_flow_action_mark *)
4617 				actions->conf;
4618 			mcp_res =
4619 				flow_mreg_add_copy_action(dev, mark->id, error);
4620 			if (!mcp_res)
4621 				return -rte_errno;
4622 			flow->rix_mreg_copy = mcp_res->idx;
4623 			return 0;
4624 		default:
4625 			break;
4626 		}
4627 	}
4628 	return 0;
4629 }
4630 
4631 #define MLX5_MAX_SPLIT_ACTIONS 24
4632 #define MLX5_MAX_SPLIT_ITEMS 24
4633 
4634 /**
4635  * Split the hairpin flow.
4636  * Since HW can't support encap and push-vlan on Rx, we move these
4637  * actions to Tx.
4638  * If the count action is after the encap then we also
4639  * move the count action. in this case the count will also measure
4640  * the outer bytes.
4641  *
4642  * @param dev
4643  *   Pointer to Ethernet device.
4644  * @param[in] actions
4645  *   Associated actions (list terminated by the END action).
4646  * @param[out] actions_rx
4647  *   Rx flow actions.
4648  * @param[out] actions_tx
4649  *   Tx flow actions..
4650  * @param[out] pattern_tx
4651  *   The pattern items for the Tx flow.
4652  * @param[out] flow_id
4653  *   The flow ID connected to this flow.
4654  *
4655  * @return
4656  *   0 on success.
4657  */
4658 static int
4659 flow_hairpin_split(struct rte_eth_dev *dev,
4660 		   const struct rte_flow_action actions[],
4661 		   struct rte_flow_action actions_rx[],
4662 		   struct rte_flow_action actions_tx[],
4663 		   struct rte_flow_item pattern_tx[],
4664 		   uint32_t flow_id)
4665 {
4666 	const struct rte_flow_action_raw_encap *raw_encap;
4667 	const struct rte_flow_action_raw_decap *raw_decap;
4668 	struct mlx5_rte_flow_action_set_tag *set_tag;
4669 	struct rte_flow_action *tag_action;
4670 	struct mlx5_rte_flow_item_tag *tag_item;
4671 	struct rte_flow_item *item;
4672 	char *addr;
4673 	int encap = 0;
4674 
4675 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4676 		switch (actions->type) {
4677 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4678 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4679 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4680 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4681 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4682 			rte_memcpy(actions_tx, actions,
4683 			       sizeof(struct rte_flow_action));
4684 			actions_tx++;
4685 			break;
4686 		case RTE_FLOW_ACTION_TYPE_COUNT:
4687 			if (encap) {
4688 				rte_memcpy(actions_tx, actions,
4689 					   sizeof(struct rte_flow_action));
4690 				actions_tx++;
4691 			} else {
4692 				rte_memcpy(actions_rx, actions,
4693 					   sizeof(struct rte_flow_action));
4694 				actions_rx++;
4695 			}
4696 			break;
4697 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4698 			raw_encap = actions->conf;
4699 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4700 				memcpy(actions_tx, actions,
4701 				       sizeof(struct rte_flow_action));
4702 				actions_tx++;
4703 				encap = 1;
4704 			} else {
4705 				rte_memcpy(actions_rx, actions,
4706 					   sizeof(struct rte_flow_action));
4707 				actions_rx++;
4708 			}
4709 			break;
4710 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4711 			raw_decap = actions->conf;
4712 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4713 				memcpy(actions_tx, actions,
4714 				       sizeof(struct rte_flow_action));
4715 				actions_tx++;
4716 			} else {
4717 				rte_memcpy(actions_rx, actions,
4718 					   sizeof(struct rte_flow_action));
4719 				actions_rx++;
4720 			}
4721 			break;
4722 		default:
4723 			rte_memcpy(actions_rx, actions,
4724 				   sizeof(struct rte_flow_action));
4725 			actions_rx++;
4726 			break;
4727 		}
4728 	}
4729 	/* Add set meta action and end action for the Rx flow. */
4730 	tag_action = actions_rx;
4731 	tag_action->type = (enum rte_flow_action_type)
4732 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4733 	actions_rx++;
4734 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4735 	actions_rx++;
4736 	set_tag = (void *)actions_rx;
4737 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4738 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4739 		.data = flow_id,
4740 	};
4741 	MLX5_ASSERT(set_tag->id > REG_NON);
4742 	tag_action->conf = set_tag;
4743 	/* Create Tx item list. */
4744 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4745 	addr = (void *)&pattern_tx[2];
4746 	item = pattern_tx;
4747 	item->type = (enum rte_flow_item_type)
4748 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4749 	tag_item = (void *)addr;
4750 	tag_item->data = flow_id;
4751 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4752 	MLX5_ASSERT(set_tag->id > REG_NON);
4753 	item->spec = tag_item;
4754 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4755 	tag_item = (void *)addr;
4756 	tag_item->data = UINT32_MAX;
4757 	tag_item->id = UINT16_MAX;
4758 	item->mask = tag_item;
4759 	item->last = NULL;
4760 	item++;
4761 	item->type = RTE_FLOW_ITEM_TYPE_END;
4762 	return 0;
4763 }
4764 
4765 /**
4766  * The last stage of splitting chain, just creates the subflow
4767  * without any modification.
4768  *
4769  * @param[in] dev
4770  *   Pointer to Ethernet device.
4771  * @param[in] flow
4772  *   Parent flow structure pointer.
4773  * @param[in, out] sub_flow
4774  *   Pointer to return the created subflow, may be NULL.
4775  * @param[in] attr
4776  *   Flow rule attributes.
4777  * @param[in] items
4778  *   Pattern specification (list terminated by the END pattern item).
4779  * @param[in] actions
4780  *   Associated actions (list terminated by the END action).
4781  * @param[in] flow_split_info
4782  *   Pointer to flow split info structure.
4783  * @param[out] error
4784  *   Perform verbose error reporting if not NULL.
4785  * @return
4786  *   0 on success, negative value otherwise
4787  */
4788 static int
4789 flow_create_split_inner(struct rte_eth_dev *dev,
4790 			struct rte_flow *flow,
4791 			struct mlx5_flow **sub_flow,
4792 			const struct rte_flow_attr *attr,
4793 			const struct rte_flow_item items[],
4794 			const struct rte_flow_action actions[],
4795 			struct mlx5_flow_split_info *flow_split_info,
4796 			struct rte_flow_error *error)
4797 {
4798 	struct mlx5_flow *dev_flow;
4799 
4800 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4801 				    flow_split_info->flow_idx, error);
4802 	if (!dev_flow)
4803 		return -rte_errno;
4804 	dev_flow->flow = flow;
4805 	dev_flow->external = flow_split_info->external;
4806 	dev_flow->skip_scale = flow_split_info->skip_scale;
4807 	/* Subflow object was created, we must include one in the list. */
4808 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4809 		      dev_flow->handle, next);
4810 	/*
4811 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4812 	 * flow may need some user defined item layer flags, and pass the
4813 	 * Metadate rxq mark flag to suffix flow as well.
4814 	 */
4815 	if (flow_split_info->prefix_layers)
4816 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4817 	if (flow_split_info->prefix_mark)
4818 		dev_flow->handle->mark = 1;
4819 	if (sub_flow)
4820 		*sub_flow = dev_flow;
4821 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4822 	dev_flow->dv.table_id = flow_split_info->table_id;
4823 #endif
4824 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4825 }
4826 
4827 /**
4828  * Get the sub policy of a meter.
4829  *
4830  * @param[in] dev
4831  *   Pointer to Ethernet device.
4832  * @param[in] flow
4833  *   Parent flow structure pointer.
4834  * @param wks
4835  *   Pointer to thread flow work space.
4836  * @param[in] attr
4837  *   Flow rule attributes.
4838  * @param[in] items
4839  *   Pattern specification (list terminated by the END pattern item).
4840  * @param[out] error
4841  *   Perform verbose error reporting if not NULL.
4842  *
4843  * @return
4844  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4845  */
4846 static struct mlx5_flow_meter_sub_policy *
4847 get_meter_sub_policy(struct rte_eth_dev *dev,
4848 		     struct rte_flow *flow,
4849 		     struct mlx5_flow_workspace *wks,
4850 		     const struct rte_flow_attr *attr,
4851 		     const struct rte_flow_item items[],
4852 		     struct rte_flow_error *error)
4853 {
4854 	struct mlx5_flow_meter_policy *policy;
4855 	struct mlx5_flow_meter_policy *final_policy;
4856 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4857 
4858 	policy = wks->policy;
4859 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4860 	if (final_policy->is_rss || final_policy->is_queue) {
4861 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4862 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4863 		uint32_t i;
4864 
4865 		/*
4866 		 * This is a tmp dev_flow,
4867 		 * no need to register any matcher for it in translate.
4868 		 */
4869 		wks->skip_matcher_reg = 1;
4870 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4871 			struct mlx5_flow dev_flow = {0};
4872 			struct mlx5_flow_handle dev_handle = { {0} };
4873 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4874 
4875 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4876 				const struct rte_flow_action_rss *rss_act =
4877 					final_policy->act_cnt[i].rss->conf;
4878 				struct rte_flow_action rss_actions[2] = {
4879 					[0] = {
4880 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4881 					.conf = rss_act,
4882 					},
4883 					[1] = {
4884 					.type = RTE_FLOW_ACTION_TYPE_END,
4885 					.conf = NULL,
4886 					}
4887 				};
4888 
4889 				dev_flow.handle = &dev_handle;
4890 				dev_flow.ingress = attr->ingress;
4891 				dev_flow.flow = flow;
4892 				dev_flow.external = 0;
4893 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4894 				dev_flow.dv.transfer = attr->transfer;
4895 #endif
4896 				/**
4897 				 * Translate RSS action to get rss hash fields.
4898 				 */
4899 				if (flow_drv_translate(dev, &dev_flow, attr,
4900 						items, rss_actions, error))
4901 					goto exit;
4902 				rss_desc_v[i] = wks->rss_desc;
4903 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4904 				rss_desc_v[i].hash_fields =
4905 						dev_flow.hash_fields;
4906 				rss_desc_v[i].queue_num =
4907 						rss_desc_v[i].hash_fields ?
4908 						rss_desc_v[i].queue_num : 1;
4909 				rss_desc_v[i].tunnel =
4910 						!!(dev_flow.handle->layers &
4911 						   MLX5_FLOW_LAYER_TUNNEL);
4912 				/* Use the RSS queues in the containers. */
4913 				rss_desc_v[i].queue =
4914 					(uint16_t *)(uintptr_t)rss_act->queue;
4915 				rss_desc[i] = &rss_desc_v[i];
4916 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4917 				/* This is queue action. */
4918 				rss_desc_v[i] = wks->rss_desc;
4919 				rss_desc_v[i].key_len = 0;
4920 				rss_desc_v[i].hash_fields = 0;
4921 				rss_desc_v[i].queue =
4922 					&final_policy->act_cnt[i].queue;
4923 				rss_desc_v[i].queue_num = 1;
4924 				rss_desc[i] = &rss_desc_v[i];
4925 			} else {
4926 				rss_desc[i] = NULL;
4927 			}
4928 		}
4929 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4930 						flow, policy, rss_desc);
4931 	} else {
4932 		enum mlx5_meter_domain mtr_domain =
4933 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4934 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4935 						MLX5_MTR_DOMAIN_INGRESS);
4936 		sub_policy = policy->sub_policys[mtr_domain][0];
4937 	}
4938 	if (!sub_policy)
4939 		rte_flow_error_set(error, EINVAL,
4940 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4941 				   "Failed to get meter sub-policy.");
4942 exit:
4943 	return sub_policy;
4944 }
4945 
4946 /**
4947  * Split the meter flow.
4948  *
4949  * As meter flow will split to three sub flow, other than meter
4950  * action, the other actions make sense to only meter accepts
4951  * the packet. If it need to be dropped, no other additional
4952  * actions should be take.
4953  *
4954  * One kind of special action which decapsulates the L3 tunnel
4955  * header will be in the prefix sub flow, as not to take the
4956  * L3 tunnel header into account.
4957  *
4958  * @param[in] dev
4959  *   Pointer to Ethernet device.
4960  * @param[in] flow
4961  *   Parent flow structure pointer.
4962  * @param wks
4963  *   Pointer to thread flow work space.
4964  * @param[in] attr
4965  *   Flow rule attributes.
4966  * @param[in] items
4967  *   Pattern specification (list terminated by the END pattern item).
4968  * @param[out] sfx_items
4969  *   Suffix flow match items (list terminated by the END pattern item).
4970  * @param[in] actions
4971  *   Associated actions (list terminated by the END action).
4972  * @param[out] actions_sfx
4973  *   Suffix flow actions.
4974  * @param[out] actions_pre
4975  *   Prefix flow actions.
4976  * @param[out] mtr_flow_id
4977  *   Pointer to meter flow id.
4978  * @param[out] error
4979  *   Perform verbose error reporting if not NULL.
4980  *
4981  * @return
4982  *   0 on success, a negative errno value otherwise and rte_errno is set.
4983  */
4984 static int
4985 flow_meter_split_prep(struct rte_eth_dev *dev,
4986 		      struct rte_flow *flow,
4987 		      struct mlx5_flow_workspace *wks,
4988 		      const struct rte_flow_attr *attr,
4989 		      const struct rte_flow_item items[],
4990 		      struct rte_flow_item sfx_items[],
4991 		      const struct rte_flow_action actions[],
4992 		      struct rte_flow_action actions_sfx[],
4993 		      struct rte_flow_action actions_pre[],
4994 		      uint32_t *mtr_flow_id,
4995 		      struct rte_flow_error *error)
4996 {
4997 	struct mlx5_priv *priv = dev->data->dev_private;
4998 	struct mlx5_flow_meter_info *fm = wks->fm;
4999 	struct rte_flow_action *tag_action = NULL;
5000 	struct rte_flow_item *tag_item;
5001 	struct mlx5_rte_flow_action_set_tag *set_tag;
5002 	const struct rte_flow_action_raw_encap *raw_encap;
5003 	const struct rte_flow_action_raw_decap *raw_decap;
5004 	struct mlx5_rte_flow_item_tag *tag_item_spec;
5005 	struct mlx5_rte_flow_item_tag *tag_item_mask;
5006 	uint32_t tag_id = 0;
5007 	struct rte_flow_item *vlan_item_dst = NULL;
5008 	const struct rte_flow_item *vlan_item_src = NULL;
5009 	struct rte_flow_action *hw_mtr_action;
5010 	struct rte_flow_action *action_pre_head = NULL;
5011 	int32_t flow_src_port = priv->representor_id;
5012 	bool mtr_first;
5013 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5014 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5015 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5016 	uint32_t flow_id = 0;
5017 	uint32_t flow_id_reversed = 0;
5018 	uint8_t flow_id_bits = 0;
5019 	int shift;
5020 
5021 	/* Prepare the suffix subflow items. */
5022 	tag_item = sfx_items++;
5023 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5024 		struct mlx5_priv *port_priv;
5025 		const struct rte_flow_item_port_id *pid_v;
5026 		int item_type = items->type;
5027 
5028 		switch (item_type) {
5029 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5030 			pid_v = items->spec;
5031 			MLX5_ASSERT(pid_v);
5032 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5033 			if (!port_priv)
5034 				return rte_flow_error_set(error,
5035 						rte_errno,
5036 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5037 						pid_v,
5038 						"Failed to get port info.");
5039 			flow_src_port = port_priv->representor_id;
5040 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5041 			    flow_src_port != priv->representor_id) {
5042 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5043 								flow, fm,
5044 								flow_src_port,
5045 								items,
5046 								error))
5047 					return -rte_errno;
5048 			}
5049 			memcpy(sfx_items, items, sizeof(*sfx_items));
5050 			sfx_items++;
5051 			break;
5052 		case RTE_FLOW_ITEM_TYPE_VLAN:
5053 			/* Determine if copy vlan item below. */
5054 			vlan_item_src = items;
5055 			vlan_item_dst = sfx_items++;
5056 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5057 			break;
5058 		default:
5059 			break;
5060 		}
5061 	}
5062 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5063 	sfx_items++;
5064 	mtr_first = priv->sh->meter_aso_en &&
5065 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5066 	/* For ASO meter, meter must be before tag in TX direction. */
5067 	if (mtr_first) {
5068 		action_pre_head = actions_pre++;
5069 		/* Leave space for tag action. */
5070 		tag_action = actions_pre++;
5071 	}
5072 	/* Prepare the actions for prefix and suffix flow. */
5073 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5074 		struct rte_flow_action *action_cur = NULL;
5075 
5076 		switch (actions->type) {
5077 		case RTE_FLOW_ACTION_TYPE_METER:
5078 			if (mtr_first) {
5079 				action_cur = action_pre_head;
5080 			} else {
5081 				/* Leave space for tag action. */
5082 				tag_action = actions_pre++;
5083 				action_cur = actions_pre++;
5084 			}
5085 			break;
5086 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5087 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5088 			action_cur = actions_pre++;
5089 			break;
5090 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5091 			raw_encap = actions->conf;
5092 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5093 				action_cur = actions_pre++;
5094 			break;
5095 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5096 			raw_decap = actions->conf;
5097 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5098 				action_cur = actions_pre++;
5099 			break;
5100 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5101 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5102 			if (vlan_item_dst && vlan_item_src) {
5103 				memcpy(vlan_item_dst, vlan_item_src,
5104 					sizeof(*vlan_item_dst));
5105 				/*
5106 				 * Convert to internal match item, it is used
5107 				 * for vlan push and set vid.
5108 				 */
5109 				vlan_item_dst->type = (enum rte_flow_item_type)
5110 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5111 			}
5112 			break;
5113 		default:
5114 			break;
5115 		}
5116 		if (!action_cur)
5117 			action_cur = (fm->def_policy) ?
5118 					actions_sfx++ : actions_pre++;
5119 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5120 	}
5121 	/* Add end action to the actions. */
5122 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5123 	if (priv->sh->meter_aso_en) {
5124 		/**
5125 		 * For ASO meter, need to add an extra jump action explicitly,
5126 		 * to jump from meter to policer table.
5127 		 */
5128 		struct mlx5_flow_meter_sub_policy *sub_policy;
5129 		struct mlx5_flow_tbl_data_entry *tbl_data;
5130 
5131 		if (!fm->def_policy) {
5132 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5133 							  attr, items, error);
5134 			if (!sub_policy)
5135 				return -rte_errno;
5136 		} else {
5137 			enum mlx5_meter_domain mtr_domain =
5138 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5139 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5140 						MLX5_MTR_DOMAIN_INGRESS);
5141 
5142 			sub_policy =
5143 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5144 		}
5145 		tbl_data = container_of(sub_policy->tbl_rsc,
5146 					struct mlx5_flow_tbl_data_entry, tbl);
5147 		hw_mtr_action = actions_pre++;
5148 		hw_mtr_action->type = (enum rte_flow_action_type)
5149 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5150 		hw_mtr_action->conf = tbl_data->jump.action;
5151 	}
5152 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5153 	actions_pre++;
5154 	if (!tag_action)
5155 		return rte_flow_error_set(error, ENOMEM,
5156 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5157 					  NULL, "No tag action space.");
5158 	if (!mtr_flow_id) {
5159 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5160 		goto exit;
5161 	}
5162 	/* Only default-policy Meter creates mtr flow id. */
5163 	if (fm->def_policy) {
5164 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5165 		if (!tag_id)
5166 			return rte_flow_error_set(error, ENOMEM,
5167 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5168 					"Failed to allocate meter flow id.");
5169 		flow_id = tag_id - 1;
5170 		flow_id_bits = (!flow_id) ? 1 :
5171 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5172 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5173 		    mtr_reg_bits) {
5174 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5175 			return rte_flow_error_set(error, EINVAL,
5176 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5177 					"Meter flow id exceeds max limit.");
5178 		}
5179 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5180 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5181 	}
5182 	/* Build tag actions and items for meter_id/meter flow_id. */
5183 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5184 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5185 	tag_item_mask = tag_item_spec + 1;
5186 	/* Both flow_id and meter_id share the same register. */
5187 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5188 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5189 							    0, error),
5190 		.offset = mtr_id_offset,
5191 		.length = mtr_reg_bits,
5192 		.data = flow->meter,
5193 	};
5194 	/*
5195 	 * The color Reg bits used by flow_id are growing from
5196 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5197 	 */
5198 	for (shift = 0; shift < flow_id_bits; shift++)
5199 		flow_id_reversed = (flow_id_reversed << 1) |
5200 				((flow_id >> shift) & 0x1);
5201 	set_tag->data |=
5202 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5203 	tag_item_spec->id = set_tag->id;
5204 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5205 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5206 	tag_action->type = (enum rte_flow_action_type)
5207 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5208 	tag_action->conf = set_tag;
5209 	tag_item->type = (enum rte_flow_item_type)
5210 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5211 	tag_item->spec = tag_item_spec;
5212 	tag_item->last = NULL;
5213 	tag_item->mask = tag_item_mask;
5214 exit:
5215 	if (mtr_flow_id)
5216 		*mtr_flow_id = tag_id;
5217 	return 0;
5218 }
5219 
5220 /**
5221  * Split action list having QUEUE/RSS for metadata register copy.
5222  *
5223  * Once Q/RSS action is detected in user's action list, the flow action
5224  * should be split in order to copy metadata registers, which will happen in
5225  * RX_CP_TBL like,
5226  *   - CQE->flow_tag := reg_c[1] (MARK)
5227  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5228  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5229  * This is because the last action of each flow must be a terminal action
5230  * (QUEUE, RSS or DROP).
5231  *
5232  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5233  * stored and kept in the mlx5_flow structure per each sub_flow.
5234  *
5235  * The Q/RSS action is replaced with,
5236  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5237  * And the following JUMP action is added at the end,
5238  *   - JUMP, to RX_CP_TBL.
5239  *
5240  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5241  * flow_create_split_metadata() routine. The flow will look like,
5242  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5243  *
5244  * @param dev
5245  *   Pointer to Ethernet device.
5246  * @param[out] split_actions
5247  *   Pointer to store split actions to jump to CP_TBL.
5248  * @param[in] actions
5249  *   Pointer to the list of original flow actions.
5250  * @param[in] qrss
5251  *   Pointer to the Q/RSS action.
5252  * @param[in] actions_n
5253  *   Number of original actions.
5254  * @param[in] mtr_sfx
5255  *   Check if it is in meter suffix table.
5256  * @param[out] error
5257  *   Perform verbose error reporting if not NULL.
5258  *
5259  * @return
5260  *   non-zero unique flow_id on success, otherwise 0 and
5261  *   error/rte_error are set.
5262  */
5263 static uint32_t
5264 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5265 			  struct rte_flow_action *split_actions,
5266 			  const struct rte_flow_action *actions,
5267 			  const struct rte_flow_action *qrss,
5268 			  int actions_n, int mtr_sfx,
5269 			  struct rte_flow_error *error)
5270 {
5271 	struct mlx5_priv *priv = dev->data->dev_private;
5272 	struct mlx5_rte_flow_action_set_tag *set_tag;
5273 	struct rte_flow_action_jump *jump;
5274 	const int qrss_idx = qrss - actions;
5275 	uint32_t flow_id = 0;
5276 	int ret = 0;
5277 
5278 	/*
5279 	 * Given actions will be split
5280 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5281 	 * - Add jump to mreg CP_TBL.
5282 	 * As a result, there will be one more action.
5283 	 */
5284 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5285 	/* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5286 	++actions_n;
5287 	set_tag = (void *)(split_actions + actions_n);
5288 	/*
5289 	 * If we are not the meter suffix flow, add the tag action.
5290 	 * Since meter suffix flow already has the tag added.
5291 	 */
5292 	if (!mtr_sfx) {
5293 		/*
5294 		 * Allocate the new subflow ID. This one is unique within
5295 		 * device and not shared with representors. Otherwise,
5296 		 * we would have to resolve multi-thread access synch
5297 		 * issue. Each flow on the shared device is appended
5298 		 * with source vport identifier, so the resulting
5299 		 * flows will be unique in the shared (by master and
5300 		 * representors) domain even if they have coinciding
5301 		 * IDs.
5302 		 */
5303 		mlx5_ipool_malloc(priv->sh->ipool
5304 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5305 		if (!flow_id)
5306 			return rte_flow_error_set(error, ENOMEM,
5307 						  RTE_FLOW_ERROR_TYPE_ACTION,
5308 						  NULL, "can't allocate id "
5309 						  "for split Q/RSS subflow");
5310 		/* Internal SET_TAG action to set flow ID. */
5311 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5312 			.data = flow_id,
5313 		};
5314 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5315 		if (ret < 0)
5316 			return ret;
5317 		set_tag->id = ret;
5318 		/* Construct new actions array. */
5319 		/* Replace QUEUE/RSS action. */
5320 		split_actions[qrss_idx] = (struct rte_flow_action){
5321 			.type = (enum rte_flow_action_type)
5322 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5323 			.conf = set_tag,
5324 		};
5325 	} else {
5326 		/*
5327 		 * If we are the suffix flow of meter, tag already exist.
5328 		 * Set the QUEUE/RSS action to void.
5329 		 */
5330 		split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5331 	}
5332 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5333 	jump = (void *)(set_tag + 1);
5334 	*jump = (struct rte_flow_action_jump){
5335 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5336 	};
5337 	split_actions[actions_n - 2] = (struct rte_flow_action){
5338 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5339 		.conf = jump,
5340 	};
5341 	split_actions[actions_n - 1] = (struct rte_flow_action){
5342 		.type = RTE_FLOW_ACTION_TYPE_END,
5343 	};
5344 	return flow_id;
5345 }
5346 
5347 /**
5348  * Extend the given action list for Tx metadata copy.
5349  *
5350  * Copy the given action list to the ext_actions and add flow metadata register
5351  * copy action in order to copy reg_a set by WQE to reg_c[0].
5352  *
5353  * @param[out] ext_actions
5354  *   Pointer to the extended action list.
5355  * @param[in] actions
5356  *   Pointer to the list of actions.
5357  * @param[in] actions_n
5358  *   Number of actions in the list.
5359  * @param[out] error
5360  *   Perform verbose error reporting if not NULL.
5361  * @param[in] encap_idx
5362  *   The encap action inndex.
5363  *
5364  * @return
5365  *   0 on success, negative value otherwise
5366  */
5367 static int
5368 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5369 		       struct rte_flow_action *ext_actions,
5370 		       const struct rte_flow_action *actions,
5371 		       int actions_n, struct rte_flow_error *error,
5372 		       int encap_idx)
5373 {
5374 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5375 		(struct mlx5_flow_action_copy_mreg *)
5376 			(ext_actions + actions_n + 1);
5377 	int ret;
5378 
5379 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5380 	if (ret < 0)
5381 		return ret;
5382 	cp_mreg->dst = ret;
5383 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5384 	if (ret < 0)
5385 		return ret;
5386 	cp_mreg->src = ret;
5387 	if (encap_idx != 0)
5388 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5389 	if (encap_idx == actions_n - 1) {
5390 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5391 			.type = (enum rte_flow_action_type)
5392 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5393 			.conf = cp_mreg,
5394 		};
5395 		ext_actions[actions_n] = (struct rte_flow_action){
5396 			.type = RTE_FLOW_ACTION_TYPE_END,
5397 		};
5398 	} else {
5399 		ext_actions[encap_idx] = (struct rte_flow_action){
5400 			.type = (enum rte_flow_action_type)
5401 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5402 			.conf = cp_mreg,
5403 		};
5404 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5405 				sizeof(*ext_actions) * (actions_n - encap_idx));
5406 	}
5407 	return 0;
5408 }
5409 
5410 /**
5411  * Check the match action from the action list.
5412  *
5413  * @param[in] actions
5414  *   Pointer to the list of actions.
5415  * @param[in] attr
5416  *   Flow rule attributes.
5417  * @param[in] action
5418  *   The action to be check if exist.
5419  * @param[out] match_action_pos
5420  *   Pointer to the position of the matched action if exists, otherwise is -1.
5421  * @param[out] qrss_action_pos
5422  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5423  * @param[out] modify_after_mirror
5424  *   Pointer to the flag of modify action after FDB mirroring.
5425  *
5426  * @return
5427  *   > 0 the total number of actions.
5428  *   0 if not found match action in action list.
5429  */
5430 static int
5431 flow_check_match_action(const struct rte_flow_action actions[],
5432 			const struct rte_flow_attr *attr,
5433 			enum rte_flow_action_type action,
5434 			int *match_action_pos, int *qrss_action_pos,
5435 			int *modify_after_mirror)
5436 {
5437 	const struct rte_flow_action_sample *sample;
5438 	const struct rte_flow_action_raw_decap *decap;
5439 	int actions_n = 0;
5440 	uint32_t ratio = 0;
5441 	int sub_type = 0;
5442 	int flag = 0;
5443 	int fdb_mirror = 0;
5444 
5445 	*match_action_pos = -1;
5446 	*qrss_action_pos = -1;
5447 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5448 		if (actions->type == action) {
5449 			flag = 1;
5450 			*match_action_pos = actions_n;
5451 		}
5452 		switch (actions->type) {
5453 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5454 		case RTE_FLOW_ACTION_TYPE_RSS:
5455 			*qrss_action_pos = actions_n;
5456 			break;
5457 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5458 			sample = actions->conf;
5459 			ratio = sample->ratio;
5460 			sub_type = ((const struct rte_flow_action *)
5461 					(sample->actions))->type;
5462 			if (ratio == 1 && attr->transfer)
5463 				fdb_mirror = 1;
5464 			break;
5465 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5466 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5467 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5468 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5469 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5470 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5471 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5472 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5473 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5474 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5475 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5476 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5477 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5478 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5479 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5480 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5481 		case RTE_FLOW_ACTION_TYPE_FLAG:
5482 		case RTE_FLOW_ACTION_TYPE_MARK:
5483 		case RTE_FLOW_ACTION_TYPE_SET_META:
5484 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5485 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5486 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5487 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5488 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5489 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5490 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5491 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5492 		case RTE_FLOW_ACTION_TYPE_METER:
5493 			if (fdb_mirror)
5494 				*modify_after_mirror = 1;
5495 			break;
5496 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5497 			decap = actions->conf;
5498 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5499 				;
5500 			actions_n++;
5501 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5502 				const struct rte_flow_action_raw_encap *encap =
5503 								actions->conf;
5504 				if (decap->size <=
5505 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5506 				    encap->size >
5507 					MLX5_ENCAPSULATION_DECISION_SIZE)
5508 					/* L3 encap. */
5509 					break;
5510 			}
5511 			if (fdb_mirror)
5512 				*modify_after_mirror = 1;
5513 			break;
5514 		default:
5515 			break;
5516 		}
5517 		actions_n++;
5518 	}
5519 	if (flag && fdb_mirror && !*modify_after_mirror) {
5520 		/* FDB mirroring uses the destination array to implement
5521 		 * instead of FLOW_SAMPLER object.
5522 		 */
5523 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5524 			flag = 0;
5525 	}
5526 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5527 	return flag ? actions_n + 1 : 0;
5528 }
5529 
5530 #define SAMPLE_SUFFIX_ITEM 2
5531 
5532 /**
5533  * Split the sample flow.
5534  *
5535  * As sample flow will split to two sub flow, sample flow with
5536  * sample action, the other actions will move to new suffix flow.
5537  *
5538  * Also add unique tag id with tag action in the sample flow,
5539  * the same tag id will be as match in the suffix flow.
5540  *
5541  * @param dev
5542  *   Pointer to Ethernet device.
5543  * @param[in] add_tag
5544  *   Add extra tag action flag.
5545  * @param[out] sfx_items
5546  *   Suffix flow match items (list terminated by the END pattern item).
5547  * @param[in] actions
5548  *   Associated actions (list terminated by the END action).
5549  * @param[out] actions_sfx
5550  *   Suffix flow actions.
5551  * @param[out] actions_pre
5552  *   Prefix flow actions.
5553  * @param[in] actions_n
5554  *  The total number of actions.
5555  * @param[in] sample_action_pos
5556  *   The sample action position.
5557  * @param[in] qrss_action_pos
5558  *   The Queue/RSS action position.
5559  * @param[in] jump_table
5560  *   Add extra jump action flag.
5561  * @param[out] error
5562  *   Perform verbose error reporting if not NULL.
5563  *
5564  * @return
5565  *   0 on success, or unique flow_id, a negative errno value
5566  *   otherwise and rte_errno is set.
5567  */
5568 static int
5569 flow_sample_split_prep(struct rte_eth_dev *dev,
5570 		       int add_tag,
5571 		       struct rte_flow_item sfx_items[],
5572 		       const struct rte_flow_action actions[],
5573 		       struct rte_flow_action actions_sfx[],
5574 		       struct rte_flow_action actions_pre[],
5575 		       int actions_n,
5576 		       int sample_action_pos,
5577 		       int qrss_action_pos,
5578 		       int jump_table,
5579 		       struct rte_flow_error *error)
5580 {
5581 	struct mlx5_priv *priv = dev->data->dev_private;
5582 	struct mlx5_rte_flow_action_set_tag *set_tag;
5583 	struct mlx5_rte_flow_item_tag *tag_spec;
5584 	struct mlx5_rte_flow_item_tag *tag_mask;
5585 	struct rte_flow_action_jump *jump_action;
5586 	uint32_t tag_id = 0;
5587 	int index;
5588 	int append_index = 0;
5589 	int ret;
5590 
5591 	if (sample_action_pos < 0)
5592 		return rte_flow_error_set(error, EINVAL,
5593 					  RTE_FLOW_ERROR_TYPE_ACTION,
5594 					  NULL, "invalid position of sample "
5595 					  "action in list");
5596 	/* Prepare the actions for prefix and suffix flow. */
5597 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5598 		index = qrss_action_pos;
5599 		/* Put the preceding the Queue/RSS action into prefix flow. */
5600 		if (index != 0)
5601 			memcpy(actions_pre, actions,
5602 			       sizeof(struct rte_flow_action) * index);
5603 		/* Put others preceding the sample action into prefix flow. */
5604 		if (sample_action_pos > index + 1)
5605 			memcpy(actions_pre + index, actions + index + 1,
5606 			       sizeof(struct rte_flow_action) *
5607 			       (sample_action_pos - index - 1));
5608 		index = sample_action_pos - 1;
5609 		/* Put Queue/RSS action into Suffix flow. */
5610 		memcpy(actions_sfx, actions + qrss_action_pos,
5611 		       sizeof(struct rte_flow_action));
5612 		actions_sfx++;
5613 	} else {
5614 		index = sample_action_pos;
5615 		if (index != 0)
5616 			memcpy(actions_pre, actions,
5617 			       sizeof(struct rte_flow_action) * index);
5618 	}
5619 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5620 	 * For CX6DX and above, metadata registers Cx preserve their value,
5621 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5622 	 */
5623 	if (add_tag) {
5624 		/* Prepare the prefix tag action. */
5625 		append_index++;
5626 		set_tag = (void *)(actions_pre + actions_n + append_index);
5627 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5628 		if (ret < 0)
5629 			return ret;
5630 		mlx5_ipool_malloc(priv->sh->ipool
5631 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5632 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5633 			.id = ret,
5634 			.data = tag_id,
5635 		};
5636 		/* Prepare the suffix subflow items. */
5637 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5638 		tag_spec->data = tag_id;
5639 		tag_spec->id = set_tag->id;
5640 		tag_mask = tag_spec + 1;
5641 		tag_mask->data = UINT32_MAX;
5642 		sfx_items[0] = (struct rte_flow_item){
5643 			.type = (enum rte_flow_item_type)
5644 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5645 			.spec = tag_spec,
5646 			.last = NULL,
5647 			.mask = tag_mask,
5648 		};
5649 		sfx_items[1] = (struct rte_flow_item){
5650 			.type = (enum rte_flow_item_type)
5651 				RTE_FLOW_ITEM_TYPE_END,
5652 		};
5653 		/* Prepare the tag action in prefix subflow. */
5654 		actions_pre[index++] =
5655 			(struct rte_flow_action){
5656 			.type = (enum rte_flow_action_type)
5657 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5658 			.conf = set_tag,
5659 		};
5660 	}
5661 	memcpy(actions_pre + index, actions + sample_action_pos,
5662 	       sizeof(struct rte_flow_action));
5663 	index += 1;
5664 	/* For the modify action after the sample action in E-Switch mirroring,
5665 	 * Add the extra jump action in prefix subflow and jump into the next
5666 	 * table, then do the modify action in the new table.
5667 	 */
5668 	if (jump_table) {
5669 		/* Prepare the prefix jump action. */
5670 		append_index++;
5671 		jump_action = (void *)(actions_pre + actions_n + append_index);
5672 		jump_action->group = jump_table;
5673 		actions_pre[index++] =
5674 			(struct rte_flow_action){
5675 			.type = (enum rte_flow_action_type)
5676 				RTE_FLOW_ACTION_TYPE_JUMP,
5677 			.conf = jump_action,
5678 		};
5679 	}
5680 	actions_pre[index] = (struct rte_flow_action){
5681 		.type = (enum rte_flow_action_type)
5682 			RTE_FLOW_ACTION_TYPE_END,
5683 	};
5684 	/* Put the actions after sample into Suffix flow. */
5685 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5686 	       sizeof(struct rte_flow_action) *
5687 	       (actions_n - sample_action_pos - 1));
5688 	return tag_id;
5689 }
5690 
5691 /**
5692  * The splitting for metadata feature.
5693  *
5694  * - Q/RSS action on NIC Rx should be split in order to pass by
5695  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5696  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5697  *
5698  * - All the actions on NIC Tx should have a mreg copy action to
5699  *   copy reg_a from WQE to reg_c[0].
5700  *
5701  * @param dev
5702  *   Pointer to Ethernet device.
5703  * @param[in] flow
5704  *   Parent flow structure pointer.
5705  * @param[in] attr
5706  *   Flow rule attributes.
5707  * @param[in] items
5708  *   Pattern specification (list terminated by the END pattern item).
5709  * @param[in] actions
5710  *   Associated actions (list terminated by the END action).
5711  * @param[in] flow_split_info
5712  *   Pointer to flow split info structure.
5713  * @param[out] error
5714  *   Perform verbose error reporting if not NULL.
5715  * @return
5716  *   0 on success, negative value otherwise
5717  */
5718 static int
5719 flow_create_split_metadata(struct rte_eth_dev *dev,
5720 			   struct rte_flow *flow,
5721 			   const struct rte_flow_attr *attr,
5722 			   const struct rte_flow_item items[],
5723 			   const struct rte_flow_action actions[],
5724 			   struct mlx5_flow_split_info *flow_split_info,
5725 			   struct rte_flow_error *error)
5726 {
5727 	struct mlx5_priv *priv = dev->data->dev_private;
5728 	struct mlx5_dev_config *config = &priv->config;
5729 	const struct rte_flow_action *qrss = NULL;
5730 	struct rte_flow_action *ext_actions = NULL;
5731 	struct mlx5_flow *dev_flow = NULL;
5732 	uint32_t qrss_id = 0;
5733 	int mtr_sfx = 0;
5734 	size_t act_size;
5735 	int actions_n;
5736 	int encap_idx;
5737 	int ret;
5738 
5739 	/* Check whether extensive metadata feature is engaged. */
5740 	if (!config->dv_flow_en ||
5741 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5742 	    !mlx5_flow_ext_mreg_supported(dev))
5743 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5744 					       actions, flow_split_info, error);
5745 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5746 							   &encap_idx);
5747 	if (qrss) {
5748 		/* Exclude hairpin flows from splitting. */
5749 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5750 			const struct rte_flow_action_queue *queue;
5751 
5752 			queue = qrss->conf;
5753 			if (mlx5_rxq_get_type(dev, queue->index) ==
5754 			    MLX5_RXQ_TYPE_HAIRPIN)
5755 				qrss = NULL;
5756 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5757 			const struct rte_flow_action_rss *rss;
5758 
5759 			rss = qrss->conf;
5760 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5761 			    MLX5_RXQ_TYPE_HAIRPIN)
5762 				qrss = NULL;
5763 		}
5764 	}
5765 	if (qrss) {
5766 		/* Check if it is in meter suffix table. */
5767 		mtr_sfx = attr->group == (attr->transfer ?
5768 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5769 			  MLX5_FLOW_TABLE_LEVEL_METER);
5770 		/*
5771 		 * Q/RSS action on NIC Rx should be split in order to pass by
5772 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5773 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5774 		 */
5775 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5776 			   sizeof(struct rte_flow_action_set_tag) +
5777 			   sizeof(struct rte_flow_action_jump);
5778 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5779 					  SOCKET_ID_ANY);
5780 		if (!ext_actions)
5781 			return rte_flow_error_set(error, ENOMEM,
5782 						  RTE_FLOW_ERROR_TYPE_ACTION,
5783 						  NULL, "no memory to split "
5784 						  "metadata flow");
5785 		/*
5786 		 * Create the new actions list with removed Q/RSS action
5787 		 * and appended set tag and jump to register copy table
5788 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5789 		 * in advance, because it is needed for set tag action.
5790 		 */
5791 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5792 						    qrss, actions_n,
5793 						    mtr_sfx, error);
5794 		if (!mtr_sfx && !qrss_id) {
5795 			ret = -rte_errno;
5796 			goto exit;
5797 		}
5798 	} else if (attr->egress && !attr->transfer) {
5799 		/*
5800 		 * All the actions on NIC Tx should have a metadata register
5801 		 * copy action to copy reg_a from WQE to reg_c[meta]
5802 		 */
5803 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5804 			   sizeof(struct mlx5_flow_action_copy_mreg);
5805 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5806 					  SOCKET_ID_ANY);
5807 		if (!ext_actions)
5808 			return rte_flow_error_set(error, ENOMEM,
5809 						  RTE_FLOW_ERROR_TYPE_ACTION,
5810 						  NULL, "no memory to split "
5811 						  "metadata flow");
5812 		/* Create the action list appended with copy register. */
5813 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5814 					     actions_n, error, encap_idx);
5815 		if (ret < 0)
5816 			goto exit;
5817 	}
5818 	/* Add the unmodified original or prefix subflow. */
5819 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5820 				      items, ext_actions ? ext_actions :
5821 				      actions, flow_split_info, error);
5822 	if (ret < 0)
5823 		goto exit;
5824 	MLX5_ASSERT(dev_flow);
5825 	if (qrss) {
5826 		const struct rte_flow_attr q_attr = {
5827 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5828 			.ingress = 1,
5829 		};
5830 		/* Internal PMD action to set register. */
5831 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5832 			.data = qrss_id,
5833 			.id = REG_NON,
5834 		};
5835 		struct rte_flow_item q_items[] = {
5836 			{
5837 				.type = (enum rte_flow_item_type)
5838 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5839 				.spec = &q_tag_spec,
5840 				.last = NULL,
5841 				.mask = NULL,
5842 			},
5843 			{
5844 				.type = RTE_FLOW_ITEM_TYPE_END,
5845 			},
5846 		};
5847 		struct rte_flow_action q_actions[] = {
5848 			{
5849 				.type = qrss->type,
5850 				.conf = qrss->conf,
5851 			},
5852 			{
5853 				.type = RTE_FLOW_ACTION_TYPE_END,
5854 			},
5855 		};
5856 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5857 
5858 		/*
5859 		 * Configure the tag item only if there is no meter subflow.
5860 		 * Since tag is already marked in the meter suffix subflow
5861 		 * we can just use the meter suffix items as is.
5862 		 */
5863 		if (qrss_id) {
5864 			/* Not meter subflow. */
5865 			MLX5_ASSERT(!mtr_sfx);
5866 			/*
5867 			 * Put unique id in prefix flow due to it is destroyed
5868 			 * after suffix flow and id will be freed after there
5869 			 * is no actual flows with this id and identifier
5870 			 * reallocation becomes possible (for example, for
5871 			 * other flows in other threads).
5872 			 */
5873 			dev_flow->handle->split_flow_id = qrss_id;
5874 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5875 						   error);
5876 			if (ret < 0)
5877 				goto exit;
5878 			q_tag_spec.id = ret;
5879 		}
5880 		dev_flow = NULL;
5881 		/* Add suffix subflow to execute Q/RSS. */
5882 		flow_split_info->prefix_layers = layers;
5883 		flow_split_info->prefix_mark = 0;
5884 		flow_split_info->table_id = 0;
5885 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5886 					      &q_attr, mtr_sfx ? items :
5887 					      q_items, q_actions,
5888 					      flow_split_info, error);
5889 		if (ret < 0)
5890 			goto exit;
5891 		/* qrss ID should be freed if failed. */
5892 		qrss_id = 0;
5893 		MLX5_ASSERT(dev_flow);
5894 	}
5895 
5896 exit:
5897 	/*
5898 	 * We do not destroy the partially created sub_flows in case of error.
5899 	 * These ones are included into parent flow list and will be destroyed
5900 	 * by flow_drv_destroy.
5901 	 */
5902 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5903 			qrss_id);
5904 	mlx5_free(ext_actions);
5905 	return ret;
5906 }
5907 
5908 /**
5909  * Create meter internal drop flow with the original pattern.
5910  *
5911  * @param dev
5912  *   Pointer to Ethernet device.
5913  * @param[in] flow
5914  *   Parent flow structure pointer.
5915  * @param[in] attr
5916  *   Flow rule attributes.
5917  * @param[in] items
5918  *   Pattern specification (list terminated by the END pattern item).
5919  * @param[in] flow_split_info
5920  *   Pointer to flow split info structure.
5921  * @param[in] fm
5922  *   Pointer to flow meter structure.
5923  * @param[out] error
5924  *   Perform verbose error reporting if not NULL.
5925  * @return
5926  *   0 on success, negative value otherwise
5927  */
5928 static uint32_t
5929 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5930 			struct rte_flow *flow,
5931 			const struct rte_flow_attr *attr,
5932 			const struct rte_flow_item items[],
5933 			struct mlx5_flow_split_info *flow_split_info,
5934 			struct mlx5_flow_meter_info *fm,
5935 			struct rte_flow_error *error)
5936 {
5937 	struct mlx5_flow *dev_flow = NULL;
5938 	struct rte_flow_attr drop_attr = *attr;
5939 	struct rte_flow_action drop_actions[3];
5940 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5941 
5942 	MLX5_ASSERT(fm->drop_cnt);
5943 	drop_actions[0].type =
5944 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5945 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5946 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5947 	drop_actions[1].conf = NULL;
5948 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5949 	drop_actions[2].conf = NULL;
5950 	drop_split_info.external = false;
5951 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5952 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5953 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5954 	return flow_create_split_inner(dev, flow, &dev_flow,
5955 				&drop_attr, items, drop_actions,
5956 				&drop_split_info, error);
5957 }
5958 
5959 /**
5960  * The splitting for meter feature.
5961  *
5962  * - The meter flow will be split to two flows as prefix and
5963  *   suffix flow. The packets make sense only it pass the prefix
5964  *   meter action.
5965  *
5966  * - Reg_C_5 is used for the packet to match betweend prefix and
5967  *   suffix flow.
5968  *
5969  * @param dev
5970  *   Pointer to Ethernet device.
5971  * @param[in] flow
5972  *   Parent flow structure pointer.
5973  * @param[in] attr
5974  *   Flow rule attributes.
5975  * @param[in] items
5976  *   Pattern specification (list terminated by the END pattern item).
5977  * @param[in] actions
5978  *   Associated actions (list terminated by the END action).
5979  * @param[in] flow_split_info
5980  *   Pointer to flow split info structure.
5981  * @param[out] error
5982  *   Perform verbose error reporting if not NULL.
5983  * @return
5984  *   0 on success, negative value otherwise
5985  */
5986 static int
5987 flow_create_split_meter(struct rte_eth_dev *dev,
5988 			struct rte_flow *flow,
5989 			const struct rte_flow_attr *attr,
5990 			const struct rte_flow_item items[],
5991 			const struct rte_flow_action actions[],
5992 			struct mlx5_flow_split_info *flow_split_info,
5993 			struct rte_flow_error *error)
5994 {
5995 	struct mlx5_priv *priv = dev->data->dev_private;
5996 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5997 	struct rte_flow_action *sfx_actions = NULL;
5998 	struct rte_flow_action *pre_actions = NULL;
5999 	struct rte_flow_item *sfx_items = NULL;
6000 	struct mlx5_flow *dev_flow = NULL;
6001 	struct rte_flow_attr sfx_attr = *attr;
6002 	struct mlx5_flow_meter_info *fm = NULL;
6003 	uint8_t skip_scale_restore;
6004 	bool has_mtr = false;
6005 	bool has_modify = false;
6006 	bool set_mtr_reg = true;
6007 	bool is_mtr_hierarchy = false;
6008 	uint32_t meter_id = 0;
6009 	uint32_t mtr_idx = 0;
6010 	uint32_t mtr_flow_id = 0;
6011 	size_t act_size;
6012 	size_t item_size;
6013 	int actions_n = 0;
6014 	int ret = 0;
6015 
6016 	if (priv->mtr_en)
6017 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6018 						    &has_modify, &meter_id);
6019 	if (has_mtr) {
6020 		if (flow->meter) {
6021 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6022 			if (!fm)
6023 				return rte_flow_error_set(error, EINVAL,
6024 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6025 						NULL, "Meter not found.");
6026 		} else {
6027 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6028 			if (!fm)
6029 				return rte_flow_error_set(error, EINVAL,
6030 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6031 						NULL, "Meter not found.");
6032 			ret = mlx5_flow_meter_attach(priv, fm,
6033 						     &sfx_attr, error);
6034 			if (ret)
6035 				return -rte_errno;
6036 			flow->meter = mtr_idx;
6037 		}
6038 		MLX5_ASSERT(wks);
6039 		wks->fm = fm;
6040 		if (!fm->def_policy) {
6041 			wks->policy = mlx5_flow_meter_policy_find(dev,
6042 								  fm->policy_id,
6043 								  NULL);
6044 			MLX5_ASSERT(wks->policy);
6045 			if (wks->policy->is_hierarchy) {
6046 				wks->final_policy =
6047 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6048 								wks->policy);
6049 				if (!wks->final_policy)
6050 					return rte_flow_error_set(error,
6051 					EINVAL,
6052 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6053 				"Failed to find terminal policy of hierarchy.");
6054 				is_mtr_hierarchy = true;
6055 			}
6056 		}
6057 		/*
6058 		 * If it isn't default-policy Meter, and
6059 		 * 1. There's no action in flow to change
6060 		 *    packet (modify/encap/decap etc.), OR
6061 		 * 2. No drop count needed for this meter.
6062 		 * 3. It's not meter hierarchy.
6063 		 * Then no need to use regC to save meter id anymore.
6064 		 */
6065 		if (!fm->def_policy && !is_mtr_hierarchy &&
6066 		    (!has_modify || !fm->drop_cnt))
6067 			set_mtr_reg = false;
6068 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6069 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6070 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6071 		/* Suffix items: tag, vlan, port id, end. */
6072 #define METER_SUFFIX_ITEM 4
6073 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6074 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6075 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6076 					  0, SOCKET_ID_ANY);
6077 		if (!sfx_actions)
6078 			return rte_flow_error_set(error, ENOMEM,
6079 						  RTE_FLOW_ERROR_TYPE_ACTION,
6080 						  NULL, "no memory to split "
6081 						  "meter flow");
6082 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6083 			     act_size);
6084 		/* There's no suffix flow for meter of non-default policy. */
6085 		if (!fm->def_policy)
6086 			pre_actions = sfx_actions + 1;
6087 		else
6088 			pre_actions = sfx_actions + actions_n;
6089 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6090 					    items, sfx_items, actions,
6091 					    sfx_actions, pre_actions,
6092 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6093 					    error);
6094 		if (ret) {
6095 			ret = -rte_errno;
6096 			goto exit;
6097 		}
6098 		/* Add the prefix subflow. */
6099 		skip_scale_restore = flow_split_info->skip_scale;
6100 		flow_split_info->skip_scale |=
6101 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6102 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6103 					      attr, items, pre_actions,
6104 					      flow_split_info, error);
6105 		flow_split_info->skip_scale = skip_scale_restore;
6106 		if (ret) {
6107 			if (mtr_flow_id)
6108 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6109 			ret = -rte_errno;
6110 			goto exit;
6111 		}
6112 		if (mtr_flow_id) {
6113 			dev_flow->handle->split_flow_id = mtr_flow_id;
6114 			dev_flow->handle->is_meter_flow_id = 1;
6115 		}
6116 		if (!fm->def_policy) {
6117 			if (!set_mtr_reg && fm->drop_cnt)
6118 				ret =
6119 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6120 							&sfx_attr, items,
6121 							flow_split_info,
6122 							fm, error);
6123 			goto exit;
6124 		}
6125 		/* Setting the sfx group atrr. */
6126 		sfx_attr.group = sfx_attr.transfer ?
6127 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6128 				 MLX5_FLOW_TABLE_LEVEL_METER;
6129 		flow_split_info->prefix_layers =
6130 				flow_get_prefix_layer_flags(dev_flow);
6131 		flow_split_info->prefix_mark |= dev_flow->handle->mark;
6132 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6133 	}
6134 	/* Add the prefix subflow. */
6135 	ret = flow_create_split_metadata(dev, flow,
6136 					 &sfx_attr, sfx_items ?
6137 					 sfx_items : items,
6138 					 sfx_actions ? sfx_actions : actions,
6139 					 flow_split_info, error);
6140 exit:
6141 	if (sfx_actions)
6142 		mlx5_free(sfx_actions);
6143 	return ret;
6144 }
6145 
6146 /**
6147  * The splitting for sample feature.
6148  *
6149  * Once Sample action is detected in the action list, the flow actions should
6150  * be split into prefix sub flow and suffix sub flow.
6151  *
6152  * The original items remain in the prefix sub flow, all actions preceding the
6153  * sample action and the sample action itself will be copied to the prefix
6154  * sub flow, the actions following the sample action will be copied to the
6155  * suffix sub flow, Queue action always be located in the suffix sub flow.
6156  *
6157  * In order to make the packet from prefix sub flow matches with suffix sub
6158  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6159  * flow uses tag item with the unique flow id.
6160  *
6161  * @param dev
6162  *   Pointer to Ethernet device.
6163  * @param[in] flow
6164  *   Parent flow structure pointer.
6165  * @param[in] attr
6166  *   Flow rule attributes.
6167  * @param[in] items
6168  *   Pattern specification (list terminated by the END pattern item).
6169  * @param[in] actions
6170  *   Associated actions (list terminated by the END action).
6171  * @param[in] flow_split_info
6172  *   Pointer to flow split info structure.
6173  * @param[out] error
6174  *   Perform verbose error reporting if not NULL.
6175  * @return
6176  *   0 on success, negative value otherwise
6177  */
6178 static int
6179 flow_create_split_sample(struct rte_eth_dev *dev,
6180 			 struct rte_flow *flow,
6181 			 const struct rte_flow_attr *attr,
6182 			 const struct rte_flow_item items[],
6183 			 const struct rte_flow_action actions[],
6184 			 struct mlx5_flow_split_info *flow_split_info,
6185 			 struct rte_flow_error *error)
6186 {
6187 	struct mlx5_priv *priv = dev->data->dev_private;
6188 	struct rte_flow_action *sfx_actions = NULL;
6189 	struct rte_flow_action *pre_actions = NULL;
6190 	struct rte_flow_item *sfx_items = NULL;
6191 	struct mlx5_flow *dev_flow = NULL;
6192 	struct rte_flow_attr sfx_attr = *attr;
6193 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6194 	struct mlx5_flow_dv_sample_resource *sample_res;
6195 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6196 	struct mlx5_flow_tbl_resource *sfx_tbl;
6197 #endif
6198 	size_t act_size;
6199 	size_t item_size;
6200 	uint32_t fdb_tx = 0;
6201 	int32_t tag_id = 0;
6202 	int actions_n = 0;
6203 	int sample_action_pos;
6204 	int qrss_action_pos;
6205 	int add_tag = 0;
6206 	int modify_after_mirror = 0;
6207 	uint16_t jump_table = 0;
6208 	const uint32_t next_ft_step = 1;
6209 	int ret = 0;
6210 
6211 	if (priv->sampler_en)
6212 		actions_n = flow_check_match_action(actions, attr,
6213 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6214 					&sample_action_pos, &qrss_action_pos,
6215 					&modify_after_mirror);
6216 	if (actions_n) {
6217 		/* The prefix actions must includes sample, tag, end. */
6218 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6219 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6220 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6221 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6222 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6223 					  item_size), 0, SOCKET_ID_ANY);
6224 		if (!sfx_actions)
6225 			return rte_flow_error_set(error, ENOMEM,
6226 						  RTE_FLOW_ERROR_TYPE_ACTION,
6227 						  NULL, "no memory to split "
6228 						  "sample flow");
6229 		/* The representor_id is UINT16_MAX for uplink. */
6230 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6231 		/*
6232 		 * When reg_c_preserve is set, metadata registers Cx preserve
6233 		 * their value even through packet duplication.
6234 		 */
6235 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6236 		if (add_tag)
6237 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6238 					+ act_size);
6239 		if (modify_after_mirror)
6240 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6241 				     next_ft_step;
6242 		pre_actions = sfx_actions + actions_n;
6243 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6244 						actions, sfx_actions,
6245 						pre_actions, actions_n,
6246 						sample_action_pos,
6247 						qrss_action_pos, jump_table,
6248 						error);
6249 		if (tag_id < 0 || (add_tag && !tag_id)) {
6250 			ret = -rte_errno;
6251 			goto exit;
6252 		}
6253 		if (modify_after_mirror)
6254 			flow_split_info->skip_scale =
6255 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6256 		/* Add the prefix subflow. */
6257 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6258 					      items, pre_actions,
6259 					      flow_split_info, error);
6260 		if (ret) {
6261 			ret = -rte_errno;
6262 			goto exit;
6263 		}
6264 		dev_flow->handle->split_flow_id = tag_id;
6265 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6266 		if (!modify_after_mirror) {
6267 			/* Set the sfx group attr. */
6268 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6269 						dev_flow->dv.sample_res;
6270 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6271 						sample_res->normal_path_tbl;
6272 			sfx_tbl_data = container_of(sfx_tbl,
6273 						struct mlx5_flow_tbl_data_entry,
6274 						tbl);
6275 			sfx_attr.group = sfx_attr.transfer ?
6276 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6277 		} else {
6278 			MLX5_ASSERT(attr->transfer);
6279 			sfx_attr.group = jump_table;
6280 		}
6281 		flow_split_info->prefix_layers =
6282 				flow_get_prefix_layer_flags(dev_flow);
6283 		flow_split_info->prefix_mark |= dev_flow->handle->mark;
6284 		/* Suffix group level already be scaled with factor, set
6285 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6286 		 * again in translation.
6287 		 */
6288 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6289 #endif
6290 	}
6291 	/* Add the suffix subflow. */
6292 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6293 				      sfx_items ? sfx_items : items,
6294 				      sfx_actions ? sfx_actions : actions,
6295 				      flow_split_info, error);
6296 exit:
6297 	if (sfx_actions)
6298 		mlx5_free(sfx_actions);
6299 	return ret;
6300 }
6301 
6302 /**
6303  * Split the flow to subflow set. The splitters might be linked
6304  * in the chain, like this:
6305  * flow_create_split_outer() calls:
6306  *   flow_create_split_meter() calls:
6307  *     flow_create_split_metadata(meter_subflow_0) calls:
6308  *       flow_create_split_inner(metadata_subflow_0)
6309  *       flow_create_split_inner(metadata_subflow_1)
6310  *       flow_create_split_inner(metadata_subflow_2)
6311  *     flow_create_split_metadata(meter_subflow_1) calls:
6312  *       flow_create_split_inner(metadata_subflow_0)
6313  *       flow_create_split_inner(metadata_subflow_1)
6314  *       flow_create_split_inner(metadata_subflow_2)
6315  *
6316  * This provide flexible way to add new levels of flow splitting.
6317  * The all of successfully created subflows are included to the
6318  * parent flow dev_flow list.
6319  *
6320  * @param dev
6321  *   Pointer to Ethernet device.
6322  * @param[in] flow
6323  *   Parent flow structure pointer.
6324  * @param[in] attr
6325  *   Flow rule attributes.
6326  * @param[in] items
6327  *   Pattern specification (list terminated by the END pattern item).
6328  * @param[in] actions
6329  *   Associated actions (list terminated by the END action).
6330  * @param[in] flow_split_info
6331  *   Pointer to flow split info structure.
6332  * @param[out] error
6333  *   Perform verbose error reporting if not NULL.
6334  * @return
6335  *   0 on success, negative value otherwise
6336  */
6337 static int
6338 flow_create_split_outer(struct rte_eth_dev *dev,
6339 			struct rte_flow *flow,
6340 			const struct rte_flow_attr *attr,
6341 			const struct rte_flow_item items[],
6342 			const struct rte_flow_action actions[],
6343 			struct mlx5_flow_split_info *flow_split_info,
6344 			struct rte_flow_error *error)
6345 {
6346 	int ret;
6347 
6348 	ret = flow_create_split_sample(dev, flow, attr, items,
6349 				       actions, flow_split_info, error);
6350 	MLX5_ASSERT(ret <= 0);
6351 	return ret;
6352 }
6353 
6354 static inline struct mlx5_flow_tunnel *
6355 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6356 {
6357 	struct mlx5_flow_tunnel *tunnel;
6358 
6359 #pragma GCC diagnostic push
6360 #pragma GCC diagnostic ignored "-Wcast-qual"
6361 	tunnel = (typeof(tunnel))flow->tunnel;
6362 #pragma GCC diagnostic pop
6363 
6364 	return tunnel;
6365 }
6366 
6367 /**
6368  * Adjust flow RSS workspace if needed.
6369  *
6370  * @param wks
6371  *   Pointer to thread flow work space.
6372  * @param rss_desc
6373  *   Pointer to RSS descriptor.
6374  * @param[in] nrssq_num
6375  *   New RSS queue number.
6376  *
6377  * @return
6378  *   0 on success, -1 otherwise and rte_errno is set.
6379  */
6380 static int
6381 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6382 			  struct mlx5_flow_rss_desc *rss_desc,
6383 			  uint32_t nrssq_num)
6384 {
6385 	if (likely(nrssq_num <= wks->rssq_num))
6386 		return 0;
6387 	rss_desc->queue = realloc(rss_desc->queue,
6388 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6389 	if (!rss_desc->queue) {
6390 		rte_errno = ENOMEM;
6391 		return -1;
6392 	}
6393 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6394 	return 0;
6395 }
6396 
6397 /**
6398  * Create a flow and add it to @p list.
6399  *
6400  * @param dev
6401  *   Pointer to Ethernet device.
6402  * @param list
6403  *   Pointer to a TAILQ flow list. If this parameter NULL,
6404  *   no list insertion occurred, flow is just created,
6405  *   this is caller's responsibility to track the
6406  *   created flow.
6407  * @param[in] attr
6408  *   Flow rule attributes.
6409  * @param[in] items
6410  *   Pattern specification (list terminated by the END pattern item).
6411  * @param[in] actions
6412  *   Associated actions (list terminated by the END action).
6413  * @param[in] external
6414  *   This flow rule is created by request external to PMD.
6415  * @param[out] error
6416  *   Perform verbose error reporting if not NULL.
6417  *
6418  * @return
6419  *   A flow index on success, 0 otherwise and rte_errno is set.
6420  */
6421 static uint32_t
6422 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6423 		 const struct rte_flow_attr *attr,
6424 		 const struct rte_flow_item items[],
6425 		 const struct rte_flow_action original_actions[],
6426 		 bool external, struct rte_flow_error *error)
6427 {
6428 	struct mlx5_priv *priv = dev->data->dev_private;
6429 	struct rte_flow *flow = NULL;
6430 	struct mlx5_flow *dev_flow;
6431 	const struct rte_flow_action_rss *rss = NULL;
6432 	struct mlx5_translated_action_handle
6433 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6434 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6435 	union {
6436 		struct mlx5_flow_expand_rss buf;
6437 		uint8_t buffer[4096];
6438 	} expand_buffer;
6439 	union {
6440 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6441 		uint8_t buffer[2048];
6442 	} actions_rx;
6443 	union {
6444 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6445 		uint8_t buffer[2048];
6446 	} actions_hairpin_tx;
6447 	union {
6448 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6449 		uint8_t buffer[2048];
6450 	} items_tx;
6451 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6452 	struct mlx5_flow_rss_desc *rss_desc;
6453 	const struct rte_flow_action *p_actions_rx;
6454 	uint32_t i;
6455 	uint32_t idx = 0;
6456 	int hairpin_flow;
6457 	struct rte_flow_attr attr_tx = { .priority = 0 };
6458 	const struct rte_flow_action *actions;
6459 	struct rte_flow_action *translated_actions = NULL;
6460 	struct mlx5_flow_tunnel *tunnel;
6461 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6462 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6463 	struct mlx5_flow_split_info flow_split_info = {
6464 		.external = !!external,
6465 		.skip_scale = 0,
6466 		.flow_idx = 0,
6467 		.prefix_mark = 0,
6468 		.prefix_layers = 0,
6469 		.table_id = 0
6470 	};
6471 	int ret;
6472 
6473 	MLX5_ASSERT(wks);
6474 	rss_desc = &wks->rss_desc;
6475 	ret = flow_action_handles_translate(dev, original_actions,
6476 					    indir_actions,
6477 					    &indir_actions_n,
6478 					    &translated_actions, error);
6479 	if (ret < 0) {
6480 		MLX5_ASSERT(translated_actions == NULL);
6481 		return 0;
6482 	}
6483 	actions = translated_actions ? translated_actions : original_actions;
6484 	p_actions_rx = actions;
6485 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6486 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6487 				external, hairpin_flow, error);
6488 	if (ret < 0)
6489 		goto error_before_hairpin_split;
6490 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6491 	if (!flow) {
6492 		rte_errno = ENOMEM;
6493 		goto error_before_hairpin_split;
6494 	}
6495 	if (hairpin_flow > 0) {
6496 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6497 			rte_errno = EINVAL;
6498 			goto error_before_hairpin_split;
6499 		}
6500 		flow_hairpin_split(dev, actions, actions_rx.actions,
6501 				   actions_hairpin_tx.actions, items_tx.items,
6502 				   idx);
6503 		p_actions_rx = actions_rx.actions;
6504 	}
6505 	flow_split_info.flow_idx = idx;
6506 	flow->drv_type = flow_get_drv_type(dev, attr);
6507 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6508 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6509 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6510 	/* RSS Action only works on NIC RX domain */
6511 	if (attr->ingress && !attr->transfer)
6512 		rss = flow_get_rss_action(dev, p_actions_rx);
6513 	if (rss) {
6514 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6515 			return 0;
6516 		/*
6517 		 * The following information is required by
6518 		 * mlx5_flow_hashfields_adjust() in advance.
6519 		 */
6520 		rss_desc->level = rss->level;
6521 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6522 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6523 	}
6524 	flow->dev_handles = 0;
6525 	if (rss && rss->types) {
6526 		unsigned int graph_root;
6527 
6528 		graph_root = find_graph_root(rss->level);
6529 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6530 					   items, rss->types,
6531 					   mlx5_support_expansion, graph_root);
6532 		MLX5_ASSERT(ret > 0 &&
6533 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6534 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6535 			for (i = 0; i < buf->entries; ++i)
6536 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6537 		}
6538 	} else {
6539 		buf->entries = 1;
6540 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6541 	}
6542 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6543 						      indir_actions_n);
6544 	for (i = 0; i < buf->entries; ++i) {
6545 		/* Initialize flow split data. */
6546 		flow_split_info.prefix_layers = 0;
6547 		flow_split_info.prefix_mark = 0;
6548 		flow_split_info.skip_scale = 0;
6549 		/*
6550 		 * The splitter may create multiple dev_flows,
6551 		 * depending on configuration. In the simplest
6552 		 * case it just creates unmodified original flow.
6553 		 */
6554 		ret = flow_create_split_outer(dev, flow, attr,
6555 					      buf->entry[i].pattern,
6556 					      p_actions_rx, &flow_split_info,
6557 					      error);
6558 		if (ret < 0)
6559 			goto error;
6560 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6561 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6562 							   p_actions_rx,
6563 							   idx,
6564 							   wks->flows[0].tunnel,
6565 							   &default_miss_ctx,
6566 							   error);
6567 			if (ret < 0) {
6568 				mlx5_free(default_miss_ctx.queue);
6569 				goto error;
6570 			}
6571 		}
6572 	}
6573 	/* Create the tx flow. */
6574 	if (hairpin_flow) {
6575 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6576 		attr_tx.ingress = 0;
6577 		attr_tx.egress = 1;
6578 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6579 					 actions_hairpin_tx.actions,
6580 					 idx, error);
6581 		if (!dev_flow)
6582 			goto error;
6583 		dev_flow->flow = flow;
6584 		dev_flow->external = 0;
6585 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6586 			      dev_flow->handle, next);
6587 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6588 					 items_tx.items,
6589 					 actions_hairpin_tx.actions, error);
6590 		if (ret < 0)
6591 			goto error;
6592 	}
6593 	/*
6594 	 * Update the metadata register copy table. If extensive
6595 	 * metadata feature is enabled and registers are supported
6596 	 * we might create the extra rte_flow for each unique
6597 	 * MARK/FLAG action ID.
6598 	 *
6599 	 * The table is updated for ingress Flows only, because
6600 	 * the egress Flows belong to the different device and
6601 	 * copy table should be updated in peer NIC Rx domain.
6602 	 */
6603 	if (attr->ingress &&
6604 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6605 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6606 		if (ret)
6607 			goto error;
6608 	}
6609 	/*
6610 	 * If the flow is external (from application) OR device is started,
6611 	 * OR mreg discover, then apply immediately.
6612 	 */
6613 	if (external || dev->data->dev_started ||
6614 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6615 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6616 		ret = flow_drv_apply(dev, flow, error);
6617 		if (ret < 0)
6618 			goto error;
6619 	}
6620 	flow->type = type;
6621 	flow_rxq_flags_set(dev, flow);
6622 	rte_free(translated_actions);
6623 	tunnel = flow_tunnel_from_rule(wks->flows);
6624 	if (tunnel) {
6625 		flow->tunnel = 1;
6626 		flow->tunnel_id = tunnel->tunnel_id;
6627 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6628 		mlx5_free(default_miss_ctx.queue);
6629 	}
6630 	mlx5_flow_pop_thread_workspace();
6631 	return idx;
6632 error:
6633 	MLX5_ASSERT(flow);
6634 	ret = rte_errno; /* Save rte_errno before cleanup. */
6635 	flow_mreg_del_copy_action(dev, flow);
6636 	flow_drv_destroy(dev, flow);
6637 	if (rss_desc->shared_rss)
6638 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6639 			mlx5_ipool_get
6640 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6641 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6642 	mlx5_ipool_free(priv->flows[type], idx);
6643 	rte_errno = ret; /* Restore rte_errno. */
6644 	ret = rte_errno;
6645 	rte_errno = ret;
6646 	mlx5_flow_pop_thread_workspace();
6647 error_before_hairpin_split:
6648 	rte_free(translated_actions);
6649 	return 0;
6650 }
6651 
6652 /**
6653  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6654  * incoming packets to table 1.
6655  *
6656  * Other flow rules, requested for group n, will be created in
6657  * e-switch table n+1.
6658  * Jump action to e-switch group n will be created to group n+1.
6659  *
6660  * Used when working in switchdev mode, to utilise advantages of table 1
6661  * and above.
6662  *
6663  * @param dev
6664  *   Pointer to Ethernet device.
6665  *
6666  * @return
6667  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6668  */
6669 struct rte_flow *
6670 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6671 {
6672 	const struct rte_flow_attr attr = {
6673 		.group = 0,
6674 		.priority = 0,
6675 		.ingress = 1,
6676 		.egress = 0,
6677 		.transfer = 1,
6678 	};
6679 	const struct rte_flow_item pattern = {
6680 		.type = RTE_FLOW_ITEM_TYPE_END,
6681 	};
6682 	struct rte_flow_action_jump jump = {
6683 		.group = 1,
6684 	};
6685 	const struct rte_flow_action actions[] = {
6686 		{
6687 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6688 			.conf = &jump,
6689 		},
6690 		{
6691 			.type = RTE_FLOW_ACTION_TYPE_END,
6692 		},
6693 	};
6694 	struct rte_flow_error error;
6695 
6696 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6697 						   &attr, &pattern,
6698 						   actions, false, &error);
6699 }
6700 
6701 /**
6702  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6703  * and sq number, directs all packets to peer vport.
6704  *
6705  * @param dev
6706  *   Pointer to Ethernet device.
6707  * @param txq
6708  *   Txq index.
6709  *
6710  * @return
6711  *   Flow ID on success, 0 otherwise and rte_errno is set.
6712  */
6713 uint32_t
6714 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6715 {
6716 	struct rte_flow_attr attr = {
6717 		.group = 0,
6718 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6719 		.ingress = 1,
6720 		.egress = 0,
6721 		.transfer = 1,
6722 	};
6723 	struct rte_flow_item_port_id port_spec = {
6724 		.id = MLX5_PORT_ESW_MGR,
6725 	};
6726 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6727 		.queue = txq,
6728 	};
6729 	struct rte_flow_item pattern[] = {
6730 		{
6731 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6732 			.spec = &port_spec,
6733 		},
6734 		{
6735 			.type = (enum rte_flow_item_type)
6736 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6737 			.spec = &txq_spec,
6738 		},
6739 		{
6740 			.type = RTE_FLOW_ITEM_TYPE_END,
6741 		},
6742 	};
6743 	struct rte_flow_action_jump jump = {
6744 		.group = 1,
6745 	};
6746 	struct rte_flow_action_port_id port = {
6747 		.id = dev->data->port_id,
6748 	};
6749 	struct rte_flow_action actions[] = {
6750 		{
6751 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6752 			.conf = &jump,
6753 		},
6754 		{
6755 			.type = RTE_FLOW_ACTION_TYPE_END,
6756 		},
6757 	};
6758 	struct rte_flow_error error;
6759 
6760 	/*
6761 	 * Creates group 0, highest priority jump flow.
6762 	 * Matches txq to bypass kernel packets.
6763 	 */
6764 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6765 			     false, &error) == 0)
6766 		return 0;
6767 	/* Create group 1, lowest priority redirect flow for txq. */
6768 	attr.group = 1;
6769 	actions[0].conf = &port;
6770 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6771 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6772 				actions, false, &error);
6773 }
6774 
6775 /**
6776  * Validate a flow supported by the NIC.
6777  *
6778  * @see rte_flow_validate()
6779  * @see rte_flow_ops
6780  */
6781 int
6782 mlx5_flow_validate(struct rte_eth_dev *dev,
6783 		   const struct rte_flow_attr *attr,
6784 		   const struct rte_flow_item items[],
6785 		   const struct rte_flow_action original_actions[],
6786 		   struct rte_flow_error *error)
6787 {
6788 	int hairpin_flow;
6789 	struct mlx5_translated_action_handle
6790 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6791 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6792 	const struct rte_flow_action *actions;
6793 	struct rte_flow_action *translated_actions = NULL;
6794 	int ret = flow_action_handles_translate(dev, original_actions,
6795 						indir_actions,
6796 						&indir_actions_n,
6797 						&translated_actions, error);
6798 
6799 	if (ret)
6800 		return ret;
6801 	actions = translated_actions ? translated_actions : original_actions;
6802 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6803 	ret = flow_drv_validate(dev, attr, items, actions,
6804 				true, hairpin_flow, error);
6805 	rte_free(translated_actions);
6806 	return ret;
6807 }
6808 
6809 /**
6810  * Create a flow.
6811  *
6812  * @see rte_flow_create()
6813  * @see rte_flow_ops
6814  */
6815 struct rte_flow *
6816 mlx5_flow_create(struct rte_eth_dev *dev,
6817 		 const struct rte_flow_attr *attr,
6818 		 const struct rte_flow_item items[],
6819 		 const struct rte_flow_action actions[],
6820 		 struct rte_flow_error *error)
6821 {
6822 	/*
6823 	 * If the device is not started yet, it is not allowed to created a
6824 	 * flow from application. PMD default flows and traffic control flows
6825 	 * are not affected.
6826 	 */
6827 	if (unlikely(!dev->data->dev_started)) {
6828 		DRV_LOG(DEBUG, "port %u is not started when "
6829 			"inserting a flow", dev->data->port_id);
6830 		rte_flow_error_set(error, ENODEV,
6831 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6832 				   NULL,
6833 				   "port not started");
6834 		return NULL;
6835 	}
6836 
6837 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6838 						   attr, items, actions,
6839 						   true, error);
6840 }
6841 
6842 /**
6843  * Destroy a flow in a list.
6844  *
6845  * @param dev
6846  *   Pointer to Ethernet device.
6847  * @param[in] flow_idx
6848  *   Index of flow to destroy.
6849  */
6850 static void
6851 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6852 		  uint32_t flow_idx)
6853 {
6854 	struct mlx5_priv *priv = dev->data->dev_private;
6855 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6856 
6857 	if (!flow)
6858 		return;
6859 	MLX5_ASSERT(flow->type == type);
6860 	/*
6861 	 * Update RX queue flags only if port is started, otherwise it is
6862 	 * already clean.
6863 	 */
6864 	if (dev->data->dev_started)
6865 		flow_rxq_flags_trim(dev, flow);
6866 	flow_drv_destroy(dev, flow);
6867 	if (flow->tunnel) {
6868 		struct mlx5_flow_tunnel *tunnel;
6869 
6870 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6871 		RTE_VERIFY(tunnel);
6872 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6873 			mlx5_flow_tunnel_free(dev, tunnel);
6874 	}
6875 	flow_mreg_del_copy_action(dev, flow);
6876 	mlx5_ipool_free(priv->flows[type], flow_idx);
6877 }
6878 
6879 /**
6880  * Destroy all flows.
6881  *
6882  * @param dev
6883  *   Pointer to Ethernet device.
6884  * @param type
6885  *   Flow type to be flushed.
6886  * @param active
6887  *   If flushing is called avtively.
6888  */
6889 void
6890 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6891 		     bool active)
6892 {
6893 	struct mlx5_priv *priv = dev->data->dev_private;
6894 	uint32_t num_flushed = 0, fidx = 1;
6895 	struct rte_flow *flow;
6896 
6897 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6898 		flow_list_destroy(dev, type, fidx);
6899 		num_flushed++;
6900 	}
6901 	if (active) {
6902 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6903 			dev->data->port_id, num_flushed);
6904 	}
6905 }
6906 
6907 /**
6908  * Stop all default actions for flows.
6909  *
6910  * @param dev
6911  *   Pointer to Ethernet device.
6912  */
6913 void
6914 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6915 {
6916 	flow_mreg_del_default_copy_action(dev);
6917 	flow_rxq_flags_clear(dev);
6918 }
6919 
6920 /**
6921  * Start all default actions for flows.
6922  *
6923  * @param dev
6924  *   Pointer to Ethernet device.
6925  * @return
6926  *   0 on success, a negative errno value otherwise and rte_errno is set.
6927  */
6928 int
6929 mlx5_flow_start_default(struct rte_eth_dev *dev)
6930 {
6931 	struct rte_flow_error error;
6932 
6933 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6934 	return flow_mreg_add_default_copy_action(dev, &error);
6935 }
6936 
6937 /**
6938  * Release key of thread specific flow workspace data.
6939  */
6940 void
6941 flow_release_workspace(void *data)
6942 {
6943 	struct mlx5_flow_workspace *wks = data;
6944 	struct mlx5_flow_workspace *next;
6945 
6946 	while (wks) {
6947 		next = wks->next;
6948 		free(wks->rss_desc.queue);
6949 		free(wks);
6950 		wks = next;
6951 	}
6952 }
6953 
6954 /**
6955  * Get thread specific current flow workspace.
6956  *
6957  * @return pointer to thread specific flow workspace data, NULL on error.
6958  */
6959 struct mlx5_flow_workspace*
6960 mlx5_flow_get_thread_workspace(void)
6961 {
6962 	struct mlx5_flow_workspace *data;
6963 
6964 	data = mlx5_flow_os_get_specific_workspace();
6965 	MLX5_ASSERT(data && data->inuse);
6966 	if (!data || !data->inuse)
6967 		DRV_LOG(ERR, "flow workspace not initialized.");
6968 	return data;
6969 }
6970 
6971 /**
6972  * Allocate and init new flow workspace.
6973  *
6974  * @return pointer to flow workspace data, NULL on error.
6975  */
6976 static struct mlx5_flow_workspace*
6977 flow_alloc_thread_workspace(void)
6978 {
6979 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6980 
6981 	if (!data) {
6982 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6983 			"memory.");
6984 		return NULL;
6985 	}
6986 	data->rss_desc.queue = calloc(1,
6987 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6988 	if (!data->rss_desc.queue)
6989 		goto err;
6990 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6991 	return data;
6992 err:
6993 	if (data->rss_desc.queue)
6994 		free(data->rss_desc.queue);
6995 	free(data);
6996 	return NULL;
6997 }
6998 
6999 /**
7000  * Get new thread specific flow workspace.
7001  *
7002  * If current workspace inuse, create new one and set as current.
7003  *
7004  * @return pointer to thread specific flow workspace data, NULL on error.
7005  */
7006 static struct mlx5_flow_workspace*
7007 mlx5_flow_push_thread_workspace(void)
7008 {
7009 	struct mlx5_flow_workspace *curr;
7010 	struct mlx5_flow_workspace *data;
7011 
7012 	curr = mlx5_flow_os_get_specific_workspace();
7013 	if (!curr) {
7014 		data = flow_alloc_thread_workspace();
7015 		if (!data)
7016 			return NULL;
7017 	} else if (!curr->inuse) {
7018 		data = curr;
7019 	} else if (curr->next) {
7020 		data = curr->next;
7021 	} else {
7022 		data = flow_alloc_thread_workspace();
7023 		if (!data)
7024 			return NULL;
7025 		curr->next = data;
7026 		data->prev = curr;
7027 	}
7028 	data->inuse = 1;
7029 	data->flow_idx = 0;
7030 	/* Set as current workspace */
7031 	if (mlx5_flow_os_set_specific_workspace(data))
7032 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7033 	return data;
7034 }
7035 
7036 /**
7037  * Close current thread specific flow workspace.
7038  *
7039  * If previous workspace available, set it as current.
7040  *
7041  * @return pointer to thread specific flow workspace data, NULL on error.
7042  */
7043 static void
7044 mlx5_flow_pop_thread_workspace(void)
7045 {
7046 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7047 
7048 	if (!data)
7049 		return;
7050 	if (!data->inuse) {
7051 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7052 		return;
7053 	}
7054 	data->inuse = 0;
7055 	if (!data->prev)
7056 		return;
7057 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7058 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7059 }
7060 
7061 /**
7062  * Verify the flow list is empty
7063  *
7064  * @param dev
7065  *  Pointer to Ethernet device.
7066  *
7067  * @return the number of flows not released.
7068  */
7069 int
7070 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7071 {
7072 	struct mlx5_priv *priv = dev->data->dev_private;
7073 	struct rte_flow *flow;
7074 	uint32_t idx = 0;
7075 	int ret = 0, i;
7076 
7077 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7078 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7079 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7080 				dev->data->port_id, (void *)flow);
7081 			ret++;
7082 		}
7083 	}
7084 	return ret;
7085 }
7086 
7087 /**
7088  * Enable default hairpin egress flow.
7089  *
7090  * @param dev
7091  *   Pointer to Ethernet device.
7092  * @param queue
7093  *   The queue index.
7094  *
7095  * @return
7096  *   0 on success, a negative errno value otherwise and rte_errno is set.
7097  */
7098 int
7099 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7100 			    uint32_t queue)
7101 {
7102 	const struct rte_flow_attr attr = {
7103 		.egress = 1,
7104 		.priority = 0,
7105 	};
7106 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7107 		.queue = queue,
7108 	};
7109 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7110 		.queue = UINT32_MAX,
7111 	};
7112 	struct rte_flow_item items[] = {
7113 		{
7114 			.type = (enum rte_flow_item_type)
7115 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7116 			.spec = &queue_spec,
7117 			.last = NULL,
7118 			.mask = &queue_mask,
7119 		},
7120 		{
7121 			.type = RTE_FLOW_ITEM_TYPE_END,
7122 		},
7123 	};
7124 	struct rte_flow_action_jump jump = {
7125 		.group = MLX5_HAIRPIN_TX_TABLE,
7126 	};
7127 	struct rte_flow_action actions[2];
7128 	uint32_t flow_idx;
7129 	struct rte_flow_error error;
7130 
7131 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7132 	actions[0].conf = &jump;
7133 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7134 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7135 				    &attr, items, actions, false, &error);
7136 	if (!flow_idx) {
7137 		DRV_LOG(DEBUG,
7138 			"Failed to create ctrl flow: rte_errno(%d),"
7139 			" type(%d), message(%s)",
7140 			rte_errno, error.type,
7141 			error.message ? error.message : " (no stated reason)");
7142 		return -rte_errno;
7143 	}
7144 	return 0;
7145 }
7146 
7147 /**
7148  * Enable a control flow configured from the control plane.
7149  *
7150  * @param dev
7151  *   Pointer to Ethernet device.
7152  * @param eth_spec
7153  *   An Ethernet flow spec to apply.
7154  * @param eth_mask
7155  *   An Ethernet flow mask to apply.
7156  * @param vlan_spec
7157  *   A VLAN flow spec to apply.
7158  * @param vlan_mask
7159  *   A VLAN flow mask to apply.
7160  *
7161  * @return
7162  *   0 on success, a negative errno value otherwise and rte_errno is set.
7163  */
7164 int
7165 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7166 		    struct rte_flow_item_eth *eth_spec,
7167 		    struct rte_flow_item_eth *eth_mask,
7168 		    struct rte_flow_item_vlan *vlan_spec,
7169 		    struct rte_flow_item_vlan *vlan_mask)
7170 {
7171 	struct mlx5_priv *priv = dev->data->dev_private;
7172 	const struct rte_flow_attr attr = {
7173 		.ingress = 1,
7174 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7175 	};
7176 	struct rte_flow_item items[] = {
7177 		{
7178 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7179 			.spec = eth_spec,
7180 			.last = NULL,
7181 			.mask = eth_mask,
7182 		},
7183 		{
7184 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7185 					      RTE_FLOW_ITEM_TYPE_END,
7186 			.spec = vlan_spec,
7187 			.last = NULL,
7188 			.mask = vlan_mask,
7189 		},
7190 		{
7191 			.type = RTE_FLOW_ITEM_TYPE_END,
7192 		},
7193 	};
7194 	uint16_t queue[priv->reta_idx_n];
7195 	struct rte_flow_action_rss action_rss = {
7196 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7197 		.level = 0,
7198 		.types = priv->rss_conf.rss_hf,
7199 		.key_len = priv->rss_conf.rss_key_len,
7200 		.queue_num = priv->reta_idx_n,
7201 		.key = priv->rss_conf.rss_key,
7202 		.queue = queue,
7203 	};
7204 	struct rte_flow_action actions[] = {
7205 		{
7206 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7207 			.conf = &action_rss,
7208 		},
7209 		{
7210 			.type = RTE_FLOW_ACTION_TYPE_END,
7211 		},
7212 	};
7213 	uint32_t flow_idx;
7214 	struct rte_flow_error error;
7215 	unsigned int i;
7216 
7217 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7218 		return 0;
7219 	}
7220 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7221 		action_rss.types = 0;
7222 	for (i = 0; i != priv->reta_idx_n; ++i)
7223 		queue[i] = (*priv->reta_idx)[i];
7224 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7225 				    &attr, items, actions, false, &error);
7226 	if (!flow_idx)
7227 		return -rte_errno;
7228 	return 0;
7229 }
7230 
7231 /**
7232  * Enable a flow control configured from the control plane.
7233  *
7234  * @param dev
7235  *   Pointer to Ethernet device.
7236  * @param eth_spec
7237  *   An Ethernet flow spec to apply.
7238  * @param eth_mask
7239  *   An Ethernet flow mask to apply.
7240  *
7241  * @return
7242  *   0 on success, a negative errno value otherwise and rte_errno is set.
7243  */
7244 int
7245 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7246 	       struct rte_flow_item_eth *eth_spec,
7247 	       struct rte_flow_item_eth *eth_mask)
7248 {
7249 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7250 }
7251 
7252 /**
7253  * Create default miss flow rule matching lacp traffic
7254  *
7255  * @param dev
7256  *   Pointer to Ethernet device.
7257  * @param eth_spec
7258  *   An Ethernet flow spec to apply.
7259  *
7260  * @return
7261  *   0 on success, a negative errno value otherwise and rte_errno is set.
7262  */
7263 int
7264 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7265 {
7266 	/*
7267 	 * The LACP matching is done by only using ether type since using
7268 	 * a multicast dst mac causes kernel to give low priority to this flow.
7269 	 */
7270 	static const struct rte_flow_item_eth lacp_spec = {
7271 		.type = RTE_BE16(0x8809),
7272 	};
7273 	static const struct rte_flow_item_eth lacp_mask = {
7274 		.type = 0xffff,
7275 	};
7276 	const struct rte_flow_attr attr = {
7277 		.ingress = 1,
7278 	};
7279 	struct rte_flow_item items[] = {
7280 		{
7281 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7282 			.spec = &lacp_spec,
7283 			.mask = &lacp_mask,
7284 		},
7285 		{
7286 			.type = RTE_FLOW_ITEM_TYPE_END,
7287 		},
7288 	};
7289 	struct rte_flow_action actions[] = {
7290 		{
7291 			.type = (enum rte_flow_action_type)
7292 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7293 		},
7294 		{
7295 			.type = RTE_FLOW_ACTION_TYPE_END,
7296 		},
7297 	};
7298 	struct rte_flow_error error;
7299 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7300 					&attr, items, actions,
7301 					false, &error);
7302 
7303 	if (!flow_idx)
7304 		return -rte_errno;
7305 	return 0;
7306 }
7307 
7308 /**
7309  * Destroy a flow.
7310  *
7311  * @see rte_flow_destroy()
7312  * @see rte_flow_ops
7313  */
7314 int
7315 mlx5_flow_destroy(struct rte_eth_dev *dev,
7316 		  struct rte_flow *flow,
7317 		  struct rte_flow_error *error __rte_unused)
7318 {
7319 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7320 				(uintptr_t)(void *)flow);
7321 	return 0;
7322 }
7323 
7324 /**
7325  * Destroy all flows.
7326  *
7327  * @see rte_flow_flush()
7328  * @see rte_flow_ops
7329  */
7330 int
7331 mlx5_flow_flush(struct rte_eth_dev *dev,
7332 		struct rte_flow_error *error __rte_unused)
7333 {
7334 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7335 	return 0;
7336 }
7337 
7338 /**
7339  * Isolated mode.
7340  *
7341  * @see rte_flow_isolate()
7342  * @see rte_flow_ops
7343  */
7344 int
7345 mlx5_flow_isolate(struct rte_eth_dev *dev,
7346 		  int enable,
7347 		  struct rte_flow_error *error)
7348 {
7349 	struct mlx5_priv *priv = dev->data->dev_private;
7350 
7351 	if (dev->data->dev_started) {
7352 		rte_flow_error_set(error, EBUSY,
7353 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7354 				   NULL,
7355 				   "port must be stopped first");
7356 		return -rte_errno;
7357 	}
7358 	priv->isolated = !!enable;
7359 	if (enable)
7360 		dev->dev_ops = &mlx5_dev_ops_isolate;
7361 	else
7362 		dev->dev_ops = &mlx5_dev_ops;
7363 
7364 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7365 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7366 
7367 	return 0;
7368 }
7369 
7370 /**
7371  * Query a flow.
7372  *
7373  * @see rte_flow_query()
7374  * @see rte_flow_ops
7375  */
7376 static int
7377 flow_drv_query(struct rte_eth_dev *dev,
7378 	       uint32_t flow_idx,
7379 	       const struct rte_flow_action *actions,
7380 	       void *data,
7381 	       struct rte_flow_error *error)
7382 {
7383 	struct mlx5_priv *priv = dev->data->dev_private;
7384 	const struct mlx5_flow_driver_ops *fops;
7385 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7386 					       flow_idx);
7387 	enum mlx5_flow_drv_type ftype;
7388 
7389 	if (!flow) {
7390 		return rte_flow_error_set(error, ENOENT,
7391 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7392 			  NULL,
7393 			  "invalid flow handle");
7394 	}
7395 	ftype = flow->drv_type;
7396 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7397 	fops = flow_get_drv_ops(ftype);
7398 
7399 	return fops->query(dev, flow, actions, data, error);
7400 }
7401 
7402 /**
7403  * Query a flow.
7404  *
7405  * @see rte_flow_query()
7406  * @see rte_flow_ops
7407  */
7408 int
7409 mlx5_flow_query(struct rte_eth_dev *dev,
7410 		struct rte_flow *flow,
7411 		const struct rte_flow_action *actions,
7412 		void *data,
7413 		struct rte_flow_error *error)
7414 {
7415 	int ret;
7416 
7417 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7418 			     error);
7419 	if (ret < 0)
7420 		return ret;
7421 	return 0;
7422 }
7423 
7424 /**
7425  * Get rte_flow callbacks.
7426  *
7427  * @param dev
7428  *   Pointer to Ethernet device structure.
7429  * @param ops
7430  *   Pointer to operation-specific structure.
7431  *
7432  * @return 0
7433  */
7434 int
7435 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7436 		  const struct rte_flow_ops **ops)
7437 {
7438 	*ops = &mlx5_flow_ops;
7439 	return 0;
7440 }
7441 
7442 /**
7443  * Validate meter policy actions.
7444  * Dispatcher for action type specific validation.
7445  *
7446  * @param[in] dev
7447  *   Pointer to the Ethernet device structure.
7448  * @param[in] action
7449  *   The meter policy action object to validate.
7450  * @param[in] attr
7451  *   Attributes of flow to determine steering domain.
7452  * @param[out] is_rss
7453  *   Is RSS or not.
7454  * @param[out] domain_bitmap
7455  *   Domain bitmap.
7456  * @param[out] is_def_policy
7457  *   Is default policy or not.
7458  * @param[out] error
7459  *   Perform verbose error reporting if not NULL. Initialized in case of
7460  *   error only.
7461  *
7462  * @return
7463  *   0 on success, otherwise negative errno value.
7464  */
7465 int
7466 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7467 			const struct rte_flow_action *actions[RTE_COLORS],
7468 			struct rte_flow_attr *attr,
7469 			bool *is_rss,
7470 			uint8_t *domain_bitmap,
7471 			uint8_t *policy_mode,
7472 			struct rte_mtr_error *error)
7473 {
7474 	const struct mlx5_flow_driver_ops *fops;
7475 
7476 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7477 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7478 				       domain_bitmap, policy_mode, error);
7479 }
7480 
7481 /**
7482  * Destroy the meter table set.
7483  *
7484  * @param[in] dev
7485  *   Pointer to Ethernet device.
7486  * @param[in] mtr_policy
7487  *   Meter policy struct.
7488  */
7489 void
7490 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7491 		      struct mlx5_flow_meter_policy *mtr_policy)
7492 {
7493 	const struct mlx5_flow_driver_ops *fops;
7494 
7495 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7496 	fops->destroy_mtr_acts(dev, mtr_policy);
7497 }
7498 
7499 /**
7500  * Create policy action, lock free,
7501  * (mutex should be acquired by caller).
7502  * Dispatcher for action type specific call.
7503  *
7504  * @param[in] dev
7505  *   Pointer to the Ethernet device structure.
7506  * @param[in] mtr_policy
7507  *   Meter policy struct.
7508  * @param[in] action
7509  *   Action specification used to create meter actions.
7510  * @param[out] error
7511  *   Perform verbose error reporting if not NULL. Initialized in case of
7512  *   error only.
7513  *
7514  * @return
7515  *   0 on success, otherwise negative errno value.
7516  */
7517 int
7518 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7519 		      struct mlx5_flow_meter_policy *mtr_policy,
7520 		      const struct rte_flow_action *actions[RTE_COLORS],
7521 		      struct rte_mtr_error *error)
7522 {
7523 	const struct mlx5_flow_driver_ops *fops;
7524 
7525 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7526 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7527 }
7528 
7529 /**
7530  * Create policy rules, lock free,
7531  * (mutex should be acquired by caller).
7532  * Dispatcher for action type specific call.
7533  *
7534  * @param[in] dev
7535  *   Pointer to the Ethernet device structure.
7536  * @param[in] mtr_policy
7537  *   Meter policy struct.
7538  *
7539  * @return
7540  *   0 on success, -1 otherwise.
7541  */
7542 int
7543 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7544 			     struct mlx5_flow_meter_policy *mtr_policy)
7545 {
7546 	const struct mlx5_flow_driver_ops *fops;
7547 
7548 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7549 	return fops->create_policy_rules(dev, mtr_policy);
7550 }
7551 
7552 /**
7553  * Destroy policy rules, lock free,
7554  * (mutex should be acquired by caller).
7555  * Dispatcher for action type specific call.
7556  *
7557  * @param[in] dev
7558  *   Pointer to the Ethernet device structure.
7559  * @param[in] mtr_policy
7560  *   Meter policy struct.
7561  */
7562 void
7563 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7564 			     struct mlx5_flow_meter_policy *mtr_policy)
7565 {
7566 	const struct mlx5_flow_driver_ops *fops;
7567 
7568 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7569 	fops->destroy_policy_rules(dev, mtr_policy);
7570 }
7571 
7572 /**
7573  * Destroy the default policy table set.
7574  *
7575  * @param[in] dev
7576  *   Pointer to Ethernet device.
7577  */
7578 void
7579 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7580 {
7581 	const struct mlx5_flow_driver_ops *fops;
7582 
7583 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7584 	fops->destroy_def_policy(dev);
7585 }
7586 
7587 /**
7588  * Destroy the default policy table set.
7589  *
7590  * @param[in] dev
7591  *   Pointer to Ethernet device.
7592  *
7593  * @return
7594  *   0 on success, -1 otherwise.
7595  */
7596 int
7597 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7598 {
7599 	const struct mlx5_flow_driver_ops *fops;
7600 
7601 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7602 	return fops->create_def_policy(dev);
7603 }
7604 
7605 /**
7606  * Create the needed meter and suffix tables.
7607  *
7608  * @param[in] dev
7609  *   Pointer to Ethernet device.
7610  *
7611  * @return
7612  *   0 on success, -1 otherwise.
7613  */
7614 int
7615 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7616 			struct mlx5_flow_meter_info *fm,
7617 			uint32_t mtr_idx,
7618 			uint8_t domain_bitmap)
7619 {
7620 	const struct mlx5_flow_driver_ops *fops;
7621 
7622 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7623 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7624 }
7625 
7626 /**
7627  * Destroy the meter table set.
7628  *
7629  * @param[in] dev
7630  *   Pointer to Ethernet device.
7631  * @param[in] tbl
7632  *   Pointer to the meter table set.
7633  */
7634 void
7635 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7636 			   struct mlx5_flow_meter_info *fm)
7637 {
7638 	const struct mlx5_flow_driver_ops *fops;
7639 
7640 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7641 	fops->destroy_mtr_tbls(dev, fm);
7642 }
7643 
7644 /**
7645  * Destroy the global meter drop table.
7646  *
7647  * @param[in] dev
7648  *   Pointer to Ethernet device.
7649  */
7650 void
7651 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7652 {
7653 	const struct mlx5_flow_driver_ops *fops;
7654 
7655 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7656 	fops->destroy_mtr_drop_tbls(dev);
7657 }
7658 
7659 /**
7660  * Destroy the sub policy table with RX queue.
7661  *
7662  * @param[in] dev
7663  *   Pointer to Ethernet device.
7664  * @param[in] mtr_policy
7665  *   Pointer to meter policy table.
7666  */
7667 void
7668 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7669 		struct mlx5_flow_meter_policy *mtr_policy)
7670 {
7671 	const struct mlx5_flow_driver_ops *fops;
7672 
7673 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7674 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7675 }
7676 
7677 /**
7678  * Allocate the needed aso flow meter id.
7679  *
7680  * @param[in] dev
7681  *   Pointer to Ethernet device.
7682  *
7683  * @return
7684  *   Index to aso flow meter on success, NULL otherwise.
7685  */
7686 uint32_t
7687 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7688 {
7689 	const struct mlx5_flow_driver_ops *fops;
7690 
7691 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7692 	return fops->create_meter(dev);
7693 }
7694 
7695 /**
7696  * Free the aso flow meter id.
7697  *
7698  * @param[in] dev
7699  *   Pointer to Ethernet device.
7700  * @param[in] mtr_idx
7701  *  Index to aso flow meter to be free.
7702  *
7703  * @return
7704  *   0 on success.
7705  */
7706 void
7707 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7708 {
7709 	const struct mlx5_flow_driver_ops *fops;
7710 
7711 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7712 	fops->free_meter(dev, mtr_idx);
7713 }
7714 
7715 /**
7716  * Allocate a counter.
7717  *
7718  * @param[in] dev
7719  *   Pointer to Ethernet device structure.
7720  *
7721  * @return
7722  *   Index to allocated counter  on success, 0 otherwise.
7723  */
7724 uint32_t
7725 mlx5_counter_alloc(struct rte_eth_dev *dev)
7726 {
7727 	const struct mlx5_flow_driver_ops *fops;
7728 	struct rte_flow_attr attr = { .transfer = 0 };
7729 
7730 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7731 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7732 		return fops->counter_alloc(dev);
7733 	}
7734 	DRV_LOG(ERR,
7735 		"port %u counter allocate is not supported.",
7736 		 dev->data->port_id);
7737 	return 0;
7738 }
7739 
7740 /**
7741  * Free a counter.
7742  *
7743  * @param[in] dev
7744  *   Pointer to Ethernet device structure.
7745  * @param[in] cnt
7746  *   Index to counter to be free.
7747  */
7748 void
7749 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7750 {
7751 	const struct mlx5_flow_driver_ops *fops;
7752 	struct rte_flow_attr attr = { .transfer = 0 };
7753 
7754 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7755 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7756 		fops->counter_free(dev, cnt);
7757 		return;
7758 	}
7759 	DRV_LOG(ERR,
7760 		"port %u counter free is not supported.",
7761 		 dev->data->port_id);
7762 }
7763 
7764 /**
7765  * Query counter statistics.
7766  *
7767  * @param[in] dev
7768  *   Pointer to Ethernet device structure.
7769  * @param[in] cnt
7770  *   Index to counter to query.
7771  * @param[in] clear
7772  *   Set to clear counter statistics.
7773  * @param[out] pkts
7774  *   The counter hits packets number to save.
7775  * @param[out] bytes
7776  *   The counter hits bytes number to save.
7777  *
7778  * @return
7779  *   0 on success, a negative errno value otherwise.
7780  */
7781 int
7782 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7783 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7784 {
7785 	const struct mlx5_flow_driver_ops *fops;
7786 	struct rte_flow_attr attr = { .transfer = 0 };
7787 
7788 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7789 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7790 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7791 	}
7792 	DRV_LOG(ERR,
7793 		"port %u counter query is not supported.",
7794 		 dev->data->port_id);
7795 	return -ENOTSUP;
7796 }
7797 
7798 /**
7799  * Allocate a new memory for the counter values wrapped by all the needed
7800  * management.
7801  *
7802  * @param[in] sh
7803  *   Pointer to mlx5_dev_ctx_shared object.
7804  *
7805  * @return
7806  *   0 on success, a negative errno value otherwise.
7807  */
7808 static int
7809 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7810 {
7811 	struct mlx5_counter_stats_mem_mng *mem_mng;
7812 	volatile struct flow_counter_stats *raw_data;
7813 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7814 	int size = (sizeof(struct flow_counter_stats) *
7815 			MLX5_COUNTERS_PER_POOL +
7816 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7817 			sizeof(struct mlx5_counter_stats_mem_mng);
7818 	size_t pgsize = rte_mem_page_size();
7819 	uint8_t *mem;
7820 	int ret;
7821 	int i;
7822 
7823 	if (pgsize == (size_t)-1) {
7824 		DRV_LOG(ERR, "Failed to get mem page size");
7825 		rte_errno = ENOMEM;
7826 		return -ENOMEM;
7827 	}
7828 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7829 	if (!mem) {
7830 		rte_errno = ENOMEM;
7831 		return -ENOMEM;
7832 	}
7833 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7834 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7835 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7836 					  sh->cdev->pdn, mem, size,
7837 					  &mem_mng->wm);
7838 	if (ret) {
7839 		rte_errno = errno;
7840 		mlx5_free(mem);
7841 		return -rte_errno;
7842 	}
7843 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7844 	raw_data = (volatile struct flow_counter_stats *)mem;
7845 	for (i = 0; i < raws_n; ++i) {
7846 		mem_mng->raws[i].mem_mng = mem_mng;
7847 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7848 	}
7849 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7850 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7851 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7852 				 next);
7853 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7854 	sh->cmng.mem_mng = mem_mng;
7855 	return 0;
7856 }
7857 
7858 /**
7859  * Set the statistic memory to the new counter pool.
7860  *
7861  * @param[in] sh
7862  *   Pointer to mlx5_dev_ctx_shared object.
7863  * @param[in] pool
7864  *   Pointer to the pool to set the statistic memory.
7865  *
7866  * @return
7867  *   0 on success, a negative errno value otherwise.
7868  */
7869 static int
7870 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7871 			       struct mlx5_flow_counter_pool *pool)
7872 {
7873 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7874 	/* Resize statistic memory once used out. */
7875 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7876 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7877 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7878 		return -1;
7879 	}
7880 	rte_spinlock_lock(&pool->sl);
7881 	pool->raw = cmng->mem_mng->raws + pool->index %
7882 		    MLX5_CNT_CONTAINER_RESIZE;
7883 	rte_spinlock_unlock(&pool->sl);
7884 	pool->raw_hw = NULL;
7885 	return 0;
7886 }
7887 
7888 #define MLX5_POOL_QUERY_FREQ_US 1000000
7889 
7890 /**
7891  * Set the periodic procedure for triggering asynchronous batch queries for all
7892  * the counter pools.
7893  *
7894  * @param[in] sh
7895  *   Pointer to mlx5_dev_ctx_shared object.
7896  */
7897 void
7898 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7899 {
7900 	uint32_t pools_n, us;
7901 
7902 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7903 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7904 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7905 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7906 		sh->cmng.query_thread_on = 0;
7907 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7908 	} else {
7909 		sh->cmng.query_thread_on = 1;
7910 	}
7911 }
7912 
7913 /**
7914  * The periodic procedure for triggering asynchronous batch queries for all the
7915  * counter pools. This function is probably called by the host thread.
7916  *
7917  * @param[in] arg
7918  *   The parameter for the alarm process.
7919  */
7920 void
7921 mlx5_flow_query_alarm(void *arg)
7922 {
7923 	struct mlx5_dev_ctx_shared *sh = arg;
7924 	int ret;
7925 	uint16_t pool_index = sh->cmng.pool_index;
7926 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7927 	struct mlx5_flow_counter_pool *pool;
7928 	uint16_t n_valid;
7929 
7930 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7931 		goto set_alarm;
7932 	rte_spinlock_lock(&cmng->pool_update_sl);
7933 	pool = cmng->pools[pool_index];
7934 	n_valid = cmng->n_valid;
7935 	rte_spinlock_unlock(&cmng->pool_update_sl);
7936 	/* Set the statistic memory to the new created pool. */
7937 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7938 		goto set_alarm;
7939 	if (pool->raw_hw)
7940 		/* There is a pool query in progress. */
7941 		goto set_alarm;
7942 	pool->raw_hw =
7943 		LIST_FIRST(&sh->cmng.free_stat_raws);
7944 	if (!pool->raw_hw)
7945 		/* No free counter statistics raw memory. */
7946 		goto set_alarm;
7947 	/*
7948 	 * Identify the counters released between query trigger and query
7949 	 * handle more efficiently. The counter released in this gap period
7950 	 * should wait for a new round of query as the new arrived packets
7951 	 * will not be taken into account.
7952 	 */
7953 	pool->query_gen++;
7954 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7955 					       MLX5_COUNTERS_PER_POOL,
7956 					       NULL, NULL,
7957 					       pool->raw_hw->mem_mng->wm.lkey,
7958 					       (void *)(uintptr_t)
7959 					       pool->raw_hw->data,
7960 					       sh->devx_comp,
7961 					       (uint64_t)(uintptr_t)pool);
7962 	if (ret) {
7963 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7964 			" %d", pool->min_dcs->id);
7965 		pool->raw_hw = NULL;
7966 		goto set_alarm;
7967 	}
7968 	LIST_REMOVE(pool->raw_hw, next);
7969 	sh->cmng.pending_queries++;
7970 	pool_index++;
7971 	if (pool_index >= n_valid)
7972 		pool_index = 0;
7973 set_alarm:
7974 	sh->cmng.pool_index = pool_index;
7975 	mlx5_set_query_alarm(sh);
7976 }
7977 
7978 /**
7979  * Check and callback event for new aged flow in the counter pool
7980  *
7981  * @param[in] sh
7982  *   Pointer to mlx5_dev_ctx_shared object.
7983  * @param[in] pool
7984  *   Pointer to Current counter pool.
7985  */
7986 static void
7987 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7988 		   struct mlx5_flow_counter_pool *pool)
7989 {
7990 	struct mlx5_priv *priv;
7991 	struct mlx5_flow_counter *cnt;
7992 	struct mlx5_age_info *age_info;
7993 	struct mlx5_age_param *age_param;
7994 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7995 	struct mlx5_counter_stats_raw *prev = pool->raw;
7996 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7997 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7998 	uint16_t expected = AGE_CANDIDATE;
7999 	uint32_t i;
8000 
8001 	pool->time_of_last_age_check = curr_time;
8002 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8003 		cnt = MLX5_POOL_GET_CNT(pool, i);
8004 		age_param = MLX5_CNT_TO_AGE(cnt);
8005 		if (__atomic_load_n(&age_param->state,
8006 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
8007 			continue;
8008 		if (cur->data[i].hits != prev->data[i].hits) {
8009 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
8010 					 __ATOMIC_RELAXED);
8011 			continue;
8012 		}
8013 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8014 				       time_delta,
8015 				       __ATOMIC_RELAXED) <= age_param->timeout)
8016 			continue;
8017 		/**
8018 		 * Hold the lock first, or if between the
8019 		 * state AGE_TMOUT and tailq operation the
8020 		 * release happened, the release procedure
8021 		 * may delete a non-existent tailq node.
8022 		 */
8023 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8024 		age_info = GET_PORT_AGE_INFO(priv);
8025 		rte_spinlock_lock(&age_info->aged_sl);
8026 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8027 						AGE_TMOUT, false,
8028 						__ATOMIC_RELAXED,
8029 						__ATOMIC_RELAXED)) {
8030 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8031 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8032 		}
8033 		rte_spinlock_unlock(&age_info->aged_sl);
8034 	}
8035 	mlx5_age_event_prepare(sh);
8036 }
8037 
8038 /**
8039  * Handler for the HW respond about ready values from an asynchronous batch
8040  * query. This function is probably called by the host thread.
8041  *
8042  * @param[in] sh
8043  *   The pointer to the shared device context.
8044  * @param[in] async_id
8045  *   The Devx async ID.
8046  * @param[in] status
8047  *   The status of the completion.
8048  */
8049 void
8050 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8051 				  uint64_t async_id, int status)
8052 {
8053 	struct mlx5_flow_counter_pool *pool =
8054 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8055 	struct mlx5_counter_stats_raw *raw_to_free;
8056 	uint8_t query_gen = pool->query_gen ^ 1;
8057 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8058 	enum mlx5_counter_type cnt_type =
8059 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8060 				MLX5_COUNTER_TYPE_ORIGIN;
8061 
8062 	if (unlikely(status)) {
8063 		raw_to_free = pool->raw_hw;
8064 	} else {
8065 		raw_to_free = pool->raw;
8066 		if (pool->is_aged)
8067 			mlx5_flow_aging_check(sh, pool);
8068 		rte_spinlock_lock(&pool->sl);
8069 		pool->raw = pool->raw_hw;
8070 		rte_spinlock_unlock(&pool->sl);
8071 		/* Be sure the new raw counters data is updated in memory. */
8072 		rte_io_wmb();
8073 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8074 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8075 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8076 				     &pool->counters[query_gen], next);
8077 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8078 		}
8079 	}
8080 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8081 	pool->raw_hw = NULL;
8082 	sh->cmng.pending_queries--;
8083 }
8084 
8085 static int
8086 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8087 		    const struct flow_grp_info *grp_info,
8088 		    struct rte_flow_error *error)
8089 {
8090 	if (grp_info->transfer && grp_info->external &&
8091 	    grp_info->fdb_def_rule) {
8092 		if (group == UINT32_MAX)
8093 			return rte_flow_error_set
8094 						(error, EINVAL,
8095 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8096 						 NULL,
8097 						 "group index not supported");
8098 		*table = group + 1;
8099 	} else {
8100 		*table = group;
8101 	}
8102 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8103 	return 0;
8104 }
8105 
8106 /**
8107  * Translate the rte_flow group index to HW table value.
8108  *
8109  * If tunnel offload is disabled, all group ids converted to flow table
8110  * id using the standard method.
8111  * If tunnel offload is enabled, group id can be converted using the
8112  * standard or tunnel conversion method. Group conversion method
8113  * selection depends on flags in `grp_info` parameter:
8114  * - Internal (grp_info.external == 0) groups conversion uses the
8115  *   standard method.
8116  * - Group ids in JUMP action converted with the tunnel conversion.
8117  * - Group id in rule attribute conversion depends on a rule type and
8118  *   group id value:
8119  *   ** non zero group attributes converted with the tunnel method
8120  *   ** zero group attribute in non-tunnel rule is converted using the
8121  *      standard method - there's only one root table
8122  *   ** zero group attribute in steer tunnel rule is converted with the
8123  *      standard method - single root table
8124  *   ** zero group attribute in match tunnel rule is a special OvS
8125  *      case: that value is used for portability reasons. That group
8126  *      id is converted with the tunnel conversion method.
8127  *
8128  * @param[in] dev
8129  *   Port device
8130  * @param[in] tunnel
8131  *   PMD tunnel offload object
8132  * @param[in] group
8133  *   rte_flow group index value.
8134  * @param[out] table
8135  *   HW table value.
8136  * @param[in] grp_info
8137  *   flags used for conversion
8138  * @param[out] error
8139  *   Pointer to error structure.
8140  *
8141  * @return
8142  *   0 on success, a negative errno value otherwise and rte_errno is set.
8143  */
8144 int
8145 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8146 			 const struct mlx5_flow_tunnel *tunnel,
8147 			 uint32_t group, uint32_t *table,
8148 			 const struct flow_grp_info *grp_info,
8149 			 struct rte_flow_error *error)
8150 {
8151 	int ret;
8152 	bool standard_translation;
8153 
8154 	if (!grp_info->skip_scale && grp_info->external &&
8155 	    group < MLX5_MAX_TABLES_EXTERNAL)
8156 		group *= MLX5_FLOW_TABLE_FACTOR;
8157 	if (is_tunnel_offload_active(dev)) {
8158 		standard_translation = !grp_info->external ||
8159 					grp_info->std_tbl_fix;
8160 	} else {
8161 		standard_translation = true;
8162 	}
8163 	DRV_LOG(DEBUG,
8164 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8165 		dev->data->port_id, group, grp_info->transfer,
8166 		grp_info->external, grp_info->fdb_def_rule,
8167 		standard_translation ? "STANDARD" : "TUNNEL");
8168 	if (standard_translation)
8169 		ret = flow_group_to_table(dev->data->port_id, group, table,
8170 					  grp_info, error);
8171 	else
8172 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8173 						      table, error);
8174 
8175 	return ret;
8176 }
8177 
8178 /**
8179  * Discover availability of metadata reg_c's.
8180  *
8181  * Iteratively use test flows to check availability.
8182  *
8183  * @param[in] dev
8184  *   Pointer to the Ethernet device structure.
8185  *
8186  * @return
8187  *   0 on success, a negative errno value otherwise and rte_errno is set.
8188  */
8189 int
8190 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8191 {
8192 	struct mlx5_priv *priv = dev->data->dev_private;
8193 	enum modify_reg idx;
8194 	int n = 0;
8195 
8196 	/* reg_c[0] and reg_c[1] are reserved. */
8197 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8198 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8199 	/* Discover availability of other reg_c's. */
8200 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8201 		struct rte_flow_attr attr = {
8202 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8203 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8204 			.ingress = 1,
8205 		};
8206 		struct rte_flow_item items[] = {
8207 			[0] = {
8208 				.type = RTE_FLOW_ITEM_TYPE_END,
8209 			},
8210 		};
8211 		struct rte_flow_action actions[] = {
8212 			[0] = {
8213 				.type = (enum rte_flow_action_type)
8214 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8215 				.conf = &(struct mlx5_flow_action_copy_mreg){
8216 					.src = REG_C_1,
8217 					.dst = idx,
8218 				},
8219 			},
8220 			[1] = {
8221 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8222 				.conf = &(struct rte_flow_action_jump){
8223 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8224 				},
8225 			},
8226 			[2] = {
8227 				.type = RTE_FLOW_ACTION_TYPE_END,
8228 			},
8229 		};
8230 		uint32_t flow_idx;
8231 		struct rte_flow *flow;
8232 		struct rte_flow_error error;
8233 
8234 		if (!priv->config.dv_flow_en)
8235 			break;
8236 		/* Create internal flow, validation skips copy action. */
8237 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8238 					items, actions, false, &error);
8239 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8240 				      flow_idx);
8241 		if (!flow)
8242 			continue;
8243 		priv->sh->flow_mreg_c[n++] = idx;
8244 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8245 	}
8246 	for (; n < MLX5_MREG_C_NUM; ++n)
8247 		priv->sh->flow_mreg_c[n] = REG_NON;
8248 	priv->sh->metadata_regc_check_flag = 1;
8249 	return 0;
8250 }
8251 
8252 int
8253 save_dump_file(const uint8_t *data, uint32_t size,
8254 	uint32_t type, uint64_t id, void *arg, FILE *file)
8255 {
8256 	char line[BUF_SIZE];
8257 	uint32_t out = 0;
8258 	uint32_t k;
8259 	uint32_t actions_num;
8260 	struct rte_flow_query_count *count;
8261 
8262 	memset(line, 0, BUF_SIZE);
8263 	switch (type) {
8264 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8265 		actions_num = *(uint32_t *)(arg);
8266 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8267 				type, id, actions_num);
8268 		break;
8269 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8270 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8271 				type, id);
8272 		break;
8273 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8274 		count = (struct rte_flow_query_count *)arg;
8275 		fprintf(file,
8276 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8277 			type, id, count->hits, count->bytes);
8278 		return 0;
8279 	default:
8280 		return -1;
8281 	}
8282 
8283 	for (k = 0; k < size; k++) {
8284 		/* Make sure we do not overrun the line buffer length. */
8285 		if (out >= BUF_SIZE - 4) {
8286 			line[out] = '\0';
8287 			break;
8288 		}
8289 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8290 				(data[k]) & 0xff);
8291 	}
8292 	fprintf(file, "%s\n", line);
8293 	return 0;
8294 }
8295 
8296 int
8297 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8298 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8299 {
8300 	struct rte_flow_action action[2];
8301 	enum mlx5_flow_drv_type ftype;
8302 	const struct mlx5_flow_driver_ops *fops;
8303 
8304 	if (!flow) {
8305 		return rte_flow_error_set(error, ENOENT,
8306 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8307 				NULL,
8308 				"invalid flow handle");
8309 	}
8310 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8311 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8312 	if (flow->counter) {
8313 		memset(count, 0, sizeof(struct rte_flow_query_count));
8314 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8315 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8316 						ftype < MLX5_FLOW_TYPE_MAX);
8317 		fops = flow_get_drv_ops(ftype);
8318 		return fops->query(dev, flow, action, count, error);
8319 	}
8320 	return -1;
8321 }
8322 
8323 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8324 /**
8325  * Dump flow ipool data to file
8326  *
8327  * @param[in] dev
8328  *   The pointer to Ethernet device.
8329  * @param[in] file
8330  *   A pointer to a file for output.
8331  * @param[out] error
8332  *   Perform verbose error reporting if not NULL. PMDs initialize this
8333  *   structure in case of error only.
8334  * @return
8335  *   0 on success, a negative value otherwise.
8336  */
8337 int
8338 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8339 	struct rte_flow *flow, FILE *file,
8340 	struct rte_flow_error *error)
8341 {
8342 	struct mlx5_priv *priv = dev->data->dev_private;
8343 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8344 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8345 	uint32_t handle_idx;
8346 	struct mlx5_flow_handle *dh;
8347 	struct rte_flow_query_count count;
8348 	uint32_t actions_num;
8349 	const uint8_t *data;
8350 	size_t size;
8351 	uint64_t id;
8352 	uint32_t type;
8353 	void *action = NULL;
8354 
8355 	if (!flow) {
8356 		return rte_flow_error_set(error, ENOENT,
8357 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8358 				NULL,
8359 				"invalid flow handle");
8360 	}
8361 	handle_idx = flow->dev_handles;
8362 	while (handle_idx) {
8363 		dh = mlx5_ipool_get(priv->sh->ipool
8364 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8365 		if (!dh)
8366 			continue;
8367 		handle_idx = dh->next.next;
8368 
8369 		/* query counter */
8370 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8371 		flow_dv_query_count_ptr(dev, flow->counter,
8372 						&action, error);
8373 		if (action) {
8374 			id = (uint64_t)(uintptr_t)action;
8375 			if (!mlx5_flow_query_counter(dev, flow, &count, error))
8376 				save_dump_file(NULL, 0, type,
8377 						id, (void *)&count, file);
8378 		}
8379 		/* Get modify_hdr and encap_decap buf from ipools. */
8380 		encap_decap = NULL;
8381 		modify_hdr = dh->dvh.modify_hdr;
8382 
8383 		if (dh->dvh.rix_encap_decap) {
8384 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8385 						[MLX5_IPOOL_DECAP_ENCAP],
8386 						dh->dvh.rix_encap_decap);
8387 		}
8388 		if (modify_hdr) {
8389 			data = (const uint8_t *)modify_hdr->actions;
8390 			size = (size_t)(modify_hdr->actions_num) * 8;
8391 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8392 			actions_num = modify_hdr->actions_num;
8393 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8394 			save_dump_file(data, size, type, id,
8395 						(void *)(&actions_num), file);
8396 		}
8397 		if (encap_decap) {
8398 			data = encap_decap->buf;
8399 			size = encap_decap->size;
8400 			id = (uint64_t)(uintptr_t)encap_decap->action;
8401 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8402 			save_dump_file(data, size, type,
8403 						id, NULL, file);
8404 		}
8405 	}
8406 	return 0;
8407 }
8408 
8409 /**
8410  * Dump all flow's encap_decap/modify_hdr/counter data to file
8411  *
8412  * @param[in] dev
8413  *   The pointer to Ethernet device.
8414  * @param[in] file
8415  *   A pointer to a file for output.
8416  * @param[out] error
8417  *   Perform verbose error reporting if not NULL. PMDs initialize this
8418  *   structure in case of error only.
8419  * @return
8420  *   0 on success, a negative value otherwise.
8421  */
8422 static int
8423 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8424 	FILE *file, struct rte_flow_error *error)
8425 {
8426 	struct mlx5_priv *priv = dev->data->dev_private;
8427 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8428 	struct mlx5_hlist *h;
8429 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8430 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8431 	struct rte_flow_query_count count;
8432 	uint32_t actions_num;
8433 	const uint8_t *data;
8434 	size_t size;
8435 	uint64_t id;
8436 	uint32_t type;
8437 	uint32_t i;
8438 	uint32_t j;
8439 	struct mlx5_list_inconst *l_inconst;
8440 	struct mlx5_list_entry *e;
8441 	int lcore_index;
8442 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8443 	uint32_t max;
8444 	void *action;
8445 
8446 	/* encap_decap hlist is lcore_share, get global core cache. */
8447 	i = MLX5_LIST_GLOBAL;
8448 	h = sh->encaps_decaps;
8449 	if (h) {
8450 		for (j = 0; j <= h->mask; j++) {
8451 			l_inconst = &h->buckets[j].l;
8452 			if (!l_inconst || !l_inconst->cache[i])
8453 				continue;
8454 
8455 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8456 			while (e) {
8457 				encap_decap =
8458 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8459 				data = encap_decap->buf;
8460 				size = encap_decap->size;
8461 				id = (uint64_t)(uintptr_t)encap_decap->action;
8462 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8463 				save_dump_file(data, size, type,
8464 					id, NULL, file);
8465 				e = LIST_NEXT(e, next);
8466 			}
8467 		}
8468 	}
8469 
8470 	/* get modify_hdr */
8471 	h = sh->modify_cmds;
8472 	if (h) {
8473 		lcore_index = rte_lcore_index(rte_lcore_id());
8474 		if (unlikely(lcore_index == -1)) {
8475 			lcore_index = MLX5_LIST_NLCORE;
8476 			rte_spinlock_lock(&h->l_const.lcore_lock);
8477 		}
8478 		i = lcore_index;
8479 
8480 		for (j = 0; j <= h->mask; j++) {
8481 			l_inconst = &h->buckets[j].l;
8482 			if (!l_inconst || !l_inconst->cache[i])
8483 				continue;
8484 
8485 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8486 			while (e) {
8487 				modify_hdr =
8488 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8489 				data = (const uint8_t *)modify_hdr->actions;
8490 				size = (size_t)(modify_hdr->actions_num) * 8;
8491 				actions_num = modify_hdr->actions_num;
8492 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8493 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8494 				save_dump_file(data, size, type, id,
8495 						(void *)(&actions_num), file);
8496 				e = LIST_NEXT(e, next);
8497 			}
8498 		}
8499 
8500 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8501 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8502 	}
8503 
8504 	/* get counter */
8505 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8506 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8507 	for (j = 1; j <= max; j++) {
8508 		action = NULL;
8509 		flow_dv_query_count_ptr(dev, j, &action, error);
8510 		if (action) {
8511 			if (!flow_dv_query_count(dev, j, &count, error)) {
8512 				type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8513 				id = (uint64_t)(uintptr_t)action;
8514 				save_dump_file(NULL, 0, type,
8515 						id, (void *)&count, file);
8516 			}
8517 		}
8518 	}
8519 	return 0;
8520 }
8521 #endif
8522 
8523 /**
8524  * Dump flow raw hw data to file
8525  *
8526  * @param[in] dev
8527  *    The pointer to Ethernet device.
8528  * @param[in] file
8529  *   A pointer to a file for output.
8530  * @param[out] error
8531  *   Perform verbose error reporting if not NULL. PMDs initialize this
8532  *   structure in case of error only.
8533  * @return
8534  *   0 on success, a nagative value otherwise.
8535  */
8536 int
8537 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8538 		   FILE *file,
8539 		   struct rte_flow_error *error __rte_unused)
8540 {
8541 	struct mlx5_priv *priv = dev->data->dev_private;
8542 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8543 	uint32_t handle_idx;
8544 	int ret;
8545 	struct mlx5_flow_handle *dh;
8546 	struct rte_flow *flow;
8547 
8548 	if (!priv->config.dv_flow_en) {
8549 		if (fputs("device dv flow disabled\n", file) <= 0)
8550 			return -errno;
8551 		return -ENOTSUP;
8552 	}
8553 
8554 	/* dump all */
8555 	if (!flow_idx) {
8556 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8557 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8558 			return -EINVAL;
8559 #endif
8560 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8561 					sh->rx_domain,
8562 					sh->tx_domain, file);
8563 	}
8564 	/* dump one */
8565 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8566 			(uintptr_t)(void *)flow_idx);
8567 	if (!flow)
8568 		return -EINVAL;
8569 
8570 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8571 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8572 #endif
8573 	handle_idx = flow->dev_handles;
8574 	while (handle_idx) {
8575 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8576 				handle_idx);
8577 		if (!dh)
8578 			return -ENOENT;
8579 		if (dh->drv_flow) {
8580 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8581 					file);
8582 			if (ret)
8583 				return -ENOENT;
8584 		}
8585 		handle_idx = dh->next.next;
8586 	}
8587 	return 0;
8588 }
8589 
8590 /**
8591  * Get aged-out flows.
8592  *
8593  * @param[in] dev
8594  *   Pointer to the Ethernet device structure.
8595  * @param[in] context
8596  *   The address of an array of pointers to the aged-out flows contexts.
8597  * @param[in] nb_countexts
8598  *   The length of context array pointers.
8599  * @param[out] error
8600  *   Perform verbose error reporting if not NULL. Initialized in case of
8601  *   error only.
8602  *
8603  * @return
8604  *   how many contexts get in success, otherwise negative errno value.
8605  *   if nb_contexts is 0, return the amount of all aged contexts.
8606  *   if nb_contexts is not 0 , return the amount of aged flows reported
8607  *   in the context array.
8608  */
8609 int
8610 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8611 			uint32_t nb_contexts, struct rte_flow_error *error)
8612 {
8613 	const struct mlx5_flow_driver_ops *fops;
8614 	struct rte_flow_attr attr = { .transfer = 0 };
8615 
8616 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8617 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8618 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8619 						    error);
8620 	}
8621 	DRV_LOG(ERR,
8622 		"port %u get aged flows is not supported.",
8623 		 dev->data->port_id);
8624 	return -ENOTSUP;
8625 }
8626 
8627 /* Wrapper for driver action_validate op callback */
8628 static int
8629 flow_drv_action_validate(struct rte_eth_dev *dev,
8630 			 const struct rte_flow_indir_action_conf *conf,
8631 			 const struct rte_flow_action *action,
8632 			 const struct mlx5_flow_driver_ops *fops,
8633 			 struct rte_flow_error *error)
8634 {
8635 	static const char err_msg[] = "indirect action validation unsupported";
8636 
8637 	if (!fops->action_validate) {
8638 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8639 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8640 				   NULL, err_msg);
8641 		return -rte_errno;
8642 	}
8643 	return fops->action_validate(dev, conf, action, error);
8644 }
8645 
8646 /**
8647  * Destroys the shared action by handle.
8648  *
8649  * @param dev
8650  *   Pointer to Ethernet device structure.
8651  * @param[in] handle
8652  *   Handle for the indirect action object to be destroyed.
8653  * @param[out] error
8654  *   Perform verbose error reporting if not NULL. PMDs initialize this
8655  *   structure in case of error only.
8656  *
8657  * @return
8658  *   0 on success, a negative errno value otherwise and rte_errno is set.
8659  *
8660  * @note: wrapper for driver action_create op callback.
8661  */
8662 static int
8663 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8664 			   struct rte_flow_action_handle *handle,
8665 			   struct rte_flow_error *error)
8666 {
8667 	static const char err_msg[] = "indirect action destruction unsupported";
8668 	struct rte_flow_attr attr = { .transfer = 0 };
8669 	const struct mlx5_flow_driver_ops *fops =
8670 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8671 
8672 	if (!fops->action_destroy) {
8673 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8674 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8675 				   NULL, err_msg);
8676 		return -rte_errno;
8677 	}
8678 	return fops->action_destroy(dev, handle, error);
8679 }
8680 
8681 /* Wrapper for driver action_destroy op callback */
8682 static int
8683 flow_drv_action_update(struct rte_eth_dev *dev,
8684 		       struct rte_flow_action_handle *handle,
8685 		       const void *update,
8686 		       const struct mlx5_flow_driver_ops *fops,
8687 		       struct rte_flow_error *error)
8688 {
8689 	static const char err_msg[] = "indirect action update unsupported";
8690 
8691 	if (!fops->action_update) {
8692 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8693 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8694 				   NULL, err_msg);
8695 		return -rte_errno;
8696 	}
8697 	return fops->action_update(dev, handle, update, error);
8698 }
8699 
8700 /* Wrapper for driver action_destroy op callback */
8701 static int
8702 flow_drv_action_query(struct rte_eth_dev *dev,
8703 		      const struct rte_flow_action_handle *handle,
8704 		      void *data,
8705 		      const struct mlx5_flow_driver_ops *fops,
8706 		      struct rte_flow_error *error)
8707 {
8708 	static const char err_msg[] = "indirect action query unsupported";
8709 
8710 	if (!fops->action_query) {
8711 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8712 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8713 				   NULL, err_msg);
8714 		return -rte_errno;
8715 	}
8716 	return fops->action_query(dev, handle, data, error);
8717 }
8718 
8719 /**
8720  * Create indirect action for reuse in multiple flow rules.
8721  *
8722  * @param dev
8723  *   Pointer to Ethernet device structure.
8724  * @param conf
8725  *   Pointer to indirect action object configuration.
8726  * @param[in] action
8727  *   Action configuration for indirect action object creation.
8728  * @param[out] error
8729  *   Perform verbose error reporting if not NULL. PMDs initialize this
8730  *   structure in case of error only.
8731  * @return
8732  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8733  */
8734 static struct rte_flow_action_handle *
8735 mlx5_action_handle_create(struct rte_eth_dev *dev,
8736 			  const struct rte_flow_indir_action_conf *conf,
8737 			  const struct rte_flow_action *action,
8738 			  struct rte_flow_error *error)
8739 {
8740 	static const char err_msg[] = "indirect action creation unsupported";
8741 	struct rte_flow_attr attr = { .transfer = 0 };
8742 	const struct mlx5_flow_driver_ops *fops =
8743 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8744 
8745 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8746 		return NULL;
8747 	if (!fops->action_create) {
8748 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8749 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8750 				   NULL, err_msg);
8751 		return NULL;
8752 	}
8753 	return fops->action_create(dev, conf, action, error);
8754 }
8755 
8756 /**
8757  * Updates inplace the indirect action configuration pointed by *handle*
8758  * with the configuration provided as *update* argument.
8759  * The update of the indirect action configuration effects all flow rules
8760  * reusing the action via handle.
8761  *
8762  * @param dev
8763  *   Pointer to Ethernet device structure.
8764  * @param[in] handle
8765  *   Handle for the indirect action to be updated.
8766  * @param[in] update
8767  *   Action specification used to modify the action pointed by handle.
8768  *   *update* could be of same type with the action pointed by the *handle*
8769  *   handle argument, or some other structures like a wrapper, depending on
8770  *   the indirect action type.
8771  * @param[out] error
8772  *   Perform verbose error reporting if not NULL. PMDs initialize this
8773  *   structure in case of error only.
8774  *
8775  * @return
8776  *   0 on success, a negative errno value otherwise and rte_errno is set.
8777  */
8778 static int
8779 mlx5_action_handle_update(struct rte_eth_dev *dev,
8780 		struct rte_flow_action_handle *handle,
8781 		const void *update,
8782 		struct rte_flow_error *error)
8783 {
8784 	struct rte_flow_attr attr = { .transfer = 0 };
8785 	const struct mlx5_flow_driver_ops *fops =
8786 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8787 	int ret;
8788 
8789 	ret = flow_drv_action_validate(dev, NULL,
8790 			(const struct rte_flow_action *)update, fops, error);
8791 	if (ret)
8792 		return ret;
8793 	return flow_drv_action_update(dev, handle, update, fops,
8794 				      error);
8795 }
8796 
8797 /**
8798  * Query the indirect action by handle.
8799  *
8800  * This function allows retrieving action-specific data such as counters.
8801  * Data is gathered by special action which may be present/referenced in
8802  * more than one flow rule definition.
8803  *
8804  * see @RTE_FLOW_ACTION_TYPE_COUNT
8805  *
8806  * @param dev
8807  *   Pointer to Ethernet device structure.
8808  * @param[in] handle
8809  *   Handle for the indirect action to query.
8810  * @param[in, out] data
8811  *   Pointer to storage for the associated query data type.
8812  * @param[out] error
8813  *   Perform verbose error reporting if not NULL. PMDs initialize this
8814  *   structure in case of error only.
8815  *
8816  * @return
8817  *   0 on success, a negative errno value otherwise and rte_errno is set.
8818  */
8819 static int
8820 mlx5_action_handle_query(struct rte_eth_dev *dev,
8821 			 const struct rte_flow_action_handle *handle,
8822 			 void *data,
8823 			 struct rte_flow_error *error)
8824 {
8825 	struct rte_flow_attr attr = { .transfer = 0 };
8826 	const struct mlx5_flow_driver_ops *fops =
8827 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8828 
8829 	return flow_drv_action_query(dev, handle, data, fops, error);
8830 }
8831 
8832 /**
8833  * Destroy all indirect actions (shared RSS).
8834  *
8835  * @param dev
8836  *   Pointer to Ethernet device.
8837  *
8838  * @return
8839  *   0 on success, a negative errno value otherwise and rte_errno is set.
8840  */
8841 int
8842 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8843 {
8844 	struct rte_flow_error error;
8845 	struct mlx5_priv *priv = dev->data->dev_private;
8846 	struct mlx5_shared_action_rss *shared_rss;
8847 	int ret = 0;
8848 	uint32_t idx;
8849 
8850 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8851 		      priv->rss_shared_actions, idx, shared_rss, next) {
8852 		ret |= mlx5_action_handle_destroy(dev,
8853 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8854 	}
8855 	return ret;
8856 }
8857 
8858 /**
8859  * Validate existing indirect actions against current device configuration
8860  * and attach them to device resources.
8861  *
8862  * @param dev
8863  *   Pointer to Ethernet device.
8864  *
8865  * @return
8866  *   0 on success, a negative errno value otherwise and rte_errno is set.
8867  */
8868 int
8869 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8870 {
8871 	struct mlx5_priv *priv = dev->data->dev_private;
8872 	struct mlx5_indexed_pool *ipool =
8873 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8874 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8875 	int ret = 0;
8876 	uint32_t idx;
8877 
8878 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8879 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8880 		const char *message;
8881 		uint32_t queue_idx;
8882 
8883 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8884 					       ind_tbl->queues_n,
8885 					       &message, &queue_idx);
8886 		if (ret != 0) {
8887 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8888 				dev->data->port_id, ind_tbl->queues[queue_idx],
8889 				message);
8890 			break;
8891 		}
8892 	}
8893 	if (ret != 0)
8894 		return ret;
8895 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8896 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8897 
8898 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8899 		if (ret != 0) {
8900 			DRV_LOG(ERR, "Port %u could not attach "
8901 				"indirection table obj %p",
8902 				dev->data->port_id, (void *)ind_tbl);
8903 			goto error;
8904 		}
8905 	}
8906 	return 0;
8907 error:
8908 	shared_rss_last = shared_rss;
8909 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8910 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8911 
8912 		if (shared_rss == shared_rss_last)
8913 			break;
8914 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8915 			DRV_LOG(CRIT, "Port %u could not detach "
8916 				"indirection table obj %p on rollback",
8917 				dev->data->port_id, (void *)ind_tbl);
8918 	}
8919 	return ret;
8920 }
8921 
8922 /**
8923  * Detach indirect actions of the device from its resources.
8924  *
8925  * @param dev
8926  *   Pointer to Ethernet device.
8927  *
8928  * @return
8929  *   0 on success, a negative errno value otherwise and rte_errno is set.
8930  */
8931 int
8932 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8933 {
8934 	struct mlx5_priv *priv = dev->data->dev_private;
8935 	struct mlx5_indexed_pool *ipool =
8936 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8937 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8938 	int ret = 0;
8939 	uint32_t idx;
8940 
8941 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8942 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8943 
8944 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8945 		if (ret != 0) {
8946 			DRV_LOG(ERR, "Port %u could not detach "
8947 				"indirection table obj %p",
8948 				dev->data->port_id, (void *)ind_tbl);
8949 			goto error;
8950 		}
8951 	}
8952 	return 0;
8953 error:
8954 	shared_rss_last = shared_rss;
8955 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8956 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8957 
8958 		if (shared_rss == shared_rss_last)
8959 			break;
8960 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8961 			DRV_LOG(CRIT, "Port %u could not attach "
8962 				"indirection table obj %p on rollback",
8963 				dev->data->port_id, (void *)ind_tbl);
8964 	}
8965 	return ret;
8966 }
8967 
8968 #ifndef HAVE_MLX5DV_DR
8969 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8970 #else
8971 #define MLX5_DOMAIN_SYNC_FLOW \
8972 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8973 #endif
8974 
8975 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8976 {
8977 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8978 	const struct mlx5_flow_driver_ops *fops;
8979 	int ret;
8980 	struct rte_flow_attr attr = { .transfer = 0 };
8981 
8982 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8983 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8984 	if (ret > 0)
8985 		ret = -ret;
8986 	return ret;
8987 }
8988 
8989 const struct mlx5_flow_tunnel *
8990 mlx5_get_tof(const struct rte_flow_item *item,
8991 	     const struct rte_flow_action *action,
8992 	     enum mlx5_tof_rule_type *rule_type)
8993 {
8994 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8995 		if (item->type == (typeof(item->type))
8996 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8997 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8998 			return flow_items_to_tunnel(item);
8999 		}
9000 	}
9001 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9002 		if (action->type == (typeof(action->type))
9003 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9004 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9005 			return flow_actions_to_tunnel(action);
9006 		}
9007 	}
9008 	return NULL;
9009 }
9010 
9011 /**
9012  * tunnel offload functionalilty is defined for DV environment only
9013  */
9014 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9015 __extension__
9016 union tunnel_offload_mark {
9017 	uint32_t val;
9018 	struct {
9019 		uint32_t app_reserve:8;
9020 		uint32_t table_id:15;
9021 		uint32_t transfer:1;
9022 		uint32_t _unused_:8;
9023 	};
9024 };
9025 
9026 static bool
9027 mlx5_access_tunnel_offload_db
9028 	(struct rte_eth_dev *dev,
9029 	 bool (*match)(struct rte_eth_dev *,
9030 		       struct mlx5_flow_tunnel *, const void *),
9031 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9032 	 void (*miss)(struct rte_eth_dev *, void *),
9033 	 void *ctx, bool lock_op);
9034 
9035 static int
9036 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9037 			     struct rte_flow *flow,
9038 			     const struct rte_flow_attr *attr,
9039 			     const struct rte_flow_action *app_actions,
9040 			     uint32_t flow_idx,
9041 			     const struct mlx5_flow_tunnel *tunnel,
9042 			     struct tunnel_default_miss_ctx *ctx,
9043 			     struct rte_flow_error *error)
9044 {
9045 	struct mlx5_priv *priv = dev->data->dev_private;
9046 	struct mlx5_flow *dev_flow;
9047 	struct rte_flow_attr miss_attr = *attr;
9048 	const struct rte_flow_item miss_items[2] = {
9049 		{
9050 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9051 			.spec = NULL,
9052 			.last = NULL,
9053 			.mask = NULL
9054 		},
9055 		{
9056 			.type = RTE_FLOW_ITEM_TYPE_END,
9057 			.spec = NULL,
9058 			.last = NULL,
9059 			.mask = NULL
9060 		}
9061 	};
9062 	union tunnel_offload_mark mark_id;
9063 	struct rte_flow_action_mark miss_mark;
9064 	struct rte_flow_action miss_actions[3] = {
9065 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9066 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9067 	};
9068 	const struct rte_flow_action_jump *jump_data;
9069 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9070 	struct flow_grp_info grp_info = {
9071 		.external = 1,
9072 		.transfer = attr->transfer,
9073 		.fdb_def_rule = !!priv->fdb_def_rule,
9074 		.std_tbl_fix = 0,
9075 	};
9076 	int ret;
9077 
9078 	if (!attr->transfer) {
9079 		uint32_t q_size;
9080 
9081 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9082 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9083 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9084 					 0, SOCKET_ID_ANY);
9085 		if (!ctx->queue)
9086 			return rte_flow_error_set
9087 				(error, ENOMEM,
9088 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9089 				NULL, "invalid default miss RSS");
9090 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9091 		ctx->action_rss.level = 0,
9092 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9093 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9094 		ctx->action_rss.queue_num = priv->reta_idx_n,
9095 		ctx->action_rss.key = priv->rss_conf.rss_key,
9096 		ctx->action_rss.queue = ctx->queue;
9097 		if (!priv->reta_idx_n || !priv->rxqs_n)
9098 			return rte_flow_error_set
9099 				(error, EINVAL,
9100 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9101 				NULL, "invalid port configuration");
9102 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9103 			ctx->action_rss.types = 0;
9104 		for (i = 0; i != priv->reta_idx_n; ++i)
9105 			ctx->queue[i] = (*priv->reta_idx)[i];
9106 	} else {
9107 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9108 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9109 	}
9110 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9111 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9112 	jump_data = app_actions->conf;
9113 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9114 	miss_attr.group = jump_data->group;
9115 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9116 				       &flow_table, &grp_info, error);
9117 	if (ret)
9118 		return rte_flow_error_set(error, EINVAL,
9119 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9120 					  NULL, "invalid tunnel id");
9121 	mark_id.app_reserve = 0;
9122 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9123 	mark_id.transfer = !!attr->transfer;
9124 	mark_id._unused_ = 0;
9125 	miss_mark.id = mark_id.val;
9126 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9127 				    miss_items, miss_actions, flow_idx, error);
9128 	if (!dev_flow)
9129 		return -rte_errno;
9130 	dev_flow->flow = flow;
9131 	dev_flow->external = true;
9132 	dev_flow->tunnel = tunnel;
9133 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9134 	/* Subflow object was created, we must include one in the list. */
9135 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9136 		      dev_flow->handle, next);
9137 	DRV_LOG(DEBUG,
9138 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9139 		dev->data->port_id, tunnel->app_tunnel.type,
9140 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9141 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9142 				  miss_actions, error);
9143 	if (!ret)
9144 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9145 						  error);
9146 
9147 	return ret;
9148 }
9149 
9150 static const struct mlx5_flow_tbl_data_entry  *
9151 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9152 {
9153 	struct mlx5_priv *priv = dev->data->dev_private;
9154 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9155 	struct mlx5_list_entry *he;
9156 	union tunnel_offload_mark mbits = { .val = mark };
9157 	union mlx5_flow_tbl_key table_key = {
9158 		{
9159 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9160 			.id = 0,
9161 			.reserved = 0,
9162 			.dummy = 0,
9163 			.is_fdb = !!mbits.transfer,
9164 			.is_egress = 0,
9165 		}
9166 	};
9167 	struct mlx5_flow_cb_ctx ctx = {
9168 		.data = &table_key.v64,
9169 	};
9170 
9171 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9172 	return he ?
9173 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9174 }
9175 
9176 static void
9177 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9178 				   struct mlx5_list_entry *entry)
9179 {
9180 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9181 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9182 
9183 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9184 			tunnel_flow_tbl_to_id(tte->flow_table));
9185 	mlx5_free(tte);
9186 }
9187 
9188 static int
9189 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9190 				  struct mlx5_list_entry *entry, void *cb_ctx)
9191 {
9192 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9193 	union tunnel_tbl_key tbl = {
9194 		.val = *(uint64_t *)(ctx->data),
9195 	};
9196 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9197 
9198 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9199 }
9200 
9201 static struct mlx5_list_entry *
9202 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9203 {
9204 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9205 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9206 	struct tunnel_tbl_entry *tte;
9207 	union tunnel_tbl_key tbl = {
9208 		.val = *(uint64_t *)(ctx->data),
9209 	};
9210 
9211 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9212 			  sizeof(*tte), 0,
9213 			  SOCKET_ID_ANY);
9214 	if (!tte)
9215 		goto err;
9216 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9217 			  &tte->flow_table);
9218 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9219 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9220 			tte->flow_table);
9221 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9222 				tte->flow_table);
9223 		goto err;
9224 	} else if (!tte->flow_table) {
9225 		goto err;
9226 	}
9227 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9228 	tte->tunnel_id = tbl.tunnel_id;
9229 	tte->group = tbl.group;
9230 	return &tte->hash;
9231 err:
9232 	if (tte)
9233 		mlx5_free(tte);
9234 	return NULL;
9235 }
9236 
9237 static struct mlx5_list_entry *
9238 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9239 				  struct mlx5_list_entry *oentry,
9240 				  void *cb_ctx __rte_unused)
9241 {
9242 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9243 						   0, SOCKET_ID_ANY);
9244 
9245 	if (!tte)
9246 		return NULL;
9247 	memcpy(tte, oentry, sizeof(*tte));
9248 	return &tte->hash;
9249 }
9250 
9251 static void
9252 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9253 				       struct mlx5_list_entry *entry)
9254 {
9255 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9256 
9257 	mlx5_free(tte);
9258 }
9259 
9260 static uint32_t
9261 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9262 				const struct mlx5_flow_tunnel *tunnel,
9263 				uint32_t group, uint32_t *table,
9264 				struct rte_flow_error *error)
9265 {
9266 	struct mlx5_list_entry *he;
9267 	struct tunnel_tbl_entry *tte;
9268 	union tunnel_tbl_key key = {
9269 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9270 		.group = group
9271 	};
9272 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9273 	struct mlx5_hlist *group_hash;
9274 	struct mlx5_flow_cb_ctx ctx = {
9275 		.data = &key.val,
9276 	};
9277 
9278 	group_hash = tunnel ? tunnel->groups : thub->groups;
9279 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9280 	if (!he)
9281 		return rte_flow_error_set(error, EINVAL,
9282 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9283 					  NULL,
9284 					  "tunnel group index not supported");
9285 	tte = container_of(he, typeof(*tte), hash);
9286 	*table = tte->flow_table;
9287 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9288 		dev->data->port_id, key.tunnel_id, group, *table);
9289 	return 0;
9290 }
9291 
9292 static void
9293 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9294 		      struct mlx5_flow_tunnel *tunnel)
9295 {
9296 	struct mlx5_priv *priv = dev->data->dev_private;
9297 	struct mlx5_indexed_pool *ipool;
9298 
9299 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9300 		dev->data->port_id, tunnel->tunnel_id);
9301 	LIST_REMOVE(tunnel, chain);
9302 	mlx5_hlist_destroy(tunnel->groups);
9303 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9304 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9305 }
9306 
9307 static bool
9308 mlx5_access_tunnel_offload_db
9309 	(struct rte_eth_dev *dev,
9310 	 bool (*match)(struct rte_eth_dev *,
9311 		       struct mlx5_flow_tunnel *, const void *),
9312 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9313 	 void (*miss)(struct rte_eth_dev *, void *),
9314 	 void *ctx, bool lock_op)
9315 {
9316 	bool verdict = false;
9317 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9318 	struct mlx5_flow_tunnel *tunnel;
9319 
9320 	rte_spinlock_lock(&thub->sl);
9321 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9322 		verdict = match(dev, tunnel, (const void *)ctx);
9323 		if (verdict)
9324 			break;
9325 	}
9326 	if (!lock_op)
9327 		rte_spinlock_unlock(&thub->sl);
9328 	if (verdict && hit)
9329 		hit(dev, tunnel, ctx);
9330 	if (!verdict && miss)
9331 		miss(dev, ctx);
9332 	if (lock_op)
9333 		rte_spinlock_unlock(&thub->sl);
9334 
9335 	return verdict;
9336 }
9337 
9338 struct tunnel_db_find_tunnel_id_ctx {
9339 	uint32_t tunnel_id;
9340 	struct mlx5_flow_tunnel *tunnel;
9341 };
9342 
9343 static bool
9344 find_tunnel_id_match(struct rte_eth_dev *dev,
9345 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9346 {
9347 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9348 
9349 	RTE_SET_USED(dev);
9350 	return tunnel->tunnel_id == ctx->tunnel_id;
9351 }
9352 
9353 static void
9354 find_tunnel_id_hit(struct rte_eth_dev *dev,
9355 		   struct mlx5_flow_tunnel *tunnel, void *x)
9356 {
9357 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9358 	RTE_SET_USED(dev);
9359 	ctx->tunnel = tunnel;
9360 }
9361 
9362 static struct mlx5_flow_tunnel *
9363 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9364 {
9365 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9366 		.tunnel_id = id,
9367 	};
9368 
9369 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9370 				      find_tunnel_id_hit, NULL, &ctx, true);
9371 
9372 	return ctx.tunnel;
9373 }
9374 
9375 static struct mlx5_flow_tunnel *
9376 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9377 			  const struct rte_flow_tunnel *app_tunnel)
9378 {
9379 	struct mlx5_priv *priv = dev->data->dev_private;
9380 	struct mlx5_indexed_pool *ipool;
9381 	struct mlx5_flow_tunnel *tunnel;
9382 	uint32_t id;
9383 
9384 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9385 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9386 	if (!tunnel)
9387 		return NULL;
9388 	if (id >= MLX5_MAX_TUNNELS) {
9389 		mlx5_ipool_free(ipool, id);
9390 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9391 		return NULL;
9392 	}
9393 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9394 					   priv->sh,
9395 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9396 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9397 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9398 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9399 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9400 	if (!tunnel->groups) {
9401 		mlx5_ipool_free(ipool, id);
9402 		return NULL;
9403 	}
9404 	/* initiate new PMD tunnel */
9405 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9406 	tunnel->tunnel_id = id;
9407 	tunnel->action.type = (typeof(tunnel->action.type))
9408 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9409 	tunnel->action.conf = tunnel;
9410 	tunnel->item.type = (typeof(tunnel->item.type))
9411 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9412 	tunnel->item.spec = tunnel;
9413 	tunnel->item.last = NULL;
9414 	tunnel->item.mask = NULL;
9415 
9416 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9417 		dev->data->port_id, tunnel->tunnel_id);
9418 
9419 	return tunnel;
9420 }
9421 
9422 struct tunnel_db_get_tunnel_ctx {
9423 	const struct rte_flow_tunnel *app_tunnel;
9424 	struct mlx5_flow_tunnel *tunnel;
9425 };
9426 
9427 static bool get_tunnel_match(struct rte_eth_dev *dev,
9428 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9429 {
9430 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9431 
9432 	RTE_SET_USED(dev);
9433 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9434 		       sizeof(*ctx->app_tunnel));
9435 }
9436 
9437 static void get_tunnel_hit(struct rte_eth_dev *dev,
9438 			   struct mlx5_flow_tunnel *tunnel, void *x)
9439 {
9440 	/* called under tunnel spinlock protection */
9441 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9442 
9443 	RTE_SET_USED(dev);
9444 	tunnel->refctn++;
9445 	ctx->tunnel = tunnel;
9446 }
9447 
9448 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9449 {
9450 	/* called under tunnel spinlock protection */
9451 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9452 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9453 
9454 	rte_spinlock_unlock(&thub->sl);
9455 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9456 	rte_spinlock_lock(&thub->sl);
9457 	if (ctx->tunnel) {
9458 		ctx->tunnel->refctn = 1;
9459 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9460 	}
9461 }
9462 
9463 
9464 static int
9465 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9466 		     const struct rte_flow_tunnel *app_tunnel,
9467 		     struct mlx5_flow_tunnel **tunnel)
9468 {
9469 	struct tunnel_db_get_tunnel_ctx ctx = {
9470 		.app_tunnel = app_tunnel,
9471 	};
9472 
9473 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9474 				      get_tunnel_miss, &ctx, true);
9475 	*tunnel = ctx.tunnel;
9476 	return ctx.tunnel ? 0 : -ENOMEM;
9477 }
9478 
9479 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9480 {
9481 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9482 
9483 	if (!thub)
9484 		return;
9485 	if (!LIST_EMPTY(&thub->tunnels))
9486 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9487 	mlx5_hlist_destroy(thub->groups);
9488 	mlx5_free(thub);
9489 }
9490 
9491 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9492 {
9493 	int err;
9494 	struct mlx5_flow_tunnel_hub *thub;
9495 
9496 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9497 			   0, SOCKET_ID_ANY);
9498 	if (!thub)
9499 		return -ENOMEM;
9500 	LIST_INIT(&thub->tunnels);
9501 	rte_spinlock_init(&thub->sl);
9502 	thub->groups = mlx5_hlist_create("flow groups", 64,
9503 					 false, true, sh,
9504 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9505 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9506 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9507 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9508 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9509 	if (!thub->groups) {
9510 		err = -rte_errno;
9511 		goto err;
9512 	}
9513 	sh->tunnel_hub = thub;
9514 
9515 	return 0;
9516 
9517 err:
9518 	if (thub->groups)
9519 		mlx5_hlist_destroy(thub->groups);
9520 	if (thub)
9521 		mlx5_free(thub);
9522 	return err;
9523 }
9524 
9525 static inline int
9526 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9527 			  struct rte_flow_tunnel *tunnel,
9528 			  struct rte_flow_error *error)
9529 {
9530 	struct mlx5_priv *priv = dev->data->dev_private;
9531 
9532 	if (!priv->config.dv_flow_en)
9533 		return rte_flow_error_set(error, ENOTSUP,
9534 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9535 					  "flow DV interface is off");
9536 	if (!is_tunnel_offload_active(dev))
9537 		return rte_flow_error_set(error, ENOTSUP,
9538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9539 					  "tunnel offload was not activated");
9540 	if (!tunnel)
9541 		return rte_flow_error_set(error, EINVAL,
9542 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9543 					  "no application tunnel");
9544 	switch (tunnel->type) {
9545 	default:
9546 		return rte_flow_error_set(error, EINVAL,
9547 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9548 					  "unsupported tunnel type");
9549 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9550 	case RTE_FLOW_ITEM_TYPE_GRE:
9551 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9552 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9553 		break;
9554 	}
9555 	return 0;
9556 }
9557 
9558 static int
9559 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9560 		    struct rte_flow_tunnel *app_tunnel,
9561 		    struct rte_flow_action **actions,
9562 		    uint32_t *num_of_actions,
9563 		    struct rte_flow_error *error)
9564 {
9565 	struct mlx5_flow_tunnel *tunnel;
9566 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9567 
9568 	if (ret)
9569 		return ret;
9570 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9571 	if (ret < 0) {
9572 		return rte_flow_error_set(error, ret,
9573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9574 					  "failed to initialize pmd tunnel");
9575 	}
9576 	*actions = &tunnel->action;
9577 	*num_of_actions = 1;
9578 	return 0;
9579 }
9580 
9581 static int
9582 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9583 		       struct rte_flow_tunnel *app_tunnel,
9584 		       struct rte_flow_item **items,
9585 		       uint32_t *num_of_items,
9586 		       struct rte_flow_error *error)
9587 {
9588 	struct mlx5_flow_tunnel *tunnel;
9589 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9590 
9591 	if (ret)
9592 		return ret;
9593 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9594 	if (ret < 0) {
9595 		return rte_flow_error_set(error, ret,
9596 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9597 					  "failed to initialize pmd tunnel");
9598 	}
9599 	*items = &tunnel->item;
9600 	*num_of_items = 1;
9601 	return 0;
9602 }
9603 
9604 struct tunnel_db_element_release_ctx {
9605 	struct rte_flow_item *items;
9606 	struct rte_flow_action *actions;
9607 	uint32_t num_elements;
9608 	struct rte_flow_error *error;
9609 	int ret;
9610 };
9611 
9612 static bool
9613 tunnel_element_release_match(struct rte_eth_dev *dev,
9614 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9615 {
9616 	const struct tunnel_db_element_release_ctx *ctx = x;
9617 
9618 	RTE_SET_USED(dev);
9619 	if (ctx->num_elements != 1)
9620 		return false;
9621 	else if (ctx->items)
9622 		return ctx->items == &tunnel->item;
9623 	else if (ctx->actions)
9624 		return ctx->actions == &tunnel->action;
9625 
9626 	return false;
9627 }
9628 
9629 static void
9630 tunnel_element_release_hit(struct rte_eth_dev *dev,
9631 			   struct mlx5_flow_tunnel *tunnel, void *x)
9632 {
9633 	struct tunnel_db_element_release_ctx *ctx = x;
9634 	ctx->ret = 0;
9635 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9636 		mlx5_flow_tunnel_free(dev, tunnel);
9637 }
9638 
9639 static void
9640 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9641 {
9642 	struct tunnel_db_element_release_ctx *ctx = x;
9643 	RTE_SET_USED(dev);
9644 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9645 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9646 				      "invalid argument");
9647 }
9648 
9649 static int
9650 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9651 		       struct rte_flow_item *pmd_items,
9652 		       uint32_t num_items, struct rte_flow_error *err)
9653 {
9654 	struct tunnel_db_element_release_ctx ctx = {
9655 		.items = pmd_items,
9656 		.actions = NULL,
9657 		.num_elements = num_items,
9658 		.error = err,
9659 	};
9660 
9661 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9662 				      tunnel_element_release_hit,
9663 				      tunnel_element_release_miss, &ctx, false);
9664 
9665 	return ctx.ret;
9666 }
9667 
9668 static int
9669 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9670 			 struct rte_flow_action *pmd_actions,
9671 			 uint32_t num_actions, struct rte_flow_error *err)
9672 {
9673 	struct tunnel_db_element_release_ctx ctx = {
9674 		.items = NULL,
9675 		.actions = pmd_actions,
9676 		.num_elements = num_actions,
9677 		.error = err,
9678 	};
9679 
9680 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9681 				      tunnel_element_release_hit,
9682 				      tunnel_element_release_miss, &ctx, false);
9683 
9684 	return ctx.ret;
9685 }
9686 
9687 static int
9688 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9689 				  struct rte_mbuf *m,
9690 				  struct rte_flow_restore_info *info,
9691 				  struct rte_flow_error *err)
9692 {
9693 	uint64_t ol_flags = m->ol_flags;
9694 	const struct mlx5_flow_tbl_data_entry *tble;
9695 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9696 
9697 	if (!is_tunnel_offload_active(dev)) {
9698 		info->flags = 0;
9699 		return 0;
9700 	}
9701 
9702 	if ((ol_flags & mask) != mask)
9703 		goto err;
9704 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9705 	if (!tble) {
9706 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9707 			dev->data->port_id, m->hash.fdir.hi);
9708 		goto err;
9709 	}
9710 	MLX5_ASSERT(tble->tunnel);
9711 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9712 	info->group_id = tble->group_id;
9713 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9714 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9715 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9716 
9717 	return 0;
9718 
9719 err:
9720 	return rte_flow_error_set(err, EINVAL,
9721 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9722 				  "failed to get restore info");
9723 }
9724 
9725 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9726 static int
9727 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9728 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9729 			   __rte_unused struct rte_flow_action **actions,
9730 			   __rte_unused uint32_t *num_of_actions,
9731 			   __rte_unused struct rte_flow_error *error)
9732 {
9733 	return -ENOTSUP;
9734 }
9735 
9736 static int
9737 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9738 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9739 		       __rte_unused struct rte_flow_item **items,
9740 		       __rte_unused uint32_t *num_of_items,
9741 		       __rte_unused struct rte_flow_error *error)
9742 {
9743 	return -ENOTSUP;
9744 }
9745 
9746 static int
9747 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9748 			      __rte_unused struct rte_flow_item *pmd_items,
9749 			      __rte_unused uint32_t num_items,
9750 			      __rte_unused struct rte_flow_error *err)
9751 {
9752 	return -ENOTSUP;
9753 }
9754 
9755 static int
9756 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9757 				__rte_unused struct rte_flow_action *pmd_action,
9758 				__rte_unused uint32_t num_actions,
9759 				__rte_unused struct rte_flow_error *err)
9760 {
9761 	return -ENOTSUP;
9762 }
9763 
9764 static int
9765 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9766 				  __rte_unused struct rte_mbuf *m,
9767 				  __rte_unused struct rte_flow_restore_info *i,
9768 				  __rte_unused struct rte_flow_error *err)
9769 {
9770 	return -ENOTSUP;
9771 }
9772 
9773 static int
9774 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9775 			     __rte_unused struct rte_flow *flow,
9776 			     __rte_unused const struct rte_flow_attr *attr,
9777 			     __rte_unused const struct rte_flow_action *actions,
9778 			     __rte_unused uint32_t flow_idx,
9779 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9780 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9781 			     __rte_unused struct rte_flow_error *error)
9782 {
9783 	return -ENOTSUP;
9784 }
9785 
9786 static struct mlx5_flow_tunnel *
9787 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9788 		    __rte_unused uint32_t id)
9789 {
9790 	return NULL;
9791 }
9792 
9793 static void
9794 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9795 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9796 {
9797 }
9798 
9799 static uint32_t
9800 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9801 				__rte_unused const struct mlx5_flow_tunnel *t,
9802 				__rte_unused uint32_t group,
9803 				__rte_unused uint32_t *table,
9804 				struct rte_flow_error *error)
9805 {
9806 	return rte_flow_error_set(error, ENOTSUP,
9807 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9808 				  "tunnel offload requires DV support");
9809 }
9810 
9811 void
9812 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9813 			__rte_unused  uint16_t port_id)
9814 {
9815 }
9816 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9817 
9818 /* Flex flow item API */
9819 static struct rte_flow_item_flex_handle *
9820 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9821 			   const struct rte_flow_item_flex_conf *conf,
9822 			   struct rte_flow_error *error)
9823 {
9824 	static const char err_msg[] = "flex item creation unsupported";
9825 	struct rte_flow_attr attr = { .transfer = 0 };
9826 	const struct mlx5_flow_driver_ops *fops =
9827 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9828 
9829 	if (!fops->item_create) {
9830 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9831 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9832 				   NULL, err_msg);
9833 		return NULL;
9834 	}
9835 	return fops->item_create(dev, conf, error);
9836 }
9837 
9838 static int
9839 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9840 			    const struct rte_flow_item_flex_handle *handle,
9841 			    struct rte_flow_error *error)
9842 {
9843 	static const char err_msg[] = "flex item release unsupported";
9844 	struct rte_flow_attr attr = { .transfer = 0 };
9845 	const struct mlx5_flow_driver_ops *fops =
9846 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9847 
9848 	if (!fops->item_release) {
9849 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9850 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9851 				   NULL, err_msg);
9852 		return -rte_errno;
9853 	}
9854 	return fops->item_release(dev, handle, error);
9855 }
9856 
9857 static void
9858 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9859 {
9860 	int ret;
9861 	struct rte_flow_error error;
9862 
9863 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9864 		char *item_name;
9865 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9866 				    sizeof(item_name),
9867 				    (void *)(uintptr_t)item->type, &error);
9868 		if (ret > 0)
9869 			printf("%s ", item_name);
9870 		else
9871 			printf("%d\n", (int)item->type);
9872 	}
9873 	printf("END\n");
9874 }
9875 
9876 static int
9877 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9878 {
9879 	const struct rte_flow_item_udp *spec = udp_item->spec;
9880 	const struct rte_flow_item_udp *mask = udp_item->mask;
9881 	uint16_t udp_dport = 0;
9882 
9883 	if (spec != NULL) {
9884 		if (!mask)
9885 			mask = &rte_flow_item_udp_mask;
9886 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9887 				mask->hdr.dst_port);
9888 	}
9889 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9890 }
9891 
9892 static const struct mlx5_flow_expand_node *
9893 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9894 		unsigned int item_idx,
9895 		const struct mlx5_flow_expand_node graph[],
9896 		const struct mlx5_flow_expand_node *node)
9897 {
9898 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9899 
9900 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9901 			node != NULL &&
9902 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9903 		/*
9904 		 * The expansion node is VXLAN and it is also the last
9905 		 * expandable item in the pattern, so need to continue
9906 		 * expansion of the inner tunnel.
9907 		 */
9908 		MLX5_ASSERT(item_idx > 0);
9909 		prev_item = pattern + item_idx - 1;
9910 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9911 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9912 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9913 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9914 	}
9915 	return node;
9916 }
9917 
9918 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9919 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9920 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9921 };
9922 
9923 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9924 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9925 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9926 	{ 9, 10, 11 }, { 12, 13, 14 },
9927 };
9928 
9929 /**
9930  * Discover the number of available flow priorities.
9931  *
9932  * @param dev
9933  *   Ethernet device.
9934  *
9935  * @return
9936  *   On success, number of available flow priorities.
9937  *   On failure, a negative errno-style code and rte_errno is set.
9938  */
9939 int
9940 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9941 {
9942 	static const uint16_t vprio[] = {8, 16};
9943 	const struct mlx5_priv *priv = dev->data->dev_private;
9944 	const struct mlx5_flow_driver_ops *fops;
9945 	enum mlx5_flow_drv_type type;
9946 	int ret;
9947 
9948 	type = mlx5_flow_os_get_type();
9949 	if (type == MLX5_FLOW_TYPE_MAX) {
9950 		type = MLX5_FLOW_TYPE_VERBS;
9951 		if (priv->sh->devx && priv->config.dv_flow_en)
9952 			type = MLX5_FLOW_TYPE_DV;
9953 	}
9954 	fops = flow_get_drv_ops(type);
9955 	if (fops->discover_priorities == NULL) {
9956 		DRV_LOG(ERR, "Priority discovery not supported");
9957 		rte_errno = ENOTSUP;
9958 		return -rte_errno;
9959 	}
9960 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9961 	if (ret < 0)
9962 		return ret;
9963 	switch (ret) {
9964 	case 8:
9965 		ret = RTE_DIM(priority_map_3);
9966 		break;
9967 	case 16:
9968 		ret = RTE_DIM(priority_map_5);
9969 		break;
9970 	default:
9971 		rte_errno = ENOTSUP;
9972 		DRV_LOG(ERR,
9973 			"port %u maximum priority: %d expected 8/16",
9974 			dev->data->port_id, ret);
9975 		return -rte_errno;
9976 	}
9977 	DRV_LOG(INFO, "port %u supported flow priorities:"
9978 		" 0-%d for ingress or egress root table,"
9979 		" 0-%d for non-root table or transfer root table.",
9980 		dev->data->port_id, ret - 2,
9981 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9982 	return ret;
9983 }
9984 
9985 /**
9986  * Adjust flow priority based on the highest layer and the request priority.
9987  *
9988  * @param[in] dev
9989  *   Pointer to the Ethernet device structure.
9990  * @param[in] priority
9991  *   The rule base priority.
9992  * @param[in] subpriority
9993  *   The priority based on the items.
9994  *
9995  * @return
9996  *   The new priority.
9997  */
9998 uint32_t
9999 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10000 			  uint32_t subpriority)
10001 {
10002 	uint32_t res = 0;
10003 	struct mlx5_priv *priv = dev->data->dev_private;
10004 
10005 	switch (priv->sh->flow_max_priority) {
10006 	case RTE_DIM(priority_map_3):
10007 		res = priority_map_3[priority][subpriority];
10008 		break;
10009 	case RTE_DIM(priority_map_5):
10010 		res = priority_map_5[priority][subpriority];
10011 		break;
10012 	}
10013 	return  res;
10014 }
10015