xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision aaf3b44c6618f60ec878237e3a3dcad0912bedf4)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rx.h"
32 #include "mlx5_tx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 struct tunnel_default_miss_ctx {
37 	uint16_t *queue;
38 	__extension__
39 	union {
40 		struct rte_flow_action_rss action_rss;
41 		struct rte_flow_action_queue miss_queue;
42 		struct rte_flow_action_jump miss_jump;
43 		uint8_t raw[0];
44 	};
45 };
46 
47 static int
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 			     struct rte_flow *flow,
50 			     const struct rte_flow_attr *attr,
51 			     const struct rte_flow_action *app_actions,
52 			     uint32_t flow_idx,
53 			     const struct mlx5_flow_tunnel *tunnel,
54 			     struct tunnel_default_miss_ctx *ctx,
55 			     struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
58 static void
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
60 static uint32_t
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 				const struct mlx5_flow_tunnel *tunnel,
63 				uint32_t group, uint32_t *table,
64 				struct rte_flow_error *error);
65 
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
68 
69 
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
72 
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
74 
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
79 #endif
80 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
81 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
82 };
83 
84 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
85 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
86 	(const int []){ \
87 		__VA_ARGS__, 0, \
88 	}
89 
90 /** Node object of input graph for mlx5_flow_expand_rss(). */
91 struct mlx5_flow_expand_node {
92 	const int *const next;
93 	/**<
94 	 * List of next node indexes. Index 0 is interpreted as a terminator.
95 	 */
96 	const enum rte_flow_item_type type;
97 	/**< Pattern item type of current node. */
98 	uint64_t rss_types;
99 	/**<
100 	 * RSS types bit-field associated with this node
101 	 * (see RTE_ETH_RSS_* definitions).
102 	 */
103 	uint64_t node_flags;
104 	/**<
105 	 *  Bit-fields that define how the node is used in the expansion.
106 	 * (see MLX5_EXPANSION_NODE_* definitions).
107 	 */
108 };
109 
110 /* Optional expand field. The expansion alg will not go deeper. */
111 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
112 
113 /* The node is not added implicitly as expansion to the flow pattern.
114  * If the node type does not match the flow pattern item type, the
115  * expansion alg will go deeper to its next items.
116  * In the current implementation, the list of next nodes indexes can
117  * have up to one node with this flag set and it has to be the last
118  * node index (before the list terminator).
119  */
120 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
121 
122 /** Object returned by mlx5_flow_expand_rss(). */
123 struct mlx5_flow_expand_rss {
124 	uint32_t entries;
125 	/**< Number of entries @p patterns and @p priorities. */
126 	struct {
127 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
128 		uint32_t priority; /**< Priority offset for each expansion. */
129 	} entry[];
130 };
131 
132 static void
133 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
134 
135 static const struct mlx5_flow_expand_node *
136 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
137 		unsigned int item_idx,
138 		const struct mlx5_flow_expand_node graph[],
139 		const struct mlx5_flow_expand_node *node);
140 
141 static bool
142 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
143 {
144 	switch (item->type) {
145 	case RTE_FLOW_ITEM_TYPE_ETH:
146 	case RTE_FLOW_ITEM_TYPE_VLAN:
147 	case RTE_FLOW_ITEM_TYPE_IPV4:
148 	case RTE_FLOW_ITEM_TYPE_IPV6:
149 	case RTE_FLOW_ITEM_TYPE_UDP:
150 	case RTE_FLOW_ITEM_TYPE_TCP:
151 	case RTE_FLOW_ITEM_TYPE_VXLAN:
152 	case RTE_FLOW_ITEM_TYPE_NVGRE:
153 	case RTE_FLOW_ITEM_TYPE_GRE:
154 	case RTE_FLOW_ITEM_TYPE_GENEVE:
155 	case RTE_FLOW_ITEM_TYPE_MPLS:
156 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
157 	case RTE_FLOW_ITEM_TYPE_GRE_KEY:
158 	case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
159 	case RTE_FLOW_ITEM_TYPE_GTP:
160 		return true;
161 	default:
162 		break;
163 	}
164 	return false;
165 }
166 
167 static enum rte_flow_item_type
168 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
169 {
170 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
171 	uint16_t ether_type = 0;
172 	uint16_t ether_type_m;
173 	uint8_t ip_next_proto = 0;
174 	uint8_t ip_next_proto_m;
175 
176 	if (item == NULL || item->spec == NULL)
177 		return ret;
178 	switch (item->type) {
179 	case RTE_FLOW_ITEM_TYPE_ETH:
180 		if (item->mask)
181 			ether_type_m = ((const struct rte_flow_item_eth *)
182 						(item->mask))->type;
183 		else
184 			ether_type_m = rte_flow_item_eth_mask.type;
185 		if (ether_type_m != RTE_BE16(0xFFFF))
186 			break;
187 		ether_type = ((const struct rte_flow_item_eth *)
188 				(item->spec))->type;
189 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
190 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
191 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
192 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
193 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
194 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
195 		else
196 			ret = RTE_FLOW_ITEM_TYPE_END;
197 		break;
198 	case RTE_FLOW_ITEM_TYPE_VLAN:
199 		if (item->mask)
200 			ether_type_m = ((const struct rte_flow_item_vlan *)
201 						(item->mask))->inner_type;
202 		else
203 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
204 		if (ether_type_m != RTE_BE16(0xFFFF))
205 			break;
206 		ether_type = ((const struct rte_flow_item_vlan *)
207 				(item->spec))->inner_type;
208 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
209 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
210 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
211 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
212 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
213 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
214 		else
215 			ret = RTE_FLOW_ITEM_TYPE_END;
216 		break;
217 	case RTE_FLOW_ITEM_TYPE_IPV4:
218 		if (item->mask)
219 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
220 					(item->mask))->hdr.next_proto_id;
221 		else
222 			ip_next_proto_m =
223 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
224 		if (ip_next_proto_m != 0xFF)
225 			break;
226 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
227 				(item->spec))->hdr.next_proto_id;
228 		if (ip_next_proto == IPPROTO_UDP)
229 			ret = RTE_FLOW_ITEM_TYPE_UDP;
230 		else if (ip_next_proto == IPPROTO_TCP)
231 			ret = RTE_FLOW_ITEM_TYPE_TCP;
232 		else if (ip_next_proto == IPPROTO_IP)
233 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
234 		else if (ip_next_proto == IPPROTO_IPV6)
235 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
236 		else
237 			ret = RTE_FLOW_ITEM_TYPE_END;
238 		break;
239 	case RTE_FLOW_ITEM_TYPE_IPV6:
240 		if (item->mask)
241 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
242 						(item->mask))->hdr.proto;
243 		else
244 			ip_next_proto_m =
245 				rte_flow_item_ipv6_mask.hdr.proto;
246 		if (ip_next_proto_m != 0xFF)
247 			break;
248 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
249 				(item->spec))->hdr.proto;
250 		if (ip_next_proto == IPPROTO_UDP)
251 			ret = RTE_FLOW_ITEM_TYPE_UDP;
252 		else if (ip_next_proto == IPPROTO_TCP)
253 			ret = RTE_FLOW_ITEM_TYPE_TCP;
254 		else if (ip_next_proto == IPPROTO_IP)
255 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
256 		else if (ip_next_proto == IPPROTO_IPV6)
257 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
258 		else
259 			ret = RTE_FLOW_ITEM_TYPE_END;
260 		break;
261 	default:
262 		ret = RTE_FLOW_ITEM_TYPE_VOID;
263 		break;
264 	}
265 	return ret;
266 }
267 
268 static const int *
269 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
270 		const int *next_node)
271 {
272 	const struct mlx5_flow_expand_node *node = NULL;
273 	const int *next = next_node;
274 
275 	while (next && *next) {
276 		/*
277 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
278 		 * flag set, because they were not found in the flow pattern.
279 		 */
280 		node = &graph[*next];
281 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
282 			break;
283 		next = node->next;
284 	}
285 	return next;
286 }
287 
288 #define MLX5_RSS_EXP_ELT_N 16
289 
290 /**
291  * Expand RSS flows into several possible flows according to the RSS hash
292  * fields requested and the driver capabilities.
293  *
294  * @param[out] buf
295  *   Buffer to store the result expansion.
296  * @param[in] size
297  *   Buffer size in bytes. If 0, @p buf can be NULL.
298  * @param[in] pattern
299  *   User flow pattern.
300  * @param[in] types
301  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
302  * @param[in] graph
303  *   Input graph to expand @p pattern according to @p types.
304  * @param[in] graph_root_index
305  *   Index of root node in @p graph, typically 0.
306  *
307  * @return
308  *   A positive value representing the size of @p buf in bytes regardless of
309  *   @p size on success, a negative errno value otherwise and rte_errno is
310  *   set, the following errors are defined:
311  *
312  *   -E2BIG: graph-depth @p graph is too deep.
313  *   -EINVAL: @p size has not enough space for expanded pattern.
314  */
315 static int
316 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
317 		     const struct rte_flow_item *pattern, uint64_t types,
318 		     const struct mlx5_flow_expand_node graph[],
319 		     int graph_root_index)
320 {
321 	const struct rte_flow_item *item;
322 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
323 	const int *next_node;
324 	const int *stack[MLX5_RSS_EXP_ELT_N];
325 	int stack_pos = 0;
326 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
327 	unsigned int i, item_idx, last_expand_item_idx = 0;
328 	size_t lsize;
329 	size_t user_pattern_size = 0;
330 	void *addr = NULL;
331 	const struct mlx5_flow_expand_node *next = NULL;
332 	struct rte_flow_item missed_item;
333 	int missed = 0;
334 	int elt = 0;
335 	const struct rte_flow_item *last_expand_item = NULL;
336 
337 	memset(&missed_item, 0, sizeof(missed_item));
338 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
339 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
340 	if (lsize > size)
341 		return -EINVAL;
342 	buf->entry[0].priority = 0;
343 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
344 	buf->entries = 0;
345 	addr = buf->entry[0].pattern;
346 	for (item = pattern, item_idx = 0;
347 			item->type != RTE_FLOW_ITEM_TYPE_END;
348 			item++, item_idx++) {
349 		if (!mlx5_flow_is_rss_expandable_item(item)) {
350 			user_pattern_size += sizeof(*item);
351 			continue;
352 		}
353 		last_expand_item = item;
354 		last_expand_item_idx = item_idx;
355 		i = 0;
356 		while (node->next && node->next[i]) {
357 			next = &graph[node->next[i]];
358 			if (next->type == item->type)
359 				break;
360 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
361 				node = next;
362 				i = 0;
363 			} else {
364 				++i;
365 			}
366 		}
367 		if (next)
368 			node = next;
369 		user_pattern_size += sizeof(*item);
370 	}
371 	user_pattern_size += sizeof(*item); /* Handle END item. */
372 	lsize += user_pattern_size;
373 	if (lsize > size)
374 		return -EINVAL;
375 	/* Copy the user pattern in the first entry of the buffer. */
376 	rte_memcpy(addr, pattern, user_pattern_size);
377 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
378 	buf->entries = 1;
379 	/* Start expanding. */
380 	memset(flow_items, 0, sizeof(flow_items));
381 	user_pattern_size -= sizeof(*item);
382 	/*
383 	 * Check if the last valid item has spec set, need complete pattern,
384 	 * and the pattern can be used for expansion.
385 	 */
386 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
387 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
388 		/* Item type END indicates expansion is not required. */
389 		return lsize;
390 	}
391 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
392 		next = NULL;
393 		missed = 1;
394 		i = 0;
395 		while (node->next && node->next[i]) {
396 			next = &graph[node->next[i]];
397 			if (next->type == missed_item.type) {
398 				flow_items[0].type = missed_item.type;
399 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
400 				break;
401 			}
402 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
403 				node = next;
404 				i = 0;
405 			} else {
406 				++i;
407 			}
408 			next = NULL;
409 		}
410 	}
411 	if (next && missed) {
412 		elt = 2; /* missed item + item end. */
413 		node = next;
414 		lsize += elt * sizeof(*item) + user_pattern_size;
415 		if (lsize > size)
416 			return -EINVAL;
417 		if (node->rss_types & types) {
418 			buf->entry[buf->entries].priority = 1;
419 			buf->entry[buf->entries].pattern = addr;
420 			buf->entries++;
421 			rte_memcpy(addr, buf->entry[0].pattern,
422 				   user_pattern_size);
423 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
424 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
425 			addr = (void *)(((uintptr_t)addr) +
426 					elt * sizeof(*item));
427 		}
428 	} else if (last_expand_item != NULL) {
429 		node = mlx5_flow_expand_rss_adjust_node(pattern,
430 				last_expand_item_idx, graph, node);
431 	}
432 	memset(flow_items, 0, sizeof(flow_items));
433 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
434 			node->next);
435 	stack[stack_pos] = next_node;
436 	node = next_node ? &graph[*next_node] : NULL;
437 	while (node) {
438 		flow_items[stack_pos].type = node->type;
439 		if (node->rss_types & types) {
440 			size_t n;
441 			/*
442 			 * compute the number of items to copy from the
443 			 * expansion and copy it.
444 			 * When the stack_pos is 0, there are 1 element in it,
445 			 * plus the addition END item.
446 			 */
447 			elt = stack_pos + 2;
448 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
449 			lsize += elt * sizeof(*item) + user_pattern_size;
450 			if (lsize > size)
451 				return -EINVAL;
452 			n = elt * sizeof(*item);
453 			buf->entry[buf->entries].priority =
454 				stack_pos + 1 + missed;
455 			buf->entry[buf->entries].pattern = addr;
456 			buf->entries++;
457 			rte_memcpy(addr, buf->entry[0].pattern,
458 				   user_pattern_size);
459 			addr = (void *)(((uintptr_t)addr) +
460 					user_pattern_size);
461 			rte_memcpy(addr, &missed_item,
462 				   missed * sizeof(*item));
463 			addr = (void *)(((uintptr_t)addr) +
464 				missed * sizeof(*item));
465 			rte_memcpy(addr, flow_items, n);
466 			addr = (void *)(((uintptr_t)addr) + n);
467 		}
468 		/* Go deeper. */
469 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
470 				node->next) {
471 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
472 					node->next);
473 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
474 				rte_errno = E2BIG;
475 				return -rte_errno;
476 			}
477 			stack[stack_pos] = next_node;
478 		} else if (*(next_node + 1)) {
479 			/* Follow up with the next possibility. */
480 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
481 					++next_node);
482 		} else if (!stack_pos) {
483 			/*
484 			 * Completing the traverse over the different paths.
485 			 * The next_node is advanced to the terminator.
486 			 */
487 			++next_node;
488 		} else {
489 			/* Move to the next path. */
490 			while (stack_pos) {
491 				next_node = stack[--stack_pos];
492 				next_node++;
493 				if (*next_node)
494 					break;
495 			}
496 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
497 					next_node);
498 			stack[stack_pos] = next_node;
499 		}
500 		node = next_node && *next_node ? &graph[*next_node] : NULL;
501 	};
502 	return lsize;
503 }
504 
505 enum mlx5_expansion {
506 	MLX5_EXPANSION_ROOT,
507 	MLX5_EXPANSION_ROOT_OUTER,
508 	MLX5_EXPANSION_OUTER_ETH,
509 	MLX5_EXPANSION_OUTER_VLAN,
510 	MLX5_EXPANSION_OUTER_IPV4,
511 	MLX5_EXPANSION_OUTER_IPV4_UDP,
512 	MLX5_EXPANSION_OUTER_IPV4_TCP,
513 	MLX5_EXPANSION_OUTER_IPV6,
514 	MLX5_EXPANSION_OUTER_IPV6_UDP,
515 	MLX5_EXPANSION_OUTER_IPV6_TCP,
516 	MLX5_EXPANSION_VXLAN,
517 	MLX5_EXPANSION_STD_VXLAN,
518 	MLX5_EXPANSION_L3_VXLAN,
519 	MLX5_EXPANSION_VXLAN_GPE,
520 	MLX5_EXPANSION_GRE,
521 	MLX5_EXPANSION_NVGRE,
522 	MLX5_EXPANSION_GRE_KEY,
523 	MLX5_EXPANSION_MPLS,
524 	MLX5_EXPANSION_ETH,
525 	MLX5_EXPANSION_VLAN,
526 	MLX5_EXPANSION_IPV4,
527 	MLX5_EXPANSION_IPV4_UDP,
528 	MLX5_EXPANSION_IPV4_TCP,
529 	MLX5_EXPANSION_IPV6,
530 	MLX5_EXPANSION_IPV6_UDP,
531 	MLX5_EXPANSION_IPV6_TCP,
532 	MLX5_EXPANSION_IPV6_FRAG_EXT,
533 	MLX5_EXPANSION_GTP
534 };
535 
536 /** Supported expansion of items. */
537 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
538 	[MLX5_EXPANSION_ROOT] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
540 						  MLX5_EXPANSION_IPV4,
541 						  MLX5_EXPANSION_IPV6),
542 		.type = RTE_FLOW_ITEM_TYPE_END,
543 	},
544 	[MLX5_EXPANSION_ROOT_OUTER] = {
545 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
546 						  MLX5_EXPANSION_OUTER_IPV4,
547 						  MLX5_EXPANSION_OUTER_IPV6),
548 		.type = RTE_FLOW_ITEM_TYPE_END,
549 	},
550 	[MLX5_EXPANSION_OUTER_ETH] = {
551 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
552 		.type = RTE_FLOW_ITEM_TYPE_ETH,
553 		.rss_types = 0,
554 	},
555 	[MLX5_EXPANSION_OUTER_VLAN] = {
556 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
557 						  MLX5_EXPANSION_OUTER_IPV6),
558 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
559 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
560 	},
561 	[MLX5_EXPANSION_OUTER_IPV4] = {
562 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
563 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
564 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
565 			 MLX5_EXPANSION_GRE,
566 			 MLX5_EXPANSION_NVGRE,
567 			 MLX5_EXPANSION_IPV4,
568 			 MLX5_EXPANSION_IPV6),
569 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
570 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
571 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
572 	},
573 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
574 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
575 						  MLX5_EXPANSION_VXLAN_GPE,
576 						  MLX5_EXPANSION_MPLS,
577 						  MLX5_EXPANSION_GTP),
578 		.type = RTE_FLOW_ITEM_TYPE_UDP,
579 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
580 	},
581 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
582 		.type = RTE_FLOW_ITEM_TYPE_TCP,
583 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
584 	},
585 	[MLX5_EXPANSION_OUTER_IPV6] = {
586 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
587 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
588 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
589 			 MLX5_EXPANSION_IPV4,
590 			 MLX5_EXPANSION_IPV6,
591 			 MLX5_EXPANSION_GRE,
592 			 MLX5_EXPANSION_NVGRE),
593 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
594 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
595 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
596 	},
597 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
598 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
599 						  MLX5_EXPANSION_VXLAN_GPE,
600 						  MLX5_EXPANSION_MPLS,
601 						  MLX5_EXPANSION_GTP),
602 		.type = RTE_FLOW_ITEM_TYPE_UDP,
603 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
604 	},
605 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
606 		.type = RTE_FLOW_ITEM_TYPE_TCP,
607 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
608 	},
609 	[MLX5_EXPANSION_VXLAN] = {
610 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
611 						  MLX5_EXPANSION_IPV4,
612 						  MLX5_EXPANSION_IPV6),
613 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
614 	},
615 	[MLX5_EXPANSION_STD_VXLAN] = {
616 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
617 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
618 	},
619 	[MLX5_EXPANSION_L3_VXLAN] = {
620 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
621 					MLX5_EXPANSION_IPV6),
622 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
623 	},
624 	[MLX5_EXPANSION_VXLAN_GPE] = {
625 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
626 						  MLX5_EXPANSION_IPV4,
627 						  MLX5_EXPANSION_IPV6),
628 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
629 	},
630 	[MLX5_EXPANSION_GRE] = {
631 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
632 						  MLX5_EXPANSION_IPV6,
633 						  MLX5_EXPANSION_GRE_KEY,
634 						  MLX5_EXPANSION_MPLS),
635 		.type = RTE_FLOW_ITEM_TYPE_GRE,
636 	},
637 	[MLX5_EXPANSION_GRE_KEY] = {
638 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
639 						  MLX5_EXPANSION_IPV6,
640 						  MLX5_EXPANSION_MPLS),
641 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
642 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
643 	},
644 	[MLX5_EXPANSION_NVGRE] = {
645 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
646 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
647 	},
648 	[MLX5_EXPANSION_MPLS] = {
649 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
650 						  MLX5_EXPANSION_IPV6,
651 						  MLX5_EXPANSION_ETH),
652 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
653 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
654 	},
655 	[MLX5_EXPANSION_ETH] = {
656 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
657 		.type = RTE_FLOW_ITEM_TYPE_ETH,
658 	},
659 	[MLX5_EXPANSION_VLAN] = {
660 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
661 						  MLX5_EXPANSION_IPV6),
662 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
663 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
664 	},
665 	[MLX5_EXPANSION_IPV4] = {
666 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
667 						  MLX5_EXPANSION_IPV4_TCP),
668 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
669 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
670 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
671 	},
672 	[MLX5_EXPANSION_IPV4_UDP] = {
673 		.type = RTE_FLOW_ITEM_TYPE_UDP,
674 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
675 	},
676 	[MLX5_EXPANSION_IPV4_TCP] = {
677 		.type = RTE_FLOW_ITEM_TYPE_TCP,
678 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
679 	},
680 	[MLX5_EXPANSION_IPV6] = {
681 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
682 						  MLX5_EXPANSION_IPV6_TCP,
683 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
684 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
685 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
686 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
687 	},
688 	[MLX5_EXPANSION_IPV6_UDP] = {
689 		.type = RTE_FLOW_ITEM_TYPE_UDP,
690 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
691 	},
692 	[MLX5_EXPANSION_IPV6_TCP] = {
693 		.type = RTE_FLOW_ITEM_TYPE_TCP,
694 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
695 	},
696 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
697 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
698 	},
699 	[MLX5_EXPANSION_GTP] = {
700 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
701 						  MLX5_EXPANSION_IPV6),
702 		.type = RTE_FLOW_ITEM_TYPE_GTP,
703 	},
704 };
705 
706 static struct rte_flow_action_handle *
707 mlx5_action_handle_create(struct rte_eth_dev *dev,
708 			  const struct rte_flow_indir_action_conf *conf,
709 			  const struct rte_flow_action *action,
710 			  struct rte_flow_error *error);
711 static int mlx5_action_handle_destroy
712 				(struct rte_eth_dev *dev,
713 				 struct rte_flow_action_handle *handle,
714 				 struct rte_flow_error *error);
715 static int mlx5_action_handle_update
716 				(struct rte_eth_dev *dev,
717 				 struct rte_flow_action_handle *handle,
718 				 const void *update,
719 				 struct rte_flow_error *error);
720 static int mlx5_action_handle_query
721 				(struct rte_eth_dev *dev,
722 				 const struct rte_flow_action_handle *handle,
723 				 void *data,
724 				 struct rte_flow_error *error);
725 static int
726 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
727 		    struct rte_flow_tunnel *app_tunnel,
728 		    struct rte_flow_action **actions,
729 		    uint32_t *num_of_actions,
730 		    struct rte_flow_error *error);
731 static int
732 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
733 		       struct rte_flow_tunnel *app_tunnel,
734 		       struct rte_flow_item **items,
735 		       uint32_t *num_of_items,
736 		       struct rte_flow_error *error);
737 static int
738 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
739 			      struct rte_flow_item *pmd_items,
740 			      uint32_t num_items, struct rte_flow_error *err);
741 static int
742 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
743 				struct rte_flow_action *pmd_actions,
744 				uint32_t num_actions,
745 				struct rte_flow_error *err);
746 static int
747 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
748 				  struct rte_mbuf *m,
749 				  struct rte_flow_restore_info *info,
750 				  struct rte_flow_error *err);
751 
752 static const struct rte_flow_ops mlx5_flow_ops = {
753 	.validate = mlx5_flow_validate,
754 	.create = mlx5_flow_create,
755 	.destroy = mlx5_flow_destroy,
756 	.flush = mlx5_flow_flush,
757 	.isolate = mlx5_flow_isolate,
758 	.query = mlx5_flow_query,
759 	.dev_dump = mlx5_flow_dev_dump,
760 	.get_aged_flows = mlx5_flow_get_aged_flows,
761 	.action_handle_create = mlx5_action_handle_create,
762 	.action_handle_destroy = mlx5_action_handle_destroy,
763 	.action_handle_update = mlx5_action_handle_update,
764 	.action_handle_query = mlx5_action_handle_query,
765 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
766 	.tunnel_match = mlx5_flow_tunnel_match,
767 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
768 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
769 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
770 };
771 
772 /* Tunnel information. */
773 struct mlx5_flow_tunnel_info {
774 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
775 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
776 };
777 
778 static struct mlx5_flow_tunnel_info tunnels_info[] = {
779 	{
780 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
781 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
782 	},
783 	{
784 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
785 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
786 	},
787 	{
788 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
789 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
790 	},
791 	{
792 		.tunnel = MLX5_FLOW_LAYER_GRE,
793 		.ptype = RTE_PTYPE_TUNNEL_GRE,
794 	},
795 	{
796 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
797 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
798 	},
799 	{
800 		.tunnel = MLX5_FLOW_LAYER_MPLS,
801 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
802 	},
803 	{
804 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
805 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
806 	},
807 	{
808 		.tunnel = MLX5_FLOW_LAYER_IPIP,
809 		.ptype = RTE_PTYPE_TUNNEL_IP,
810 	},
811 	{
812 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
813 		.ptype = RTE_PTYPE_TUNNEL_IP,
814 	},
815 	{
816 		.tunnel = MLX5_FLOW_LAYER_GTP,
817 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
818 	},
819 };
820 
821 
822 
823 /**
824  * Translate tag ID to register.
825  *
826  * @param[in] dev
827  *   Pointer to the Ethernet device structure.
828  * @param[in] feature
829  *   The feature that request the register.
830  * @param[in] id
831  *   The request register ID.
832  * @param[out] error
833  *   Error description in case of any.
834  *
835  * @return
836  *   The request register on success, a negative errno
837  *   value otherwise and rte_errno is set.
838  */
839 int
840 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
841 		     enum mlx5_feature_name feature,
842 		     uint32_t id,
843 		     struct rte_flow_error *error)
844 {
845 	struct mlx5_priv *priv = dev->data->dev_private;
846 	struct mlx5_dev_config *config = &priv->config;
847 	enum modify_reg start_reg;
848 	bool skip_mtr_reg = false;
849 
850 	switch (feature) {
851 	case MLX5_HAIRPIN_RX:
852 		return REG_B;
853 	case MLX5_HAIRPIN_TX:
854 		return REG_A;
855 	case MLX5_METADATA_RX:
856 		switch (config->dv_xmeta_en) {
857 		case MLX5_XMETA_MODE_LEGACY:
858 			return REG_B;
859 		case MLX5_XMETA_MODE_META16:
860 			return REG_C_0;
861 		case MLX5_XMETA_MODE_META32:
862 			return REG_C_1;
863 		}
864 		break;
865 	case MLX5_METADATA_TX:
866 		return REG_A;
867 	case MLX5_METADATA_FDB:
868 		switch (config->dv_xmeta_en) {
869 		case MLX5_XMETA_MODE_LEGACY:
870 			return REG_NON;
871 		case MLX5_XMETA_MODE_META16:
872 			return REG_C_0;
873 		case MLX5_XMETA_MODE_META32:
874 			return REG_C_1;
875 		}
876 		break;
877 	case MLX5_FLOW_MARK:
878 		switch (config->dv_xmeta_en) {
879 		case MLX5_XMETA_MODE_LEGACY:
880 			return REG_NON;
881 		case MLX5_XMETA_MODE_META16:
882 			return REG_C_1;
883 		case MLX5_XMETA_MODE_META32:
884 			return REG_C_0;
885 		}
886 		break;
887 	case MLX5_MTR_ID:
888 		/*
889 		 * If meter color and meter id share one register, flow match
890 		 * should use the meter color register for match.
891 		 */
892 		if (priv->mtr_reg_share)
893 			return priv->mtr_color_reg;
894 		else
895 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
896 			       REG_C_3;
897 	case MLX5_MTR_COLOR:
898 	case MLX5_ASO_FLOW_HIT:
899 	case MLX5_ASO_CONNTRACK:
900 		/* All features use the same REG_C. */
901 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
902 		return priv->mtr_color_reg;
903 	case MLX5_COPY_MARK:
904 		/*
905 		 * Metadata COPY_MARK register using is in meter suffix sub
906 		 * flow while with meter. It's safe to share the same register.
907 		 */
908 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
909 	case MLX5_APP_TAG:
910 		/*
911 		 * If meter is enable, it will engage the register for color
912 		 * match and flow match. If meter color match is not using the
913 		 * REG_C_2, need to skip the REG_C_x be used by meter color
914 		 * match.
915 		 * If meter is disable, free to use all available registers.
916 		 */
917 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
918 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
919 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
920 		if (id > (uint32_t)(REG_C_7 - start_reg))
921 			return rte_flow_error_set(error, EINVAL,
922 						  RTE_FLOW_ERROR_TYPE_ITEM,
923 						  NULL, "invalid tag id");
924 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
925 			return rte_flow_error_set(error, ENOTSUP,
926 						  RTE_FLOW_ERROR_TYPE_ITEM,
927 						  NULL, "unsupported tag id");
928 		/*
929 		 * This case means meter is using the REG_C_x great than 2.
930 		 * Take care not to conflict with meter color REG_C_x.
931 		 * If the available index REG_C_y >= REG_C_x, skip the
932 		 * color register.
933 		 */
934 		if (skip_mtr_reg && priv->sh->flow_mreg_c
935 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
936 			if (id >= (uint32_t)(REG_C_7 - start_reg))
937 				return rte_flow_error_set(error, EINVAL,
938 						       RTE_FLOW_ERROR_TYPE_ITEM,
939 							NULL, "invalid tag id");
940 			if (priv->sh->flow_mreg_c
941 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
942 				return priv->sh->flow_mreg_c
943 					       [id + 1 + start_reg - REG_C_0];
944 			return rte_flow_error_set(error, ENOTSUP,
945 						  RTE_FLOW_ERROR_TYPE_ITEM,
946 						  NULL, "unsupported tag id");
947 		}
948 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
949 	}
950 	MLX5_ASSERT(false);
951 	return rte_flow_error_set(error, EINVAL,
952 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
953 				  NULL, "invalid feature name");
954 }
955 
956 /**
957  * Check extensive flow metadata register support.
958  *
959  * @param dev
960  *   Pointer to rte_eth_dev structure.
961  *
962  * @return
963  *   True if device supports extensive flow metadata register, otherwise false.
964  */
965 bool
966 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
967 {
968 	struct mlx5_priv *priv = dev->data->dev_private;
969 
970 	/*
971 	 * Having available reg_c can be regarded inclusively as supporting
972 	 * extensive flow metadata register, which could mean,
973 	 * - metadata register copy action by modify header.
974 	 * - 16 modify header actions is supported.
975 	 * - reg_c's are preserved across different domain (FDB and NIC) on
976 	 *   packet loopback by flow lookup miss.
977 	 */
978 	return priv->sh->flow_mreg_c[2] != REG_NON;
979 }
980 
981 /**
982  * Get the lowest priority.
983  *
984  * @param[in] dev
985  *   Pointer to the Ethernet device structure.
986  * @param[in] attributes
987  *   Pointer to device flow rule attributes.
988  *
989  * @return
990  *   The value of lowest priority of flow.
991  */
992 uint32_t
993 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
994 			  const struct rte_flow_attr *attr)
995 {
996 	struct mlx5_priv *priv = dev->data->dev_private;
997 
998 	if (!attr->group && !attr->transfer)
999 		return priv->sh->flow_max_priority - 2;
1000 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1001 }
1002 
1003 /**
1004  * Calculate matcher priority of the flow.
1005  *
1006  * @param[in] dev
1007  *   Pointer to the Ethernet device structure.
1008  * @param[in] attr
1009  *   Pointer to device flow rule attributes.
1010  * @param[in] subpriority
1011  *   The priority based on the items.
1012  * @param[in] external
1013  *   Flow is user flow.
1014  * @return
1015  *   The matcher priority of the flow.
1016  */
1017 uint16_t
1018 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1019 			  const struct rte_flow_attr *attr,
1020 			  uint32_t subpriority, bool external)
1021 {
1022 	uint16_t priority = (uint16_t)attr->priority;
1023 	struct mlx5_priv *priv = dev->data->dev_private;
1024 
1025 	if (!attr->group && !attr->transfer) {
1026 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1027 			priority = priv->sh->flow_max_priority - 1;
1028 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1029 	} else if (!external && attr->transfer && attr->group == 0 &&
1030 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1031 		return (priv->sh->flow_max_priority - 1) * 3;
1032 	}
1033 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1034 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1035 	return priority * 3 + subpriority;
1036 }
1037 
1038 /**
1039  * Verify the @p item specifications (spec, last, mask) are compatible with the
1040  * NIC capabilities.
1041  *
1042  * @param[in] item
1043  *   Item specification.
1044  * @param[in] mask
1045  *   @p item->mask or flow default bit-masks.
1046  * @param[in] nic_mask
1047  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1048  * @param[in] size
1049  *   Bit-masks size in bytes.
1050  * @param[in] range_accepted
1051  *   True if range of values is accepted for specific fields, false otherwise.
1052  * @param[out] error
1053  *   Pointer to error structure.
1054  *
1055  * @return
1056  *   0 on success, a negative errno value otherwise and rte_errno is set.
1057  */
1058 int
1059 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1060 			  const uint8_t *mask,
1061 			  const uint8_t *nic_mask,
1062 			  unsigned int size,
1063 			  bool range_accepted,
1064 			  struct rte_flow_error *error)
1065 {
1066 	unsigned int i;
1067 
1068 	MLX5_ASSERT(nic_mask);
1069 	for (i = 0; i < size; ++i)
1070 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1071 			return rte_flow_error_set(error, ENOTSUP,
1072 						  RTE_FLOW_ERROR_TYPE_ITEM,
1073 						  item,
1074 						  "mask enables non supported"
1075 						  " bits");
1076 	if (!item->spec && (item->mask || item->last))
1077 		return rte_flow_error_set(error, EINVAL,
1078 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1079 					  "mask/last without a spec is not"
1080 					  " supported");
1081 	if (item->spec && item->last && !range_accepted) {
1082 		uint8_t spec[size];
1083 		uint8_t last[size];
1084 		unsigned int i;
1085 		int ret;
1086 
1087 		for (i = 0; i < size; ++i) {
1088 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1089 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1090 		}
1091 		ret = memcmp(spec, last, size);
1092 		if (ret != 0)
1093 			return rte_flow_error_set(error, EINVAL,
1094 						  RTE_FLOW_ERROR_TYPE_ITEM,
1095 						  item,
1096 						  "range is not valid");
1097 	}
1098 	return 0;
1099 }
1100 
1101 /**
1102  * Adjust the hash fields according to the @p flow information.
1103  *
1104  * @param[in] dev_flow.
1105  *   Pointer to the mlx5_flow.
1106  * @param[in] tunnel
1107  *   1 when the hash field is for a tunnel item.
1108  * @param[in] layer_types
1109  *   RTE_ETH_RSS_* types.
1110  * @param[in] hash_fields
1111  *   Item hash fields.
1112  *
1113  * @return
1114  *   The hash fields that should be used.
1115  */
1116 uint64_t
1117 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1118 			    int tunnel __rte_unused, uint64_t layer_types,
1119 			    uint64_t hash_fields)
1120 {
1121 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1122 	int rss_request_inner = rss_desc->level >= 2;
1123 
1124 	/* Check RSS hash level for tunnel. */
1125 	if (tunnel && rss_request_inner)
1126 		hash_fields |= IBV_RX_HASH_INNER;
1127 	else if (tunnel || rss_request_inner)
1128 		return 0;
1129 #endif
1130 	/* Check if requested layer matches RSS hash fields. */
1131 	if (!(rss_desc->types & layer_types))
1132 		return 0;
1133 	return hash_fields;
1134 }
1135 
1136 /**
1137  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1138  * if several tunnel rules are used on this queue, the tunnel ptype will be
1139  * cleared.
1140  *
1141  * @param rxq_ctrl
1142  *   Rx queue to update.
1143  */
1144 static void
1145 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1146 {
1147 	unsigned int i;
1148 	uint32_t tunnel_ptype = 0;
1149 
1150 	/* Look up for the ptype to use. */
1151 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1152 		if (!rxq_ctrl->flow_tunnels_n[i])
1153 			continue;
1154 		if (!tunnel_ptype) {
1155 			tunnel_ptype = tunnels_info[i].ptype;
1156 		} else {
1157 			tunnel_ptype = 0;
1158 			break;
1159 		}
1160 	}
1161 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1162 }
1163 
1164 /**
1165  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1166  * flow.
1167  *
1168  * @param[in] dev
1169  *   Pointer to the Ethernet device structure.
1170  * @param[in] dev_handle
1171  *   Pointer to device flow handle structure.
1172  */
1173 void
1174 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1175 		       struct mlx5_flow_handle *dev_handle)
1176 {
1177 	struct mlx5_priv *priv = dev->data->dev_private;
1178 	const int mark = dev_handle->mark;
1179 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1180 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1181 	unsigned int i;
1182 
1183 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1184 		struct mlx5_hrxq *hrxq;
1185 
1186 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1187 			      dev_handle->rix_hrxq);
1188 		if (hrxq)
1189 			ind_tbl = hrxq->ind_table;
1190 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1191 		struct mlx5_shared_action_rss *shared_rss;
1192 
1193 		shared_rss = mlx5_ipool_get
1194 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1195 			 dev_handle->rix_srss);
1196 		if (shared_rss)
1197 			ind_tbl = shared_rss->ind_tbl;
1198 	}
1199 	if (!ind_tbl)
1200 		return;
1201 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1202 		int idx = ind_tbl->queues[i];
1203 		struct mlx5_rxq_ctrl *rxq_ctrl =
1204 			container_of((*priv->rxqs)[idx],
1205 				     struct mlx5_rxq_ctrl, rxq);
1206 
1207 		/*
1208 		 * To support metadata register copy on Tx loopback,
1209 		 * this must be always enabled (metadata may arive
1210 		 * from other port - not from local flows only.
1211 		 */
1212 		if (priv->config.dv_flow_en &&
1213 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1214 		    mlx5_flow_ext_mreg_supported(dev)) {
1215 			rxq_ctrl->rxq.mark = 1;
1216 			rxq_ctrl->flow_mark_n = 1;
1217 		} else if (mark) {
1218 			rxq_ctrl->rxq.mark = 1;
1219 			rxq_ctrl->flow_mark_n++;
1220 		}
1221 		if (tunnel) {
1222 			unsigned int j;
1223 
1224 			/* Increase the counter matching the flow. */
1225 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1226 				if ((tunnels_info[j].tunnel &
1227 				     dev_handle->layers) ==
1228 				    tunnels_info[j].tunnel) {
1229 					rxq_ctrl->flow_tunnels_n[j]++;
1230 					break;
1231 				}
1232 			}
1233 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1234 		}
1235 	}
1236 }
1237 
1238 /**
1239  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1240  *
1241  * @param[in] dev
1242  *   Pointer to the Ethernet device structure.
1243  * @param[in] flow
1244  *   Pointer to flow structure.
1245  */
1246 static void
1247 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1248 {
1249 	struct mlx5_priv *priv = dev->data->dev_private;
1250 	uint32_t handle_idx;
1251 	struct mlx5_flow_handle *dev_handle;
1252 
1253 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1254 		       handle_idx, dev_handle, next)
1255 		flow_drv_rxq_flags_set(dev, dev_handle);
1256 }
1257 
1258 /**
1259  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1260  * device flow if no other flow uses it with the same kind of request.
1261  *
1262  * @param dev
1263  *   Pointer to Ethernet device.
1264  * @param[in] dev_handle
1265  *   Pointer to the device flow handle structure.
1266  */
1267 static void
1268 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1269 			struct mlx5_flow_handle *dev_handle)
1270 {
1271 	struct mlx5_priv *priv = dev->data->dev_private;
1272 	const int mark = dev_handle->mark;
1273 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1274 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1275 	unsigned int i;
1276 
1277 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1278 		struct mlx5_hrxq *hrxq;
1279 
1280 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1281 			      dev_handle->rix_hrxq);
1282 		if (hrxq)
1283 			ind_tbl = hrxq->ind_table;
1284 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1285 		struct mlx5_shared_action_rss *shared_rss;
1286 
1287 		shared_rss = mlx5_ipool_get
1288 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1289 			 dev_handle->rix_srss);
1290 		if (shared_rss)
1291 			ind_tbl = shared_rss->ind_tbl;
1292 	}
1293 	if (!ind_tbl)
1294 		return;
1295 	MLX5_ASSERT(dev->data->dev_started);
1296 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1297 		int idx = ind_tbl->queues[i];
1298 		struct mlx5_rxq_ctrl *rxq_ctrl =
1299 			container_of((*priv->rxqs)[idx],
1300 				     struct mlx5_rxq_ctrl, rxq);
1301 
1302 		if (priv->config.dv_flow_en &&
1303 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1304 		    mlx5_flow_ext_mreg_supported(dev)) {
1305 			rxq_ctrl->rxq.mark = 1;
1306 			rxq_ctrl->flow_mark_n = 1;
1307 		} else if (mark) {
1308 			rxq_ctrl->flow_mark_n--;
1309 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1310 		}
1311 		if (tunnel) {
1312 			unsigned int j;
1313 
1314 			/* Decrease the counter matching the flow. */
1315 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1316 				if ((tunnels_info[j].tunnel &
1317 				     dev_handle->layers) ==
1318 				    tunnels_info[j].tunnel) {
1319 					rxq_ctrl->flow_tunnels_n[j]--;
1320 					break;
1321 				}
1322 			}
1323 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1324 		}
1325 	}
1326 }
1327 
1328 /**
1329  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1330  * @p flow if no other flow uses it with the same kind of request.
1331  *
1332  * @param dev
1333  *   Pointer to Ethernet device.
1334  * @param[in] flow
1335  *   Pointer to the flow.
1336  */
1337 static void
1338 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1339 {
1340 	struct mlx5_priv *priv = dev->data->dev_private;
1341 	uint32_t handle_idx;
1342 	struct mlx5_flow_handle *dev_handle;
1343 
1344 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1345 		       handle_idx, dev_handle, next)
1346 		flow_drv_rxq_flags_trim(dev, dev_handle);
1347 }
1348 
1349 /**
1350  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1351  *
1352  * @param dev
1353  *   Pointer to Ethernet device.
1354  */
1355 static void
1356 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1357 {
1358 	struct mlx5_priv *priv = dev->data->dev_private;
1359 	unsigned int i;
1360 
1361 	for (i = 0; i != priv->rxqs_n; ++i) {
1362 		struct mlx5_rxq_ctrl *rxq_ctrl;
1363 		unsigned int j;
1364 
1365 		if (!(*priv->rxqs)[i])
1366 			continue;
1367 		rxq_ctrl = container_of((*priv->rxqs)[i],
1368 					struct mlx5_rxq_ctrl, rxq);
1369 		rxq_ctrl->flow_mark_n = 0;
1370 		rxq_ctrl->rxq.mark = 0;
1371 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1372 			rxq_ctrl->flow_tunnels_n[j] = 0;
1373 		rxq_ctrl->rxq.tunnel = 0;
1374 	}
1375 }
1376 
1377 /**
1378  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1379  *
1380  * @param[in] dev
1381  *   Pointer to the Ethernet device structure.
1382  */
1383 void
1384 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1385 {
1386 	struct mlx5_priv *priv = dev->data->dev_private;
1387 	struct mlx5_rxq_data *data;
1388 	unsigned int i;
1389 
1390 	for (i = 0; i != priv->rxqs_n; ++i) {
1391 		if (!(*priv->rxqs)[i])
1392 			continue;
1393 		data = (*priv->rxqs)[i];
1394 		if (!rte_flow_dynf_metadata_avail()) {
1395 			data->dynf_meta = 0;
1396 			data->flow_meta_mask = 0;
1397 			data->flow_meta_offset = -1;
1398 			data->flow_meta_port_mask = 0;
1399 		} else {
1400 			data->dynf_meta = 1;
1401 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1402 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1403 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1404 		}
1405 	}
1406 }
1407 
1408 /*
1409  * return a pointer to the desired action in the list of actions.
1410  *
1411  * @param[in] actions
1412  *   The list of actions to search the action in.
1413  * @param[in] action
1414  *   The action to find.
1415  *
1416  * @return
1417  *   Pointer to the action in the list, if found. NULL otherwise.
1418  */
1419 const struct rte_flow_action *
1420 mlx5_flow_find_action(const struct rte_flow_action *actions,
1421 		      enum rte_flow_action_type action)
1422 {
1423 	if (actions == NULL)
1424 		return NULL;
1425 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1426 		if (actions->type == action)
1427 			return actions;
1428 	return NULL;
1429 }
1430 
1431 /*
1432  * Validate the flag action.
1433  *
1434  * @param[in] action_flags
1435  *   Bit-fields that holds the actions detected until now.
1436  * @param[in] attr
1437  *   Attributes of flow that includes this action.
1438  * @param[out] error
1439  *   Pointer to error structure.
1440  *
1441  * @return
1442  *   0 on success, a negative errno value otherwise and rte_errno is set.
1443  */
1444 int
1445 mlx5_flow_validate_action_flag(uint64_t action_flags,
1446 			       const struct rte_flow_attr *attr,
1447 			       struct rte_flow_error *error)
1448 {
1449 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1450 		return rte_flow_error_set(error, EINVAL,
1451 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1452 					  "can't mark and flag in same flow");
1453 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1454 		return rte_flow_error_set(error, EINVAL,
1455 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1456 					  "can't have 2 flag"
1457 					  " actions in same flow");
1458 	if (attr->egress)
1459 		return rte_flow_error_set(error, ENOTSUP,
1460 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1461 					  "flag action not supported for "
1462 					  "egress");
1463 	return 0;
1464 }
1465 
1466 /*
1467  * Validate the mark action.
1468  *
1469  * @param[in] action
1470  *   Pointer to the queue action.
1471  * @param[in] action_flags
1472  *   Bit-fields that holds the actions detected until now.
1473  * @param[in] attr
1474  *   Attributes of flow that includes this action.
1475  * @param[out] error
1476  *   Pointer to error structure.
1477  *
1478  * @return
1479  *   0 on success, a negative errno value otherwise and rte_errno is set.
1480  */
1481 int
1482 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1483 			       uint64_t action_flags,
1484 			       const struct rte_flow_attr *attr,
1485 			       struct rte_flow_error *error)
1486 {
1487 	const struct rte_flow_action_mark *mark = action->conf;
1488 
1489 	if (!mark)
1490 		return rte_flow_error_set(error, EINVAL,
1491 					  RTE_FLOW_ERROR_TYPE_ACTION,
1492 					  action,
1493 					  "configuration cannot be null");
1494 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1495 		return rte_flow_error_set(error, EINVAL,
1496 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1497 					  &mark->id,
1498 					  "mark id must in 0 <= id < "
1499 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1500 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1501 		return rte_flow_error_set(error, EINVAL,
1502 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1503 					  "can't flag and mark in same flow");
1504 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1505 		return rte_flow_error_set(error, EINVAL,
1506 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1507 					  "can't have 2 mark actions in same"
1508 					  " flow");
1509 	if (attr->egress)
1510 		return rte_flow_error_set(error, ENOTSUP,
1511 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1512 					  "mark action not supported for "
1513 					  "egress");
1514 	return 0;
1515 }
1516 
1517 /*
1518  * Validate the drop action.
1519  *
1520  * @param[in] action_flags
1521  *   Bit-fields that holds the actions detected until now.
1522  * @param[in] attr
1523  *   Attributes of flow that includes this action.
1524  * @param[out] error
1525  *   Pointer to error structure.
1526  *
1527  * @return
1528  *   0 on success, a negative errno value otherwise and rte_errno is set.
1529  */
1530 int
1531 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1532 			       const struct rte_flow_attr *attr,
1533 			       struct rte_flow_error *error)
1534 {
1535 	if (attr->egress)
1536 		return rte_flow_error_set(error, ENOTSUP,
1537 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1538 					  "drop action not supported for "
1539 					  "egress");
1540 	return 0;
1541 }
1542 
1543 /*
1544  * Validate the queue action.
1545  *
1546  * @param[in] action
1547  *   Pointer to the queue action.
1548  * @param[in] action_flags
1549  *   Bit-fields that holds the actions detected until now.
1550  * @param[in] dev
1551  *   Pointer to the Ethernet device structure.
1552  * @param[in] attr
1553  *   Attributes of flow that includes this action.
1554  * @param[out] error
1555  *   Pointer to error structure.
1556  *
1557  * @return
1558  *   0 on success, a negative errno value otherwise and rte_errno is set.
1559  */
1560 int
1561 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1562 				uint64_t action_flags,
1563 				struct rte_eth_dev *dev,
1564 				const struct rte_flow_attr *attr,
1565 				struct rte_flow_error *error)
1566 {
1567 	struct mlx5_priv *priv = dev->data->dev_private;
1568 	const struct rte_flow_action_queue *queue = action->conf;
1569 
1570 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1571 		return rte_flow_error_set(error, EINVAL,
1572 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1573 					  "can't have 2 fate actions in"
1574 					  " same flow");
1575 	if (!priv->rxqs_n)
1576 		return rte_flow_error_set(error, EINVAL,
1577 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1578 					  NULL, "No Rx queues configured");
1579 	if (queue->index >= priv->rxqs_n)
1580 		return rte_flow_error_set(error, EINVAL,
1581 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1582 					  &queue->index,
1583 					  "queue index out of range");
1584 	if (!(*priv->rxqs)[queue->index])
1585 		return rte_flow_error_set(error, EINVAL,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1587 					  &queue->index,
1588 					  "queue is not configured");
1589 	if (attr->egress)
1590 		return rte_flow_error_set(error, ENOTSUP,
1591 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1592 					  "queue action not supported for "
1593 					  "egress");
1594 	return 0;
1595 }
1596 
1597 /**
1598  * Validate queue numbers for device RSS.
1599  *
1600  * @param[in] dev
1601  *   Configured device.
1602  * @param[in] queues
1603  *   Array of queue numbers.
1604  * @param[in] queues_n
1605  *   Size of the @p queues array.
1606  * @param[out] error
1607  *   On error, filled with a textual error description.
1608  * @param[out] queue
1609  *   On error, filled with an offending queue index in @p queues array.
1610  *
1611  * @return
1612  *   0 on success, a negative errno code on error.
1613  */
1614 static int
1615 mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
1616 			 const uint16_t *queues, uint32_t queues_n,
1617 			 const char **error, uint32_t *queue_idx)
1618 {
1619 	const struct mlx5_priv *priv = dev->data->dev_private;
1620 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1621 	uint32_t i;
1622 
1623 	for (i = 0; i != queues_n; ++i) {
1624 		struct mlx5_rxq_ctrl *rxq_ctrl;
1625 
1626 		if (queues[i] >= priv->rxqs_n) {
1627 			*error = "queue index out of range";
1628 			*queue_idx = i;
1629 			return -EINVAL;
1630 		}
1631 		if (!(*priv->rxqs)[queues[i]]) {
1632 			*error =  "queue is not configured";
1633 			*queue_idx = i;
1634 			return -EINVAL;
1635 		}
1636 		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
1637 					struct mlx5_rxq_ctrl, rxq);
1638 		if (i == 0)
1639 			rxq_type = rxq_ctrl->type;
1640 		if (rxq_type != rxq_ctrl->type) {
1641 			*error = "combining hairpin and regular RSS queues is not supported";
1642 			*queue_idx = i;
1643 			return -ENOTSUP;
1644 		}
1645 	}
1646 	return 0;
1647 }
1648 
1649 /*
1650  * Validate the rss action.
1651  *
1652  * @param[in] dev
1653  *   Pointer to the Ethernet device structure.
1654  * @param[in] action
1655  *   Pointer to the queue action.
1656  * @param[out] error
1657  *   Pointer to error structure.
1658  *
1659  * @return
1660  *   0 on success, a negative errno value otherwise and rte_errno is set.
1661  */
1662 int
1663 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1664 			 const struct rte_flow_action *action,
1665 			 struct rte_flow_error *error)
1666 {
1667 	struct mlx5_priv *priv = dev->data->dev_private;
1668 	const struct rte_flow_action_rss *rss = action->conf;
1669 	int ret;
1670 	const char *message;
1671 	uint32_t queue_idx;
1672 
1673 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1674 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1675 		return rte_flow_error_set(error, ENOTSUP,
1676 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1677 					  &rss->func,
1678 					  "RSS hash function not supported");
1679 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1680 	if (rss->level > 2)
1681 #else
1682 	if (rss->level > 1)
1683 #endif
1684 		return rte_flow_error_set(error, ENOTSUP,
1685 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1686 					  &rss->level,
1687 					  "tunnel RSS is not supported");
1688 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1689 	if (rss->key_len == 0 && rss->key != NULL)
1690 		return rte_flow_error_set(error, ENOTSUP,
1691 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1692 					  &rss->key_len,
1693 					  "RSS hash key length 0");
1694 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1695 		return rte_flow_error_set(error, ENOTSUP,
1696 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1697 					  &rss->key_len,
1698 					  "RSS hash key too small");
1699 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1700 		return rte_flow_error_set(error, ENOTSUP,
1701 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1702 					  &rss->key_len,
1703 					  "RSS hash key too large");
1704 	if (rss->queue_num > priv->config.ind_table_max_size)
1705 		return rte_flow_error_set(error, ENOTSUP,
1706 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1707 					  &rss->queue_num,
1708 					  "number of queues too large");
1709 	if (rss->types & MLX5_RSS_HF_MASK)
1710 		return rte_flow_error_set(error, ENOTSUP,
1711 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1712 					  &rss->types,
1713 					  "some RSS protocols are not"
1714 					  " supported");
1715 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1716 	    !(rss->types & RTE_ETH_RSS_IP))
1717 		return rte_flow_error_set(error, EINVAL,
1718 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1719 					  "L3 partial RSS requested but L3 RSS"
1720 					  " type not specified");
1721 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1722 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1723 		return rte_flow_error_set(error, EINVAL,
1724 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1725 					  "L4 partial RSS requested but L4 RSS"
1726 					  " type not specified");
1727 	if (!priv->rxqs_n)
1728 		return rte_flow_error_set(error, EINVAL,
1729 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1730 					  NULL, "No Rx queues configured");
1731 	if (!rss->queue_num)
1732 		return rte_flow_error_set(error, EINVAL,
1733 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1734 					  NULL, "No queues configured");
1735 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1736 				       &message, &queue_idx);
1737 	if (ret != 0) {
1738 		return rte_flow_error_set(error, -ret,
1739 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1740 					  &rss->queue[queue_idx], message);
1741 	}
1742 	return 0;
1743 }
1744 
1745 /*
1746  * Validate the rss action.
1747  *
1748  * @param[in] action
1749  *   Pointer to the queue action.
1750  * @param[in] action_flags
1751  *   Bit-fields that holds the actions detected until now.
1752  * @param[in] dev
1753  *   Pointer to the Ethernet device structure.
1754  * @param[in] attr
1755  *   Attributes of flow that includes this action.
1756  * @param[in] item_flags
1757  *   Items that were detected.
1758  * @param[out] error
1759  *   Pointer to error structure.
1760  *
1761  * @return
1762  *   0 on success, a negative errno value otherwise and rte_errno is set.
1763  */
1764 int
1765 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1766 			      uint64_t action_flags,
1767 			      struct rte_eth_dev *dev,
1768 			      const struct rte_flow_attr *attr,
1769 			      uint64_t item_flags,
1770 			      struct rte_flow_error *error)
1771 {
1772 	const struct rte_flow_action_rss *rss = action->conf;
1773 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1774 	int ret;
1775 
1776 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1777 		return rte_flow_error_set(error, EINVAL,
1778 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1779 					  "can't have 2 fate actions"
1780 					  " in same flow");
1781 	ret = mlx5_validate_action_rss(dev, action, error);
1782 	if (ret)
1783 		return ret;
1784 	if (attr->egress)
1785 		return rte_flow_error_set(error, ENOTSUP,
1786 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1787 					  "rss action not supported for "
1788 					  "egress");
1789 	if (rss->level > 1 && !tunnel)
1790 		return rte_flow_error_set(error, EINVAL,
1791 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1792 					  "inner RSS is not supported for "
1793 					  "non-tunnel flows");
1794 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1795 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1796 		return rte_flow_error_set(error, EINVAL,
1797 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1798 					  "RSS on eCPRI is not supported now");
1799 	}
1800 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1801 	    !(item_flags &
1802 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1803 	    rss->level > 1)
1804 		return rte_flow_error_set(error, EINVAL,
1805 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1806 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1807 	return 0;
1808 }
1809 
1810 /*
1811  * Validate the default miss action.
1812  *
1813  * @param[in] action_flags
1814  *   Bit-fields that holds the actions detected until now.
1815  * @param[out] error
1816  *   Pointer to error structure.
1817  *
1818  * @return
1819  *   0 on success, a negative errno value otherwise and rte_errno is set.
1820  */
1821 int
1822 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1823 				const struct rte_flow_attr *attr,
1824 				struct rte_flow_error *error)
1825 {
1826 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1827 		return rte_flow_error_set(error, EINVAL,
1828 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1829 					  "can't have 2 fate actions in"
1830 					  " same flow");
1831 	if (attr->egress)
1832 		return rte_flow_error_set(error, ENOTSUP,
1833 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1834 					  "default miss action not supported "
1835 					  "for egress");
1836 	if (attr->group)
1837 		return rte_flow_error_set(error, ENOTSUP,
1838 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1839 					  "only group 0 is supported");
1840 	if (attr->transfer)
1841 		return rte_flow_error_set(error, ENOTSUP,
1842 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1843 					  NULL, "transfer is not supported");
1844 	return 0;
1845 }
1846 
1847 /*
1848  * Validate the count action.
1849  *
1850  * @param[in] dev
1851  *   Pointer to the Ethernet device structure.
1852  * @param[in] attr
1853  *   Attributes of flow that includes this action.
1854  * @param[out] error
1855  *   Pointer to error structure.
1856  *
1857  * @return
1858  *   0 on success, a negative errno value otherwise and rte_errno is set.
1859  */
1860 int
1861 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1862 				const struct rte_flow_attr *attr,
1863 				struct rte_flow_error *error)
1864 {
1865 	if (attr->egress)
1866 		return rte_flow_error_set(error, ENOTSUP,
1867 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1868 					  "count action not supported for "
1869 					  "egress");
1870 	return 0;
1871 }
1872 
1873 /*
1874  * Validate the ASO CT action.
1875  *
1876  * @param[in] dev
1877  *   Pointer to the Ethernet device structure.
1878  * @param[in] conntrack
1879  *   Pointer to the CT action profile.
1880  * @param[out] error
1881  *   Pointer to error structure.
1882  *
1883  * @return
1884  *   0 on success, a negative errno value otherwise and rte_errno is set.
1885  */
1886 int
1887 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1888 			const struct rte_flow_action_conntrack *conntrack,
1889 			struct rte_flow_error *error)
1890 {
1891 	RTE_SET_USED(dev);
1892 
1893 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1894 		return rte_flow_error_set(error, EINVAL,
1895 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1896 					  "Invalid CT state");
1897 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1898 		return rte_flow_error_set(error, EINVAL,
1899 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1900 					  "Invalid last TCP packet flag");
1901 	return 0;
1902 }
1903 
1904 /**
1905  * Verify the @p attributes will be correctly understood by the NIC and store
1906  * them in the @p flow if everything is correct.
1907  *
1908  * @param[in] dev
1909  *   Pointer to the Ethernet device structure.
1910  * @param[in] attributes
1911  *   Pointer to flow attributes
1912  * @param[out] error
1913  *   Pointer to error structure.
1914  *
1915  * @return
1916  *   0 on success, a negative errno value otherwise and rte_errno is set.
1917  */
1918 int
1919 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1920 			      const struct rte_flow_attr *attributes,
1921 			      struct rte_flow_error *error)
1922 {
1923 	struct mlx5_priv *priv = dev->data->dev_private;
1924 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1925 
1926 	if (attributes->group)
1927 		return rte_flow_error_set(error, ENOTSUP,
1928 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1929 					  NULL, "groups is not supported");
1930 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1931 	    attributes->priority >= priority_max)
1932 		return rte_flow_error_set(error, ENOTSUP,
1933 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1934 					  NULL, "priority out of range");
1935 	if (attributes->egress)
1936 		return rte_flow_error_set(error, ENOTSUP,
1937 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1938 					  "egress is not supported");
1939 	if (attributes->transfer && !priv->config.dv_esw_en)
1940 		return rte_flow_error_set(error, ENOTSUP,
1941 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1942 					  NULL, "transfer is not supported");
1943 	if (!attributes->ingress)
1944 		return rte_flow_error_set(error, EINVAL,
1945 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1946 					  NULL,
1947 					  "ingress attribute is mandatory");
1948 	return 0;
1949 }
1950 
1951 /**
1952  * Validate ICMP6 item.
1953  *
1954  * @param[in] item
1955  *   Item specification.
1956  * @param[in] item_flags
1957  *   Bit-fields that holds the items detected until now.
1958  * @param[in] ext_vlan_sup
1959  *   Whether extended VLAN features are supported or not.
1960  * @param[out] error
1961  *   Pointer to error structure.
1962  *
1963  * @return
1964  *   0 on success, a negative errno value otherwise and rte_errno is set.
1965  */
1966 int
1967 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1968 			       uint64_t item_flags,
1969 			       uint8_t target_protocol,
1970 			       struct rte_flow_error *error)
1971 {
1972 	const struct rte_flow_item_icmp6 *mask = item->mask;
1973 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1974 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1975 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1976 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1977 				      MLX5_FLOW_LAYER_OUTER_L4;
1978 	int ret;
1979 
1980 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1981 		return rte_flow_error_set(error, EINVAL,
1982 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1983 					  "protocol filtering not compatible"
1984 					  " with ICMP6 layer");
1985 	if (!(item_flags & l3m))
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1988 					  "IPv6 is mandatory to filter on"
1989 					  " ICMP6");
1990 	if (item_flags & l4m)
1991 		return rte_flow_error_set(error, EINVAL,
1992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 					  "multiple L4 layers not supported");
1994 	if (!mask)
1995 		mask = &rte_flow_item_icmp6_mask;
1996 	ret = mlx5_flow_item_acceptable
1997 		(item, (const uint8_t *)mask,
1998 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1999 		 sizeof(struct rte_flow_item_icmp6),
2000 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2001 	if (ret < 0)
2002 		return ret;
2003 	return 0;
2004 }
2005 
2006 /**
2007  * Validate ICMP item.
2008  *
2009  * @param[in] item
2010  *   Item specification.
2011  * @param[in] item_flags
2012  *   Bit-fields that holds the items detected until now.
2013  * @param[out] error
2014  *   Pointer to error structure.
2015  *
2016  * @return
2017  *   0 on success, a negative errno value otherwise and rte_errno is set.
2018  */
2019 int
2020 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2021 			     uint64_t item_flags,
2022 			     uint8_t target_protocol,
2023 			     struct rte_flow_error *error)
2024 {
2025 	const struct rte_flow_item_icmp *mask = item->mask;
2026 	const struct rte_flow_item_icmp nic_mask = {
2027 		.hdr.icmp_type = 0xff,
2028 		.hdr.icmp_code = 0xff,
2029 		.hdr.icmp_ident = RTE_BE16(0xffff),
2030 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2031 	};
2032 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2034 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2035 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2036 				      MLX5_FLOW_LAYER_OUTER_L4;
2037 	int ret;
2038 
2039 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2040 		return rte_flow_error_set(error, EINVAL,
2041 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2042 					  "protocol filtering not compatible"
2043 					  " with ICMP layer");
2044 	if (!(item_flags & l3m))
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "IPv4 is mandatory to filter"
2048 					  " on ICMP");
2049 	if (item_flags & l4m)
2050 		return rte_flow_error_set(error, EINVAL,
2051 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 					  "multiple L4 layers not supported");
2053 	if (!mask)
2054 		mask = &nic_mask;
2055 	ret = mlx5_flow_item_acceptable
2056 		(item, (const uint8_t *)mask,
2057 		 (const uint8_t *)&nic_mask,
2058 		 sizeof(struct rte_flow_item_icmp),
2059 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2060 	if (ret < 0)
2061 		return ret;
2062 	return 0;
2063 }
2064 
2065 /**
2066  * Validate Ethernet item.
2067  *
2068  * @param[in] item
2069  *   Item specification.
2070  * @param[in] item_flags
2071  *   Bit-fields that holds the items detected until now.
2072  * @param[out] error
2073  *   Pointer to error structure.
2074  *
2075  * @return
2076  *   0 on success, a negative errno value otherwise and rte_errno is set.
2077  */
2078 int
2079 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2080 			    uint64_t item_flags, bool ext_vlan_sup,
2081 			    struct rte_flow_error *error)
2082 {
2083 	const struct rte_flow_item_eth *mask = item->mask;
2084 	const struct rte_flow_item_eth nic_mask = {
2085 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2086 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2087 		.type = RTE_BE16(0xffff),
2088 		.has_vlan = ext_vlan_sup ? 1 : 0,
2089 	};
2090 	int ret;
2091 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2092 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2093 				       MLX5_FLOW_LAYER_OUTER_L2;
2094 
2095 	if (item_flags & ethm)
2096 		return rte_flow_error_set(error, ENOTSUP,
2097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 					  "multiple L2 layers not supported");
2099 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2100 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2101 		return rte_flow_error_set(error, EINVAL,
2102 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2103 					  "L2 layer should not follow "
2104 					  "L3 layers");
2105 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2106 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2107 		return rte_flow_error_set(error, EINVAL,
2108 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2109 					  "L2 layer should not follow VLAN");
2110 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2111 		return rte_flow_error_set(error, EINVAL,
2112 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2113 					  "L2 layer should not follow GTP");
2114 	if (!mask)
2115 		mask = &rte_flow_item_eth_mask;
2116 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2117 					(const uint8_t *)&nic_mask,
2118 					sizeof(struct rte_flow_item_eth),
2119 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2120 	return ret;
2121 }
2122 
2123 /**
2124  * Validate VLAN item.
2125  *
2126  * @param[in] item
2127  *   Item specification.
2128  * @param[in] item_flags
2129  *   Bit-fields that holds the items detected until now.
2130  * @param[in] dev
2131  *   Ethernet device flow is being created on.
2132  * @param[out] error
2133  *   Pointer to error structure.
2134  *
2135  * @return
2136  *   0 on success, a negative errno value otherwise and rte_errno is set.
2137  */
2138 int
2139 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2140 			     uint64_t item_flags,
2141 			     struct rte_eth_dev *dev,
2142 			     struct rte_flow_error *error)
2143 {
2144 	const struct rte_flow_item_vlan *spec = item->spec;
2145 	const struct rte_flow_item_vlan *mask = item->mask;
2146 	const struct rte_flow_item_vlan nic_mask = {
2147 		.tci = RTE_BE16(UINT16_MAX),
2148 		.inner_type = RTE_BE16(UINT16_MAX),
2149 	};
2150 	uint16_t vlan_tag = 0;
2151 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2152 	int ret;
2153 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2154 					MLX5_FLOW_LAYER_INNER_L4) :
2155 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2156 					MLX5_FLOW_LAYER_OUTER_L4);
2157 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2158 					MLX5_FLOW_LAYER_OUTER_VLAN;
2159 
2160 	if (item_flags & vlanm)
2161 		return rte_flow_error_set(error, EINVAL,
2162 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2163 					  "multiple VLAN layers not supported");
2164 	else if ((item_flags & l34m) != 0)
2165 		return rte_flow_error_set(error, EINVAL,
2166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 					  "VLAN cannot follow L3/L4 layer");
2168 	if (!mask)
2169 		mask = &rte_flow_item_vlan_mask;
2170 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2171 					(const uint8_t *)&nic_mask,
2172 					sizeof(struct rte_flow_item_vlan),
2173 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2174 	if (ret)
2175 		return ret;
2176 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2177 		struct mlx5_priv *priv = dev->data->dev_private;
2178 
2179 		if (priv->vmwa_context) {
2180 			/*
2181 			 * Non-NULL context means we have a virtual machine
2182 			 * and SR-IOV enabled, we have to create VLAN interface
2183 			 * to make hypervisor to setup E-Switch vport
2184 			 * context correctly. We avoid creating the multiple
2185 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2186 			 */
2187 			return rte_flow_error_set(error, EINVAL,
2188 						  RTE_FLOW_ERROR_TYPE_ITEM,
2189 						  item,
2190 						  "VLAN tag mask is not"
2191 						  " supported in virtual"
2192 						  " environment");
2193 		}
2194 	}
2195 	if (spec) {
2196 		vlan_tag = spec->tci;
2197 		vlan_tag &= mask->tci;
2198 	}
2199 	/*
2200 	 * From verbs perspective an empty VLAN is equivalent
2201 	 * to a packet without VLAN layer.
2202 	 */
2203 	if (!vlan_tag)
2204 		return rte_flow_error_set(error, EINVAL,
2205 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2206 					  item->spec,
2207 					  "VLAN cannot be empty");
2208 	return 0;
2209 }
2210 
2211 /**
2212  * Validate IPV4 item.
2213  *
2214  * @param[in] item
2215  *   Item specification.
2216  * @param[in] item_flags
2217  *   Bit-fields that holds the items detected until now.
2218  * @param[in] last_item
2219  *   Previous validated item in the pattern items.
2220  * @param[in] ether_type
2221  *   Type in the ethernet layer header (including dot1q).
2222  * @param[in] acc_mask
2223  *   Acceptable mask, if NULL default internal default mask
2224  *   will be used to check whether item fields are supported.
2225  * @param[in] range_accepted
2226  *   True if range of values is accepted for specific fields, false otherwise.
2227  * @param[out] error
2228  *   Pointer to error structure.
2229  *
2230  * @return
2231  *   0 on success, a negative errno value otherwise and rte_errno is set.
2232  */
2233 int
2234 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2235 			     uint64_t item_flags,
2236 			     uint64_t last_item,
2237 			     uint16_t ether_type,
2238 			     const struct rte_flow_item_ipv4 *acc_mask,
2239 			     bool range_accepted,
2240 			     struct rte_flow_error *error)
2241 {
2242 	const struct rte_flow_item_ipv4 *mask = item->mask;
2243 	const struct rte_flow_item_ipv4 *spec = item->spec;
2244 	const struct rte_flow_item_ipv4 nic_mask = {
2245 		.hdr = {
2246 			.src_addr = RTE_BE32(0xffffffff),
2247 			.dst_addr = RTE_BE32(0xffffffff),
2248 			.type_of_service = 0xff,
2249 			.next_proto_id = 0xff,
2250 		},
2251 	};
2252 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2253 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2254 				      MLX5_FLOW_LAYER_OUTER_L3;
2255 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2256 				      MLX5_FLOW_LAYER_OUTER_L4;
2257 	int ret;
2258 	uint8_t next_proto = 0xFF;
2259 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2260 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2261 				  MLX5_FLOW_LAYER_INNER_VLAN);
2262 
2263 	if ((last_item & l2_vlan) && ether_type &&
2264 	    ether_type != RTE_ETHER_TYPE_IPV4)
2265 		return rte_flow_error_set(error, EINVAL,
2266 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2267 					  "IPv4 cannot follow L2/VLAN layer "
2268 					  "which ether type is not IPv4");
2269 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2270 		if (mask && spec)
2271 			next_proto = mask->hdr.next_proto_id &
2272 				     spec->hdr.next_proto_id;
2273 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2274 			return rte_flow_error_set(error, EINVAL,
2275 						  RTE_FLOW_ERROR_TYPE_ITEM,
2276 						  item,
2277 						  "multiple tunnel "
2278 						  "not supported");
2279 	}
2280 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2281 		return rte_flow_error_set(error, EINVAL,
2282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2283 					  "wrong tunnel type - IPv6 specified "
2284 					  "but IPv4 item provided");
2285 	if (item_flags & l3m)
2286 		return rte_flow_error_set(error, ENOTSUP,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "multiple L3 layers not supported");
2289 	else if (item_flags & l4m)
2290 		return rte_flow_error_set(error, EINVAL,
2291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2292 					  "L3 cannot follow an L4 layer.");
2293 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2294 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2295 		return rte_flow_error_set(error, EINVAL,
2296 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2297 					  "L3 cannot follow an NVGRE layer.");
2298 	if (!mask)
2299 		mask = &rte_flow_item_ipv4_mask;
2300 	else if (mask->hdr.next_proto_id != 0 &&
2301 		 mask->hdr.next_proto_id != 0xff)
2302 		return rte_flow_error_set(error, EINVAL,
2303 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2304 					  "partial mask is not supported"
2305 					  " for protocol");
2306 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2307 					acc_mask ? (const uint8_t *)acc_mask
2308 						 : (const uint8_t *)&nic_mask,
2309 					sizeof(struct rte_flow_item_ipv4),
2310 					range_accepted, error);
2311 	if (ret < 0)
2312 		return ret;
2313 	return 0;
2314 }
2315 
2316 /**
2317  * Validate IPV6 item.
2318  *
2319  * @param[in] item
2320  *   Item specification.
2321  * @param[in] item_flags
2322  *   Bit-fields that holds the items detected until now.
2323  * @param[in] last_item
2324  *   Previous validated item in the pattern items.
2325  * @param[in] ether_type
2326  *   Type in the ethernet layer header (including dot1q).
2327  * @param[in] acc_mask
2328  *   Acceptable mask, if NULL default internal default mask
2329  *   will be used to check whether item fields are supported.
2330  * @param[out] error
2331  *   Pointer to error structure.
2332  *
2333  * @return
2334  *   0 on success, a negative errno value otherwise and rte_errno is set.
2335  */
2336 int
2337 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2338 			     uint64_t item_flags,
2339 			     uint64_t last_item,
2340 			     uint16_t ether_type,
2341 			     const struct rte_flow_item_ipv6 *acc_mask,
2342 			     struct rte_flow_error *error)
2343 {
2344 	const struct rte_flow_item_ipv6 *mask = item->mask;
2345 	const struct rte_flow_item_ipv6 *spec = item->spec;
2346 	const struct rte_flow_item_ipv6 nic_mask = {
2347 		.hdr = {
2348 			.src_addr =
2349 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2350 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2351 			.dst_addr =
2352 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2353 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2354 			.vtc_flow = RTE_BE32(0xffffffff),
2355 			.proto = 0xff,
2356 		},
2357 	};
2358 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2359 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2360 				      MLX5_FLOW_LAYER_OUTER_L3;
2361 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2362 				      MLX5_FLOW_LAYER_OUTER_L4;
2363 	int ret;
2364 	uint8_t next_proto = 0xFF;
2365 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2366 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2367 				  MLX5_FLOW_LAYER_INNER_VLAN);
2368 
2369 	if ((last_item & l2_vlan) && ether_type &&
2370 	    ether_type != RTE_ETHER_TYPE_IPV6)
2371 		return rte_flow_error_set(error, EINVAL,
2372 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2373 					  "IPv6 cannot follow L2/VLAN layer "
2374 					  "which ether type is not IPv6");
2375 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2376 		next_proto = spec->hdr.proto;
2377 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2378 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2379 			return rte_flow_error_set(error, EINVAL,
2380 						  RTE_FLOW_ERROR_TYPE_ITEM,
2381 						  item,
2382 						  "multiple tunnel "
2383 						  "not supported");
2384 	}
2385 	if (next_proto == IPPROTO_HOPOPTS  ||
2386 	    next_proto == IPPROTO_ROUTING  ||
2387 	    next_proto == IPPROTO_FRAGMENT ||
2388 	    next_proto == IPPROTO_ESP	   ||
2389 	    next_proto == IPPROTO_AH	   ||
2390 	    next_proto == IPPROTO_DSTOPTS)
2391 		return rte_flow_error_set(error, EINVAL,
2392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2393 					  "IPv6 proto (next header) should "
2394 					  "not be set as extension header");
2395 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2396 		return rte_flow_error_set(error, EINVAL,
2397 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2398 					  "wrong tunnel type - IPv4 specified "
2399 					  "but IPv6 item provided");
2400 	if (item_flags & l3m)
2401 		return rte_flow_error_set(error, ENOTSUP,
2402 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2403 					  "multiple L3 layers not supported");
2404 	else if (item_flags & l4m)
2405 		return rte_flow_error_set(error, EINVAL,
2406 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2407 					  "L3 cannot follow an L4 layer.");
2408 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2409 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2410 		return rte_flow_error_set(error, EINVAL,
2411 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2412 					  "L3 cannot follow an NVGRE layer.");
2413 	if (!mask)
2414 		mask = &rte_flow_item_ipv6_mask;
2415 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2416 					acc_mask ? (const uint8_t *)acc_mask
2417 						 : (const uint8_t *)&nic_mask,
2418 					sizeof(struct rte_flow_item_ipv6),
2419 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2420 	if (ret < 0)
2421 		return ret;
2422 	return 0;
2423 }
2424 
2425 /**
2426  * Validate UDP item.
2427  *
2428  * @param[in] item
2429  *   Item specification.
2430  * @param[in] item_flags
2431  *   Bit-fields that holds the items detected until now.
2432  * @param[in] target_protocol
2433  *   The next protocol in the previous item.
2434  * @param[in] flow_mask
2435  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2436  * @param[out] error
2437  *   Pointer to error structure.
2438  *
2439  * @return
2440  *   0 on success, a negative errno value otherwise and rte_errno is set.
2441  */
2442 int
2443 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2444 			    uint64_t item_flags,
2445 			    uint8_t target_protocol,
2446 			    struct rte_flow_error *error)
2447 {
2448 	const struct rte_flow_item_udp *mask = item->mask;
2449 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2450 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2451 				      MLX5_FLOW_LAYER_OUTER_L3;
2452 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2453 				      MLX5_FLOW_LAYER_OUTER_L4;
2454 	int ret;
2455 
2456 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2457 		return rte_flow_error_set(error, EINVAL,
2458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2459 					  "protocol filtering not compatible"
2460 					  " with UDP layer");
2461 	if (!(item_flags & l3m))
2462 		return rte_flow_error_set(error, EINVAL,
2463 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2464 					  "L3 is mandatory to filter on L4");
2465 	if (item_flags & l4m)
2466 		return rte_flow_error_set(error, EINVAL,
2467 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2468 					  "multiple L4 layers not supported");
2469 	if (!mask)
2470 		mask = &rte_flow_item_udp_mask;
2471 	ret = mlx5_flow_item_acceptable
2472 		(item, (const uint8_t *)mask,
2473 		 (const uint8_t *)&rte_flow_item_udp_mask,
2474 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2475 		 error);
2476 	if (ret < 0)
2477 		return ret;
2478 	return 0;
2479 }
2480 
2481 /**
2482  * Validate TCP item.
2483  *
2484  * @param[in] item
2485  *   Item specification.
2486  * @param[in] item_flags
2487  *   Bit-fields that holds the items detected until now.
2488  * @param[in] target_protocol
2489  *   The next protocol in the previous item.
2490  * @param[out] error
2491  *   Pointer to error structure.
2492  *
2493  * @return
2494  *   0 on success, a negative errno value otherwise and rte_errno is set.
2495  */
2496 int
2497 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2498 			    uint64_t item_flags,
2499 			    uint8_t target_protocol,
2500 			    const struct rte_flow_item_tcp *flow_mask,
2501 			    struct rte_flow_error *error)
2502 {
2503 	const struct rte_flow_item_tcp *mask = item->mask;
2504 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2505 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2506 				      MLX5_FLOW_LAYER_OUTER_L3;
2507 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2508 				      MLX5_FLOW_LAYER_OUTER_L4;
2509 	int ret;
2510 
2511 	MLX5_ASSERT(flow_mask);
2512 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2513 		return rte_flow_error_set(error, EINVAL,
2514 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2515 					  "protocol filtering not compatible"
2516 					  " with TCP layer");
2517 	if (!(item_flags & l3m))
2518 		return rte_flow_error_set(error, EINVAL,
2519 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2520 					  "L3 is mandatory to filter on L4");
2521 	if (item_flags & l4m)
2522 		return rte_flow_error_set(error, EINVAL,
2523 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2524 					  "multiple L4 layers not supported");
2525 	if (!mask)
2526 		mask = &rte_flow_item_tcp_mask;
2527 	ret = mlx5_flow_item_acceptable
2528 		(item, (const uint8_t *)mask,
2529 		 (const uint8_t *)flow_mask,
2530 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2531 		 error);
2532 	if (ret < 0)
2533 		return ret;
2534 	return 0;
2535 }
2536 
2537 /**
2538  * Validate VXLAN item.
2539  *
2540  * @param[in] dev
2541  *   Pointer to the Ethernet device structure.
2542  * @param[in] udp_dport
2543  *   UDP destination port
2544  * @param[in] item
2545  *   Item specification.
2546  * @param[in] item_flags
2547  *   Bit-fields that holds the items detected until now.
2548  * @param[in] attr
2549  *   Flow rule attributes.
2550  * @param[out] error
2551  *   Pointer to error structure.
2552  *
2553  * @return
2554  *   0 on success, a negative errno value otherwise and rte_errno is set.
2555  */
2556 int
2557 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2558 			      uint16_t udp_dport,
2559 			      const struct rte_flow_item *item,
2560 			      uint64_t item_flags,
2561 			      const struct rte_flow_attr *attr,
2562 			      struct rte_flow_error *error)
2563 {
2564 	const struct rte_flow_item_vxlan *spec = item->spec;
2565 	const struct rte_flow_item_vxlan *mask = item->mask;
2566 	int ret;
2567 	struct mlx5_priv *priv = dev->data->dev_private;
2568 	union vni {
2569 		uint32_t vlan_id;
2570 		uint8_t vni[4];
2571 	} id = { .vlan_id = 0, };
2572 	const struct rte_flow_item_vxlan nic_mask = {
2573 		.vni = "\xff\xff\xff",
2574 		.rsvd1 = 0xff,
2575 	};
2576 	const struct rte_flow_item_vxlan *valid_mask;
2577 
2578 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2579 		return rte_flow_error_set(error, ENOTSUP,
2580 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2581 					  "multiple tunnel layers not"
2582 					  " supported");
2583 	valid_mask = &rte_flow_item_vxlan_mask;
2584 	/*
2585 	 * Verify only UDPv4 is present as defined in
2586 	 * https://tools.ietf.org/html/rfc7348
2587 	 */
2588 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2589 		return rte_flow_error_set(error, EINVAL,
2590 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2591 					  "no outer UDP layer found");
2592 	if (!mask)
2593 		mask = &rte_flow_item_vxlan_mask;
2594 
2595 	if (priv->sh->steering_format_version !=
2596 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2597 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2598 		/* FDB domain & NIC domain non-zero group */
2599 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2600 			valid_mask = &nic_mask;
2601 		/* Group zero in NIC domain */
2602 		if (!attr->group && !attr->transfer &&
2603 		    priv->sh->tunnel_header_0_1)
2604 			valid_mask = &nic_mask;
2605 	}
2606 	ret = mlx5_flow_item_acceptable
2607 		(item, (const uint8_t *)mask,
2608 		 (const uint8_t *)valid_mask,
2609 		 sizeof(struct rte_flow_item_vxlan),
2610 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2611 	if (ret < 0)
2612 		return ret;
2613 	if (spec) {
2614 		memcpy(&id.vni[1], spec->vni, 3);
2615 		memcpy(&id.vni[1], mask->vni, 3);
2616 	}
2617 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2618 		return rte_flow_error_set(error, ENOTSUP,
2619 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2620 					  "VXLAN tunnel must be fully defined");
2621 	return 0;
2622 }
2623 
2624 /**
2625  * Validate VXLAN_GPE item.
2626  *
2627  * @param[in] item
2628  *   Item specification.
2629  * @param[in] item_flags
2630  *   Bit-fields that holds the items detected until now.
2631  * @param[in] priv
2632  *   Pointer to the private data structure.
2633  * @param[in] target_protocol
2634  *   The next protocol in the previous item.
2635  * @param[out] error
2636  *   Pointer to error structure.
2637  *
2638  * @return
2639  *   0 on success, a negative errno value otherwise and rte_errno is set.
2640  */
2641 int
2642 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2643 				  uint64_t item_flags,
2644 				  struct rte_eth_dev *dev,
2645 				  struct rte_flow_error *error)
2646 {
2647 	struct mlx5_priv *priv = dev->data->dev_private;
2648 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2649 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2650 	int ret;
2651 	union vni {
2652 		uint32_t vlan_id;
2653 		uint8_t vni[4];
2654 	} id = { .vlan_id = 0, };
2655 
2656 	if (!priv->config.l3_vxlan_en)
2657 		return rte_flow_error_set(error, ENOTSUP,
2658 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2659 					  "L3 VXLAN is not enabled by device"
2660 					  " parameter and/or not configured in"
2661 					  " firmware");
2662 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2663 		return rte_flow_error_set(error, ENOTSUP,
2664 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2665 					  "multiple tunnel layers not"
2666 					  " supported");
2667 	/*
2668 	 * Verify only UDPv4 is present as defined in
2669 	 * https://tools.ietf.org/html/rfc7348
2670 	 */
2671 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2672 		return rte_flow_error_set(error, EINVAL,
2673 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2674 					  "no outer UDP layer found");
2675 	if (!mask)
2676 		mask = &rte_flow_item_vxlan_gpe_mask;
2677 	ret = mlx5_flow_item_acceptable
2678 		(item, (const uint8_t *)mask,
2679 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2680 		 sizeof(struct rte_flow_item_vxlan_gpe),
2681 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2682 	if (ret < 0)
2683 		return ret;
2684 	if (spec) {
2685 		if (spec->protocol)
2686 			return rte_flow_error_set(error, ENOTSUP,
2687 						  RTE_FLOW_ERROR_TYPE_ITEM,
2688 						  item,
2689 						  "VxLAN-GPE protocol"
2690 						  " not supported");
2691 		memcpy(&id.vni[1], spec->vni, 3);
2692 		memcpy(&id.vni[1], mask->vni, 3);
2693 	}
2694 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2695 		return rte_flow_error_set(error, ENOTSUP,
2696 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2697 					  "VXLAN-GPE tunnel must be fully"
2698 					  " defined");
2699 	return 0;
2700 }
2701 /**
2702  * Validate GRE Key item.
2703  *
2704  * @param[in] item
2705  *   Item specification.
2706  * @param[in] item_flags
2707  *   Bit flags to mark detected items.
2708  * @param[in] gre_item
2709  *   Pointer to gre_item
2710  * @param[out] error
2711  *   Pointer to error structure.
2712  *
2713  * @return
2714  *   0 on success, a negative errno value otherwise and rte_errno is set.
2715  */
2716 int
2717 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2718 				uint64_t item_flags,
2719 				const struct rte_flow_item *gre_item,
2720 				struct rte_flow_error *error)
2721 {
2722 	const rte_be32_t *mask = item->mask;
2723 	int ret = 0;
2724 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2725 	const struct rte_flow_item_gre *gre_spec;
2726 	const struct rte_flow_item_gre *gre_mask;
2727 
2728 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2729 		return rte_flow_error_set(error, ENOTSUP,
2730 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2731 					  "Multiple GRE key not support");
2732 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2733 		return rte_flow_error_set(error, ENOTSUP,
2734 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2735 					  "No preceding GRE header");
2736 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2737 		return rte_flow_error_set(error, ENOTSUP,
2738 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2739 					  "GRE key following a wrong item");
2740 	gre_mask = gre_item->mask;
2741 	if (!gre_mask)
2742 		gre_mask = &rte_flow_item_gre_mask;
2743 	gre_spec = gre_item->spec;
2744 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2745 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2746 		return rte_flow_error_set(error, EINVAL,
2747 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2748 					  "Key bit must be on");
2749 
2750 	if (!mask)
2751 		mask = &gre_key_default_mask;
2752 	ret = mlx5_flow_item_acceptable
2753 		(item, (const uint8_t *)mask,
2754 		 (const uint8_t *)&gre_key_default_mask,
2755 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2756 	return ret;
2757 }
2758 
2759 /**
2760  * Validate GRE item.
2761  *
2762  * @param[in] item
2763  *   Item specification.
2764  * @param[in] item_flags
2765  *   Bit flags to mark detected items.
2766  * @param[in] target_protocol
2767  *   The next protocol in the previous item.
2768  * @param[out] error
2769  *   Pointer to error structure.
2770  *
2771  * @return
2772  *   0 on success, a negative errno value otherwise and rte_errno is set.
2773  */
2774 int
2775 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2776 			    uint64_t item_flags,
2777 			    uint8_t target_protocol,
2778 			    struct rte_flow_error *error)
2779 {
2780 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2781 	const struct rte_flow_item_gre *mask = item->mask;
2782 	int ret;
2783 	const struct rte_flow_item_gre nic_mask = {
2784 		.c_rsvd0_ver = RTE_BE16(0xB000),
2785 		.protocol = RTE_BE16(UINT16_MAX),
2786 	};
2787 
2788 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2789 		return rte_flow_error_set(error, EINVAL,
2790 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2791 					  "protocol filtering not compatible"
2792 					  " with this GRE layer");
2793 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2794 		return rte_flow_error_set(error, ENOTSUP,
2795 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2796 					  "multiple tunnel layers not"
2797 					  " supported");
2798 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2799 		return rte_flow_error_set(error, ENOTSUP,
2800 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2801 					  "L3 Layer is missing");
2802 	if (!mask)
2803 		mask = &rte_flow_item_gre_mask;
2804 	ret = mlx5_flow_item_acceptable
2805 		(item, (const uint8_t *)mask,
2806 		 (const uint8_t *)&nic_mask,
2807 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2808 		 error);
2809 	if (ret < 0)
2810 		return ret;
2811 #ifndef HAVE_MLX5DV_DR
2812 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2813 	if (spec && (spec->protocol & mask->protocol))
2814 		return rte_flow_error_set(error, ENOTSUP,
2815 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2816 					  "without MPLS support the"
2817 					  " specification cannot be used for"
2818 					  " filtering");
2819 #endif
2820 #endif
2821 	return 0;
2822 }
2823 
2824 /**
2825  * Validate Geneve item.
2826  *
2827  * @param[in] item
2828  *   Item specification.
2829  * @param[in] itemFlags
2830  *   Bit-fields that holds the items detected until now.
2831  * @param[in] enPriv
2832  *   Pointer to the private data structure.
2833  * @param[out] error
2834  *   Pointer to error structure.
2835  *
2836  * @return
2837  *   0 on success, a negative errno value otherwise and rte_errno is set.
2838  */
2839 
2840 int
2841 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2842 			       uint64_t item_flags,
2843 			       struct rte_eth_dev *dev,
2844 			       struct rte_flow_error *error)
2845 {
2846 	struct mlx5_priv *priv = dev->data->dev_private;
2847 	const struct rte_flow_item_geneve *spec = item->spec;
2848 	const struct rte_flow_item_geneve *mask = item->mask;
2849 	int ret;
2850 	uint16_t gbhdr;
2851 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2852 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2853 	const struct rte_flow_item_geneve nic_mask = {
2854 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2855 		.vni = "\xff\xff\xff",
2856 		.protocol = RTE_BE16(UINT16_MAX),
2857 	};
2858 
2859 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2860 		return rte_flow_error_set(error, ENOTSUP,
2861 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2862 					  "L3 Geneve is not enabled by device"
2863 					  " parameter and/or not configured in"
2864 					  " firmware");
2865 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2866 		return rte_flow_error_set(error, ENOTSUP,
2867 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2868 					  "multiple tunnel layers not"
2869 					  " supported");
2870 	/*
2871 	 * Verify only UDPv4 is present as defined in
2872 	 * https://tools.ietf.org/html/rfc7348
2873 	 */
2874 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2875 		return rte_flow_error_set(error, EINVAL,
2876 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2877 					  "no outer UDP layer found");
2878 	if (!mask)
2879 		mask = &rte_flow_item_geneve_mask;
2880 	ret = mlx5_flow_item_acceptable
2881 				  (item, (const uint8_t *)mask,
2882 				   (const uint8_t *)&nic_mask,
2883 				   sizeof(struct rte_flow_item_geneve),
2884 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2885 	if (ret)
2886 		return ret;
2887 	if (spec) {
2888 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2889 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2890 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2891 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2892 			return rte_flow_error_set(error, ENOTSUP,
2893 						  RTE_FLOW_ERROR_TYPE_ITEM,
2894 						  item,
2895 						  "Geneve protocol unsupported"
2896 						  " fields are being used");
2897 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2898 			return rte_flow_error_set
2899 					(error, ENOTSUP,
2900 					 RTE_FLOW_ERROR_TYPE_ITEM,
2901 					 item,
2902 					 "Unsupported Geneve options length");
2903 	}
2904 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2905 		return rte_flow_error_set
2906 				    (error, ENOTSUP,
2907 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2908 				     "Geneve tunnel must be fully defined");
2909 	return 0;
2910 }
2911 
2912 /**
2913  * Validate Geneve TLV option item.
2914  *
2915  * @param[in] item
2916  *   Item specification.
2917  * @param[in] last_item
2918  *   Previous validated item in the pattern items.
2919  * @param[in] geneve_item
2920  *   Previous GENEVE item specification.
2921  * @param[in] dev
2922  *   Pointer to the rte_eth_dev structure.
2923  * @param[out] error
2924  *   Pointer to error structure.
2925  *
2926  * @return
2927  *   0 on success, a negative errno value otherwise and rte_errno is set.
2928  */
2929 int
2930 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2931 				   uint64_t last_item,
2932 				   const struct rte_flow_item *geneve_item,
2933 				   struct rte_eth_dev *dev,
2934 				   struct rte_flow_error *error)
2935 {
2936 	struct mlx5_priv *priv = dev->data->dev_private;
2937 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2938 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2939 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2940 	uint8_t data_max_supported =
2941 			hca_attr->max_geneve_tlv_option_data_len * 4;
2942 	struct mlx5_dev_config *config = &priv->config;
2943 	const struct rte_flow_item_geneve *geneve_spec;
2944 	const struct rte_flow_item_geneve *geneve_mask;
2945 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2946 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2947 	unsigned int i;
2948 	unsigned int data_len;
2949 	uint8_t tlv_option_len;
2950 	uint16_t optlen_m, optlen_v;
2951 	const struct rte_flow_item_geneve_opt full_mask = {
2952 		.option_class = RTE_BE16(0xffff),
2953 		.option_type = 0xff,
2954 		.option_len = 0x1f,
2955 	};
2956 
2957 	if (!mask)
2958 		mask = &rte_flow_item_geneve_opt_mask;
2959 	if (!spec)
2960 		return rte_flow_error_set
2961 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2962 			"Geneve TLV opt class/type/length must be specified");
2963 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2964 		return rte_flow_error_set
2965 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2966 			"Geneve TLV opt length exceeeds the limit (31)");
2967 	/* Check if class type and length masks are full. */
2968 	if (full_mask.option_class != mask->option_class ||
2969 	    full_mask.option_type != mask->option_type ||
2970 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2971 		return rte_flow_error_set
2972 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2973 			"Geneve TLV opt class/type/length masks must be full");
2974 	/* Check if length is supported */
2975 	if ((uint32_t)spec->option_len >
2976 			config->hca_attr.max_geneve_tlv_option_data_len)
2977 		return rte_flow_error_set
2978 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2979 			"Geneve TLV opt length not supported");
2980 	if (config->hca_attr.max_geneve_tlv_options > 1)
2981 		DRV_LOG(DEBUG,
2982 			"max_geneve_tlv_options supports more than 1 option");
2983 	/* Check GENEVE item preceding. */
2984 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2985 		return rte_flow_error_set
2986 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2987 			"Geneve opt item must be preceded with Geneve item");
2988 	geneve_spec = geneve_item->spec;
2989 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2990 					  &rte_flow_item_geneve_mask;
2991 	/* Check if GENEVE TLV option size doesn't exceed option length */
2992 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2993 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2994 		tlv_option_len = spec->option_len & mask->option_len;
2995 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2996 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2997 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2998 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2999 		if ((optlen_v & optlen_m) <= tlv_option_len)
3000 			return rte_flow_error_set
3001 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3002 				 "GENEVE TLV option length exceeds optlen");
3003 	}
3004 	/* Check if length is 0 or data is 0. */
3005 	if (spec->data == NULL || spec->option_len == 0)
3006 		return rte_flow_error_set
3007 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3008 			"Geneve TLV opt with zero data/length not supported");
3009 	/* Check not all data & mask are 0. */
3010 	data_len = spec->option_len * 4;
3011 	if (mask->data == NULL) {
3012 		for (i = 0; i < data_len; i++)
3013 			if (spec->data[i])
3014 				break;
3015 		if (i == data_len)
3016 			return rte_flow_error_set(error, ENOTSUP,
3017 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3018 				"Can't match on Geneve option data 0");
3019 	} else {
3020 		for (i = 0; i < data_len; i++)
3021 			if (spec->data[i] & mask->data[i])
3022 				break;
3023 		if (i == data_len)
3024 			return rte_flow_error_set(error, ENOTSUP,
3025 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3026 				"Can't match on Geneve option data and mask 0");
3027 		/* Check data mask supported. */
3028 		for (i = data_max_supported; i < data_len ; i++)
3029 			if (mask->data[i])
3030 				return rte_flow_error_set(error, ENOTSUP,
3031 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3032 					"Data mask is of unsupported size");
3033 	}
3034 	/* Check GENEVE option is supported in NIC. */
3035 	if (!config->hca_attr.geneve_tlv_opt)
3036 		return rte_flow_error_set
3037 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3038 			"Geneve TLV opt not supported");
3039 	/* Check if we already have geneve option with different type/class. */
3040 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3041 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3042 	if (geneve_opt_resource != NULL)
3043 		if (geneve_opt_resource->option_class != spec->option_class ||
3044 		    geneve_opt_resource->option_type != spec->option_type ||
3045 		    geneve_opt_resource->length != spec->option_len) {
3046 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3047 			return rte_flow_error_set(error, ENOTSUP,
3048 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3049 				"Only one Geneve TLV option supported");
3050 		}
3051 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3052 	return 0;
3053 }
3054 
3055 /**
3056  * Validate MPLS item.
3057  *
3058  * @param[in] dev
3059  *   Pointer to the rte_eth_dev structure.
3060  * @param[in] item
3061  *   Item specification.
3062  * @param[in] item_flags
3063  *   Bit-fields that holds the items detected until now.
3064  * @param[in] prev_layer
3065  *   The protocol layer indicated in previous item.
3066  * @param[out] error
3067  *   Pointer to error structure.
3068  *
3069  * @return
3070  *   0 on success, a negative errno value otherwise and rte_errno is set.
3071  */
3072 int
3073 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3074 			     const struct rte_flow_item *item __rte_unused,
3075 			     uint64_t item_flags __rte_unused,
3076 			     uint64_t prev_layer __rte_unused,
3077 			     struct rte_flow_error *error)
3078 {
3079 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3080 	const struct rte_flow_item_mpls *mask = item->mask;
3081 	struct mlx5_priv *priv = dev->data->dev_private;
3082 	int ret;
3083 
3084 	if (!priv->config.mpls_en)
3085 		return rte_flow_error_set(error, ENOTSUP,
3086 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3087 					  "MPLS not supported or"
3088 					  " disabled in firmware"
3089 					  " configuration.");
3090 	/* MPLS over UDP, GRE is allowed */
3091 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3092 			    MLX5_FLOW_LAYER_GRE |
3093 			    MLX5_FLOW_LAYER_GRE_KEY)))
3094 		return rte_flow_error_set(error, EINVAL,
3095 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3096 					  "protocol filtering not compatible"
3097 					  " with MPLS layer");
3098 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3099 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3100 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3101 		return rte_flow_error_set(error, ENOTSUP,
3102 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3103 					  "multiple tunnel layers not"
3104 					  " supported");
3105 	if (!mask)
3106 		mask = &rte_flow_item_mpls_mask;
3107 	ret = mlx5_flow_item_acceptable
3108 		(item, (const uint8_t *)mask,
3109 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3110 		 sizeof(struct rte_flow_item_mpls),
3111 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3112 	if (ret < 0)
3113 		return ret;
3114 	return 0;
3115 #else
3116 	return rte_flow_error_set(error, ENOTSUP,
3117 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3118 				  "MPLS is not supported by Verbs, please"
3119 				  " update.");
3120 #endif
3121 }
3122 
3123 /**
3124  * Validate NVGRE item.
3125  *
3126  * @param[in] item
3127  *   Item specification.
3128  * @param[in] item_flags
3129  *   Bit flags to mark detected items.
3130  * @param[in] target_protocol
3131  *   The next protocol in the previous item.
3132  * @param[out] error
3133  *   Pointer to error structure.
3134  *
3135  * @return
3136  *   0 on success, a negative errno value otherwise and rte_errno is set.
3137  */
3138 int
3139 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3140 			      uint64_t item_flags,
3141 			      uint8_t target_protocol,
3142 			      struct rte_flow_error *error)
3143 {
3144 	const struct rte_flow_item_nvgre *mask = item->mask;
3145 	int ret;
3146 
3147 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3148 		return rte_flow_error_set(error, EINVAL,
3149 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3150 					  "protocol filtering not compatible"
3151 					  " with this GRE layer");
3152 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3153 		return rte_flow_error_set(error, ENOTSUP,
3154 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3155 					  "multiple tunnel layers not"
3156 					  " supported");
3157 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3158 		return rte_flow_error_set(error, ENOTSUP,
3159 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3160 					  "L3 Layer is missing");
3161 	if (!mask)
3162 		mask = &rte_flow_item_nvgre_mask;
3163 	ret = mlx5_flow_item_acceptable
3164 		(item, (const uint8_t *)mask,
3165 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3166 		 sizeof(struct rte_flow_item_nvgre),
3167 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3168 	if (ret < 0)
3169 		return ret;
3170 	return 0;
3171 }
3172 
3173 /**
3174  * Validate eCPRI item.
3175  *
3176  * @param[in] item
3177  *   Item specification.
3178  * @param[in] item_flags
3179  *   Bit-fields that holds the items detected until now.
3180  * @param[in] last_item
3181  *   Previous validated item in the pattern items.
3182  * @param[in] ether_type
3183  *   Type in the ethernet layer header (including dot1q).
3184  * @param[in] acc_mask
3185  *   Acceptable mask, if NULL default internal default mask
3186  *   will be used to check whether item fields are supported.
3187  * @param[out] error
3188  *   Pointer to error structure.
3189  *
3190  * @return
3191  *   0 on success, a negative errno value otherwise and rte_errno is set.
3192  */
3193 int
3194 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3195 			      uint64_t item_flags,
3196 			      uint64_t last_item,
3197 			      uint16_t ether_type,
3198 			      const struct rte_flow_item_ecpri *acc_mask,
3199 			      struct rte_flow_error *error)
3200 {
3201 	const struct rte_flow_item_ecpri *mask = item->mask;
3202 	const struct rte_flow_item_ecpri nic_mask = {
3203 		.hdr = {
3204 			.common = {
3205 				.u32 =
3206 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3207 					.type = 0xFF,
3208 					}).u32),
3209 			},
3210 			.dummy[0] = 0xFFFFFFFF,
3211 		},
3212 	};
3213 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3214 					MLX5_FLOW_LAYER_OUTER_VLAN);
3215 	struct rte_flow_item_ecpri mask_lo;
3216 
3217 	if (!(last_item & outer_l2_vlan) &&
3218 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3219 		return rte_flow_error_set(error, EINVAL,
3220 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3221 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3222 	if ((last_item & outer_l2_vlan) && ether_type &&
3223 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3224 		return rte_flow_error_set(error, EINVAL,
3225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3226 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3227 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3228 		return rte_flow_error_set(error, EINVAL,
3229 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3230 					  "eCPRI with tunnel is not supported right now");
3231 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3232 		return rte_flow_error_set(error, ENOTSUP,
3233 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3234 					  "multiple L3 layers not supported");
3235 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3236 		return rte_flow_error_set(error, EINVAL,
3237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3238 					  "eCPRI cannot coexist with a TCP layer");
3239 	/* In specification, eCPRI could be over UDP layer. */
3240 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3241 		return rte_flow_error_set(error, EINVAL,
3242 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3243 					  "eCPRI over UDP layer is not yet supported right now");
3244 	/* Mask for type field in common header could be zero. */
3245 	if (!mask)
3246 		mask = &rte_flow_item_ecpri_mask;
3247 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3248 	/* Input mask is in big-endian format. */
3249 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3250 		return rte_flow_error_set(error, EINVAL,
3251 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3252 					  "partial mask is not supported for protocol");
3253 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3254 		return rte_flow_error_set(error, EINVAL,
3255 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3256 					  "message header mask must be after a type mask");
3257 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3258 					 acc_mask ? (const uint8_t *)acc_mask
3259 						  : (const uint8_t *)&nic_mask,
3260 					 sizeof(struct rte_flow_item_ecpri),
3261 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3262 }
3263 
3264 static int
3265 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3266 		   const struct rte_flow_attr *attr __rte_unused,
3267 		   const struct rte_flow_item items[] __rte_unused,
3268 		   const struct rte_flow_action actions[] __rte_unused,
3269 		   bool external __rte_unused,
3270 		   int hairpin __rte_unused,
3271 		   struct rte_flow_error *error)
3272 {
3273 	return rte_flow_error_set(error, ENOTSUP,
3274 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3275 }
3276 
3277 static struct mlx5_flow *
3278 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3279 		  const struct rte_flow_attr *attr __rte_unused,
3280 		  const struct rte_flow_item items[] __rte_unused,
3281 		  const struct rte_flow_action actions[] __rte_unused,
3282 		  struct rte_flow_error *error)
3283 {
3284 	rte_flow_error_set(error, ENOTSUP,
3285 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3286 	return NULL;
3287 }
3288 
3289 static int
3290 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3291 		    struct mlx5_flow *dev_flow __rte_unused,
3292 		    const struct rte_flow_attr *attr __rte_unused,
3293 		    const struct rte_flow_item items[] __rte_unused,
3294 		    const struct rte_flow_action actions[] __rte_unused,
3295 		    struct rte_flow_error *error)
3296 {
3297 	return rte_flow_error_set(error, ENOTSUP,
3298 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3299 }
3300 
3301 static int
3302 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3303 		struct rte_flow *flow __rte_unused,
3304 		struct rte_flow_error *error)
3305 {
3306 	return rte_flow_error_set(error, ENOTSUP,
3307 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3308 }
3309 
3310 static void
3311 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3312 		 struct rte_flow *flow __rte_unused)
3313 {
3314 }
3315 
3316 static void
3317 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3318 		  struct rte_flow *flow __rte_unused)
3319 {
3320 }
3321 
3322 static int
3323 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3324 		struct rte_flow *flow __rte_unused,
3325 		const struct rte_flow_action *actions __rte_unused,
3326 		void *data __rte_unused,
3327 		struct rte_flow_error *error)
3328 {
3329 	return rte_flow_error_set(error, ENOTSUP,
3330 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3331 }
3332 
3333 static int
3334 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3335 		      uint32_t domains __rte_unused,
3336 		      uint32_t flags __rte_unused)
3337 {
3338 	return 0;
3339 }
3340 
3341 /* Void driver to protect from null pointer reference. */
3342 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3343 	.validate = flow_null_validate,
3344 	.prepare = flow_null_prepare,
3345 	.translate = flow_null_translate,
3346 	.apply = flow_null_apply,
3347 	.remove = flow_null_remove,
3348 	.destroy = flow_null_destroy,
3349 	.query = flow_null_query,
3350 	.sync_domain = flow_null_sync_domain,
3351 };
3352 
3353 /**
3354  * Select flow driver type according to flow attributes and device
3355  * configuration.
3356  *
3357  * @param[in] dev
3358  *   Pointer to the dev structure.
3359  * @param[in] attr
3360  *   Pointer to the flow attributes.
3361  *
3362  * @return
3363  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3364  */
3365 static enum mlx5_flow_drv_type
3366 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3367 {
3368 	struct mlx5_priv *priv = dev->data->dev_private;
3369 	/* The OS can determine first a specific flow type (DV, VERBS) */
3370 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3371 
3372 	if (type != MLX5_FLOW_TYPE_MAX)
3373 		return type;
3374 	/* If no OS specific type - continue with DV/VERBS selection */
3375 	if (attr->transfer && priv->config.dv_esw_en)
3376 		type = MLX5_FLOW_TYPE_DV;
3377 	if (!attr->transfer)
3378 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3379 						 MLX5_FLOW_TYPE_VERBS;
3380 	return type;
3381 }
3382 
3383 #define flow_get_drv_ops(type) flow_drv_ops[type]
3384 
3385 /**
3386  * Flow driver validation API. This abstracts calling driver specific functions.
3387  * The type of flow driver is determined according to flow attributes.
3388  *
3389  * @param[in] dev
3390  *   Pointer to the dev structure.
3391  * @param[in] attr
3392  *   Pointer to the flow attributes.
3393  * @param[in] items
3394  *   Pointer to the list of items.
3395  * @param[in] actions
3396  *   Pointer to the list of actions.
3397  * @param[in] external
3398  *   This flow rule is created by request external to PMD.
3399  * @param[in] hairpin
3400  *   Number of hairpin TX actions, 0 means classic flow.
3401  * @param[out] error
3402  *   Pointer to the error structure.
3403  *
3404  * @return
3405  *   0 on success, a negative errno value otherwise and rte_errno is set.
3406  */
3407 static inline int
3408 flow_drv_validate(struct rte_eth_dev *dev,
3409 		  const struct rte_flow_attr *attr,
3410 		  const struct rte_flow_item items[],
3411 		  const struct rte_flow_action actions[],
3412 		  bool external, int hairpin, struct rte_flow_error *error)
3413 {
3414 	const struct mlx5_flow_driver_ops *fops;
3415 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3416 
3417 	fops = flow_get_drv_ops(type);
3418 	return fops->validate(dev, attr, items, actions, external,
3419 			      hairpin, error);
3420 }
3421 
3422 /**
3423  * Flow driver preparation API. This abstracts calling driver specific
3424  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3425  * calculates the size of memory required for device flow, allocates the memory,
3426  * initializes the device flow and returns the pointer.
3427  *
3428  * @note
3429  *   This function initializes device flow structure such as dv or verbs in
3430  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3431  *   rest. For example, adding returning device flow to flow->dev_flow list and
3432  *   setting backward reference to the flow should be done out of this function.
3433  *   layers field is not filled either.
3434  *
3435  * @param[in] dev
3436  *   Pointer to the dev structure.
3437  * @param[in] attr
3438  *   Pointer to the flow attributes.
3439  * @param[in] items
3440  *   Pointer to the list of items.
3441  * @param[in] actions
3442  *   Pointer to the list of actions.
3443  * @param[in] flow_idx
3444  *   This memory pool index to the flow.
3445  * @param[out] error
3446  *   Pointer to the error structure.
3447  *
3448  * @return
3449  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3450  */
3451 static inline struct mlx5_flow *
3452 flow_drv_prepare(struct rte_eth_dev *dev,
3453 		 const struct rte_flow *flow,
3454 		 const struct rte_flow_attr *attr,
3455 		 const struct rte_flow_item items[],
3456 		 const struct rte_flow_action actions[],
3457 		 uint32_t flow_idx,
3458 		 struct rte_flow_error *error)
3459 {
3460 	const struct mlx5_flow_driver_ops *fops;
3461 	enum mlx5_flow_drv_type type = flow->drv_type;
3462 	struct mlx5_flow *mlx5_flow = NULL;
3463 
3464 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3465 	fops = flow_get_drv_ops(type);
3466 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3467 	if (mlx5_flow)
3468 		mlx5_flow->flow_idx = flow_idx;
3469 	return mlx5_flow;
3470 }
3471 
3472 /**
3473  * Flow driver translation API. This abstracts calling driver specific
3474  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3475  * translates a generic flow into a driver flow. flow_drv_prepare() must
3476  * precede.
3477  *
3478  * @note
3479  *   dev_flow->layers could be filled as a result of parsing during translation
3480  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3481  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3482  *   flow->actions could be overwritten even though all the expanded dev_flows
3483  *   have the same actions.
3484  *
3485  * @param[in] dev
3486  *   Pointer to the rte dev structure.
3487  * @param[in, out] dev_flow
3488  *   Pointer to the mlx5 flow.
3489  * @param[in] attr
3490  *   Pointer to the flow attributes.
3491  * @param[in] items
3492  *   Pointer to the list of items.
3493  * @param[in] actions
3494  *   Pointer to the list of actions.
3495  * @param[out] error
3496  *   Pointer to the error structure.
3497  *
3498  * @return
3499  *   0 on success, a negative errno value otherwise and rte_errno is set.
3500  */
3501 static inline int
3502 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3503 		   const struct rte_flow_attr *attr,
3504 		   const struct rte_flow_item items[],
3505 		   const struct rte_flow_action actions[],
3506 		   struct rte_flow_error *error)
3507 {
3508 	const struct mlx5_flow_driver_ops *fops;
3509 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3510 
3511 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3512 	fops = flow_get_drv_ops(type);
3513 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3514 }
3515 
3516 /**
3517  * Flow driver apply API. This abstracts calling driver specific functions.
3518  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3519  * translated driver flows on to device. flow_drv_translate() must precede.
3520  *
3521  * @param[in] dev
3522  *   Pointer to Ethernet device structure.
3523  * @param[in, out] flow
3524  *   Pointer to flow structure.
3525  * @param[out] error
3526  *   Pointer to error structure.
3527  *
3528  * @return
3529  *   0 on success, a negative errno value otherwise and rte_errno is set.
3530  */
3531 static inline int
3532 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3533 	       struct rte_flow_error *error)
3534 {
3535 	const struct mlx5_flow_driver_ops *fops;
3536 	enum mlx5_flow_drv_type type = flow->drv_type;
3537 
3538 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3539 	fops = flow_get_drv_ops(type);
3540 	return fops->apply(dev, flow, error);
3541 }
3542 
3543 /**
3544  * Flow driver destroy API. This abstracts calling driver specific functions.
3545  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3546  * on device and releases resources of the flow.
3547  *
3548  * @param[in] dev
3549  *   Pointer to Ethernet device.
3550  * @param[in, out] flow
3551  *   Pointer to flow structure.
3552  */
3553 static inline void
3554 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3555 {
3556 	const struct mlx5_flow_driver_ops *fops;
3557 	enum mlx5_flow_drv_type type = flow->drv_type;
3558 
3559 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3560 	fops = flow_get_drv_ops(type);
3561 	fops->destroy(dev, flow);
3562 }
3563 
3564 /**
3565  * Flow driver find RSS policy tbl API. This abstracts calling driver
3566  * specific functions. Parent flow (rte_flow) should have driver
3567  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3568  *
3569  * @param[in] dev
3570  *   Pointer to Ethernet device.
3571  * @param[in, out] flow
3572  *   Pointer to flow structure.
3573  * @param[in] policy
3574  *   Pointer to meter policy table.
3575  * @param[in] rss_desc
3576  *   Pointer to rss_desc
3577  */
3578 static struct mlx5_flow_meter_sub_policy *
3579 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3580 		struct rte_flow *flow,
3581 		struct mlx5_flow_meter_policy *policy,
3582 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3583 {
3584 	const struct mlx5_flow_driver_ops *fops;
3585 	enum mlx5_flow_drv_type type = flow->drv_type;
3586 
3587 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3588 	fops = flow_get_drv_ops(type);
3589 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3590 }
3591 
3592 /**
3593  * Flow driver color tag rule API. This abstracts calling driver
3594  * specific functions. Parent flow (rte_flow) should have driver
3595  * type (drv_type). It will create the color tag rules in hierarchy meter.
3596  *
3597  * @param[in] dev
3598  *   Pointer to Ethernet device.
3599  * @param[in, out] flow
3600  *   Pointer to flow structure.
3601  * @param[in] fm
3602  *   Pointer to flow meter structure.
3603  * @param[in] src_port
3604  *   The src port this extra rule should use.
3605  * @param[in] item
3606  *   The src port id match item.
3607  * @param[out] error
3608  *   Pointer to error structure.
3609  */
3610 static int
3611 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3612 		struct rte_flow *flow,
3613 		struct mlx5_flow_meter_info *fm,
3614 		int32_t src_port,
3615 		const struct rte_flow_item *item,
3616 		struct rte_flow_error *error)
3617 {
3618 	const struct mlx5_flow_driver_ops *fops;
3619 	enum mlx5_flow_drv_type type = flow->drv_type;
3620 
3621 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3622 	fops = flow_get_drv_ops(type);
3623 	return fops->meter_hierarchy_rule_create(dev, fm,
3624 						src_port, item, error);
3625 }
3626 
3627 /**
3628  * Get RSS action from the action list.
3629  *
3630  * @param[in] dev
3631  *   Pointer to Ethernet device.
3632  * @param[in] actions
3633  *   Pointer to the list of actions.
3634  * @param[in] flow
3635  *   Parent flow structure pointer.
3636  *
3637  * @return
3638  *   Pointer to the RSS action if exist, else return NULL.
3639  */
3640 static const struct rte_flow_action_rss*
3641 flow_get_rss_action(struct rte_eth_dev *dev,
3642 		    const struct rte_flow_action actions[])
3643 {
3644 	struct mlx5_priv *priv = dev->data->dev_private;
3645 	const struct rte_flow_action_rss *rss = NULL;
3646 	struct mlx5_meter_policy_action_container *acg;
3647 	struct mlx5_meter_policy_action_container *acy;
3648 
3649 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3650 		switch (actions->type) {
3651 		case RTE_FLOW_ACTION_TYPE_RSS:
3652 			rss = actions->conf;
3653 			break;
3654 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3655 		{
3656 			const struct rte_flow_action_sample *sample =
3657 								actions->conf;
3658 			const struct rte_flow_action *act = sample->actions;
3659 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3660 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3661 					rss = act->conf;
3662 			break;
3663 		}
3664 		case RTE_FLOW_ACTION_TYPE_METER:
3665 		{
3666 			uint32_t mtr_idx;
3667 			struct mlx5_flow_meter_info *fm;
3668 			struct mlx5_flow_meter_policy *policy;
3669 			const struct rte_flow_action_meter *mtr = actions->conf;
3670 
3671 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3672 			if (fm && !fm->def_policy) {
3673 				policy = mlx5_flow_meter_policy_find(dev,
3674 						fm->policy_id, NULL);
3675 				MLX5_ASSERT(policy);
3676 				if (policy->is_hierarchy) {
3677 					policy =
3678 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3679 									policy);
3680 					if (!policy)
3681 						return NULL;
3682 				}
3683 				if (policy->is_rss) {
3684 					acg =
3685 					&policy->act_cnt[RTE_COLOR_GREEN];
3686 					acy =
3687 					&policy->act_cnt[RTE_COLOR_YELLOW];
3688 					if (acg->fate_action ==
3689 					    MLX5_FLOW_FATE_SHARED_RSS)
3690 						rss = acg->rss->conf;
3691 					else if (acy->fate_action ==
3692 						 MLX5_FLOW_FATE_SHARED_RSS)
3693 						rss = acy->rss->conf;
3694 				}
3695 			}
3696 			break;
3697 		}
3698 		default:
3699 			break;
3700 		}
3701 	}
3702 	return rss;
3703 }
3704 
3705 /**
3706  * Get ASO age action by index.
3707  *
3708  * @param[in] dev
3709  *   Pointer to the Ethernet device structure.
3710  * @param[in] age_idx
3711  *   Index to the ASO age action.
3712  *
3713  * @return
3714  *   The specified ASO age action.
3715  */
3716 struct mlx5_aso_age_action*
3717 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3718 {
3719 	uint16_t pool_idx = age_idx & UINT16_MAX;
3720 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3721 	struct mlx5_priv *priv = dev->data->dev_private;
3722 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3723 	struct mlx5_aso_age_pool *pool;
3724 
3725 	rte_rwlock_read_lock(&mng->resize_rwl);
3726 	pool = mng->pools[pool_idx];
3727 	rte_rwlock_read_unlock(&mng->resize_rwl);
3728 	return &pool->actions[offset - 1];
3729 }
3730 
3731 /* maps indirect action to translated direct in some actions array */
3732 struct mlx5_translated_action_handle {
3733 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3734 	int index; /**< Index in related array of rte_flow_action. */
3735 };
3736 
3737 /**
3738  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3739  * direct action if translation possible.
3740  * This functionality used to run same execution path for both direct and
3741  * indirect actions on flow create. All necessary preparations for indirect
3742  * action handling should be performed on *handle* actions list returned
3743  * from this call.
3744  *
3745  * @param[in] dev
3746  *   Pointer to Ethernet device.
3747  * @param[in] actions
3748  *   List of actions to translate.
3749  * @param[out] handle
3750  *   List to store translated indirect action object handles.
3751  * @param[in, out] indir_n
3752  *   Size of *handle* array. On return should be updated with number of
3753  *   indirect actions retrieved from the *actions* list.
3754  * @param[out] translated_actions
3755  *   List of actions where all indirect actions were translated to direct
3756  *   if possible. NULL if no translation took place.
3757  * @param[out] error
3758  *   Pointer to the error structure.
3759  *
3760  * @return
3761  *   0 on success, a negative errno value otherwise and rte_errno is set.
3762  */
3763 static int
3764 flow_action_handles_translate(struct rte_eth_dev *dev,
3765 			      const struct rte_flow_action actions[],
3766 			      struct mlx5_translated_action_handle *handle,
3767 			      int *indir_n,
3768 			      struct rte_flow_action **translated_actions,
3769 			      struct rte_flow_error *error)
3770 {
3771 	struct mlx5_priv *priv = dev->data->dev_private;
3772 	struct rte_flow_action *translated = NULL;
3773 	size_t actions_size;
3774 	int n;
3775 	int copied_n = 0;
3776 	struct mlx5_translated_action_handle *handle_end = NULL;
3777 
3778 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3779 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3780 			continue;
3781 		if (copied_n == *indir_n) {
3782 			return rte_flow_error_set
3783 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3784 				 NULL, "too many shared actions");
3785 		}
3786 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3787 			   sizeof(actions[n].conf));
3788 		handle[copied_n].index = n;
3789 		copied_n++;
3790 	}
3791 	n++;
3792 	*indir_n = copied_n;
3793 	if (!copied_n)
3794 		return 0;
3795 	actions_size = sizeof(struct rte_flow_action) * n;
3796 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3797 	if (!translated) {
3798 		rte_errno = ENOMEM;
3799 		return -ENOMEM;
3800 	}
3801 	memcpy(translated, actions, actions_size);
3802 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3803 		struct mlx5_shared_action_rss *shared_rss;
3804 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3805 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3806 		uint32_t idx = act_idx &
3807 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3808 
3809 		switch (type) {
3810 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3811 			shared_rss = mlx5_ipool_get
3812 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3813 			translated[handle->index].type =
3814 				RTE_FLOW_ACTION_TYPE_RSS;
3815 			translated[handle->index].conf =
3816 				&shared_rss->origin;
3817 			break;
3818 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3819 			translated[handle->index].type =
3820 						(enum rte_flow_action_type)
3821 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3822 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3823 			break;
3824 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3825 			if (priv->sh->flow_hit_aso_en) {
3826 				translated[handle->index].type =
3827 					(enum rte_flow_action_type)
3828 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3829 				translated[handle->index].conf =
3830 							 (void *)(uintptr_t)idx;
3831 				break;
3832 			}
3833 			/* Fall-through */
3834 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3835 			if (priv->sh->ct_aso_en) {
3836 				translated[handle->index].type =
3837 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3838 				translated[handle->index].conf =
3839 							 (void *)(uintptr_t)idx;
3840 				break;
3841 			}
3842 			/* Fall-through */
3843 		default:
3844 			mlx5_free(translated);
3845 			return rte_flow_error_set
3846 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3847 				 NULL, "invalid indirect action type");
3848 		}
3849 	}
3850 	*translated_actions = translated;
3851 	return 0;
3852 }
3853 
3854 /**
3855  * Get Shared RSS action from the action list.
3856  *
3857  * @param[in] dev
3858  *   Pointer to Ethernet device.
3859  * @param[in] shared
3860  *   Pointer to the list of actions.
3861  * @param[in] shared_n
3862  *   Actions list length.
3863  *
3864  * @return
3865  *   The MLX5 RSS action ID if exists, otherwise return 0.
3866  */
3867 static uint32_t
3868 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3869 			   struct mlx5_translated_action_handle *handle,
3870 			   int shared_n)
3871 {
3872 	struct mlx5_translated_action_handle *handle_end;
3873 	struct mlx5_priv *priv = dev->data->dev_private;
3874 	struct mlx5_shared_action_rss *shared_rss;
3875 
3876 
3877 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3878 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3879 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3880 		uint32_t idx = act_idx &
3881 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3882 		switch (type) {
3883 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3884 			shared_rss = mlx5_ipool_get
3885 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3886 									   idx);
3887 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3888 					   __ATOMIC_RELAXED);
3889 			return idx;
3890 		default:
3891 			break;
3892 		}
3893 	}
3894 	return 0;
3895 }
3896 
3897 static unsigned int
3898 find_graph_root(uint32_t rss_level)
3899 {
3900 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3901 			       MLX5_EXPANSION_ROOT_OUTER;
3902 }
3903 
3904 /**
3905  *  Get layer flags from the prefix flow.
3906  *
3907  *  Some flows may be split to several subflows, the prefix subflow gets the
3908  *  match items and the suffix sub flow gets the actions.
3909  *  Some actions need the user defined match item flags to get the detail for
3910  *  the action.
3911  *  This function helps the suffix flow to get the item layer flags from prefix
3912  *  subflow.
3913  *
3914  * @param[in] dev_flow
3915  *   Pointer the created preifx subflow.
3916  *
3917  * @return
3918  *   The layers get from prefix subflow.
3919  */
3920 static inline uint64_t
3921 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3922 {
3923 	uint64_t layers = 0;
3924 
3925 	/*
3926 	 * Layers bits could be localization, but usually the compiler will
3927 	 * help to do the optimization work for source code.
3928 	 * If no decap actions, use the layers directly.
3929 	 */
3930 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3931 		return dev_flow->handle->layers;
3932 	/* Convert L3 layers with decap action. */
3933 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3934 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3935 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3936 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3937 	/* Convert L4 layers with decap action.  */
3938 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3939 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3940 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3941 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3942 	return layers;
3943 }
3944 
3945 /**
3946  * Get metadata split action information.
3947  *
3948  * @param[in] actions
3949  *   Pointer to the list of actions.
3950  * @param[out] qrss
3951  *   Pointer to the return pointer.
3952  * @param[out] qrss_type
3953  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3954  *   if no QUEUE/RSS is found.
3955  * @param[out] encap_idx
3956  *   Pointer to the index of the encap action if exists, otherwise the last
3957  *   action index.
3958  *
3959  * @return
3960  *   Total number of actions.
3961  */
3962 static int
3963 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3964 				       const struct rte_flow_action **qrss,
3965 				       int *encap_idx)
3966 {
3967 	const struct rte_flow_action_raw_encap *raw_encap;
3968 	int actions_n = 0;
3969 	int raw_decap_idx = -1;
3970 
3971 	*encap_idx = -1;
3972 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3973 		switch (actions->type) {
3974 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3975 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3976 			*encap_idx = actions_n;
3977 			break;
3978 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3979 			raw_decap_idx = actions_n;
3980 			break;
3981 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3982 			raw_encap = actions->conf;
3983 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3984 				*encap_idx = raw_decap_idx != -1 ?
3985 						      raw_decap_idx : actions_n;
3986 			break;
3987 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3988 		case RTE_FLOW_ACTION_TYPE_RSS:
3989 			*qrss = actions;
3990 			break;
3991 		default:
3992 			break;
3993 		}
3994 		actions_n++;
3995 	}
3996 	if (*encap_idx == -1)
3997 		*encap_idx = actions_n;
3998 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3999 	return actions_n + 1;
4000 }
4001 
4002 /**
4003  * Check if the action will change packet.
4004  *
4005  * @param dev
4006  *   Pointer to Ethernet device.
4007  * @param[in] type
4008  *   action type.
4009  *
4010  * @return
4011  *   true if action will change packet, false otherwise.
4012  */
4013 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4014 					  enum rte_flow_action_type type)
4015 {
4016 	struct mlx5_priv *priv = dev->data->dev_private;
4017 
4018 	switch (type) {
4019 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4020 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4021 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4022 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4023 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4024 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4025 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4026 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4027 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4028 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4029 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4030 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4031 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4032 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4033 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4034 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4035 	case RTE_FLOW_ACTION_TYPE_SET_META:
4036 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4037 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4038 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4039 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4040 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4041 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4042 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4043 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4044 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4045 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4046 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4047 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4048 		return true;
4049 	case RTE_FLOW_ACTION_TYPE_FLAG:
4050 	case RTE_FLOW_ACTION_TYPE_MARK:
4051 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4052 			return true;
4053 		else
4054 			return false;
4055 	default:
4056 		return false;
4057 	}
4058 }
4059 
4060 /**
4061  * Check meter action from the action list.
4062  *
4063  * @param dev
4064  *   Pointer to Ethernet device.
4065  * @param[in] actions
4066  *   Pointer to the list of actions.
4067  * @param[out] has_mtr
4068  *   Pointer to the meter exist flag.
4069  * @param[out] has_modify
4070  *   Pointer to the flag showing there's packet change action.
4071  * @param[out] meter_id
4072  *   Pointer to the meter id.
4073  *
4074  * @return
4075  *   Total number of actions.
4076  */
4077 static int
4078 flow_check_meter_action(struct rte_eth_dev *dev,
4079 			const struct rte_flow_action actions[],
4080 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4081 {
4082 	const struct rte_flow_action_meter *mtr = NULL;
4083 	int actions_n = 0;
4084 
4085 	MLX5_ASSERT(has_mtr);
4086 	*has_mtr = false;
4087 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4088 		switch (actions->type) {
4089 		case RTE_FLOW_ACTION_TYPE_METER:
4090 			mtr = actions->conf;
4091 			*meter_id = mtr->mtr_id;
4092 			*has_mtr = true;
4093 			break;
4094 		default:
4095 			break;
4096 		}
4097 		if (!*has_mtr)
4098 			*has_modify |= flow_check_modify_action_type(dev,
4099 								actions->type);
4100 		actions_n++;
4101 	}
4102 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4103 	return actions_n + 1;
4104 }
4105 
4106 /**
4107  * Check if the flow should be split due to hairpin.
4108  * The reason for the split is that in current HW we can't
4109  * support encap and push-vlan on Rx, so if a flow contains
4110  * these actions we move it to Tx.
4111  *
4112  * @param dev
4113  *   Pointer to Ethernet device.
4114  * @param[in] attr
4115  *   Flow rule attributes.
4116  * @param[in] actions
4117  *   Associated actions (list terminated by the END action).
4118  *
4119  * @return
4120  *   > 0 the number of actions and the flow should be split,
4121  *   0 when no split required.
4122  */
4123 static int
4124 flow_check_hairpin_split(struct rte_eth_dev *dev,
4125 			 const struct rte_flow_attr *attr,
4126 			 const struct rte_flow_action actions[])
4127 {
4128 	int queue_action = 0;
4129 	int action_n = 0;
4130 	int split = 0;
4131 	const struct rte_flow_action_queue *queue;
4132 	const struct rte_flow_action_rss *rss;
4133 	const struct rte_flow_action_raw_encap *raw_encap;
4134 	const struct rte_eth_hairpin_conf *conf;
4135 
4136 	if (!attr->ingress)
4137 		return 0;
4138 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4139 		switch (actions->type) {
4140 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4141 			queue = actions->conf;
4142 			if (queue == NULL)
4143 				return 0;
4144 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4145 			if (conf == NULL || conf->tx_explicit != 0)
4146 				return 0;
4147 			queue_action = 1;
4148 			action_n++;
4149 			break;
4150 		case RTE_FLOW_ACTION_TYPE_RSS:
4151 			rss = actions->conf;
4152 			if (rss == NULL || rss->queue_num == 0)
4153 				return 0;
4154 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4155 			if (conf == NULL || conf->tx_explicit != 0)
4156 				return 0;
4157 			queue_action = 1;
4158 			action_n++;
4159 			break;
4160 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4161 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4162 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4163 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4164 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4165 			split++;
4166 			action_n++;
4167 			break;
4168 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4169 			raw_encap = actions->conf;
4170 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4171 				split++;
4172 			action_n++;
4173 			break;
4174 		default:
4175 			action_n++;
4176 			break;
4177 		}
4178 	}
4179 	if (split && queue_action)
4180 		return action_n;
4181 	return 0;
4182 }
4183 
4184 /* Declare flow create/destroy prototype in advance. */
4185 static uint32_t
4186 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4187 		 const struct rte_flow_attr *attr,
4188 		 const struct rte_flow_item items[],
4189 		 const struct rte_flow_action actions[],
4190 		 bool external, struct rte_flow_error *error);
4191 
4192 static void
4193 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4194 		  uint32_t flow_idx);
4195 
4196 int
4197 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4198 		      struct mlx5_list_entry *entry, void *cb_ctx)
4199 {
4200 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4201 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4202 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4203 
4204 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4205 }
4206 
4207 struct mlx5_list_entry *
4208 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4209 {
4210 	struct rte_eth_dev *dev = tool_ctx;
4211 	struct mlx5_priv *priv = dev->data->dev_private;
4212 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4213 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4214 	struct rte_flow_error *error = ctx->error;
4215 	uint32_t idx = 0;
4216 	int ret;
4217 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4218 	struct rte_flow_attr attr = {
4219 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4220 		.ingress = 1,
4221 	};
4222 	struct mlx5_rte_flow_item_tag tag_spec = {
4223 		.data = mark_id,
4224 	};
4225 	struct rte_flow_item items[] = {
4226 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4227 	};
4228 	struct rte_flow_action_mark ftag = {
4229 		.id = mark_id,
4230 	};
4231 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4232 		.dst = REG_B,
4233 		.src = REG_NON,
4234 	};
4235 	struct rte_flow_action_jump jump = {
4236 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4237 	};
4238 	struct rte_flow_action actions[] = {
4239 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4240 	};
4241 
4242 	/* Fill the register fileds in the flow. */
4243 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4244 	if (ret < 0)
4245 		return NULL;
4246 	tag_spec.id = ret;
4247 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4248 	if (ret < 0)
4249 		return NULL;
4250 	cp_mreg.src = ret;
4251 	/* Provide the full width of FLAG specific value. */
4252 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4253 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4254 	/* Build a new flow. */
4255 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4256 		items[0] = (struct rte_flow_item){
4257 			.type = (enum rte_flow_item_type)
4258 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4259 			.spec = &tag_spec,
4260 		};
4261 		items[1] = (struct rte_flow_item){
4262 			.type = RTE_FLOW_ITEM_TYPE_END,
4263 		};
4264 		actions[0] = (struct rte_flow_action){
4265 			.type = (enum rte_flow_action_type)
4266 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4267 			.conf = &ftag,
4268 		};
4269 		actions[1] = (struct rte_flow_action){
4270 			.type = (enum rte_flow_action_type)
4271 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4272 			.conf = &cp_mreg,
4273 		};
4274 		actions[2] = (struct rte_flow_action){
4275 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4276 			.conf = &jump,
4277 		};
4278 		actions[3] = (struct rte_flow_action){
4279 			.type = RTE_FLOW_ACTION_TYPE_END,
4280 		};
4281 	} else {
4282 		/* Default rule, wildcard match. */
4283 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4284 		items[0] = (struct rte_flow_item){
4285 			.type = RTE_FLOW_ITEM_TYPE_END,
4286 		};
4287 		actions[0] = (struct rte_flow_action){
4288 			.type = (enum rte_flow_action_type)
4289 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4290 			.conf = &cp_mreg,
4291 		};
4292 		actions[1] = (struct rte_flow_action){
4293 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4294 			.conf = &jump,
4295 		};
4296 		actions[2] = (struct rte_flow_action){
4297 			.type = RTE_FLOW_ACTION_TYPE_END,
4298 		};
4299 	}
4300 	/* Build a new entry. */
4301 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4302 	if (!mcp_res) {
4303 		rte_errno = ENOMEM;
4304 		return NULL;
4305 	}
4306 	mcp_res->idx = idx;
4307 	mcp_res->mark_id = mark_id;
4308 	/*
4309 	 * The copy Flows are not included in any list. There
4310 	 * ones are referenced from other Flows and can not
4311 	 * be applied, removed, deleted in ardbitrary order
4312 	 * by list traversing.
4313 	 */
4314 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4315 					&attr, items, actions, false, error);
4316 	if (!mcp_res->rix_flow) {
4317 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4318 		return NULL;
4319 	}
4320 	return &mcp_res->hlist_ent;
4321 }
4322 
4323 struct mlx5_list_entry *
4324 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4325 		      void *cb_ctx __rte_unused)
4326 {
4327 	struct rte_eth_dev *dev = tool_ctx;
4328 	struct mlx5_priv *priv = dev->data->dev_private;
4329 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4330 	uint32_t idx = 0;
4331 
4332 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4333 	if (!mcp_res) {
4334 		rte_errno = ENOMEM;
4335 		return NULL;
4336 	}
4337 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4338 	mcp_res->idx = idx;
4339 	return &mcp_res->hlist_ent;
4340 }
4341 
4342 void
4343 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4344 {
4345 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4346 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4347 	struct rte_eth_dev *dev = tool_ctx;
4348 	struct mlx5_priv *priv = dev->data->dev_private;
4349 
4350 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4351 }
4352 
4353 /**
4354  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4355  *
4356  * As mark_id is unique, if there's already a registered flow for the mark_id,
4357  * return by increasing the reference counter of the resource. Otherwise, create
4358  * the resource (mcp_res) and flow.
4359  *
4360  * Flow looks like,
4361  *   - If ingress port is ANY and reg_c[1] is mark_id,
4362  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4363  *
4364  * For default flow (zero mark_id), flow is like,
4365  *   - If ingress port is ANY,
4366  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4367  *
4368  * @param dev
4369  *   Pointer to Ethernet device.
4370  * @param mark_id
4371  *   ID of MARK action, zero means default flow for META.
4372  * @param[out] error
4373  *   Perform verbose error reporting if not NULL.
4374  *
4375  * @return
4376  *   Associated resource on success, NULL otherwise and rte_errno is set.
4377  */
4378 static struct mlx5_flow_mreg_copy_resource *
4379 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4380 			  struct rte_flow_error *error)
4381 {
4382 	struct mlx5_priv *priv = dev->data->dev_private;
4383 	struct mlx5_list_entry *entry;
4384 	struct mlx5_flow_cb_ctx ctx = {
4385 		.dev = dev,
4386 		.error = error,
4387 		.data = &mark_id,
4388 	};
4389 
4390 	/* Check if already registered. */
4391 	MLX5_ASSERT(priv->mreg_cp_tbl);
4392 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4393 	if (!entry)
4394 		return NULL;
4395 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4396 			    hlist_ent);
4397 }
4398 
4399 void
4400 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4401 {
4402 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4403 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4404 	struct rte_eth_dev *dev = tool_ctx;
4405 	struct mlx5_priv *priv = dev->data->dev_private;
4406 
4407 	MLX5_ASSERT(mcp_res->rix_flow);
4408 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4409 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4410 }
4411 
4412 /**
4413  * Release flow in RX_CP_TBL.
4414  *
4415  * @param dev
4416  *   Pointer to Ethernet device.
4417  * @flow
4418  *   Parent flow for wich copying is provided.
4419  */
4420 static void
4421 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4422 			  struct rte_flow *flow)
4423 {
4424 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4425 	struct mlx5_priv *priv = dev->data->dev_private;
4426 
4427 	if (!flow->rix_mreg_copy)
4428 		return;
4429 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4430 				 flow->rix_mreg_copy);
4431 	if (!mcp_res || !priv->mreg_cp_tbl)
4432 		return;
4433 	MLX5_ASSERT(mcp_res->rix_flow);
4434 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4435 	flow->rix_mreg_copy = 0;
4436 }
4437 
4438 /**
4439  * Remove the default copy action from RX_CP_TBL.
4440  *
4441  * This functions is called in the mlx5_dev_start(). No thread safe
4442  * is guaranteed.
4443  *
4444  * @param dev
4445  *   Pointer to Ethernet device.
4446  */
4447 static void
4448 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4449 {
4450 	struct mlx5_list_entry *entry;
4451 	struct mlx5_priv *priv = dev->data->dev_private;
4452 	struct mlx5_flow_cb_ctx ctx;
4453 	uint32_t mark_id;
4454 
4455 	/* Check if default flow is registered. */
4456 	if (!priv->mreg_cp_tbl)
4457 		return;
4458 	mark_id = MLX5_DEFAULT_COPY_ID;
4459 	ctx.data = &mark_id;
4460 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4461 	if (!entry)
4462 		return;
4463 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4464 }
4465 
4466 /**
4467  * Add the default copy action in in RX_CP_TBL.
4468  *
4469  * This functions is called in the mlx5_dev_start(). No thread safe
4470  * is guaranteed.
4471  *
4472  * @param dev
4473  *   Pointer to Ethernet device.
4474  * @param[out] error
4475  *   Perform verbose error reporting if not NULL.
4476  *
4477  * @return
4478  *   0 for success, negative value otherwise and rte_errno is set.
4479  */
4480 static int
4481 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4482 				  struct rte_flow_error *error)
4483 {
4484 	struct mlx5_priv *priv = dev->data->dev_private;
4485 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4486 	struct mlx5_flow_cb_ctx ctx;
4487 	uint32_t mark_id;
4488 
4489 	/* Check whether extensive metadata feature is engaged. */
4490 	if (!priv->config.dv_flow_en ||
4491 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4492 	    !mlx5_flow_ext_mreg_supported(dev) ||
4493 	    !priv->sh->dv_regc0_mask)
4494 		return 0;
4495 	/*
4496 	 * Add default mreg copy flow may be called multiple time, but
4497 	 * only be called once in stop. Avoid register it twice.
4498 	 */
4499 	mark_id = MLX5_DEFAULT_COPY_ID;
4500 	ctx.data = &mark_id;
4501 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4502 		return 0;
4503 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4504 	if (!mcp_res)
4505 		return -rte_errno;
4506 	return 0;
4507 }
4508 
4509 /**
4510  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4511  *
4512  * All the flow having Q/RSS action should be split by
4513  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4514  * performs the following,
4515  *   - CQE->flow_tag := reg_c[1] (MARK)
4516  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4517  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4518  * but there should be a flow per each MARK ID set by MARK action.
4519  *
4520  * For the aforementioned reason, if there's a MARK action in flow's action
4521  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4522  * the MARK ID to CQE's flow_tag like,
4523  *   - If reg_c[1] is mark_id,
4524  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4525  *
4526  * For SET_META action which stores value in reg_c[0], as the destination is
4527  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4528  * MARK ID means the default flow. The default flow looks like,
4529  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4530  *
4531  * @param dev
4532  *   Pointer to Ethernet device.
4533  * @param flow
4534  *   Pointer to flow structure.
4535  * @param[in] actions
4536  *   Pointer to the list of actions.
4537  * @param[out] error
4538  *   Perform verbose error reporting if not NULL.
4539  *
4540  * @return
4541  *   0 on success, negative value otherwise and rte_errno is set.
4542  */
4543 static int
4544 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4545 			    struct rte_flow *flow,
4546 			    const struct rte_flow_action *actions,
4547 			    struct rte_flow_error *error)
4548 {
4549 	struct mlx5_priv *priv = dev->data->dev_private;
4550 	struct mlx5_dev_config *config = &priv->config;
4551 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4552 	const struct rte_flow_action_mark *mark;
4553 
4554 	/* Check whether extensive metadata feature is engaged. */
4555 	if (!config->dv_flow_en ||
4556 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4557 	    !mlx5_flow_ext_mreg_supported(dev) ||
4558 	    !priv->sh->dv_regc0_mask)
4559 		return 0;
4560 	/* Find MARK action. */
4561 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4562 		switch (actions->type) {
4563 		case RTE_FLOW_ACTION_TYPE_FLAG:
4564 			mcp_res = flow_mreg_add_copy_action
4565 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4566 			if (!mcp_res)
4567 				return -rte_errno;
4568 			flow->rix_mreg_copy = mcp_res->idx;
4569 			return 0;
4570 		case RTE_FLOW_ACTION_TYPE_MARK:
4571 			mark = (const struct rte_flow_action_mark *)
4572 				actions->conf;
4573 			mcp_res =
4574 				flow_mreg_add_copy_action(dev, mark->id, error);
4575 			if (!mcp_res)
4576 				return -rte_errno;
4577 			flow->rix_mreg_copy = mcp_res->idx;
4578 			return 0;
4579 		default:
4580 			break;
4581 		}
4582 	}
4583 	return 0;
4584 }
4585 
4586 #define MLX5_MAX_SPLIT_ACTIONS 24
4587 #define MLX5_MAX_SPLIT_ITEMS 24
4588 
4589 /**
4590  * Split the hairpin flow.
4591  * Since HW can't support encap and push-vlan on Rx, we move these
4592  * actions to Tx.
4593  * If the count action is after the encap then we also
4594  * move the count action. in this case the count will also measure
4595  * the outer bytes.
4596  *
4597  * @param dev
4598  *   Pointer to Ethernet device.
4599  * @param[in] actions
4600  *   Associated actions (list terminated by the END action).
4601  * @param[out] actions_rx
4602  *   Rx flow actions.
4603  * @param[out] actions_tx
4604  *   Tx flow actions..
4605  * @param[out] pattern_tx
4606  *   The pattern items for the Tx flow.
4607  * @param[out] flow_id
4608  *   The flow ID connected to this flow.
4609  *
4610  * @return
4611  *   0 on success.
4612  */
4613 static int
4614 flow_hairpin_split(struct rte_eth_dev *dev,
4615 		   const struct rte_flow_action actions[],
4616 		   struct rte_flow_action actions_rx[],
4617 		   struct rte_flow_action actions_tx[],
4618 		   struct rte_flow_item pattern_tx[],
4619 		   uint32_t flow_id)
4620 {
4621 	const struct rte_flow_action_raw_encap *raw_encap;
4622 	const struct rte_flow_action_raw_decap *raw_decap;
4623 	struct mlx5_rte_flow_action_set_tag *set_tag;
4624 	struct rte_flow_action *tag_action;
4625 	struct mlx5_rte_flow_item_tag *tag_item;
4626 	struct rte_flow_item *item;
4627 	char *addr;
4628 	int encap = 0;
4629 
4630 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4631 		switch (actions->type) {
4632 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4633 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4634 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4635 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4636 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4637 			rte_memcpy(actions_tx, actions,
4638 			       sizeof(struct rte_flow_action));
4639 			actions_tx++;
4640 			break;
4641 		case RTE_FLOW_ACTION_TYPE_COUNT:
4642 			if (encap) {
4643 				rte_memcpy(actions_tx, actions,
4644 					   sizeof(struct rte_flow_action));
4645 				actions_tx++;
4646 			} else {
4647 				rte_memcpy(actions_rx, actions,
4648 					   sizeof(struct rte_flow_action));
4649 				actions_rx++;
4650 			}
4651 			break;
4652 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4653 			raw_encap = actions->conf;
4654 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4655 				memcpy(actions_tx, actions,
4656 				       sizeof(struct rte_flow_action));
4657 				actions_tx++;
4658 				encap = 1;
4659 			} else {
4660 				rte_memcpy(actions_rx, actions,
4661 					   sizeof(struct rte_flow_action));
4662 				actions_rx++;
4663 			}
4664 			break;
4665 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4666 			raw_decap = actions->conf;
4667 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4668 				memcpy(actions_tx, actions,
4669 				       sizeof(struct rte_flow_action));
4670 				actions_tx++;
4671 			} else {
4672 				rte_memcpy(actions_rx, actions,
4673 					   sizeof(struct rte_flow_action));
4674 				actions_rx++;
4675 			}
4676 			break;
4677 		default:
4678 			rte_memcpy(actions_rx, actions,
4679 				   sizeof(struct rte_flow_action));
4680 			actions_rx++;
4681 			break;
4682 		}
4683 	}
4684 	/* Add set meta action and end action for the Rx flow. */
4685 	tag_action = actions_rx;
4686 	tag_action->type = (enum rte_flow_action_type)
4687 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4688 	actions_rx++;
4689 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4690 	actions_rx++;
4691 	set_tag = (void *)actions_rx;
4692 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4693 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4694 		.data = flow_id,
4695 	};
4696 	MLX5_ASSERT(set_tag->id > REG_NON);
4697 	tag_action->conf = set_tag;
4698 	/* Create Tx item list. */
4699 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4700 	addr = (void *)&pattern_tx[2];
4701 	item = pattern_tx;
4702 	item->type = (enum rte_flow_item_type)
4703 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4704 	tag_item = (void *)addr;
4705 	tag_item->data = flow_id;
4706 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4707 	MLX5_ASSERT(set_tag->id > REG_NON);
4708 	item->spec = tag_item;
4709 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4710 	tag_item = (void *)addr;
4711 	tag_item->data = UINT32_MAX;
4712 	tag_item->id = UINT16_MAX;
4713 	item->mask = tag_item;
4714 	item->last = NULL;
4715 	item++;
4716 	item->type = RTE_FLOW_ITEM_TYPE_END;
4717 	return 0;
4718 }
4719 
4720 /**
4721  * The last stage of splitting chain, just creates the subflow
4722  * without any modification.
4723  *
4724  * @param[in] dev
4725  *   Pointer to Ethernet device.
4726  * @param[in] flow
4727  *   Parent flow structure pointer.
4728  * @param[in, out] sub_flow
4729  *   Pointer to return the created subflow, may be NULL.
4730  * @param[in] attr
4731  *   Flow rule attributes.
4732  * @param[in] items
4733  *   Pattern specification (list terminated by the END pattern item).
4734  * @param[in] actions
4735  *   Associated actions (list terminated by the END action).
4736  * @param[in] flow_split_info
4737  *   Pointer to flow split info structure.
4738  * @param[out] error
4739  *   Perform verbose error reporting if not NULL.
4740  * @return
4741  *   0 on success, negative value otherwise
4742  */
4743 static int
4744 flow_create_split_inner(struct rte_eth_dev *dev,
4745 			struct rte_flow *flow,
4746 			struct mlx5_flow **sub_flow,
4747 			const struct rte_flow_attr *attr,
4748 			const struct rte_flow_item items[],
4749 			const struct rte_flow_action actions[],
4750 			struct mlx5_flow_split_info *flow_split_info,
4751 			struct rte_flow_error *error)
4752 {
4753 	struct mlx5_flow *dev_flow;
4754 
4755 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4756 				    flow_split_info->flow_idx, error);
4757 	if (!dev_flow)
4758 		return -rte_errno;
4759 	dev_flow->flow = flow;
4760 	dev_flow->external = flow_split_info->external;
4761 	dev_flow->skip_scale = flow_split_info->skip_scale;
4762 	/* Subflow object was created, we must include one in the list. */
4763 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4764 		      dev_flow->handle, next);
4765 	/*
4766 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4767 	 * flow may need some user defined item layer flags, and pass the
4768 	 * Metadate rxq mark flag to suffix flow as well.
4769 	 */
4770 	if (flow_split_info->prefix_layers)
4771 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4772 	if (flow_split_info->prefix_mark)
4773 		dev_flow->handle->mark = 1;
4774 	if (sub_flow)
4775 		*sub_flow = dev_flow;
4776 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4777 	dev_flow->dv.table_id = flow_split_info->table_id;
4778 #endif
4779 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4780 }
4781 
4782 /**
4783  * Get the sub policy of a meter.
4784  *
4785  * @param[in] dev
4786  *   Pointer to Ethernet device.
4787  * @param[in] flow
4788  *   Parent flow structure pointer.
4789  * @param wks
4790  *   Pointer to thread flow work space.
4791  * @param[in] attr
4792  *   Flow rule attributes.
4793  * @param[in] items
4794  *   Pattern specification (list terminated by the END pattern item).
4795  * @param[out] error
4796  *   Perform verbose error reporting if not NULL.
4797  *
4798  * @return
4799  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4800  */
4801 static struct mlx5_flow_meter_sub_policy *
4802 get_meter_sub_policy(struct rte_eth_dev *dev,
4803 		     struct rte_flow *flow,
4804 		     struct mlx5_flow_workspace *wks,
4805 		     const struct rte_flow_attr *attr,
4806 		     const struct rte_flow_item items[],
4807 		     struct rte_flow_error *error)
4808 {
4809 	struct mlx5_flow_meter_policy *policy;
4810 	struct mlx5_flow_meter_policy *final_policy;
4811 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4812 
4813 	policy = wks->policy;
4814 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4815 	if (final_policy->is_rss || final_policy->is_queue) {
4816 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4817 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4818 		uint32_t i;
4819 
4820 		/*
4821 		 * This is a tmp dev_flow,
4822 		 * no need to register any matcher for it in translate.
4823 		 */
4824 		wks->skip_matcher_reg = 1;
4825 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4826 			struct mlx5_flow dev_flow = {0};
4827 			struct mlx5_flow_handle dev_handle = { {0} };
4828 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4829 
4830 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4831 				const struct rte_flow_action_rss *rss_act =
4832 					final_policy->act_cnt[i].rss->conf;
4833 				struct rte_flow_action rss_actions[2] = {
4834 					[0] = {
4835 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4836 					.conf = rss_act,
4837 					},
4838 					[1] = {
4839 					.type = RTE_FLOW_ACTION_TYPE_END,
4840 					.conf = NULL,
4841 					}
4842 				};
4843 
4844 				dev_flow.handle = &dev_handle;
4845 				dev_flow.ingress = attr->ingress;
4846 				dev_flow.flow = flow;
4847 				dev_flow.external = 0;
4848 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4849 				dev_flow.dv.transfer = attr->transfer;
4850 #endif
4851 				/**
4852 				 * Translate RSS action to get rss hash fields.
4853 				 */
4854 				if (flow_drv_translate(dev, &dev_flow, attr,
4855 						items, rss_actions, error))
4856 					goto exit;
4857 				rss_desc_v[i] = wks->rss_desc;
4858 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4859 				rss_desc_v[i].hash_fields =
4860 						dev_flow.hash_fields;
4861 				rss_desc_v[i].queue_num =
4862 						rss_desc_v[i].hash_fields ?
4863 						rss_desc_v[i].queue_num : 1;
4864 				rss_desc_v[i].tunnel =
4865 						!!(dev_flow.handle->layers &
4866 						   MLX5_FLOW_LAYER_TUNNEL);
4867 				/* Use the RSS queues in the containers. */
4868 				rss_desc_v[i].queue =
4869 					(uint16_t *)(uintptr_t)rss_act->queue;
4870 				rss_desc[i] = &rss_desc_v[i];
4871 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4872 				/* This is queue action. */
4873 				rss_desc_v[i] = wks->rss_desc;
4874 				rss_desc_v[i].key_len = 0;
4875 				rss_desc_v[i].hash_fields = 0;
4876 				rss_desc_v[i].queue =
4877 					&final_policy->act_cnt[i].queue;
4878 				rss_desc_v[i].queue_num = 1;
4879 				rss_desc[i] = &rss_desc_v[i];
4880 			} else {
4881 				rss_desc[i] = NULL;
4882 			}
4883 		}
4884 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4885 						flow, policy, rss_desc);
4886 	} else {
4887 		enum mlx5_meter_domain mtr_domain =
4888 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4889 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4890 						MLX5_MTR_DOMAIN_INGRESS);
4891 		sub_policy = policy->sub_policys[mtr_domain][0];
4892 	}
4893 	if (!sub_policy)
4894 		rte_flow_error_set(error, EINVAL,
4895 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4896 				   "Failed to get meter sub-policy.");
4897 exit:
4898 	return sub_policy;
4899 }
4900 
4901 /**
4902  * Split the meter flow.
4903  *
4904  * As meter flow will split to three sub flow, other than meter
4905  * action, the other actions make sense to only meter accepts
4906  * the packet. If it need to be dropped, no other additional
4907  * actions should be take.
4908  *
4909  * One kind of special action which decapsulates the L3 tunnel
4910  * header will be in the prefix sub flow, as not to take the
4911  * L3 tunnel header into account.
4912  *
4913  * @param[in] dev
4914  *   Pointer to Ethernet device.
4915  * @param[in] flow
4916  *   Parent flow structure pointer.
4917  * @param wks
4918  *   Pointer to thread flow work space.
4919  * @param[in] attr
4920  *   Flow rule attributes.
4921  * @param[in] items
4922  *   Pattern specification (list terminated by the END pattern item).
4923  * @param[out] sfx_items
4924  *   Suffix flow match items (list terminated by the END pattern item).
4925  * @param[in] actions
4926  *   Associated actions (list terminated by the END action).
4927  * @param[out] actions_sfx
4928  *   Suffix flow actions.
4929  * @param[out] actions_pre
4930  *   Prefix flow actions.
4931  * @param[out] mtr_flow_id
4932  *   Pointer to meter flow id.
4933  * @param[out] error
4934  *   Perform verbose error reporting if not NULL.
4935  *
4936  * @return
4937  *   0 on success, a negative errno value otherwise and rte_errno is set.
4938  */
4939 static int
4940 flow_meter_split_prep(struct rte_eth_dev *dev,
4941 		      struct rte_flow *flow,
4942 		      struct mlx5_flow_workspace *wks,
4943 		      const struct rte_flow_attr *attr,
4944 		      const struct rte_flow_item items[],
4945 		      struct rte_flow_item sfx_items[],
4946 		      const struct rte_flow_action actions[],
4947 		      struct rte_flow_action actions_sfx[],
4948 		      struct rte_flow_action actions_pre[],
4949 		      uint32_t *mtr_flow_id,
4950 		      struct rte_flow_error *error)
4951 {
4952 	struct mlx5_priv *priv = dev->data->dev_private;
4953 	struct mlx5_flow_meter_info *fm = wks->fm;
4954 	struct rte_flow_action *tag_action = NULL;
4955 	struct rte_flow_item *tag_item;
4956 	struct mlx5_rte_flow_action_set_tag *set_tag;
4957 	const struct rte_flow_action_raw_encap *raw_encap;
4958 	const struct rte_flow_action_raw_decap *raw_decap;
4959 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4960 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4961 	uint32_t tag_id = 0;
4962 	struct rte_flow_item *vlan_item_dst = NULL;
4963 	const struct rte_flow_item *vlan_item_src = NULL;
4964 	struct rte_flow_action *hw_mtr_action;
4965 	struct rte_flow_action *action_pre_head = NULL;
4966 	int32_t flow_src_port = priv->representor_id;
4967 	bool mtr_first;
4968 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4969 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4970 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4971 	uint32_t flow_id = 0;
4972 	uint32_t flow_id_reversed = 0;
4973 	uint8_t flow_id_bits = 0;
4974 	int shift;
4975 
4976 	/* Prepare the suffix subflow items. */
4977 	tag_item = sfx_items++;
4978 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4979 		struct mlx5_priv *port_priv;
4980 		const struct rte_flow_item_port_id *pid_v;
4981 		int item_type = items->type;
4982 
4983 		switch (item_type) {
4984 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4985 			pid_v = items->spec;
4986 			MLX5_ASSERT(pid_v);
4987 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4988 			if (!port_priv)
4989 				return rte_flow_error_set(error,
4990 						rte_errno,
4991 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4992 						pid_v,
4993 						"Failed to get port info.");
4994 			flow_src_port = port_priv->representor_id;
4995 			if (!fm->def_policy && wks->policy->is_hierarchy &&
4996 			    flow_src_port != priv->representor_id) {
4997 				if (flow_drv_mtr_hierarchy_rule_create(dev,
4998 								flow, fm,
4999 								flow_src_port,
5000 								items,
5001 								error))
5002 					return -rte_errno;
5003 			}
5004 			memcpy(sfx_items, items, sizeof(*sfx_items));
5005 			sfx_items++;
5006 			break;
5007 		case RTE_FLOW_ITEM_TYPE_VLAN:
5008 			/* Determine if copy vlan item below. */
5009 			vlan_item_src = items;
5010 			vlan_item_dst = sfx_items++;
5011 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5012 			break;
5013 		default:
5014 			break;
5015 		}
5016 	}
5017 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5018 	sfx_items++;
5019 	mtr_first = priv->sh->meter_aso_en &&
5020 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5021 	/* For ASO meter, meter must be before tag in TX direction. */
5022 	if (mtr_first) {
5023 		action_pre_head = actions_pre++;
5024 		/* Leave space for tag action. */
5025 		tag_action = actions_pre++;
5026 	}
5027 	/* Prepare the actions for prefix and suffix flow. */
5028 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5029 		struct rte_flow_action *action_cur = NULL;
5030 
5031 		switch (actions->type) {
5032 		case RTE_FLOW_ACTION_TYPE_METER:
5033 			if (mtr_first) {
5034 				action_cur = action_pre_head;
5035 			} else {
5036 				/* Leave space for tag action. */
5037 				tag_action = actions_pre++;
5038 				action_cur = actions_pre++;
5039 			}
5040 			break;
5041 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5042 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5043 			action_cur = actions_pre++;
5044 			break;
5045 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5046 			raw_encap = actions->conf;
5047 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5048 				action_cur = actions_pre++;
5049 			break;
5050 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5051 			raw_decap = actions->conf;
5052 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5053 				action_cur = actions_pre++;
5054 			break;
5055 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5056 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5057 			if (vlan_item_dst && vlan_item_src) {
5058 				memcpy(vlan_item_dst, vlan_item_src,
5059 					sizeof(*vlan_item_dst));
5060 				/*
5061 				 * Convert to internal match item, it is used
5062 				 * for vlan push and set vid.
5063 				 */
5064 				vlan_item_dst->type = (enum rte_flow_item_type)
5065 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5066 			}
5067 			break;
5068 		default:
5069 			break;
5070 		}
5071 		if (!action_cur)
5072 			action_cur = (fm->def_policy) ?
5073 					actions_sfx++ : actions_pre++;
5074 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5075 	}
5076 	/* Add end action to the actions. */
5077 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5078 	if (priv->sh->meter_aso_en) {
5079 		/**
5080 		 * For ASO meter, need to add an extra jump action explicitly,
5081 		 * to jump from meter to policer table.
5082 		 */
5083 		struct mlx5_flow_meter_sub_policy *sub_policy;
5084 		struct mlx5_flow_tbl_data_entry *tbl_data;
5085 
5086 		if (!fm->def_policy) {
5087 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5088 							  attr, items, error);
5089 			if (!sub_policy)
5090 				return -rte_errno;
5091 		} else {
5092 			enum mlx5_meter_domain mtr_domain =
5093 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5094 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5095 						MLX5_MTR_DOMAIN_INGRESS);
5096 
5097 			sub_policy =
5098 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5099 		}
5100 		tbl_data = container_of(sub_policy->tbl_rsc,
5101 					struct mlx5_flow_tbl_data_entry, tbl);
5102 		hw_mtr_action = actions_pre++;
5103 		hw_mtr_action->type = (enum rte_flow_action_type)
5104 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5105 		hw_mtr_action->conf = tbl_data->jump.action;
5106 	}
5107 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5108 	actions_pre++;
5109 	if (!tag_action)
5110 		return rte_flow_error_set(error, ENOMEM,
5111 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5112 					  NULL, "No tag action space.");
5113 	if (!mtr_flow_id) {
5114 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5115 		goto exit;
5116 	}
5117 	/* Only default-policy Meter creates mtr flow id. */
5118 	if (fm->def_policy) {
5119 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5120 		if (!tag_id)
5121 			return rte_flow_error_set(error, ENOMEM,
5122 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5123 					"Failed to allocate meter flow id.");
5124 		flow_id = tag_id - 1;
5125 		flow_id_bits = (!flow_id) ? 1 :
5126 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5127 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5128 		    mtr_reg_bits) {
5129 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5130 			return rte_flow_error_set(error, EINVAL,
5131 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5132 					"Meter flow id exceeds max limit.");
5133 		}
5134 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5135 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5136 	}
5137 	/* Build tag actions and items for meter_id/meter flow_id. */
5138 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5139 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5140 	tag_item_mask = tag_item_spec + 1;
5141 	/* Both flow_id and meter_id share the same register. */
5142 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5143 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5144 							    0, error),
5145 		.offset = mtr_id_offset,
5146 		.length = mtr_reg_bits,
5147 		.data = flow->meter,
5148 	};
5149 	/*
5150 	 * The color Reg bits used by flow_id are growing from
5151 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5152 	 */
5153 	for (shift = 0; shift < flow_id_bits; shift++)
5154 		flow_id_reversed = (flow_id_reversed << 1) |
5155 				((flow_id >> shift) & 0x1);
5156 	set_tag->data |=
5157 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5158 	tag_item_spec->id = set_tag->id;
5159 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5160 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5161 	tag_action->type = (enum rte_flow_action_type)
5162 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5163 	tag_action->conf = set_tag;
5164 	tag_item->type = (enum rte_flow_item_type)
5165 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5166 	tag_item->spec = tag_item_spec;
5167 	tag_item->last = NULL;
5168 	tag_item->mask = tag_item_mask;
5169 exit:
5170 	if (mtr_flow_id)
5171 		*mtr_flow_id = tag_id;
5172 	return 0;
5173 }
5174 
5175 /**
5176  * Split action list having QUEUE/RSS for metadata register copy.
5177  *
5178  * Once Q/RSS action is detected in user's action list, the flow action
5179  * should be split in order to copy metadata registers, which will happen in
5180  * RX_CP_TBL like,
5181  *   - CQE->flow_tag := reg_c[1] (MARK)
5182  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5183  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5184  * This is because the last action of each flow must be a terminal action
5185  * (QUEUE, RSS or DROP).
5186  *
5187  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5188  * stored and kept in the mlx5_flow structure per each sub_flow.
5189  *
5190  * The Q/RSS action is replaced with,
5191  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5192  * And the following JUMP action is added at the end,
5193  *   - JUMP, to RX_CP_TBL.
5194  *
5195  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5196  * flow_create_split_metadata() routine. The flow will look like,
5197  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5198  *
5199  * @param dev
5200  *   Pointer to Ethernet device.
5201  * @param[out] split_actions
5202  *   Pointer to store split actions to jump to CP_TBL.
5203  * @param[in] actions
5204  *   Pointer to the list of original flow actions.
5205  * @param[in] qrss
5206  *   Pointer to the Q/RSS action.
5207  * @param[in] actions_n
5208  *   Number of original actions.
5209  * @param[out] error
5210  *   Perform verbose error reporting if not NULL.
5211  *
5212  * @return
5213  *   non-zero unique flow_id on success, otherwise 0 and
5214  *   error/rte_error are set.
5215  */
5216 static uint32_t
5217 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5218 			  struct rte_flow_action *split_actions,
5219 			  const struct rte_flow_action *actions,
5220 			  const struct rte_flow_action *qrss,
5221 			  int actions_n, struct rte_flow_error *error)
5222 {
5223 	struct mlx5_priv *priv = dev->data->dev_private;
5224 	struct mlx5_rte_flow_action_set_tag *set_tag;
5225 	struct rte_flow_action_jump *jump;
5226 	const int qrss_idx = qrss - actions;
5227 	uint32_t flow_id = 0;
5228 	int ret = 0;
5229 
5230 	/*
5231 	 * Given actions will be split
5232 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5233 	 * - Add jump to mreg CP_TBL.
5234 	 * As a result, there will be one more action.
5235 	 */
5236 	++actions_n;
5237 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5238 	set_tag = (void *)(split_actions + actions_n);
5239 	/*
5240 	 * If tag action is not set to void(it means we are not the meter
5241 	 * suffix flow), add the tag action. Since meter suffix flow already
5242 	 * has the tag added.
5243 	 */
5244 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5245 		/*
5246 		 * Allocate the new subflow ID. This one is unique within
5247 		 * device and not shared with representors. Otherwise,
5248 		 * we would have to resolve multi-thread access synch
5249 		 * issue. Each flow on the shared device is appended
5250 		 * with source vport identifier, so the resulting
5251 		 * flows will be unique in the shared (by master and
5252 		 * representors) domain even if they have coinciding
5253 		 * IDs.
5254 		 */
5255 		mlx5_ipool_malloc(priv->sh->ipool
5256 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5257 		if (!flow_id)
5258 			return rte_flow_error_set(error, ENOMEM,
5259 						  RTE_FLOW_ERROR_TYPE_ACTION,
5260 						  NULL, "can't allocate id "
5261 						  "for split Q/RSS subflow");
5262 		/* Internal SET_TAG action to set flow ID. */
5263 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5264 			.data = flow_id,
5265 		};
5266 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5267 		if (ret < 0)
5268 			return ret;
5269 		set_tag->id = ret;
5270 		/* Construct new actions array. */
5271 		/* Replace QUEUE/RSS action. */
5272 		split_actions[qrss_idx] = (struct rte_flow_action){
5273 			.type = (enum rte_flow_action_type)
5274 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5275 			.conf = set_tag,
5276 		};
5277 	}
5278 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5279 	jump = (void *)(set_tag + 1);
5280 	*jump = (struct rte_flow_action_jump){
5281 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5282 	};
5283 	split_actions[actions_n - 2] = (struct rte_flow_action){
5284 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5285 		.conf = jump,
5286 	};
5287 	split_actions[actions_n - 1] = (struct rte_flow_action){
5288 		.type = RTE_FLOW_ACTION_TYPE_END,
5289 	};
5290 	return flow_id;
5291 }
5292 
5293 /**
5294  * Extend the given action list for Tx metadata copy.
5295  *
5296  * Copy the given action list to the ext_actions and add flow metadata register
5297  * copy action in order to copy reg_a set by WQE to reg_c[0].
5298  *
5299  * @param[out] ext_actions
5300  *   Pointer to the extended action list.
5301  * @param[in] actions
5302  *   Pointer to the list of actions.
5303  * @param[in] actions_n
5304  *   Number of actions in the list.
5305  * @param[out] error
5306  *   Perform verbose error reporting if not NULL.
5307  * @param[in] encap_idx
5308  *   The encap action inndex.
5309  *
5310  * @return
5311  *   0 on success, negative value otherwise
5312  */
5313 static int
5314 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5315 		       struct rte_flow_action *ext_actions,
5316 		       const struct rte_flow_action *actions,
5317 		       int actions_n, struct rte_flow_error *error,
5318 		       int encap_idx)
5319 {
5320 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5321 		(struct mlx5_flow_action_copy_mreg *)
5322 			(ext_actions + actions_n + 1);
5323 	int ret;
5324 
5325 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5326 	if (ret < 0)
5327 		return ret;
5328 	cp_mreg->dst = ret;
5329 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5330 	if (ret < 0)
5331 		return ret;
5332 	cp_mreg->src = ret;
5333 	if (encap_idx != 0)
5334 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5335 	if (encap_idx == actions_n - 1) {
5336 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5337 			.type = (enum rte_flow_action_type)
5338 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5339 			.conf = cp_mreg,
5340 		};
5341 		ext_actions[actions_n] = (struct rte_flow_action){
5342 			.type = RTE_FLOW_ACTION_TYPE_END,
5343 		};
5344 	} else {
5345 		ext_actions[encap_idx] = (struct rte_flow_action){
5346 			.type = (enum rte_flow_action_type)
5347 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5348 			.conf = cp_mreg,
5349 		};
5350 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5351 				sizeof(*ext_actions) * (actions_n - encap_idx));
5352 	}
5353 	return 0;
5354 }
5355 
5356 /**
5357  * Check the match action from the action list.
5358  *
5359  * @param[in] actions
5360  *   Pointer to the list of actions.
5361  * @param[in] attr
5362  *   Flow rule attributes.
5363  * @param[in] action
5364  *   The action to be check if exist.
5365  * @param[out] match_action_pos
5366  *   Pointer to the position of the matched action if exists, otherwise is -1.
5367  * @param[out] qrss_action_pos
5368  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5369  * @param[out] modify_after_mirror
5370  *   Pointer to the flag of modify action after FDB mirroring.
5371  *
5372  * @return
5373  *   > 0 the total number of actions.
5374  *   0 if not found match action in action list.
5375  */
5376 static int
5377 flow_check_match_action(const struct rte_flow_action actions[],
5378 			const struct rte_flow_attr *attr,
5379 			enum rte_flow_action_type action,
5380 			int *match_action_pos, int *qrss_action_pos,
5381 			int *modify_after_mirror)
5382 {
5383 	const struct rte_flow_action_sample *sample;
5384 	const struct rte_flow_action_raw_decap *decap;
5385 	int actions_n = 0;
5386 	uint32_t ratio = 0;
5387 	int sub_type = 0;
5388 	int flag = 0;
5389 	int fdb_mirror = 0;
5390 
5391 	*match_action_pos = -1;
5392 	*qrss_action_pos = -1;
5393 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5394 		if (actions->type == action) {
5395 			flag = 1;
5396 			*match_action_pos = actions_n;
5397 		}
5398 		switch (actions->type) {
5399 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5400 		case RTE_FLOW_ACTION_TYPE_RSS:
5401 			*qrss_action_pos = actions_n;
5402 			break;
5403 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5404 			sample = actions->conf;
5405 			ratio = sample->ratio;
5406 			sub_type = ((const struct rte_flow_action *)
5407 					(sample->actions))->type;
5408 			if (ratio == 1 && attr->transfer)
5409 				fdb_mirror = 1;
5410 			break;
5411 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5412 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5413 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5414 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5415 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5416 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5417 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5418 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5419 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5420 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5421 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5422 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5423 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5424 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5425 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5426 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5427 		case RTE_FLOW_ACTION_TYPE_FLAG:
5428 		case RTE_FLOW_ACTION_TYPE_MARK:
5429 		case RTE_FLOW_ACTION_TYPE_SET_META:
5430 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5431 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5432 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5433 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5434 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5435 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5436 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5437 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5438 		case RTE_FLOW_ACTION_TYPE_METER:
5439 			if (fdb_mirror)
5440 				*modify_after_mirror = 1;
5441 			break;
5442 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5443 			decap = actions->conf;
5444 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5445 				;
5446 			actions_n++;
5447 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5448 				const struct rte_flow_action_raw_encap *encap =
5449 								actions->conf;
5450 				if (decap->size <=
5451 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5452 				    encap->size >
5453 					MLX5_ENCAPSULATION_DECISION_SIZE)
5454 					/* L3 encap. */
5455 					break;
5456 			}
5457 			if (fdb_mirror)
5458 				*modify_after_mirror = 1;
5459 			break;
5460 		default:
5461 			break;
5462 		}
5463 		actions_n++;
5464 	}
5465 	if (flag && fdb_mirror && !*modify_after_mirror) {
5466 		/* FDB mirroring uses the destination array to implement
5467 		 * instead of FLOW_SAMPLER object.
5468 		 */
5469 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5470 			flag = 0;
5471 	}
5472 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5473 	return flag ? actions_n + 1 : 0;
5474 }
5475 
5476 #define SAMPLE_SUFFIX_ITEM 2
5477 
5478 /**
5479  * Split the sample flow.
5480  *
5481  * As sample flow will split to two sub flow, sample flow with
5482  * sample action, the other actions will move to new suffix flow.
5483  *
5484  * Also add unique tag id with tag action in the sample flow,
5485  * the same tag id will be as match in the suffix flow.
5486  *
5487  * @param dev
5488  *   Pointer to Ethernet device.
5489  * @param[in] add_tag
5490  *   Add extra tag action flag.
5491  * @param[out] sfx_items
5492  *   Suffix flow match items (list terminated by the END pattern item).
5493  * @param[in] actions
5494  *   Associated actions (list terminated by the END action).
5495  * @param[out] actions_sfx
5496  *   Suffix flow actions.
5497  * @param[out] actions_pre
5498  *   Prefix flow actions.
5499  * @param[in] actions_n
5500  *  The total number of actions.
5501  * @param[in] sample_action_pos
5502  *   The sample action position.
5503  * @param[in] qrss_action_pos
5504  *   The Queue/RSS action position.
5505  * @param[in] jump_table
5506  *   Add extra jump action flag.
5507  * @param[out] error
5508  *   Perform verbose error reporting if not NULL.
5509  *
5510  * @return
5511  *   0 on success, or unique flow_id, a negative errno value
5512  *   otherwise and rte_errno is set.
5513  */
5514 static int
5515 flow_sample_split_prep(struct rte_eth_dev *dev,
5516 		       int add_tag,
5517 		       struct rte_flow_item sfx_items[],
5518 		       const struct rte_flow_action actions[],
5519 		       struct rte_flow_action actions_sfx[],
5520 		       struct rte_flow_action actions_pre[],
5521 		       int actions_n,
5522 		       int sample_action_pos,
5523 		       int qrss_action_pos,
5524 		       int jump_table,
5525 		       struct rte_flow_error *error)
5526 {
5527 	struct mlx5_priv *priv = dev->data->dev_private;
5528 	struct mlx5_rte_flow_action_set_tag *set_tag;
5529 	struct mlx5_rte_flow_item_tag *tag_spec;
5530 	struct mlx5_rte_flow_item_tag *tag_mask;
5531 	struct rte_flow_action_jump *jump_action;
5532 	uint32_t tag_id = 0;
5533 	int index;
5534 	int append_index = 0;
5535 	int ret;
5536 
5537 	if (sample_action_pos < 0)
5538 		return rte_flow_error_set(error, EINVAL,
5539 					  RTE_FLOW_ERROR_TYPE_ACTION,
5540 					  NULL, "invalid position of sample "
5541 					  "action in list");
5542 	/* Prepare the actions for prefix and suffix flow. */
5543 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5544 		index = qrss_action_pos;
5545 		/* Put the preceding the Queue/RSS action into prefix flow. */
5546 		if (index != 0)
5547 			memcpy(actions_pre, actions,
5548 			       sizeof(struct rte_flow_action) * index);
5549 		/* Put others preceding the sample action into prefix flow. */
5550 		if (sample_action_pos > index + 1)
5551 			memcpy(actions_pre + index, actions + index + 1,
5552 			       sizeof(struct rte_flow_action) *
5553 			       (sample_action_pos - index - 1));
5554 		index = sample_action_pos - 1;
5555 		/* Put Queue/RSS action into Suffix flow. */
5556 		memcpy(actions_sfx, actions + qrss_action_pos,
5557 		       sizeof(struct rte_flow_action));
5558 		actions_sfx++;
5559 	} else {
5560 		index = sample_action_pos;
5561 		if (index != 0)
5562 			memcpy(actions_pre, actions,
5563 			       sizeof(struct rte_flow_action) * index);
5564 	}
5565 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5566 	 * For CX6DX and above, metadata registers Cx preserve their value,
5567 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5568 	 */
5569 	if (add_tag) {
5570 		/* Prepare the prefix tag action. */
5571 		append_index++;
5572 		set_tag = (void *)(actions_pre + actions_n + append_index);
5573 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5574 		if (ret < 0)
5575 			return ret;
5576 		mlx5_ipool_malloc(priv->sh->ipool
5577 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5578 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5579 			.id = ret,
5580 			.data = tag_id,
5581 		};
5582 		/* Prepare the suffix subflow items. */
5583 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5584 		tag_spec->data = tag_id;
5585 		tag_spec->id = set_tag->id;
5586 		tag_mask = tag_spec + 1;
5587 		tag_mask->data = UINT32_MAX;
5588 		sfx_items[0] = (struct rte_flow_item){
5589 			.type = (enum rte_flow_item_type)
5590 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5591 			.spec = tag_spec,
5592 			.last = NULL,
5593 			.mask = tag_mask,
5594 		};
5595 		sfx_items[1] = (struct rte_flow_item){
5596 			.type = (enum rte_flow_item_type)
5597 				RTE_FLOW_ITEM_TYPE_END,
5598 		};
5599 		/* Prepare the tag action in prefix subflow. */
5600 		actions_pre[index++] =
5601 			(struct rte_flow_action){
5602 			.type = (enum rte_flow_action_type)
5603 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5604 			.conf = set_tag,
5605 		};
5606 	}
5607 	memcpy(actions_pre + index, actions + sample_action_pos,
5608 	       sizeof(struct rte_flow_action));
5609 	index += 1;
5610 	/* For the modify action after the sample action in E-Switch mirroring,
5611 	 * Add the extra jump action in prefix subflow and jump into the next
5612 	 * table, then do the modify action in the new table.
5613 	 */
5614 	if (jump_table) {
5615 		/* Prepare the prefix jump action. */
5616 		append_index++;
5617 		jump_action = (void *)(actions_pre + actions_n + append_index);
5618 		jump_action->group = jump_table;
5619 		actions_pre[index++] =
5620 			(struct rte_flow_action){
5621 			.type = (enum rte_flow_action_type)
5622 				RTE_FLOW_ACTION_TYPE_JUMP,
5623 			.conf = jump_action,
5624 		};
5625 	}
5626 	actions_pre[index] = (struct rte_flow_action){
5627 		.type = (enum rte_flow_action_type)
5628 			RTE_FLOW_ACTION_TYPE_END,
5629 	};
5630 	/* Put the actions after sample into Suffix flow. */
5631 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5632 	       sizeof(struct rte_flow_action) *
5633 	       (actions_n - sample_action_pos - 1));
5634 	return tag_id;
5635 }
5636 
5637 /**
5638  * The splitting for metadata feature.
5639  *
5640  * - Q/RSS action on NIC Rx should be split in order to pass by
5641  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5642  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5643  *
5644  * - All the actions on NIC Tx should have a mreg copy action to
5645  *   copy reg_a from WQE to reg_c[0].
5646  *
5647  * @param dev
5648  *   Pointer to Ethernet device.
5649  * @param[in] flow
5650  *   Parent flow structure pointer.
5651  * @param[in] attr
5652  *   Flow rule attributes.
5653  * @param[in] items
5654  *   Pattern specification (list terminated by the END pattern item).
5655  * @param[in] actions
5656  *   Associated actions (list terminated by the END action).
5657  * @param[in] flow_split_info
5658  *   Pointer to flow split info structure.
5659  * @param[out] error
5660  *   Perform verbose error reporting if not NULL.
5661  * @return
5662  *   0 on success, negative value otherwise
5663  */
5664 static int
5665 flow_create_split_metadata(struct rte_eth_dev *dev,
5666 			   struct rte_flow *flow,
5667 			   const struct rte_flow_attr *attr,
5668 			   const struct rte_flow_item items[],
5669 			   const struct rte_flow_action actions[],
5670 			   struct mlx5_flow_split_info *flow_split_info,
5671 			   struct rte_flow_error *error)
5672 {
5673 	struct mlx5_priv *priv = dev->data->dev_private;
5674 	struct mlx5_dev_config *config = &priv->config;
5675 	const struct rte_flow_action *qrss = NULL;
5676 	struct rte_flow_action *ext_actions = NULL;
5677 	struct mlx5_flow *dev_flow = NULL;
5678 	uint32_t qrss_id = 0;
5679 	int mtr_sfx = 0;
5680 	size_t act_size;
5681 	int actions_n;
5682 	int encap_idx;
5683 	int ret;
5684 
5685 	/* Check whether extensive metadata feature is engaged. */
5686 	if (!config->dv_flow_en ||
5687 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5688 	    !mlx5_flow_ext_mreg_supported(dev))
5689 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5690 					       actions, flow_split_info, error);
5691 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5692 							   &encap_idx);
5693 	if (qrss) {
5694 		/* Exclude hairpin flows from splitting. */
5695 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5696 			const struct rte_flow_action_queue *queue;
5697 
5698 			queue = qrss->conf;
5699 			if (mlx5_rxq_get_type(dev, queue->index) ==
5700 			    MLX5_RXQ_TYPE_HAIRPIN)
5701 				qrss = NULL;
5702 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5703 			const struct rte_flow_action_rss *rss;
5704 
5705 			rss = qrss->conf;
5706 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5707 			    MLX5_RXQ_TYPE_HAIRPIN)
5708 				qrss = NULL;
5709 		}
5710 	}
5711 	if (qrss) {
5712 		/* Check if it is in meter suffix table. */
5713 		mtr_sfx = attr->group == (attr->transfer ?
5714 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5715 			  MLX5_FLOW_TABLE_LEVEL_METER);
5716 		/*
5717 		 * Q/RSS action on NIC Rx should be split in order to pass by
5718 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5719 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5720 		 */
5721 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5722 			   sizeof(struct rte_flow_action_set_tag) +
5723 			   sizeof(struct rte_flow_action_jump);
5724 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5725 					  SOCKET_ID_ANY);
5726 		if (!ext_actions)
5727 			return rte_flow_error_set(error, ENOMEM,
5728 						  RTE_FLOW_ERROR_TYPE_ACTION,
5729 						  NULL, "no memory to split "
5730 						  "metadata flow");
5731 		/*
5732 		 * If we are the suffix flow of meter, tag already exist.
5733 		 * Set the tag action to void.
5734 		 */
5735 		if (mtr_sfx)
5736 			ext_actions[qrss - actions].type =
5737 						RTE_FLOW_ACTION_TYPE_VOID;
5738 		else
5739 			ext_actions[qrss - actions].type =
5740 						(enum rte_flow_action_type)
5741 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5742 		/*
5743 		 * Create the new actions list with removed Q/RSS action
5744 		 * and appended set tag and jump to register copy table
5745 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5746 		 * in advance, because it is needed for set tag action.
5747 		 */
5748 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5749 						    qrss, actions_n, error);
5750 		if (!mtr_sfx && !qrss_id) {
5751 			ret = -rte_errno;
5752 			goto exit;
5753 		}
5754 	} else if (attr->egress && !attr->transfer) {
5755 		/*
5756 		 * All the actions on NIC Tx should have a metadata register
5757 		 * copy action to copy reg_a from WQE to reg_c[meta]
5758 		 */
5759 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5760 			   sizeof(struct mlx5_flow_action_copy_mreg);
5761 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5762 					  SOCKET_ID_ANY);
5763 		if (!ext_actions)
5764 			return rte_flow_error_set(error, ENOMEM,
5765 						  RTE_FLOW_ERROR_TYPE_ACTION,
5766 						  NULL, "no memory to split "
5767 						  "metadata flow");
5768 		/* Create the action list appended with copy register. */
5769 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5770 					     actions_n, error, encap_idx);
5771 		if (ret < 0)
5772 			goto exit;
5773 	}
5774 	/* Add the unmodified original or prefix subflow. */
5775 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5776 				      items, ext_actions ? ext_actions :
5777 				      actions, flow_split_info, error);
5778 	if (ret < 0)
5779 		goto exit;
5780 	MLX5_ASSERT(dev_flow);
5781 	if (qrss) {
5782 		const struct rte_flow_attr q_attr = {
5783 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5784 			.ingress = 1,
5785 		};
5786 		/* Internal PMD action to set register. */
5787 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5788 			.data = qrss_id,
5789 			.id = REG_NON,
5790 		};
5791 		struct rte_flow_item q_items[] = {
5792 			{
5793 				.type = (enum rte_flow_item_type)
5794 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5795 				.spec = &q_tag_spec,
5796 				.last = NULL,
5797 				.mask = NULL,
5798 			},
5799 			{
5800 				.type = RTE_FLOW_ITEM_TYPE_END,
5801 			},
5802 		};
5803 		struct rte_flow_action q_actions[] = {
5804 			{
5805 				.type = qrss->type,
5806 				.conf = qrss->conf,
5807 			},
5808 			{
5809 				.type = RTE_FLOW_ACTION_TYPE_END,
5810 			},
5811 		};
5812 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5813 
5814 		/*
5815 		 * Configure the tag item only if there is no meter subflow.
5816 		 * Since tag is already marked in the meter suffix subflow
5817 		 * we can just use the meter suffix items as is.
5818 		 */
5819 		if (qrss_id) {
5820 			/* Not meter subflow. */
5821 			MLX5_ASSERT(!mtr_sfx);
5822 			/*
5823 			 * Put unique id in prefix flow due to it is destroyed
5824 			 * after suffix flow and id will be freed after there
5825 			 * is no actual flows with this id and identifier
5826 			 * reallocation becomes possible (for example, for
5827 			 * other flows in other threads).
5828 			 */
5829 			dev_flow->handle->split_flow_id = qrss_id;
5830 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5831 						   error);
5832 			if (ret < 0)
5833 				goto exit;
5834 			q_tag_spec.id = ret;
5835 		}
5836 		dev_flow = NULL;
5837 		/* Add suffix subflow to execute Q/RSS. */
5838 		flow_split_info->prefix_layers = layers;
5839 		flow_split_info->prefix_mark = 0;
5840 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5841 					      &q_attr, mtr_sfx ? items :
5842 					      q_items, q_actions,
5843 					      flow_split_info, error);
5844 		if (ret < 0)
5845 			goto exit;
5846 		/* qrss ID should be freed if failed. */
5847 		qrss_id = 0;
5848 		MLX5_ASSERT(dev_flow);
5849 	}
5850 
5851 exit:
5852 	/*
5853 	 * We do not destroy the partially created sub_flows in case of error.
5854 	 * These ones are included into parent flow list and will be destroyed
5855 	 * by flow_drv_destroy.
5856 	 */
5857 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5858 			qrss_id);
5859 	mlx5_free(ext_actions);
5860 	return ret;
5861 }
5862 
5863 /**
5864  * Create meter internal drop flow with the original pattern.
5865  *
5866  * @param dev
5867  *   Pointer to Ethernet device.
5868  * @param[in] flow
5869  *   Parent flow structure pointer.
5870  * @param[in] attr
5871  *   Flow rule attributes.
5872  * @param[in] items
5873  *   Pattern specification (list terminated by the END pattern item).
5874  * @param[in] flow_split_info
5875  *   Pointer to flow split info structure.
5876  * @param[in] fm
5877  *   Pointer to flow meter structure.
5878  * @param[out] error
5879  *   Perform verbose error reporting if not NULL.
5880  * @return
5881  *   0 on success, negative value otherwise
5882  */
5883 static uint32_t
5884 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5885 			struct rte_flow *flow,
5886 			const struct rte_flow_attr *attr,
5887 			const struct rte_flow_item items[],
5888 			struct mlx5_flow_split_info *flow_split_info,
5889 			struct mlx5_flow_meter_info *fm,
5890 			struct rte_flow_error *error)
5891 {
5892 	struct mlx5_flow *dev_flow = NULL;
5893 	struct rte_flow_attr drop_attr = *attr;
5894 	struct rte_flow_action drop_actions[3];
5895 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5896 
5897 	MLX5_ASSERT(fm->drop_cnt);
5898 	drop_actions[0].type =
5899 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5900 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5901 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5902 	drop_actions[1].conf = NULL;
5903 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5904 	drop_actions[2].conf = NULL;
5905 	drop_split_info.external = false;
5906 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5907 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5908 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5909 	return flow_create_split_inner(dev, flow, &dev_flow,
5910 				&drop_attr, items, drop_actions,
5911 				&drop_split_info, error);
5912 }
5913 
5914 /**
5915  * The splitting for meter feature.
5916  *
5917  * - The meter flow will be split to two flows as prefix and
5918  *   suffix flow. The packets make sense only it pass the prefix
5919  *   meter action.
5920  *
5921  * - Reg_C_5 is used for the packet to match betweend prefix and
5922  *   suffix flow.
5923  *
5924  * @param dev
5925  *   Pointer to Ethernet device.
5926  * @param[in] flow
5927  *   Parent flow structure pointer.
5928  * @param[in] attr
5929  *   Flow rule attributes.
5930  * @param[in] items
5931  *   Pattern specification (list terminated by the END pattern item).
5932  * @param[in] actions
5933  *   Associated actions (list terminated by the END action).
5934  * @param[in] flow_split_info
5935  *   Pointer to flow split info structure.
5936  * @param[out] error
5937  *   Perform verbose error reporting if not NULL.
5938  * @return
5939  *   0 on success, negative value otherwise
5940  */
5941 static int
5942 flow_create_split_meter(struct rte_eth_dev *dev,
5943 			struct rte_flow *flow,
5944 			const struct rte_flow_attr *attr,
5945 			const struct rte_flow_item items[],
5946 			const struct rte_flow_action actions[],
5947 			struct mlx5_flow_split_info *flow_split_info,
5948 			struct rte_flow_error *error)
5949 {
5950 	struct mlx5_priv *priv = dev->data->dev_private;
5951 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5952 	struct rte_flow_action *sfx_actions = NULL;
5953 	struct rte_flow_action *pre_actions = NULL;
5954 	struct rte_flow_item *sfx_items = NULL;
5955 	struct mlx5_flow *dev_flow = NULL;
5956 	struct rte_flow_attr sfx_attr = *attr;
5957 	struct mlx5_flow_meter_info *fm = NULL;
5958 	uint8_t skip_scale_restore;
5959 	bool has_mtr = false;
5960 	bool has_modify = false;
5961 	bool set_mtr_reg = true;
5962 	bool is_mtr_hierarchy = false;
5963 	uint32_t meter_id = 0;
5964 	uint32_t mtr_idx = 0;
5965 	uint32_t mtr_flow_id = 0;
5966 	size_t act_size;
5967 	size_t item_size;
5968 	int actions_n = 0;
5969 	int ret = 0;
5970 
5971 	if (priv->mtr_en)
5972 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5973 						    &has_modify, &meter_id);
5974 	if (has_mtr) {
5975 		if (flow->meter) {
5976 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5977 			if (!fm)
5978 				return rte_flow_error_set(error, EINVAL,
5979 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5980 						NULL, "Meter not found.");
5981 		} else {
5982 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5983 			if (!fm)
5984 				return rte_flow_error_set(error, EINVAL,
5985 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5986 						NULL, "Meter not found.");
5987 			ret = mlx5_flow_meter_attach(priv, fm,
5988 						     &sfx_attr, error);
5989 			if (ret)
5990 				return -rte_errno;
5991 			flow->meter = mtr_idx;
5992 		}
5993 		MLX5_ASSERT(wks);
5994 		wks->fm = fm;
5995 		if (!fm->def_policy) {
5996 			wks->policy = mlx5_flow_meter_policy_find(dev,
5997 								  fm->policy_id,
5998 								  NULL);
5999 			MLX5_ASSERT(wks->policy);
6000 			if (wks->policy->is_hierarchy) {
6001 				wks->final_policy =
6002 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6003 								wks->policy);
6004 				if (!wks->final_policy)
6005 					return rte_flow_error_set(error,
6006 					EINVAL,
6007 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6008 				"Failed to find terminal policy of hierarchy.");
6009 				is_mtr_hierarchy = true;
6010 			}
6011 		}
6012 		/*
6013 		 * If it isn't default-policy Meter, and
6014 		 * 1. There's no action in flow to change
6015 		 *    packet (modify/encap/decap etc.), OR
6016 		 * 2. No drop count needed for this meter.
6017 		 * 3. It's not meter hierarchy.
6018 		 * Then no need to use regC to save meter id anymore.
6019 		 */
6020 		if (!fm->def_policy && !is_mtr_hierarchy &&
6021 		    (!has_modify || !fm->drop_cnt))
6022 			set_mtr_reg = false;
6023 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6024 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6025 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6026 		/* Suffix items: tag, vlan, port id, end. */
6027 #define METER_SUFFIX_ITEM 4
6028 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6029 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6030 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6031 					  0, SOCKET_ID_ANY);
6032 		if (!sfx_actions)
6033 			return rte_flow_error_set(error, ENOMEM,
6034 						  RTE_FLOW_ERROR_TYPE_ACTION,
6035 						  NULL, "no memory to split "
6036 						  "meter flow");
6037 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6038 			     act_size);
6039 		/* There's no suffix flow for meter of non-default policy. */
6040 		if (!fm->def_policy)
6041 			pre_actions = sfx_actions + 1;
6042 		else
6043 			pre_actions = sfx_actions + actions_n;
6044 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6045 					    items, sfx_items, actions,
6046 					    sfx_actions, pre_actions,
6047 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6048 					    error);
6049 		if (ret) {
6050 			ret = -rte_errno;
6051 			goto exit;
6052 		}
6053 		/* Add the prefix subflow. */
6054 		flow_split_info->prefix_mark = 0;
6055 		skip_scale_restore = flow_split_info->skip_scale;
6056 		flow_split_info->skip_scale |=
6057 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6058 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6059 					      attr, items, pre_actions,
6060 					      flow_split_info, error);
6061 		flow_split_info->skip_scale = skip_scale_restore;
6062 		if (ret) {
6063 			if (mtr_flow_id)
6064 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6065 			ret = -rte_errno;
6066 			goto exit;
6067 		}
6068 		if (mtr_flow_id) {
6069 			dev_flow->handle->split_flow_id = mtr_flow_id;
6070 			dev_flow->handle->is_meter_flow_id = 1;
6071 		}
6072 		if (!fm->def_policy) {
6073 			if (!set_mtr_reg && fm->drop_cnt)
6074 				ret =
6075 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6076 							&sfx_attr, items,
6077 							flow_split_info,
6078 							fm, error);
6079 			goto exit;
6080 		}
6081 		/* Setting the sfx group atrr. */
6082 		sfx_attr.group = sfx_attr.transfer ?
6083 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6084 				 MLX5_FLOW_TABLE_LEVEL_METER;
6085 		flow_split_info->prefix_layers =
6086 				flow_get_prefix_layer_flags(dev_flow);
6087 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6088 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6089 	}
6090 	/* Add the prefix subflow. */
6091 	ret = flow_create_split_metadata(dev, flow,
6092 					 &sfx_attr, sfx_items ?
6093 					 sfx_items : items,
6094 					 sfx_actions ? sfx_actions : actions,
6095 					 flow_split_info, error);
6096 exit:
6097 	if (sfx_actions)
6098 		mlx5_free(sfx_actions);
6099 	return ret;
6100 }
6101 
6102 /**
6103  * The splitting for sample feature.
6104  *
6105  * Once Sample action is detected in the action list, the flow actions should
6106  * be split into prefix sub flow and suffix sub flow.
6107  *
6108  * The original items remain in the prefix sub flow, all actions preceding the
6109  * sample action and the sample action itself will be copied to the prefix
6110  * sub flow, the actions following the sample action will be copied to the
6111  * suffix sub flow, Queue action always be located in the suffix sub flow.
6112  *
6113  * In order to make the packet from prefix sub flow matches with suffix sub
6114  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6115  * flow uses tag item with the unique flow id.
6116  *
6117  * @param dev
6118  *   Pointer to Ethernet device.
6119  * @param[in] flow
6120  *   Parent flow structure pointer.
6121  * @param[in] attr
6122  *   Flow rule attributes.
6123  * @param[in] items
6124  *   Pattern specification (list terminated by the END pattern item).
6125  * @param[in] actions
6126  *   Associated actions (list terminated by the END action).
6127  * @param[in] flow_split_info
6128  *   Pointer to flow split info structure.
6129  * @param[out] error
6130  *   Perform verbose error reporting if not NULL.
6131  * @return
6132  *   0 on success, negative value otherwise
6133  */
6134 static int
6135 flow_create_split_sample(struct rte_eth_dev *dev,
6136 			 struct rte_flow *flow,
6137 			 const struct rte_flow_attr *attr,
6138 			 const struct rte_flow_item items[],
6139 			 const struct rte_flow_action actions[],
6140 			 struct mlx5_flow_split_info *flow_split_info,
6141 			 struct rte_flow_error *error)
6142 {
6143 	struct mlx5_priv *priv = dev->data->dev_private;
6144 	struct rte_flow_action *sfx_actions = NULL;
6145 	struct rte_flow_action *pre_actions = NULL;
6146 	struct rte_flow_item *sfx_items = NULL;
6147 	struct mlx5_flow *dev_flow = NULL;
6148 	struct rte_flow_attr sfx_attr = *attr;
6149 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6150 	struct mlx5_flow_dv_sample_resource *sample_res;
6151 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6152 	struct mlx5_flow_tbl_resource *sfx_tbl;
6153 #endif
6154 	size_t act_size;
6155 	size_t item_size;
6156 	uint32_t fdb_tx = 0;
6157 	int32_t tag_id = 0;
6158 	int actions_n = 0;
6159 	int sample_action_pos;
6160 	int qrss_action_pos;
6161 	int add_tag = 0;
6162 	int modify_after_mirror = 0;
6163 	uint16_t jump_table = 0;
6164 	const uint32_t next_ft_step = 1;
6165 	int ret = 0;
6166 
6167 	if (priv->sampler_en)
6168 		actions_n = flow_check_match_action(actions, attr,
6169 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6170 					&sample_action_pos, &qrss_action_pos,
6171 					&modify_after_mirror);
6172 	if (actions_n) {
6173 		/* The prefix actions must includes sample, tag, end. */
6174 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6175 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6176 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6177 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6178 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6179 					  item_size), 0, SOCKET_ID_ANY);
6180 		if (!sfx_actions)
6181 			return rte_flow_error_set(error, ENOMEM,
6182 						  RTE_FLOW_ERROR_TYPE_ACTION,
6183 						  NULL, "no memory to split "
6184 						  "sample flow");
6185 		/* The representor_id is UINT16_MAX for uplink. */
6186 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6187 		/*
6188 		 * When reg_c_preserve is set, metadata registers Cx preserve
6189 		 * their value even through packet duplication.
6190 		 */
6191 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6192 		if (add_tag)
6193 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6194 					+ act_size);
6195 		if (modify_after_mirror)
6196 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6197 				     next_ft_step;
6198 		pre_actions = sfx_actions + actions_n;
6199 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6200 						actions, sfx_actions,
6201 						pre_actions, actions_n,
6202 						sample_action_pos,
6203 						qrss_action_pos, jump_table,
6204 						error);
6205 		if (tag_id < 0 || (add_tag && !tag_id)) {
6206 			ret = -rte_errno;
6207 			goto exit;
6208 		}
6209 		if (modify_after_mirror)
6210 			flow_split_info->skip_scale =
6211 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6212 		/* Add the prefix subflow. */
6213 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6214 					      items, pre_actions,
6215 					      flow_split_info, error);
6216 		if (ret) {
6217 			ret = -rte_errno;
6218 			goto exit;
6219 		}
6220 		dev_flow->handle->split_flow_id = tag_id;
6221 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6222 		if (!modify_after_mirror) {
6223 			/* Set the sfx group attr. */
6224 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6225 						dev_flow->dv.sample_res;
6226 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6227 						sample_res->normal_path_tbl;
6228 			sfx_tbl_data = container_of(sfx_tbl,
6229 						struct mlx5_flow_tbl_data_entry,
6230 						tbl);
6231 			sfx_attr.group = sfx_attr.transfer ?
6232 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6233 		} else {
6234 			MLX5_ASSERT(attr->transfer);
6235 			sfx_attr.group = jump_table;
6236 		}
6237 		flow_split_info->prefix_layers =
6238 				flow_get_prefix_layer_flags(dev_flow);
6239 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6240 		/* Suffix group level already be scaled with factor, set
6241 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6242 		 * again in translation.
6243 		 */
6244 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6245 #endif
6246 	}
6247 	/* Add the suffix subflow. */
6248 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6249 				      sfx_items ? sfx_items : items,
6250 				      sfx_actions ? sfx_actions : actions,
6251 				      flow_split_info, error);
6252 exit:
6253 	if (sfx_actions)
6254 		mlx5_free(sfx_actions);
6255 	return ret;
6256 }
6257 
6258 /**
6259  * Split the flow to subflow set. The splitters might be linked
6260  * in the chain, like this:
6261  * flow_create_split_outer() calls:
6262  *   flow_create_split_meter() calls:
6263  *     flow_create_split_metadata(meter_subflow_0) calls:
6264  *       flow_create_split_inner(metadata_subflow_0)
6265  *       flow_create_split_inner(metadata_subflow_1)
6266  *       flow_create_split_inner(metadata_subflow_2)
6267  *     flow_create_split_metadata(meter_subflow_1) calls:
6268  *       flow_create_split_inner(metadata_subflow_0)
6269  *       flow_create_split_inner(metadata_subflow_1)
6270  *       flow_create_split_inner(metadata_subflow_2)
6271  *
6272  * This provide flexible way to add new levels of flow splitting.
6273  * The all of successfully created subflows are included to the
6274  * parent flow dev_flow list.
6275  *
6276  * @param dev
6277  *   Pointer to Ethernet device.
6278  * @param[in] flow
6279  *   Parent flow structure pointer.
6280  * @param[in] attr
6281  *   Flow rule attributes.
6282  * @param[in] items
6283  *   Pattern specification (list terminated by the END pattern item).
6284  * @param[in] actions
6285  *   Associated actions (list terminated by the END action).
6286  * @param[in] flow_split_info
6287  *   Pointer to flow split info structure.
6288  * @param[out] error
6289  *   Perform verbose error reporting if not NULL.
6290  * @return
6291  *   0 on success, negative value otherwise
6292  */
6293 static int
6294 flow_create_split_outer(struct rte_eth_dev *dev,
6295 			struct rte_flow *flow,
6296 			const struct rte_flow_attr *attr,
6297 			const struct rte_flow_item items[],
6298 			const struct rte_flow_action actions[],
6299 			struct mlx5_flow_split_info *flow_split_info,
6300 			struct rte_flow_error *error)
6301 {
6302 	int ret;
6303 
6304 	ret = flow_create_split_sample(dev, flow, attr, items,
6305 				       actions, flow_split_info, error);
6306 	MLX5_ASSERT(ret <= 0);
6307 	return ret;
6308 }
6309 
6310 static inline struct mlx5_flow_tunnel *
6311 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6312 {
6313 	struct mlx5_flow_tunnel *tunnel;
6314 
6315 #pragma GCC diagnostic push
6316 #pragma GCC diagnostic ignored "-Wcast-qual"
6317 	tunnel = (typeof(tunnel))flow->tunnel;
6318 #pragma GCC diagnostic pop
6319 
6320 	return tunnel;
6321 }
6322 
6323 /**
6324  * Adjust flow RSS workspace if needed.
6325  *
6326  * @param wks
6327  *   Pointer to thread flow work space.
6328  * @param rss_desc
6329  *   Pointer to RSS descriptor.
6330  * @param[in] nrssq_num
6331  *   New RSS queue number.
6332  *
6333  * @return
6334  *   0 on success, -1 otherwise and rte_errno is set.
6335  */
6336 static int
6337 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6338 			  struct mlx5_flow_rss_desc *rss_desc,
6339 			  uint32_t nrssq_num)
6340 {
6341 	if (likely(nrssq_num <= wks->rssq_num))
6342 		return 0;
6343 	rss_desc->queue = realloc(rss_desc->queue,
6344 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6345 	if (!rss_desc->queue) {
6346 		rte_errno = ENOMEM;
6347 		return -1;
6348 	}
6349 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6350 	return 0;
6351 }
6352 
6353 /**
6354  * Create a flow and add it to @p list.
6355  *
6356  * @param dev
6357  *   Pointer to Ethernet device.
6358  * @param list
6359  *   Pointer to a TAILQ flow list. If this parameter NULL,
6360  *   no list insertion occurred, flow is just created,
6361  *   this is caller's responsibility to track the
6362  *   created flow.
6363  * @param[in] attr
6364  *   Flow rule attributes.
6365  * @param[in] items
6366  *   Pattern specification (list terminated by the END pattern item).
6367  * @param[in] actions
6368  *   Associated actions (list terminated by the END action).
6369  * @param[in] external
6370  *   This flow rule is created by request external to PMD.
6371  * @param[out] error
6372  *   Perform verbose error reporting if not NULL.
6373  *
6374  * @return
6375  *   A flow index on success, 0 otherwise and rte_errno is set.
6376  */
6377 static uint32_t
6378 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6379 		 const struct rte_flow_attr *attr,
6380 		 const struct rte_flow_item items[],
6381 		 const struct rte_flow_action original_actions[],
6382 		 bool external, struct rte_flow_error *error)
6383 {
6384 	struct mlx5_priv *priv = dev->data->dev_private;
6385 	struct rte_flow *flow = NULL;
6386 	struct mlx5_flow *dev_flow;
6387 	const struct rte_flow_action_rss *rss = NULL;
6388 	struct mlx5_translated_action_handle
6389 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6390 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6391 	union {
6392 		struct mlx5_flow_expand_rss buf;
6393 		uint8_t buffer[4096];
6394 	} expand_buffer;
6395 	union {
6396 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6397 		uint8_t buffer[2048];
6398 	} actions_rx;
6399 	union {
6400 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6401 		uint8_t buffer[2048];
6402 	} actions_hairpin_tx;
6403 	union {
6404 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6405 		uint8_t buffer[2048];
6406 	} items_tx;
6407 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6408 	struct mlx5_flow_rss_desc *rss_desc;
6409 	const struct rte_flow_action *p_actions_rx;
6410 	uint32_t i;
6411 	uint32_t idx = 0;
6412 	int hairpin_flow;
6413 	struct rte_flow_attr attr_tx = { .priority = 0 };
6414 	const struct rte_flow_action *actions;
6415 	struct rte_flow_action *translated_actions = NULL;
6416 	struct mlx5_flow_tunnel *tunnel;
6417 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6418 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6419 	struct mlx5_flow_split_info flow_split_info = {
6420 		.external = !!external,
6421 		.skip_scale = 0,
6422 		.flow_idx = 0,
6423 		.prefix_mark = 0,
6424 		.prefix_layers = 0,
6425 		.table_id = 0
6426 	};
6427 	int ret;
6428 
6429 	MLX5_ASSERT(wks);
6430 	rss_desc = &wks->rss_desc;
6431 	ret = flow_action_handles_translate(dev, original_actions,
6432 					    indir_actions,
6433 					    &indir_actions_n,
6434 					    &translated_actions, error);
6435 	if (ret < 0) {
6436 		MLX5_ASSERT(translated_actions == NULL);
6437 		return 0;
6438 	}
6439 	actions = translated_actions ? translated_actions : original_actions;
6440 	p_actions_rx = actions;
6441 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6442 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6443 				external, hairpin_flow, error);
6444 	if (ret < 0)
6445 		goto error_before_hairpin_split;
6446 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6447 	if (!flow) {
6448 		rte_errno = ENOMEM;
6449 		goto error_before_hairpin_split;
6450 	}
6451 	if (hairpin_flow > 0) {
6452 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6453 			rte_errno = EINVAL;
6454 			goto error_before_hairpin_split;
6455 		}
6456 		flow_hairpin_split(dev, actions, actions_rx.actions,
6457 				   actions_hairpin_tx.actions, items_tx.items,
6458 				   idx);
6459 		p_actions_rx = actions_rx.actions;
6460 	}
6461 	flow_split_info.flow_idx = idx;
6462 	flow->drv_type = flow_get_drv_type(dev, attr);
6463 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6464 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6465 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6466 	/* RSS Action only works on NIC RX domain */
6467 	if (attr->ingress && !attr->transfer)
6468 		rss = flow_get_rss_action(dev, p_actions_rx);
6469 	if (rss) {
6470 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6471 			return 0;
6472 		/*
6473 		 * The following information is required by
6474 		 * mlx5_flow_hashfields_adjust() in advance.
6475 		 */
6476 		rss_desc->level = rss->level;
6477 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6478 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6479 	}
6480 	flow->dev_handles = 0;
6481 	if (rss && rss->types) {
6482 		unsigned int graph_root;
6483 
6484 		graph_root = find_graph_root(rss->level);
6485 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6486 					   items, rss->types,
6487 					   mlx5_support_expansion, graph_root);
6488 		MLX5_ASSERT(ret > 0 &&
6489 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6490 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6491 			for (i = 0; i < buf->entries; ++i)
6492 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6493 		}
6494 	} else {
6495 		buf->entries = 1;
6496 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6497 	}
6498 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6499 						      indir_actions_n);
6500 	for (i = 0; i < buf->entries; ++i) {
6501 		/* Initialize flow split data. */
6502 		flow_split_info.prefix_layers = 0;
6503 		flow_split_info.prefix_mark = 0;
6504 		flow_split_info.skip_scale = 0;
6505 		/*
6506 		 * The splitter may create multiple dev_flows,
6507 		 * depending on configuration. In the simplest
6508 		 * case it just creates unmodified original flow.
6509 		 */
6510 		ret = flow_create_split_outer(dev, flow, attr,
6511 					      buf->entry[i].pattern,
6512 					      p_actions_rx, &flow_split_info,
6513 					      error);
6514 		if (ret < 0)
6515 			goto error;
6516 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6517 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6518 							   p_actions_rx,
6519 							   idx,
6520 							   wks->flows[0].tunnel,
6521 							   &default_miss_ctx,
6522 							   error);
6523 			if (ret < 0) {
6524 				mlx5_free(default_miss_ctx.queue);
6525 				goto error;
6526 			}
6527 		}
6528 	}
6529 	/* Create the tx flow. */
6530 	if (hairpin_flow) {
6531 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6532 		attr_tx.ingress = 0;
6533 		attr_tx.egress = 1;
6534 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6535 					 actions_hairpin_tx.actions,
6536 					 idx, error);
6537 		if (!dev_flow)
6538 			goto error;
6539 		dev_flow->flow = flow;
6540 		dev_flow->external = 0;
6541 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6542 			      dev_flow->handle, next);
6543 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6544 					 items_tx.items,
6545 					 actions_hairpin_tx.actions, error);
6546 		if (ret < 0)
6547 			goto error;
6548 	}
6549 	/*
6550 	 * Update the metadata register copy table. If extensive
6551 	 * metadata feature is enabled and registers are supported
6552 	 * we might create the extra rte_flow for each unique
6553 	 * MARK/FLAG action ID.
6554 	 *
6555 	 * The table is updated for ingress Flows only, because
6556 	 * the egress Flows belong to the different device and
6557 	 * copy table should be updated in peer NIC Rx domain.
6558 	 */
6559 	if (attr->ingress &&
6560 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6561 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6562 		if (ret)
6563 			goto error;
6564 	}
6565 	/*
6566 	 * If the flow is external (from application) OR device is started,
6567 	 * OR mreg discover, then apply immediately.
6568 	 */
6569 	if (external || dev->data->dev_started ||
6570 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6571 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6572 		ret = flow_drv_apply(dev, flow, error);
6573 		if (ret < 0)
6574 			goto error;
6575 	}
6576 	flow->type = type;
6577 	flow_rxq_flags_set(dev, flow);
6578 	rte_free(translated_actions);
6579 	tunnel = flow_tunnel_from_rule(wks->flows);
6580 	if (tunnel) {
6581 		flow->tunnel = 1;
6582 		flow->tunnel_id = tunnel->tunnel_id;
6583 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6584 		mlx5_free(default_miss_ctx.queue);
6585 	}
6586 	mlx5_flow_pop_thread_workspace();
6587 	return idx;
6588 error:
6589 	MLX5_ASSERT(flow);
6590 	ret = rte_errno; /* Save rte_errno before cleanup. */
6591 	flow_mreg_del_copy_action(dev, flow);
6592 	flow_drv_destroy(dev, flow);
6593 	if (rss_desc->shared_rss)
6594 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6595 			mlx5_ipool_get
6596 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6597 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6598 	mlx5_ipool_free(priv->flows[type], idx);
6599 	rte_errno = ret; /* Restore rte_errno. */
6600 	ret = rte_errno;
6601 	rte_errno = ret;
6602 	mlx5_flow_pop_thread_workspace();
6603 error_before_hairpin_split:
6604 	rte_free(translated_actions);
6605 	return 0;
6606 }
6607 
6608 /**
6609  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6610  * incoming packets to table 1.
6611  *
6612  * Other flow rules, requested for group n, will be created in
6613  * e-switch table n+1.
6614  * Jump action to e-switch group n will be created to group n+1.
6615  *
6616  * Used when working in switchdev mode, to utilise advantages of table 1
6617  * and above.
6618  *
6619  * @param dev
6620  *   Pointer to Ethernet device.
6621  *
6622  * @return
6623  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6624  */
6625 struct rte_flow *
6626 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6627 {
6628 	const struct rte_flow_attr attr = {
6629 		.group = 0,
6630 		.priority = 0,
6631 		.ingress = 1,
6632 		.egress = 0,
6633 		.transfer = 1,
6634 	};
6635 	const struct rte_flow_item pattern = {
6636 		.type = RTE_FLOW_ITEM_TYPE_END,
6637 	};
6638 	struct rte_flow_action_jump jump = {
6639 		.group = 1,
6640 	};
6641 	const struct rte_flow_action actions[] = {
6642 		{
6643 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6644 			.conf = &jump,
6645 		},
6646 		{
6647 			.type = RTE_FLOW_ACTION_TYPE_END,
6648 		},
6649 	};
6650 	struct rte_flow_error error;
6651 
6652 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6653 						   &attr, &pattern,
6654 						   actions, false, &error);
6655 }
6656 
6657 /**
6658  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6659  * and sq number, directs all packets to peer vport.
6660  *
6661  * @param dev
6662  *   Pointer to Ethernet device.
6663  * @param txq
6664  *   Txq index.
6665  *
6666  * @return
6667  *   Flow ID on success, 0 otherwise and rte_errno is set.
6668  */
6669 uint32_t
6670 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6671 {
6672 	struct rte_flow_attr attr = {
6673 		.group = 0,
6674 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6675 		.ingress = 1,
6676 		.egress = 0,
6677 		.transfer = 1,
6678 	};
6679 	struct rte_flow_item_port_id port_spec = {
6680 		.id = MLX5_PORT_ESW_MGR,
6681 	};
6682 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6683 		.queue = txq,
6684 	};
6685 	struct rte_flow_item pattern[] = {
6686 		{
6687 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6688 			.spec = &port_spec,
6689 		},
6690 		{
6691 			.type = (enum rte_flow_item_type)
6692 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6693 			.spec = &txq_spec,
6694 		},
6695 		{
6696 			.type = RTE_FLOW_ITEM_TYPE_END,
6697 		},
6698 	};
6699 	struct rte_flow_action_jump jump = {
6700 		.group = 1,
6701 	};
6702 	struct rte_flow_action_port_id port = {
6703 		.id = dev->data->port_id,
6704 	};
6705 	struct rte_flow_action actions[] = {
6706 		{
6707 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6708 			.conf = &jump,
6709 		},
6710 		{
6711 			.type = RTE_FLOW_ACTION_TYPE_END,
6712 		},
6713 	};
6714 	struct rte_flow_error error;
6715 
6716 	/*
6717 	 * Creates group 0, highest priority jump flow.
6718 	 * Matches txq to bypass kernel packets.
6719 	 */
6720 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6721 			     false, &error) == 0)
6722 		return 0;
6723 	/* Create group 1, lowest priority redirect flow for txq. */
6724 	attr.group = 1;
6725 	actions[0].conf = &port;
6726 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6727 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6728 				actions, false, &error);
6729 }
6730 
6731 /**
6732  * Validate a flow supported by the NIC.
6733  *
6734  * @see rte_flow_validate()
6735  * @see rte_flow_ops
6736  */
6737 int
6738 mlx5_flow_validate(struct rte_eth_dev *dev,
6739 		   const struct rte_flow_attr *attr,
6740 		   const struct rte_flow_item items[],
6741 		   const struct rte_flow_action original_actions[],
6742 		   struct rte_flow_error *error)
6743 {
6744 	int hairpin_flow;
6745 	struct mlx5_translated_action_handle
6746 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6747 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6748 	const struct rte_flow_action *actions;
6749 	struct rte_flow_action *translated_actions = NULL;
6750 	int ret = flow_action_handles_translate(dev, original_actions,
6751 						indir_actions,
6752 						&indir_actions_n,
6753 						&translated_actions, error);
6754 
6755 	if (ret)
6756 		return ret;
6757 	actions = translated_actions ? translated_actions : original_actions;
6758 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6759 	ret = flow_drv_validate(dev, attr, items, actions,
6760 				true, hairpin_flow, error);
6761 	rte_free(translated_actions);
6762 	return ret;
6763 }
6764 
6765 /**
6766  * Create a flow.
6767  *
6768  * @see rte_flow_create()
6769  * @see rte_flow_ops
6770  */
6771 struct rte_flow *
6772 mlx5_flow_create(struct rte_eth_dev *dev,
6773 		 const struct rte_flow_attr *attr,
6774 		 const struct rte_flow_item items[],
6775 		 const struct rte_flow_action actions[],
6776 		 struct rte_flow_error *error)
6777 {
6778 	/*
6779 	 * If the device is not started yet, it is not allowed to created a
6780 	 * flow from application. PMD default flows and traffic control flows
6781 	 * are not affected.
6782 	 */
6783 	if (unlikely(!dev->data->dev_started)) {
6784 		DRV_LOG(DEBUG, "port %u is not started when "
6785 			"inserting a flow", dev->data->port_id);
6786 		rte_flow_error_set(error, ENODEV,
6787 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6788 				   NULL,
6789 				   "port not started");
6790 		return NULL;
6791 	}
6792 
6793 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6794 						   attr, items, actions,
6795 						   true, error);
6796 }
6797 
6798 /**
6799  * Destroy a flow in a list.
6800  *
6801  * @param dev
6802  *   Pointer to Ethernet device.
6803  * @param[in] flow_idx
6804  *   Index of flow to destroy.
6805  */
6806 static void
6807 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6808 		  uint32_t flow_idx)
6809 {
6810 	struct mlx5_priv *priv = dev->data->dev_private;
6811 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6812 
6813 	if (!flow)
6814 		return;
6815 	MLX5_ASSERT(flow->type == type);
6816 	/*
6817 	 * Update RX queue flags only if port is started, otherwise it is
6818 	 * already clean.
6819 	 */
6820 	if (dev->data->dev_started)
6821 		flow_rxq_flags_trim(dev, flow);
6822 	flow_drv_destroy(dev, flow);
6823 	if (flow->tunnel) {
6824 		struct mlx5_flow_tunnel *tunnel;
6825 
6826 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6827 		RTE_VERIFY(tunnel);
6828 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6829 			mlx5_flow_tunnel_free(dev, tunnel);
6830 	}
6831 	flow_mreg_del_copy_action(dev, flow);
6832 	mlx5_ipool_free(priv->flows[type], flow_idx);
6833 }
6834 
6835 /**
6836  * Destroy all flows.
6837  *
6838  * @param dev
6839  *   Pointer to Ethernet device.
6840  * @param type
6841  *   Flow type to be flushed.
6842  * @param active
6843  *   If flushing is called avtively.
6844  */
6845 void
6846 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6847 		     bool active)
6848 {
6849 	struct mlx5_priv *priv = dev->data->dev_private;
6850 	uint32_t num_flushed = 0, fidx = 1;
6851 	struct rte_flow *flow;
6852 
6853 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6854 		flow_list_destroy(dev, type, fidx);
6855 		num_flushed++;
6856 	}
6857 	if (active) {
6858 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6859 			dev->data->port_id, num_flushed);
6860 	}
6861 }
6862 
6863 /**
6864  * Stop all default actions for flows.
6865  *
6866  * @param dev
6867  *   Pointer to Ethernet device.
6868  */
6869 void
6870 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6871 {
6872 	flow_mreg_del_default_copy_action(dev);
6873 	flow_rxq_flags_clear(dev);
6874 }
6875 
6876 /**
6877  * Start all default actions for flows.
6878  *
6879  * @param dev
6880  *   Pointer to Ethernet device.
6881  * @return
6882  *   0 on success, a negative errno value otherwise and rte_errno is set.
6883  */
6884 int
6885 mlx5_flow_start_default(struct rte_eth_dev *dev)
6886 {
6887 	struct rte_flow_error error;
6888 
6889 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6890 	return flow_mreg_add_default_copy_action(dev, &error);
6891 }
6892 
6893 /**
6894  * Release key of thread specific flow workspace data.
6895  */
6896 void
6897 flow_release_workspace(void *data)
6898 {
6899 	struct mlx5_flow_workspace *wks = data;
6900 	struct mlx5_flow_workspace *next;
6901 
6902 	while (wks) {
6903 		next = wks->next;
6904 		free(wks->rss_desc.queue);
6905 		free(wks);
6906 		wks = next;
6907 	}
6908 }
6909 
6910 /**
6911  * Get thread specific current flow workspace.
6912  *
6913  * @return pointer to thread specific flow workspace data, NULL on error.
6914  */
6915 struct mlx5_flow_workspace*
6916 mlx5_flow_get_thread_workspace(void)
6917 {
6918 	struct mlx5_flow_workspace *data;
6919 
6920 	data = mlx5_flow_os_get_specific_workspace();
6921 	MLX5_ASSERT(data && data->inuse);
6922 	if (!data || !data->inuse)
6923 		DRV_LOG(ERR, "flow workspace not initialized.");
6924 	return data;
6925 }
6926 
6927 /**
6928  * Allocate and init new flow workspace.
6929  *
6930  * @return pointer to flow workspace data, NULL on error.
6931  */
6932 static struct mlx5_flow_workspace*
6933 flow_alloc_thread_workspace(void)
6934 {
6935 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6936 
6937 	if (!data) {
6938 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6939 			"memory.");
6940 		return NULL;
6941 	}
6942 	data->rss_desc.queue = calloc(1,
6943 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6944 	if (!data->rss_desc.queue)
6945 		goto err;
6946 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6947 	return data;
6948 err:
6949 	if (data->rss_desc.queue)
6950 		free(data->rss_desc.queue);
6951 	free(data);
6952 	return NULL;
6953 }
6954 
6955 /**
6956  * Get new thread specific flow workspace.
6957  *
6958  * If current workspace inuse, create new one and set as current.
6959  *
6960  * @return pointer to thread specific flow workspace data, NULL on error.
6961  */
6962 static struct mlx5_flow_workspace*
6963 mlx5_flow_push_thread_workspace(void)
6964 {
6965 	struct mlx5_flow_workspace *curr;
6966 	struct mlx5_flow_workspace *data;
6967 
6968 	curr = mlx5_flow_os_get_specific_workspace();
6969 	if (!curr) {
6970 		data = flow_alloc_thread_workspace();
6971 		if (!data)
6972 			return NULL;
6973 	} else if (!curr->inuse) {
6974 		data = curr;
6975 	} else if (curr->next) {
6976 		data = curr->next;
6977 	} else {
6978 		data = flow_alloc_thread_workspace();
6979 		if (!data)
6980 			return NULL;
6981 		curr->next = data;
6982 		data->prev = curr;
6983 	}
6984 	data->inuse = 1;
6985 	data->flow_idx = 0;
6986 	/* Set as current workspace */
6987 	if (mlx5_flow_os_set_specific_workspace(data))
6988 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6989 	return data;
6990 }
6991 
6992 /**
6993  * Close current thread specific flow workspace.
6994  *
6995  * If previous workspace available, set it as current.
6996  *
6997  * @return pointer to thread specific flow workspace data, NULL on error.
6998  */
6999 static void
7000 mlx5_flow_pop_thread_workspace(void)
7001 {
7002 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7003 
7004 	if (!data)
7005 		return;
7006 	if (!data->inuse) {
7007 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7008 		return;
7009 	}
7010 	data->inuse = 0;
7011 	if (!data->prev)
7012 		return;
7013 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7014 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7015 }
7016 
7017 /**
7018  * Verify the flow list is empty
7019  *
7020  * @param dev
7021  *  Pointer to Ethernet device.
7022  *
7023  * @return the number of flows not released.
7024  */
7025 int
7026 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7027 {
7028 	struct mlx5_priv *priv = dev->data->dev_private;
7029 	struct rte_flow *flow;
7030 	uint32_t idx = 0;
7031 	int ret = 0, i;
7032 
7033 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7034 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7035 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7036 				dev->data->port_id, (void *)flow);
7037 			ret++;
7038 		}
7039 	}
7040 	return ret;
7041 }
7042 
7043 /**
7044  * Enable default hairpin egress flow.
7045  *
7046  * @param dev
7047  *   Pointer to Ethernet device.
7048  * @param queue
7049  *   The queue index.
7050  *
7051  * @return
7052  *   0 on success, a negative errno value otherwise and rte_errno is set.
7053  */
7054 int
7055 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7056 			    uint32_t queue)
7057 {
7058 	const struct rte_flow_attr attr = {
7059 		.egress = 1,
7060 		.priority = 0,
7061 	};
7062 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7063 		.queue = queue,
7064 	};
7065 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7066 		.queue = UINT32_MAX,
7067 	};
7068 	struct rte_flow_item items[] = {
7069 		{
7070 			.type = (enum rte_flow_item_type)
7071 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7072 			.spec = &queue_spec,
7073 			.last = NULL,
7074 			.mask = &queue_mask,
7075 		},
7076 		{
7077 			.type = RTE_FLOW_ITEM_TYPE_END,
7078 		},
7079 	};
7080 	struct rte_flow_action_jump jump = {
7081 		.group = MLX5_HAIRPIN_TX_TABLE,
7082 	};
7083 	struct rte_flow_action actions[2];
7084 	uint32_t flow_idx;
7085 	struct rte_flow_error error;
7086 
7087 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7088 	actions[0].conf = &jump;
7089 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7090 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7091 				    &attr, items, actions, false, &error);
7092 	if (!flow_idx) {
7093 		DRV_LOG(DEBUG,
7094 			"Failed to create ctrl flow: rte_errno(%d),"
7095 			" type(%d), message(%s)",
7096 			rte_errno, error.type,
7097 			error.message ? error.message : " (no stated reason)");
7098 		return -rte_errno;
7099 	}
7100 	return 0;
7101 }
7102 
7103 /**
7104  * Enable a control flow configured from the control plane.
7105  *
7106  * @param dev
7107  *   Pointer to Ethernet device.
7108  * @param eth_spec
7109  *   An Ethernet flow spec to apply.
7110  * @param eth_mask
7111  *   An Ethernet flow mask to apply.
7112  * @param vlan_spec
7113  *   A VLAN flow spec to apply.
7114  * @param vlan_mask
7115  *   A VLAN flow mask to apply.
7116  *
7117  * @return
7118  *   0 on success, a negative errno value otherwise and rte_errno is set.
7119  */
7120 int
7121 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7122 		    struct rte_flow_item_eth *eth_spec,
7123 		    struct rte_flow_item_eth *eth_mask,
7124 		    struct rte_flow_item_vlan *vlan_spec,
7125 		    struct rte_flow_item_vlan *vlan_mask)
7126 {
7127 	struct mlx5_priv *priv = dev->data->dev_private;
7128 	const struct rte_flow_attr attr = {
7129 		.ingress = 1,
7130 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7131 	};
7132 	struct rte_flow_item items[] = {
7133 		{
7134 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7135 			.spec = eth_spec,
7136 			.last = NULL,
7137 			.mask = eth_mask,
7138 		},
7139 		{
7140 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7141 					      RTE_FLOW_ITEM_TYPE_END,
7142 			.spec = vlan_spec,
7143 			.last = NULL,
7144 			.mask = vlan_mask,
7145 		},
7146 		{
7147 			.type = RTE_FLOW_ITEM_TYPE_END,
7148 		},
7149 	};
7150 	uint16_t queue[priv->reta_idx_n];
7151 	struct rte_flow_action_rss action_rss = {
7152 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7153 		.level = 0,
7154 		.types = priv->rss_conf.rss_hf,
7155 		.key_len = priv->rss_conf.rss_key_len,
7156 		.queue_num = priv->reta_idx_n,
7157 		.key = priv->rss_conf.rss_key,
7158 		.queue = queue,
7159 	};
7160 	struct rte_flow_action actions[] = {
7161 		{
7162 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7163 			.conf = &action_rss,
7164 		},
7165 		{
7166 			.type = RTE_FLOW_ACTION_TYPE_END,
7167 		},
7168 	};
7169 	uint32_t flow_idx;
7170 	struct rte_flow_error error;
7171 	unsigned int i;
7172 
7173 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7174 		return 0;
7175 	}
7176 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7177 		action_rss.types = 0;
7178 	for (i = 0; i != priv->reta_idx_n; ++i)
7179 		queue[i] = (*priv->reta_idx)[i];
7180 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7181 				    &attr, items, actions, false, &error);
7182 	if (!flow_idx)
7183 		return -rte_errno;
7184 	return 0;
7185 }
7186 
7187 /**
7188  * Enable a flow control configured from the control plane.
7189  *
7190  * @param dev
7191  *   Pointer to Ethernet device.
7192  * @param eth_spec
7193  *   An Ethernet flow spec to apply.
7194  * @param eth_mask
7195  *   An Ethernet flow mask to apply.
7196  *
7197  * @return
7198  *   0 on success, a negative errno value otherwise and rte_errno is set.
7199  */
7200 int
7201 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7202 	       struct rte_flow_item_eth *eth_spec,
7203 	       struct rte_flow_item_eth *eth_mask)
7204 {
7205 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7206 }
7207 
7208 /**
7209  * Create default miss flow rule matching lacp traffic
7210  *
7211  * @param dev
7212  *   Pointer to Ethernet device.
7213  * @param eth_spec
7214  *   An Ethernet flow spec to apply.
7215  *
7216  * @return
7217  *   0 on success, a negative errno value otherwise and rte_errno is set.
7218  */
7219 int
7220 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7221 {
7222 	/*
7223 	 * The LACP matching is done by only using ether type since using
7224 	 * a multicast dst mac causes kernel to give low priority to this flow.
7225 	 */
7226 	static const struct rte_flow_item_eth lacp_spec = {
7227 		.type = RTE_BE16(0x8809),
7228 	};
7229 	static const struct rte_flow_item_eth lacp_mask = {
7230 		.type = 0xffff,
7231 	};
7232 	const struct rte_flow_attr attr = {
7233 		.ingress = 1,
7234 	};
7235 	struct rte_flow_item items[] = {
7236 		{
7237 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7238 			.spec = &lacp_spec,
7239 			.mask = &lacp_mask,
7240 		},
7241 		{
7242 			.type = RTE_FLOW_ITEM_TYPE_END,
7243 		},
7244 	};
7245 	struct rte_flow_action actions[] = {
7246 		{
7247 			.type = (enum rte_flow_action_type)
7248 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7249 		},
7250 		{
7251 			.type = RTE_FLOW_ACTION_TYPE_END,
7252 		},
7253 	};
7254 	struct rte_flow_error error;
7255 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7256 					&attr, items, actions,
7257 					false, &error);
7258 
7259 	if (!flow_idx)
7260 		return -rte_errno;
7261 	return 0;
7262 }
7263 
7264 /**
7265  * Destroy a flow.
7266  *
7267  * @see rte_flow_destroy()
7268  * @see rte_flow_ops
7269  */
7270 int
7271 mlx5_flow_destroy(struct rte_eth_dev *dev,
7272 		  struct rte_flow *flow,
7273 		  struct rte_flow_error *error __rte_unused)
7274 {
7275 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7276 				(uintptr_t)(void *)flow);
7277 	return 0;
7278 }
7279 
7280 /**
7281  * Destroy all flows.
7282  *
7283  * @see rte_flow_flush()
7284  * @see rte_flow_ops
7285  */
7286 int
7287 mlx5_flow_flush(struct rte_eth_dev *dev,
7288 		struct rte_flow_error *error __rte_unused)
7289 {
7290 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7291 	return 0;
7292 }
7293 
7294 /**
7295  * Isolated mode.
7296  *
7297  * @see rte_flow_isolate()
7298  * @see rte_flow_ops
7299  */
7300 int
7301 mlx5_flow_isolate(struct rte_eth_dev *dev,
7302 		  int enable,
7303 		  struct rte_flow_error *error)
7304 {
7305 	struct mlx5_priv *priv = dev->data->dev_private;
7306 
7307 	if (dev->data->dev_started) {
7308 		rte_flow_error_set(error, EBUSY,
7309 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7310 				   NULL,
7311 				   "port must be stopped first");
7312 		return -rte_errno;
7313 	}
7314 	priv->isolated = !!enable;
7315 	if (enable)
7316 		dev->dev_ops = &mlx5_dev_ops_isolate;
7317 	else
7318 		dev->dev_ops = &mlx5_dev_ops;
7319 
7320 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7321 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7322 
7323 	return 0;
7324 }
7325 
7326 /**
7327  * Query a flow.
7328  *
7329  * @see rte_flow_query()
7330  * @see rte_flow_ops
7331  */
7332 static int
7333 flow_drv_query(struct rte_eth_dev *dev,
7334 	       uint32_t flow_idx,
7335 	       const struct rte_flow_action *actions,
7336 	       void *data,
7337 	       struct rte_flow_error *error)
7338 {
7339 	struct mlx5_priv *priv = dev->data->dev_private;
7340 	const struct mlx5_flow_driver_ops *fops;
7341 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7342 					       flow_idx);
7343 	enum mlx5_flow_drv_type ftype;
7344 
7345 	if (!flow) {
7346 		return rte_flow_error_set(error, ENOENT,
7347 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7348 			  NULL,
7349 			  "invalid flow handle");
7350 	}
7351 	ftype = flow->drv_type;
7352 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7353 	fops = flow_get_drv_ops(ftype);
7354 
7355 	return fops->query(dev, flow, actions, data, error);
7356 }
7357 
7358 /**
7359  * Query a flow.
7360  *
7361  * @see rte_flow_query()
7362  * @see rte_flow_ops
7363  */
7364 int
7365 mlx5_flow_query(struct rte_eth_dev *dev,
7366 		struct rte_flow *flow,
7367 		const struct rte_flow_action *actions,
7368 		void *data,
7369 		struct rte_flow_error *error)
7370 {
7371 	int ret;
7372 
7373 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7374 			     error);
7375 	if (ret < 0)
7376 		return ret;
7377 	return 0;
7378 }
7379 
7380 /**
7381  * Get rte_flow callbacks.
7382  *
7383  * @param dev
7384  *   Pointer to Ethernet device structure.
7385  * @param ops
7386  *   Pointer to operation-specific structure.
7387  *
7388  * @return 0
7389  */
7390 int
7391 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7392 		  const struct rte_flow_ops **ops)
7393 {
7394 	*ops = &mlx5_flow_ops;
7395 	return 0;
7396 }
7397 
7398 /**
7399  * Validate meter policy actions.
7400  * Dispatcher for action type specific validation.
7401  *
7402  * @param[in] dev
7403  *   Pointer to the Ethernet device structure.
7404  * @param[in] action
7405  *   The meter policy action object to validate.
7406  * @param[in] attr
7407  *   Attributes of flow to determine steering domain.
7408  * @param[out] is_rss
7409  *   Is RSS or not.
7410  * @param[out] domain_bitmap
7411  *   Domain bitmap.
7412  * @param[out] is_def_policy
7413  *   Is default policy or not.
7414  * @param[out] error
7415  *   Perform verbose error reporting if not NULL. Initialized in case of
7416  *   error only.
7417  *
7418  * @return
7419  *   0 on success, otherwise negative errno value.
7420  */
7421 int
7422 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7423 			const struct rte_flow_action *actions[RTE_COLORS],
7424 			struct rte_flow_attr *attr,
7425 			bool *is_rss,
7426 			uint8_t *domain_bitmap,
7427 			uint8_t *policy_mode,
7428 			struct rte_mtr_error *error)
7429 {
7430 	const struct mlx5_flow_driver_ops *fops;
7431 
7432 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7433 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7434 				       domain_bitmap, policy_mode, error);
7435 }
7436 
7437 /**
7438  * Destroy the meter table set.
7439  *
7440  * @param[in] dev
7441  *   Pointer to Ethernet device.
7442  * @param[in] mtr_policy
7443  *   Meter policy struct.
7444  */
7445 void
7446 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7447 		      struct mlx5_flow_meter_policy *mtr_policy)
7448 {
7449 	const struct mlx5_flow_driver_ops *fops;
7450 
7451 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7452 	fops->destroy_mtr_acts(dev, mtr_policy);
7453 }
7454 
7455 /**
7456  * Create policy action, lock free,
7457  * (mutex should be acquired by caller).
7458  * Dispatcher for action type specific call.
7459  *
7460  * @param[in] dev
7461  *   Pointer to the Ethernet device structure.
7462  * @param[in] mtr_policy
7463  *   Meter policy struct.
7464  * @param[in] action
7465  *   Action specification used to create meter actions.
7466  * @param[out] error
7467  *   Perform verbose error reporting if not NULL. Initialized in case of
7468  *   error only.
7469  *
7470  * @return
7471  *   0 on success, otherwise negative errno value.
7472  */
7473 int
7474 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7475 		      struct mlx5_flow_meter_policy *mtr_policy,
7476 		      const struct rte_flow_action *actions[RTE_COLORS],
7477 		      struct rte_mtr_error *error)
7478 {
7479 	const struct mlx5_flow_driver_ops *fops;
7480 
7481 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7482 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7483 }
7484 
7485 /**
7486  * Create policy rules, lock free,
7487  * (mutex should be acquired by caller).
7488  * Dispatcher for action type specific call.
7489  *
7490  * @param[in] dev
7491  *   Pointer to the Ethernet device structure.
7492  * @param[in] mtr_policy
7493  *   Meter policy struct.
7494  *
7495  * @return
7496  *   0 on success, -1 otherwise.
7497  */
7498 int
7499 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7500 			     struct mlx5_flow_meter_policy *mtr_policy)
7501 {
7502 	const struct mlx5_flow_driver_ops *fops;
7503 
7504 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7505 	return fops->create_policy_rules(dev, mtr_policy);
7506 }
7507 
7508 /**
7509  * Destroy policy rules, lock free,
7510  * (mutex should be acquired by caller).
7511  * Dispatcher for action type specific call.
7512  *
7513  * @param[in] dev
7514  *   Pointer to the Ethernet device structure.
7515  * @param[in] mtr_policy
7516  *   Meter policy struct.
7517  */
7518 void
7519 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7520 			     struct mlx5_flow_meter_policy *mtr_policy)
7521 {
7522 	const struct mlx5_flow_driver_ops *fops;
7523 
7524 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7525 	fops->destroy_policy_rules(dev, mtr_policy);
7526 }
7527 
7528 /**
7529  * Destroy the default policy table set.
7530  *
7531  * @param[in] dev
7532  *   Pointer to Ethernet device.
7533  */
7534 void
7535 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7536 {
7537 	const struct mlx5_flow_driver_ops *fops;
7538 
7539 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7540 	fops->destroy_def_policy(dev);
7541 }
7542 
7543 /**
7544  * Destroy the default policy table set.
7545  *
7546  * @param[in] dev
7547  *   Pointer to Ethernet device.
7548  *
7549  * @return
7550  *   0 on success, -1 otherwise.
7551  */
7552 int
7553 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7554 {
7555 	const struct mlx5_flow_driver_ops *fops;
7556 
7557 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7558 	return fops->create_def_policy(dev);
7559 }
7560 
7561 /**
7562  * Create the needed meter and suffix tables.
7563  *
7564  * @param[in] dev
7565  *   Pointer to Ethernet device.
7566  *
7567  * @return
7568  *   0 on success, -1 otherwise.
7569  */
7570 int
7571 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7572 			struct mlx5_flow_meter_info *fm,
7573 			uint32_t mtr_idx,
7574 			uint8_t domain_bitmap)
7575 {
7576 	const struct mlx5_flow_driver_ops *fops;
7577 
7578 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7579 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7580 }
7581 
7582 /**
7583  * Destroy the meter table set.
7584  *
7585  * @param[in] dev
7586  *   Pointer to Ethernet device.
7587  * @param[in] tbl
7588  *   Pointer to the meter table set.
7589  */
7590 void
7591 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7592 			   struct mlx5_flow_meter_info *fm)
7593 {
7594 	const struct mlx5_flow_driver_ops *fops;
7595 
7596 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7597 	fops->destroy_mtr_tbls(dev, fm);
7598 }
7599 
7600 /**
7601  * Destroy the global meter drop table.
7602  *
7603  * @param[in] dev
7604  *   Pointer to Ethernet device.
7605  */
7606 void
7607 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7608 {
7609 	const struct mlx5_flow_driver_ops *fops;
7610 
7611 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7612 	fops->destroy_mtr_drop_tbls(dev);
7613 }
7614 
7615 /**
7616  * Destroy the sub policy table with RX queue.
7617  *
7618  * @param[in] dev
7619  *   Pointer to Ethernet device.
7620  * @param[in] mtr_policy
7621  *   Pointer to meter policy table.
7622  */
7623 void
7624 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7625 		struct mlx5_flow_meter_policy *mtr_policy)
7626 {
7627 	const struct mlx5_flow_driver_ops *fops;
7628 
7629 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7630 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7631 }
7632 
7633 /**
7634  * Allocate the needed aso flow meter id.
7635  *
7636  * @param[in] dev
7637  *   Pointer to Ethernet device.
7638  *
7639  * @return
7640  *   Index to aso flow meter on success, NULL otherwise.
7641  */
7642 uint32_t
7643 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7644 {
7645 	const struct mlx5_flow_driver_ops *fops;
7646 
7647 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7648 	return fops->create_meter(dev);
7649 }
7650 
7651 /**
7652  * Free the aso flow meter id.
7653  *
7654  * @param[in] dev
7655  *   Pointer to Ethernet device.
7656  * @param[in] mtr_idx
7657  *  Index to aso flow meter to be free.
7658  *
7659  * @return
7660  *   0 on success.
7661  */
7662 void
7663 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7664 {
7665 	const struct mlx5_flow_driver_ops *fops;
7666 
7667 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7668 	fops->free_meter(dev, mtr_idx);
7669 }
7670 
7671 /**
7672  * Allocate a counter.
7673  *
7674  * @param[in] dev
7675  *   Pointer to Ethernet device structure.
7676  *
7677  * @return
7678  *   Index to allocated counter  on success, 0 otherwise.
7679  */
7680 uint32_t
7681 mlx5_counter_alloc(struct rte_eth_dev *dev)
7682 {
7683 	const struct mlx5_flow_driver_ops *fops;
7684 	struct rte_flow_attr attr = { .transfer = 0 };
7685 
7686 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7687 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7688 		return fops->counter_alloc(dev);
7689 	}
7690 	DRV_LOG(ERR,
7691 		"port %u counter allocate is not supported.",
7692 		 dev->data->port_id);
7693 	return 0;
7694 }
7695 
7696 /**
7697  * Free a counter.
7698  *
7699  * @param[in] dev
7700  *   Pointer to Ethernet device structure.
7701  * @param[in] cnt
7702  *   Index to counter to be free.
7703  */
7704 void
7705 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7706 {
7707 	const struct mlx5_flow_driver_ops *fops;
7708 	struct rte_flow_attr attr = { .transfer = 0 };
7709 
7710 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7711 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7712 		fops->counter_free(dev, cnt);
7713 		return;
7714 	}
7715 	DRV_LOG(ERR,
7716 		"port %u counter free is not supported.",
7717 		 dev->data->port_id);
7718 }
7719 
7720 /**
7721  * Query counter statistics.
7722  *
7723  * @param[in] dev
7724  *   Pointer to Ethernet device structure.
7725  * @param[in] cnt
7726  *   Index to counter to query.
7727  * @param[in] clear
7728  *   Set to clear counter statistics.
7729  * @param[out] pkts
7730  *   The counter hits packets number to save.
7731  * @param[out] bytes
7732  *   The counter hits bytes number to save.
7733  *
7734  * @return
7735  *   0 on success, a negative errno value otherwise.
7736  */
7737 int
7738 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7739 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7740 {
7741 	const struct mlx5_flow_driver_ops *fops;
7742 	struct rte_flow_attr attr = { .transfer = 0 };
7743 
7744 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7745 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7746 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7747 	}
7748 	DRV_LOG(ERR,
7749 		"port %u counter query is not supported.",
7750 		 dev->data->port_id);
7751 	return -ENOTSUP;
7752 }
7753 
7754 /**
7755  * Allocate a new memory for the counter values wrapped by all the needed
7756  * management.
7757  *
7758  * @param[in] sh
7759  *   Pointer to mlx5_dev_ctx_shared object.
7760  *
7761  * @return
7762  *   0 on success, a negative errno value otherwise.
7763  */
7764 static int
7765 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7766 {
7767 	struct mlx5_devx_mkey_attr mkey_attr;
7768 	struct mlx5_counter_stats_mem_mng *mem_mng;
7769 	volatile struct flow_counter_stats *raw_data;
7770 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7771 	int size = (sizeof(struct flow_counter_stats) *
7772 			MLX5_COUNTERS_PER_POOL +
7773 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7774 			sizeof(struct mlx5_counter_stats_mem_mng);
7775 	size_t pgsize = rte_mem_page_size();
7776 	uint8_t *mem;
7777 	int i;
7778 
7779 	if (pgsize == (size_t)-1) {
7780 		DRV_LOG(ERR, "Failed to get mem page size");
7781 		rte_errno = ENOMEM;
7782 		return -ENOMEM;
7783 	}
7784 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7785 	if (!mem) {
7786 		rte_errno = ENOMEM;
7787 		return -ENOMEM;
7788 	}
7789 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7790 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7791 	mem_mng->umem = mlx5_os_umem_reg(sh->cdev->ctx, mem, size,
7792 						 IBV_ACCESS_LOCAL_WRITE);
7793 	if (!mem_mng->umem) {
7794 		rte_errno = errno;
7795 		mlx5_free(mem);
7796 		return -rte_errno;
7797 	}
7798 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7799 	mkey_attr.addr = (uintptr_t)mem;
7800 	mkey_attr.size = size;
7801 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7802 	mkey_attr.pd = sh->cdev->pdn;
7803 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7804 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7805 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->cdev->ctx, &mkey_attr);
7806 	if (!mem_mng->dm) {
7807 		mlx5_os_umem_dereg(mem_mng->umem);
7808 		rte_errno = errno;
7809 		mlx5_free(mem);
7810 		return -rte_errno;
7811 	}
7812 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7813 	raw_data = (volatile struct flow_counter_stats *)mem;
7814 	for (i = 0; i < raws_n; ++i) {
7815 		mem_mng->raws[i].mem_mng = mem_mng;
7816 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7817 	}
7818 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7819 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7820 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7821 				 next);
7822 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7823 	sh->cmng.mem_mng = mem_mng;
7824 	return 0;
7825 }
7826 
7827 /**
7828  * Set the statistic memory to the new counter pool.
7829  *
7830  * @param[in] sh
7831  *   Pointer to mlx5_dev_ctx_shared object.
7832  * @param[in] pool
7833  *   Pointer to the pool to set the statistic memory.
7834  *
7835  * @return
7836  *   0 on success, a negative errno value otherwise.
7837  */
7838 static int
7839 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7840 			       struct mlx5_flow_counter_pool *pool)
7841 {
7842 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7843 	/* Resize statistic memory once used out. */
7844 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7845 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7846 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7847 		return -1;
7848 	}
7849 	rte_spinlock_lock(&pool->sl);
7850 	pool->raw = cmng->mem_mng->raws + pool->index %
7851 		    MLX5_CNT_CONTAINER_RESIZE;
7852 	rte_spinlock_unlock(&pool->sl);
7853 	pool->raw_hw = NULL;
7854 	return 0;
7855 }
7856 
7857 #define MLX5_POOL_QUERY_FREQ_US 1000000
7858 
7859 /**
7860  * Set the periodic procedure for triggering asynchronous batch queries for all
7861  * the counter pools.
7862  *
7863  * @param[in] sh
7864  *   Pointer to mlx5_dev_ctx_shared object.
7865  */
7866 void
7867 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7868 {
7869 	uint32_t pools_n, us;
7870 
7871 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7872 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7873 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7874 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7875 		sh->cmng.query_thread_on = 0;
7876 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7877 	} else {
7878 		sh->cmng.query_thread_on = 1;
7879 	}
7880 }
7881 
7882 /**
7883  * The periodic procedure for triggering asynchronous batch queries for all the
7884  * counter pools. This function is probably called by the host thread.
7885  *
7886  * @param[in] arg
7887  *   The parameter for the alarm process.
7888  */
7889 void
7890 mlx5_flow_query_alarm(void *arg)
7891 {
7892 	struct mlx5_dev_ctx_shared *sh = arg;
7893 	int ret;
7894 	uint16_t pool_index = sh->cmng.pool_index;
7895 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7896 	struct mlx5_flow_counter_pool *pool;
7897 	uint16_t n_valid;
7898 
7899 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7900 		goto set_alarm;
7901 	rte_spinlock_lock(&cmng->pool_update_sl);
7902 	pool = cmng->pools[pool_index];
7903 	n_valid = cmng->n_valid;
7904 	rte_spinlock_unlock(&cmng->pool_update_sl);
7905 	/* Set the statistic memory to the new created pool. */
7906 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7907 		goto set_alarm;
7908 	if (pool->raw_hw)
7909 		/* There is a pool query in progress. */
7910 		goto set_alarm;
7911 	pool->raw_hw =
7912 		LIST_FIRST(&sh->cmng.free_stat_raws);
7913 	if (!pool->raw_hw)
7914 		/* No free counter statistics raw memory. */
7915 		goto set_alarm;
7916 	/*
7917 	 * Identify the counters released between query trigger and query
7918 	 * handle more efficiently. The counter released in this gap period
7919 	 * should wait for a new round of query as the new arrived packets
7920 	 * will not be taken into account.
7921 	 */
7922 	pool->query_gen++;
7923 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7924 					       MLX5_COUNTERS_PER_POOL,
7925 					       NULL, NULL,
7926 					       pool->raw_hw->mem_mng->dm->id,
7927 					       (void *)(uintptr_t)
7928 					       pool->raw_hw->data,
7929 					       sh->devx_comp,
7930 					       (uint64_t)(uintptr_t)pool);
7931 	if (ret) {
7932 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7933 			" %d", pool->min_dcs->id);
7934 		pool->raw_hw = NULL;
7935 		goto set_alarm;
7936 	}
7937 	LIST_REMOVE(pool->raw_hw, next);
7938 	sh->cmng.pending_queries++;
7939 	pool_index++;
7940 	if (pool_index >= n_valid)
7941 		pool_index = 0;
7942 set_alarm:
7943 	sh->cmng.pool_index = pool_index;
7944 	mlx5_set_query_alarm(sh);
7945 }
7946 
7947 /**
7948  * Check and callback event for new aged flow in the counter pool
7949  *
7950  * @param[in] sh
7951  *   Pointer to mlx5_dev_ctx_shared object.
7952  * @param[in] pool
7953  *   Pointer to Current counter pool.
7954  */
7955 static void
7956 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7957 		   struct mlx5_flow_counter_pool *pool)
7958 {
7959 	struct mlx5_priv *priv;
7960 	struct mlx5_flow_counter *cnt;
7961 	struct mlx5_age_info *age_info;
7962 	struct mlx5_age_param *age_param;
7963 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7964 	struct mlx5_counter_stats_raw *prev = pool->raw;
7965 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7966 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7967 	uint16_t expected = AGE_CANDIDATE;
7968 	uint32_t i;
7969 
7970 	pool->time_of_last_age_check = curr_time;
7971 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7972 		cnt = MLX5_POOL_GET_CNT(pool, i);
7973 		age_param = MLX5_CNT_TO_AGE(cnt);
7974 		if (__atomic_load_n(&age_param->state,
7975 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7976 			continue;
7977 		if (cur->data[i].hits != prev->data[i].hits) {
7978 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7979 					 __ATOMIC_RELAXED);
7980 			continue;
7981 		}
7982 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7983 				       time_delta,
7984 				       __ATOMIC_RELAXED) <= age_param->timeout)
7985 			continue;
7986 		/**
7987 		 * Hold the lock first, or if between the
7988 		 * state AGE_TMOUT and tailq operation the
7989 		 * release happened, the release procedure
7990 		 * may delete a non-existent tailq node.
7991 		 */
7992 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7993 		age_info = GET_PORT_AGE_INFO(priv);
7994 		rte_spinlock_lock(&age_info->aged_sl);
7995 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7996 						AGE_TMOUT, false,
7997 						__ATOMIC_RELAXED,
7998 						__ATOMIC_RELAXED)) {
7999 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8000 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8001 		}
8002 		rte_spinlock_unlock(&age_info->aged_sl);
8003 	}
8004 	mlx5_age_event_prepare(sh);
8005 }
8006 
8007 /**
8008  * Handler for the HW respond about ready values from an asynchronous batch
8009  * query. This function is probably called by the host thread.
8010  *
8011  * @param[in] sh
8012  *   The pointer to the shared device context.
8013  * @param[in] async_id
8014  *   The Devx async ID.
8015  * @param[in] status
8016  *   The status of the completion.
8017  */
8018 void
8019 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8020 				  uint64_t async_id, int status)
8021 {
8022 	struct mlx5_flow_counter_pool *pool =
8023 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8024 	struct mlx5_counter_stats_raw *raw_to_free;
8025 	uint8_t query_gen = pool->query_gen ^ 1;
8026 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8027 	enum mlx5_counter_type cnt_type =
8028 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8029 				MLX5_COUNTER_TYPE_ORIGIN;
8030 
8031 	if (unlikely(status)) {
8032 		raw_to_free = pool->raw_hw;
8033 	} else {
8034 		raw_to_free = pool->raw;
8035 		if (pool->is_aged)
8036 			mlx5_flow_aging_check(sh, pool);
8037 		rte_spinlock_lock(&pool->sl);
8038 		pool->raw = pool->raw_hw;
8039 		rte_spinlock_unlock(&pool->sl);
8040 		/* Be sure the new raw counters data is updated in memory. */
8041 		rte_io_wmb();
8042 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8043 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8044 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8045 				     &pool->counters[query_gen], next);
8046 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8047 		}
8048 	}
8049 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8050 	pool->raw_hw = NULL;
8051 	sh->cmng.pending_queries--;
8052 }
8053 
8054 static int
8055 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8056 		    const struct flow_grp_info *grp_info,
8057 		    struct rte_flow_error *error)
8058 {
8059 	if (grp_info->transfer && grp_info->external &&
8060 	    grp_info->fdb_def_rule) {
8061 		if (group == UINT32_MAX)
8062 			return rte_flow_error_set
8063 						(error, EINVAL,
8064 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8065 						 NULL,
8066 						 "group index not supported");
8067 		*table = group + 1;
8068 	} else {
8069 		*table = group;
8070 	}
8071 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8072 	return 0;
8073 }
8074 
8075 /**
8076  * Translate the rte_flow group index to HW table value.
8077  *
8078  * If tunnel offload is disabled, all group ids converted to flow table
8079  * id using the standard method.
8080  * If tunnel offload is enabled, group id can be converted using the
8081  * standard or tunnel conversion method. Group conversion method
8082  * selection depends on flags in `grp_info` parameter:
8083  * - Internal (grp_info.external == 0) groups conversion uses the
8084  *   standard method.
8085  * - Group ids in JUMP action converted with the tunnel conversion.
8086  * - Group id in rule attribute conversion depends on a rule type and
8087  *   group id value:
8088  *   ** non zero group attributes converted with the tunnel method
8089  *   ** zero group attribute in non-tunnel rule is converted using the
8090  *      standard method - there's only one root table
8091  *   ** zero group attribute in steer tunnel rule is converted with the
8092  *      standard method - single root table
8093  *   ** zero group attribute in match tunnel rule is a special OvS
8094  *      case: that value is used for portability reasons. That group
8095  *      id is converted with the tunnel conversion method.
8096  *
8097  * @param[in] dev
8098  *   Port device
8099  * @param[in] tunnel
8100  *   PMD tunnel offload object
8101  * @param[in] group
8102  *   rte_flow group index value.
8103  * @param[out] table
8104  *   HW table value.
8105  * @param[in] grp_info
8106  *   flags used for conversion
8107  * @param[out] error
8108  *   Pointer to error structure.
8109  *
8110  * @return
8111  *   0 on success, a negative errno value otherwise and rte_errno is set.
8112  */
8113 int
8114 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8115 			 const struct mlx5_flow_tunnel *tunnel,
8116 			 uint32_t group, uint32_t *table,
8117 			 const struct flow_grp_info *grp_info,
8118 			 struct rte_flow_error *error)
8119 {
8120 	int ret;
8121 	bool standard_translation;
8122 
8123 	if (!grp_info->skip_scale && grp_info->external &&
8124 	    group < MLX5_MAX_TABLES_EXTERNAL)
8125 		group *= MLX5_FLOW_TABLE_FACTOR;
8126 	if (is_tunnel_offload_active(dev)) {
8127 		standard_translation = !grp_info->external ||
8128 					grp_info->std_tbl_fix;
8129 	} else {
8130 		standard_translation = true;
8131 	}
8132 	DRV_LOG(DEBUG,
8133 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8134 		dev->data->port_id, group, grp_info->transfer,
8135 		grp_info->external, grp_info->fdb_def_rule,
8136 		standard_translation ? "STANDARD" : "TUNNEL");
8137 	if (standard_translation)
8138 		ret = flow_group_to_table(dev->data->port_id, group, table,
8139 					  grp_info, error);
8140 	else
8141 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8142 						      table, error);
8143 
8144 	return ret;
8145 }
8146 
8147 /**
8148  * Discover availability of metadata reg_c's.
8149  *
8150  * Iteratively use test flows to check availability.
8151  *
8152  * @param[in] dev
8153  *   Pointer to the Ethernet device structure.
8154  *
8155  * @return
8156  *   0 on success, a negative errno value otherwise and rte_errno is set.
8157  */
8158 int
8159 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8160 {
8161 	struct mlx5_priv *priv = dev->data->dev_private;
8162 	enum modify_reg idx;
8163 	int n = 0;
8164 
8165 	/* reg_c[0] and reg_c[1] are reserved. */
8166 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8167 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8168 	/* Discover availability of other reg_c's. */
8169 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8170 		struct rte_flow_attr attr = {
8171 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8172 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8173 			.ingress = 1,
8174 		};
8175 		struct rte_flow_item items[] = {
8176 			[0] = {
8177 				.type = RTE_FLOW_ITEM_TYPE_END,
8178 			},
8179 		};
8180 		struct rte_flow_action actions[] = {
8181 			[0] = {
8182 				.type = (enum rte_flow_action_type)
8183 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8184 				.conf = &(struct mlx5_flow_action_copy_mreg){
8185 					.src = REG_C_1,
8186 					.dst = idx,
8187 				},
8188 			},
8189 			[1] = {
8190 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8191 				.conf = &(struct rte_flow_action_jump){
8192 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8193 				},
8194 			},
8195 			[2] = {
8196 				.type = RTE_FLOW_ACTION_TYPE_END,
8197 			},
8198 		};
8199 		uint32_t flow_idx;
8200 		struct rte_flow *flow;
8201 		struct rte_flow_error error;
8202 
8203 		if (!priv->config.dv_flow_en)
8204 			break;
8205 		/* Create internal flow, validation skips copy action. */
8206 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8207 					items, actions, false, &error);
8208 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8209 				      flow_idx);
8210 		if (!flow)
8211 			continue;
8212 		priv->sh->flow_mreg_c[n++] = idx;
8213 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8214 	}
8215 	for (; n < MLX5_MREG_C_NUM; ++n)
8216 		priv->sh->flow_mreg_c[n] = REG_NON;
8217 	priv->sh->metadata_regc_check_flag = 1;
8218 	return 0;
8219 }
8220 
8221 int
8222 save_dump_file(const uint8_t *data, uint32_t size,
8223 	uint32_t type, uint64_t id, void *arg, FILE *file)
8224 {
8225 	char line[BUF_SIZE];
8226 	uint32_t out = 0;
8227 	uint32_t k;
8228 	uint32_t actions_num;
8229 	struct rte_flow_query_count *count;
8230 
8231 	memset(line, 0, BUF_SIZE);
8232 	switch (type) {
8233 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8234 		actions_num = *(uint32_t *)(arg);
8235 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8236 				type, id, actions_num);
8237 		break;
8238 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8239 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8240 				type, id);
8241 		break;
8242 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8243 		count = (struct rte_flow_query_count *)arg;
8244 		fprintf(file,
8245 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8246 			type, id, count->hits, count->bytes);
8247 		return 0;
8248 	default:
8249 		return -1;
8250 	}
8251 
8252 	for (k = 0; k < size; k++) {
8253 		/* Make sure we do not overrun the line buffer length. */
8254 		if (out >= BUF_SIZE - 4) {
8255 			line[out] = '\0';
8256 			break;
8257 		}
8258 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8259 				(data[k]) & 0xff);
8260 	}
8261 	fprintf(file, "%s\n", line);
8262 	return 0;
8263 }
8264 
8265 int
8266 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8267 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8268 {
8269 	struct rte_flow_action action[2];
8270 	enum mlx5_flow_drv_type ftype;
8271 	const struct mlx5_flow_driver_ops *fops;
8272 
8273 	if (!flow) {
8274 		return rte_flow_error_set(error, ENOENT,
8275 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8276 				NULL,
8277 				"invalid flow handle");
8278 	}
8279 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8280 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8281 	if (flow->counter) {
8282 		memset(count, 0, sizeof(struct rte_flow_query_count));
8283 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8284 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8285 						ftype < MLX5_FLOW_TYPE_MAX);
8286 		fops = flow_get_drv_ops(ftype);
8287 		return fops->query(dev, flow, action, count, error);
8288 	}
8289 	return -1;
8290 }
8291 
8292 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8293 /**
8294  * Dump flow ipool data to file
8295  *
8296  * @param[in] dev
8297  *   The pointer to Ethernet device.
8298  * @param[in] file
8299  *   A pointer to a file for output.
8300  * @param[out] error
8301  *   Perform verbose error reporting if not NULL. PMDs initialize this
8302  *   structure in case of error only.
8303  * @return
8304  *   0 on success, a negative value otherwise.
8305  */
8306 int
8307 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8308 	struct rte_flow *flow, FILE *file,
8309 	struct rte_flow_error *error)
8310 {
8311 	struct mlx5_priv *priv = dev->data->dev_private;
8312 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8313 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8314 	uint32_t handle_idx;
8315 	struct mlx5_flow_handle *dh;
8316 	struct rte_flow_query_count count;
8317 	uint32_t actions_num;
8318 	const uint8_t *data;
8319 	size_t size;
8320 	uint64_t id;
8321 	uint32_t type;
8322 	void *action = NULL;
8323 
8324 	if (!flow) {
8325 		return rte_flow_error_set(error, ENOENT,
8326 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8327 				NULL,
8328 				"invalid flow handle");
8329 	}
8330 	handle_idx = flow->dev_handles;
8331 	while (handle_idx) {
8332 		dh = mlx5_ipool_get(priv->sh->ipool
8333 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8334 		if (!dh)
8335 			continue;
8336 		handle_idx = dh->next.next;
8337 
8338 		/* query counter */
8339 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8340 		flow_dv_query_count_ptr(dev, flow->counter,
8341 						&action, error);
8342 		if (action) {
8343 			id = (uint64_t)(uintptr_t)action;
8344 			if (!mlx5_flow_query_counter(dev, flow, &count, error))
8345 				save_dump_file(NULL, 0, type,
8346 						id, (void *)&count, file);
8347 		}
8348 		/* Get modify_hdr and encap_decap buf from ipools. */
8349 		encap_decap = NULL;
8350 		modify_hdr = dh->dvh.modify_hdr;
8351 
8352 		if (dh->dvh.rix_encap_decap) {
8353 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8354 						[MLX5_IPOOL_DECAP_ENCAP],
8355 						dh->dvh.rix_encap_decap);
8356 		}
8357 		if (modify_hdr) {
8358 			data = (const uint8_t *)modify_hdr->actions;
8359 			size = (size_t)(modify_hdr->actions_num) * 8;
8360 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8361 			actions_num = modify_hdr->actions_num;
8362 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8363 			save_dump_file(data, size, type, id,
8364 						(void *)(&actions_num), file);
8365 		}
8366 		if (encap_decap) {
8367 			data = encap_decap->buf;
8368 			size = encap_decap->size;
8369 			id = (uint64_t)(uintptr_t)encap_decap->action;
8370 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8371 			save_dump_file(data, size, type,
8372 						id, NULL, file);
8373 		}
8374 	}
8375 	return 0;
8376 }
8377 
8378 /**
8379  * Dump all flow's encap_decap/modify_hdr/counter data to file
8380  *
8381  * @param[in] dev
8382  *   The pointer to Ethernet device.
8383  * @param[in] file
8384  *   A pointer to a file for output.
8385  * @param[out] error
8386  *   Perform verbose error reporting if not NULL. PMDs initialize this
8387  *   structure in case of error only.
8388  * @return
8389  *   0 on success, a negative value otherwise.
8390  */
8391 static int
8392 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8393 	FILE *file, struct rte_flow_error *error)
8394 {
8395 	struct mlx5_priv *priv = dev->data->dev_private;
8396 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8397 	struct mlx5_hlist *h;
8398 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8399 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8400 	struct rte_flow_query_count count;
8401 	uint32_t actions_num;
8402 	const uint8_t *data;
8403 	size_t size;
8404 	uint64_t id;
8405 	uint32_t type;
8406 	uint32_t i;
8407 	uint32_t j;
8408 	struct mlx5_list_inconst *l_inconst;
8409 	struct mlx5_list_entry *e;
8410 	int lcore_index;
8411 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8412 	uint32_t max;
8413 	void *action;
8414 
8415 	/* encap_decap hlist is lcore_share, get global core cache. */
8416 	i = MLX5_LIST_GLOBAL;
8417 	h = sh->encaps_decaps;
8418 	if (h) {
8419 		for (j = 0; j <= h->mask; j++) {
8420 			l_inconst = &h->buckets[j].l;
8421 			if (!l_inconst || !l_inconst->cache[i])
8422 				continue;
8423 
8424 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8425 			while (e) {
8426 				encap_decap =
8427 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8428 				data = encap_decap->buf;
8429 				size = encap_decap->size;
8430 				id = (uint64_t)(uintptr_t)encap_decap->action;
8431 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8432 				save_dump_file(data, size, type,
8433 					id, NULL, file);
8434 				e = LIST_NEXT(e, next);
8435 			}
8436 		}
8437 	}
8438 
8439 	/* get modify_hdr */
8440 	h = sh->modify_cmds;
8441 	if (h) {
8442 		lcore_index = rte_lcore_index(rte_lcore_id());
8443 		if (unlikely(lcore_index == -1)) {
8444 			lcore_index = MLX5_LIST_NLCORE;
8445 			rte_spinlock_lock(&h->l_const.lcore_lock);
8446 		}
8447 		i = lcore_index;
8448 
8449 		for (j = 0; j <= h->mask; j++) {
8450 			l_inconst = &h->buckets[j].l;
8451 			if (!l_inconst || !l_inconst->cache[i])
8452 				continue;
8453 
8454 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8455 			while (e) {
8456 				modify_hdr =
8457 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8458 				data = (const uint8_t *)modify_hdr->actions;
8459 				size = (size_t)(modify_hdr->actions_num) * 8;
8460 				actions_num = modify_hdr->actions_num;
8461 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8462 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8463 				save_dump_file(data, size, type, id,
8464 						(void *)(&actions_num), file);
8465 				e = LIST_NEXT(e, next);
8466 			}
8467 		}
8468 
8469 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8470 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8471 	}
8472 
8473 	/* get counter */
8474 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8475 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8476 	for (j = 1; j <= max; j++) {
8477 		action = NULL;
8478 		flow_dv_query_count_ptr(dev, j, &action, error);
8479 		if (action) {
8480 			if (!flow_dv_query_count(dev, j, &count, error)) {
8481 				type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8482 				id = (uint64_t)(uintptr_t)action;
8483 				save_dump_file(NULL, 0, type,
8484 						id, (void *)&count, file);
8485 			}
8486 		}
8487 	}
8488 	return 0;
8489 }
8490 #endif
8491 
8492 /**
8493  * Dump flow raw hw data to file
8494  *
8495  * @param[in] dev
8496  *    The pointer to Ethernet device.
8497  * @param[in] file
8498  *   A pointer to a file for output.
8499  * @param[out] error
8500  *   Perform verbose error reporting if not NULL. PMDs initialize this
8501  *   structure in case of error only.
8502  * @return
8503  *   0 on success, a nagative value otherwise.
8504  */
8505 int
8506 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8507 		   FILE *file,
8508 		   struct rte_flow_error *error __rte_unused)
8509 {
8510 	struct mlx5_priv *priv = dev->data->dev_private;
8511 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8512 	uint32_t handle_idx;
8513 	int ret;
8514 	struct mlx5_flow_handle *dh;
8515 	struct rte_flow *flow;
8516 
8517 	if (!priv->config.dv_flow_en) {
8518 		if (fputs("device dv flow disabled\n", file) <= 0)
8519 			return -errno;
8520 		return -ENOTSUP;
8521 	}
8522 
8523 	/* dump all */
8524 	if (!flow_idx) {
8525 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8526 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8527 			return -EINVAL;
8528 #endif
8529 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8530 					sh->rx_domain,
8531 					sh->tx_domain, file);
8532 	}
8533 	/* dump one */
8534 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8535 			(uintptr_t)(void *)flow_idx);
8536 	if (!flow)
8537 		return -EINVAL;
8538 
8539 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8540 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8541 #endif
8542 	handle_idx = flow->dev_handles;
8543 	while (handle_idx) {
8544 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8545 				handle_idx);
8546 		if (!dh)
8547 			return -ENOENT;
8548 		if (dh->drv_flow) {
8549 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8550 					file);
8551 			if (ret)
8552 				return -ENOENT;
8553 		}
8554 		handle_idx = dh->next.next;
8555 	}
8556 	return 0;
8557 }
8558 
8559 /**
8560  * Get aged-out flows.
8561  *
8562  * @param[in] dev
8563  *   Pointer to the Ethernet device structure.
8564  * @param[in] context
8565  *   The address of an array of pointers to the aged-out flows contexts.
8566  * @param[in] nb_countexts
8567  *   The length of context array pointers.
8568  * @param[out] error
8569  *   Perform verbose error reporting if not NULL. Initialized in case of
8570  *   error only.
8571  *
8572  * @return
8573  *   how many contexts get in success, otherwise negative errno value.
8574  *   if nb_contexts is 0, return the amount of all aged contexts.
8575  *   if nb_contexts is not 0 , return the amount of aged flows reported
8576  *   in the context array.
8577  */
8578 int
8579 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8580 			uint32_t nb_contexts, struct rte_flow_error *error)
8581 {
8582 	const struct mlx5_flow_driver_ops *fops;
8583 	struct rte_flow_attr attr = { .transfer = 0 };
8584 
8585 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8586 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8587 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8588 						    error);
8589 	}
8590 	DRV_LOG(ERR,
8591 		"port %u get aged flows is not supported.",
8592 		 dev->data->port_id);
8593 	return -ENOTSUP;
8594 }
8595 
8596 /* Wrapper for driver action_validate op callback */
8597 static int
8598 flow_drv_action_validate(struct rte_eth_dev *dev,
8599 			 const struct rte_flow_indir_action_conf *conf,
8600 			 const struct rte_flow_action *action,
8601 			 const struct mlx5_flow_driver_ops *fops,
8602 			 struct rte_flow_error *error)
8603 {
8604 	static const char err_msg[] = "indirect action validation unsupported";
8605 
8606 	if (!fops->action_validate) {
8607 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8608 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8609 				   NULL, err_msg);
8610 		return -rte_errno;
8611 	}
8612 	return fops->action_validate(dev, conf, action, error);
8613 }
8614 
8615 /**
8616  * Destroys the shared action by handle.
8617  *
8618  * @param dev
8619  *   Pointer to Ethernet device structure.
8620  * @param[in] handle
8621  *   Handle for the indirect action object to be destroyed.
8622  * @param[out] error
8623  *   Perform verbose error reporting if not NULL. PMDs initialize this
8624  *   structure in case of error only.
8625  *
8626  * @return
8627  *   0 on success, a negative errno value otherwise and rte_errno is set.
8628  *
8629  * @note: wrapper for driver action_create op callback.
8630  */
8631 static int
8632 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8633 			   struct rte_flow_action_handle *handle,
8634 			   struct rte_flow_error *error)
8635 {
8636 	static const char err_msg[] = "indirect action destruction unsupported";
8637 	struct rte_flow_attr attr = { .transfer = 0 };
8638 	const struct mlx5_flow_driver_ops *fops =
8639 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8640 
8641 	if (!fops->action_destroy) {
8642 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8643 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8644 				   NULL, err_msg);
8645 		return -rte_errno;
8646 	}
8647 	return fops->action_destroy(dev, handle, error);
8648 }
8649 
8650 /* Wrapper for driver action_destroy op callback */
8651 static int
8652 flow_drv_action_update(struct rte_eth_dev *dev,
8653 		       struct rte_flow_action_handle *handle,
8654 		       const void *update,
8655 		       const struct mlx5_flow_driver_ops *fops,
8656 		       struct rte_flow_error *error)
8657 {
8658 	static const char err_msg[] = "indirect action update unsupported";
8659 
8660 	if (!fops->action_update) {
8661 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8662 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8663 				   NULL, err_msg);
8664 		return -rte_errno;
8665 	}
8666 	return fops->action_update(dev, handle, update, error);
8667 }
8668 
8669 /* Wrapper for driver action_destroy op callback */
8670 static int
8671 flow_drv_action_query(struct rte_eth_dev *dev,
8672 		      const struct rte_flow_action_handle *handle,
8673 		      void *data,
8674 		      const struct mlx5_flow_driver_ops *fops,
8675 		      struct rte_flow_error *error)
8676 {
8677 	static const char err_msg[] = "indirect action query unsupported";
8678 
8679 	if (!fops->action_query) {
8680 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8681 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8682 				   NULL, err_msg);
8683 		return -rte_errno;
8684 	}
8685 	return fops->action_query(dev, handle, data, error);
8686 }
8687 
8688 /**
8689  * Create indirect action for reuse in multiple flow rules.
8690  *
8691  * @param dev
8692  *   Pointer to Ethernet device structure.
8693  * @param conf
8694  *   Pointer to indirect action object configuration.
8695  * @param[in] action
8696  *   Action configuration for indirect action object creation.
8697  * @param[out] error
8698  *   Perform verbose error reporting if not NULL. PMDs initialize this
8699  *   structure in case of error only.
8700  * @return
8701  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8702  */
8703 static struct rte_flow_action_handle *
8704 mlx5_action_handle_create(struct rte_eth_dev *dev,
8705 			  const struct rte_flow_indir_action_conf *conf,
8706 			  const struct rte_flow_action *action,
8707 			  struct rte_flow_error *error)
8708 {
8709 	static const char err_msg[] = "indirect action creation unsupported";
8710 	struct rte_flow_attr attr = { .transfer = 0 };
8711 	const struct mlx5_flow_driver_ops *fops =
8712 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8713 
8714 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8715 		return NULL;
8716 	if (!fops->action_create) {
8717 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8718 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8719 				   NULL, err_msg);
8720 		return NULL;
8721 	}
8722 	return fops->action_create(dev, conf, action, error);
8723 }
8724 
8725 /**
8726  * Updates inplace the indirect action configuration pointed by *handle*
8727  * with the configuration provided as *update* argument.
8728  * The update of the indirect action configuration effects all flow rules
8729  * reusing the action via handle.
8730  *
8731  * @param dev
8732  *   Pointer to Ethernet device structure.
8733  * @param[in] handle
8734  *   Handle for the indirect action to be updated.
8735  * @param[in] update
8736  *   Action specification used to modify the action pointed by handle.
8737  *   *update* could be of same type with the action pointed by the *handle*
8738  *   handle argument, or some other structures like a wrapper, depending on
8739  *   the indirect action type.
8740  * @param[out] error
8741  *   Perform verbose error reporting if not NULL. PMDs initialize this
8742  *   structure in case of error only.
8743  *
8744  * @return
8745  *   0 on success, a negative errno value otherwise and rte_errno is set.
8746  */
8747 static int
8748 mlx5_action_handle_update(struct rte_eth_dev *dev,
8749 		struct rte_flow_action_handle *handle,
8750 		const void *update,
8751 		struct rte_flow_error *error)
8752 {
8753 	struct rte_flow_attr attr = { .transfer = 0 };
8754 	const struct mlx5_flow_driver_ops *fops =
8755 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8756 	int ret;
8757 
8758 	ret = flow_drv_action_validate(dev, NULL,
8759 			(const struct rte_flow_action *)update, fops, error);
8760 	if (ret)
8761 		return ret;
8762 	return flow_drv_action_update(dev, handle, update, fops,
8763 				      error);
8764 }
8765 
8766 /**
8767  * Query the indirect action by handle.
8768  *
8769  * This function allows retrieving action-specific data such as counters.
8770  * Data is gathered by special action which may be present/referenced in
8771  * more than one flow rule definition.
8772  *
8773  * see @RTE_FLOW_ACTION_TYPE_COUNT
8774  *
8775  * @param dev
8776  *   Pointer to Ethernet device structure.
8777  * @param[in] handle
8778  *   Handle for the indirect action to query.
8779  * @param[in, out] data
8780  *   Pointer to storage for the associated query data type.
8781  * @param[out] error
8782  *   Perform verbose error reporting if not NULL. PMDs initialize this
8783  *   structure in case of error only.
8784  *
8785  * @return
8786  *   0 on success, a negative errno value otherwise and rte_errno is set.
8787  */
8788 static int
8789 mlx5_action_handle_query(struct rte_eth_dev *dev,
8790 			 const struct rte_flow_action_handle *handle,
8791 			 void *data,
8792 			 struct rte_flow_error *error)
8793 {
8794 	struct rte_flow_attr attr = { .transfer = 0 };
8795 	const struct mlx5_flow_driver_ops *fops =
8796 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8797 
8798 	return flow_drv_action_query(dev, handle, data, fops, error);
8799 }
8800 
8801 /**
8802  * Destroy all indirect actions (shared RSS).
8803  *
8804  * @param dev
8805  *   Pointer to Ethernet device.
8806  *
8807  * @return
8808  *   0 on success, a negative errno value otherwise and rte_errno is set.
8809  */
8810 int
8811 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8812 {
8813 	struct rte_flow_error error;
8814 	struct mlx5_priv *priv = dev->data->dev_private;
8815 	struct mlx5_shared_action_rss *shared_rss;
8816 	int ret = 0;
8817 	uint32_t idx;
8818 
8819 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8820 		      priv->rss_shared_actions, idx, shared_rss, next) {
8821 		ret |= mlx5_action_handle_destroy(dev,
8822 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8823 	}
8824 	return ret;
8825 }
8826 
8827 /**
8828  * Validate existing indirect actions against current device configuration
8829  * and attach them to device resources.
8830  *
8831  * @param dev
8832  *   Pointer to Ethernet device.
8833  *
8834  * @return
8835  *   0 on success, a negative errno value otherwise and rte_errno is set.
8836  */
8837 int
8838 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8839 {
8840 	struct mlx5_priv *priv = dev->data->dev_private;
8841 	struct mlx5_indexed_pool *ipool =
8842 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8843 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8844 	int ret = 0;
8845 	uint32_t idx;
8846 
8847 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8848 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8849 		const char *message;
8850 		uint32_t queue_idx;
8851 
8852 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8853 					       ind_tbl->queues_n,
8854 					       &message, &queue_idx);
8855 		if (ret != 0) {
8856 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8857 				dev->data->port_id, ind_tbl->queues[queue_idx],
8858 				message);
8859 			break;
8860 		}
8861 	}
8862 	if (ret != 0)
8863 		return ret;
8864 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8865 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8866 
8867 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8868 		if (ret != 0) {
8869 			DRV_LOG(ERR, "Port %u could not attach "
8870 				"indirection table obj %p",
8871 				dev->data->port_id, (void *)ind_tbl);
8872 			goto error;
8873 		}
8874 	}
8875 	return 0;
8876 error:
8877 	shared_rss_last = shared_rss;
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 
8881 		if (shared_rss == shared_rss_last)
8882 			break;
8883 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8884 			DRV_LOG(CRIT, "Port %u could not detach "
8885 				"indirection table obj %p on rollback",
8886 				dev->data->port_id, (void *)ind_tbl);
8887 	}
8888 	return ret;
8889 }
8890 
8891 /**
8892  * Detach indirect actions of the device from its resources.
8893  *
8894  * @param dev
8895  *   Pointer to Ethernet device.
8896  *
8897  * @return
8898  *   0 on success, a negative errno value otherwise and rte_errno is set.
8899  */
8900 int
8901 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8902 {
8903 	struct mlx5_priv *priv = dev->data->dev_private;
8904 	struct mlx5_indexed_pool *ipool =
8905 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8906 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8907 	int ret = 0;
8908 	uint32_t idx;
8909 
8910 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8911 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8912 
8913 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8914 		if (ret != 0) {
8915 			DRV_LOG(ERR, "Port %u could not detach "
8916 				"indirection table obj %p",
8917 				dev->data->port_id, (void *)ind_tbl);
8918 			goto error;
8919 		}
8920 	}
8921 	return 0;
8922 error:
8923 	shared_rss_last = shared_rss;
8924 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8925 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8926 
8927 		if (shared_rss == shared_rss_last)
8928 			break;
8929 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8930 			DRV_LOG(CRIT, "Port %u could not attach "
8931 				"indirection table obj %p on rollback",
8932 				dev->data->port_id, (void *)ind_tbl);
8933 	}
8934 	return ret;
8935 }
8936 
8937 #ifndef HAVE_MLX5DV_DR
8938 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8939 #else
8940 #define MLX5_DOMAIN_SYNC_FLOW \
8941 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8942 #endif
8943 
8944 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8945 {
8946 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8947 	const struct mlx5_flow_driver_ops *fops;
8948 	int ret;
8949 	struct rte_flow_attr attr = { .transfer = 0 };
8950 
8951 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8952 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8953 	if (ret > 0)
8954 		ret = -ret;
8955 	return ret;
8956 }
8957 
8958 const struct mlx5_flow_tunnel *
8959 mlx5_get_tof(const struct rte_flow_item *item,
8960 	     const struct rte_flow_action *action,
8961 	     enum mlx5_tof_rule_type *rule_type)
8962 {
8963 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8964 		if (item->type == (typeof(item->type))
8965 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8966 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8967 			return flow_items_to_tunnel(item);
8968 		}
8969 	}
8970 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8971 		if (action->type == (typeof(action->type))
8972 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8973 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8974 			return flow_actions_to_tunnel(action);
8975 		}
8976 	}
8977 	return NULL;
8978 }
8979 
8980 /**
8981  * tunnel offload functionalilty is defined for DV environment only
8982  */
8983 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8984 __extension__
8985 union tunnel_offload_mark {
8986 	uint32_t val;
8987 	struct {
8988 		uint32_t app_reserve:8;
8989 		uint32_t table_id:15;
8990 		uint32_t transfer:1;
8991 		uint32_t _unused_:8;
8992 	};
8993 };
8994 
8995 static bool
8996 mlx5_access_tunnel_offload_db
8997 	(struct rte_eth_dev *dev,
8998 	 bool (*match)(struct rte_eth_dev *,
8999 		       struct mlx5_flow_tunnel *, const void *),
9000 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9001 	 void (*miss)(struct rte_eth_dev *, void *),
9002 	 void *ctx, bool lock_op);
9003 
9004 static int
9005 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9006 			     struct rte_flow *flow,
9007 			     const struct rte_flow_attr *attr,
9008 			     const struct rte_flow_action *app_actions,
9009 			     uint32_t flow_idx,
9010 			     const struct mlx5_flow_tunnel *tunnel,
9011 			     struct tunnel_default_miss_ctx *ctx,
9012 			     struct rte_flow_error *error)
9013 {
9014 	struct mlx5_priv *priv = dev->data->dev_private;
9015 	struct mlx5_flow *dev_flow;
9016 	struct rte_flow_attr miss_attr = *attr;
9017 	const struct rte_flow_item miss_items[2] = {
9018 		{
9019 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9020 			.spec = NULL,
9021 			.last = NULL,
9022 			.mask = NULL
9023 		},
9024 		{
9025 			.type = RTE_FLOW_ITEM_TYPE_END,
9026 			.spec = NULL,
9027 			.last = NULL,
9028 			.mask = NULL
9029 		}
9030 	};
9031 	union tunnel_offload_mark mark_id;
9032 	struct rte_flow_action_mark miss_mark;
9033 	struct rte_flow_action miss_actions[3] = {
9034 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9035 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9036 	};
9037 	const struct rte_flow_action_jump *jump_data;
9038 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9039 	struct flow_grp_info grp_info = {
9040 		.external = 1,
9041 		.transfer = attr->transfer,
9042 		.fdb_def_rule = !!priv->fdb_def_rule,
9043 		.std_tbl_fix = 0,
9044 	};
9045 	int ret;
9046 
9047 	if (!attr->transfer) {
9048 		uint32_t q_size;
9049 
9050 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9051 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9052 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9053 					 0, SOCKET_ID_ANY);
9054 		if (!ctx->queue)
9055 			return rte_flow_error_set
9056 				(error, ENOMEM,
9057 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9058 				NULL, "invalid default miss RSS");
9059 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9060 		ctx->action_rss.level = 0,
9061 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9062 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9063 		ctx->action_rss.queue_num = priv->reta_idx_n,
9064 		ctx->action_rss.key = priv->rss_conf.rss_key,
9065 		ctx->action_rss.queue = ctx->queue;
9066 		if (!priv->reta_idx_n || !priv->rxqs_n)
9067 			return rte_flow_error_set
9068 				(error, EINVAL,
9069 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9070 				NULL, "invalid port configuration");
9071 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9072 			ctx->action_rss.types = 0;
9073 		for (i = 0; i != priv->reta_idx_n; ++i)
9074 			ctx->queue[i] = (*priv->reta_idx)[i];
9075 	} else {
9076 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9077 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9078 	}
9079 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9080 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9081 	jump_data = app_actions->conf;
9082 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9083 	miss_attr.group = jump_data->group;
9084 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9085 				       &flow_table, &grp_info, error);
9086 	if (ret)
9087 		return rte_flow_error_set(error, EINVAL,
9088 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9089 					  NULL, "invalid tunnel id");
9090 	mark_id.app_reserve = 0;
9091 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9092 	mark_id.transfer = !!attr->transfer;
9093 	mark_id._unused_ = 0;
9094 	miss_mark.id = mark_id.val;
9095 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9096 				    miss_items, miss_actions, flow_idx, error);
9097 	if (!dev_flow)
9098 		return -rte_errno;
9099 	dev_flow->flow = flow;
9100 	dev_flow->external = true;
9101 	dev_flow->tunnel = tunnel;
9102 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9103 	/* Subflow object was created, we must include one in the list. */
9104 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9105 		      dev_flow->handle, next);
9106 	DRV_LOG(DEBUG,
9107 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9108 		dev->data->port_id, tunnel->app_tunnel.type,
9109 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9110 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9111 				  miss_actions, error);
9112 	if (!ret)
9113 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9114 						  error);
9115 
9116 	return ret;
9117 }
9118 
9119 static const struct mlx5_flow_tbl_data_entry  *
9120 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9121 {
9122 	struct mlx5_priv *priv = dev->data->dev_private;
9123 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9124 	struct mlx5_list_entry *he;
9125 	union tunnel_offload_mark mbits = { .val = mark };
9126 	union mlx5_flow_tbl_key table_key = {
9127 		{
9128 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9129 			.id = 0,
9130 			.reserved = 0,
9131 			.dummy = 0,
9132 			.is_fdb = !!mbits.transfer,
9133 			.is_egress = 0,
9134 		}
9135 	};
9136 	struct mlx5_flow_cb_ctx ctx = {
9137 		.data = &table_key.v64,
9138 	};
9139 
9140 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9141 	return he ?
9142 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9143 }
9144 
9145 static void
9146 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9147 				   struct mlx5_list_entry *entry)
9148 {
9149 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9150 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9151 
9152 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9153 			tunnel_flow_tbl_to_id(tte->flow_table));
9154 	mlx5_free(tte);
9155 }
9156 
9157 static int
9158 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9159 				  struct mlx5_list_entry *entry, void *cb_ctx)
9160 {
9161 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9162 	union tunnel_tbl_key tbl = {
9163 		.val = *(uint64_t *)(ctx->data),
9164 	};
9165 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9166 
9167 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9168 }
9169 
9170 static struct mlx5_list_entry *
9171 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9172 {
9173 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9174 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9175 	struct tunnel_tbl_entry *tte;
9176 	union tunnel_tbl_key tbl = {
9177 		.val = *(uint64_t *)(ctx->data),
9178 	};
9179 
9180 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9181 			  sizeof(*tte), 0,
9182 			  SOCKET_ID_ANY);
9183 	if (!tte)
9184 		goto err;
9185 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9186 			  &tte->flow_table);
9187 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9188 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9189 			tte->flow_table);
9190 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9191 				tte->flow_table);
9192 		goto err;
9193 	} else if (!tte->flow_table) {
9194 		goto err;
9195 	}
9196 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9197 	tte->tunnel_id = tbl.tunnel_id;
9198 	tte->group = tbl.group;
9199 	return &tte->hash;
9200 err:
9201 	if (tte)
9202 		mlx5_free(tte);
9203 	return NULL;
9204 }
9205 
9206 static struct mlx5_list_entry *
9207 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9208 				  struct mlx5_list_entry *oentry,
9209 				  void *cb_ctx __rte_unused)
9210 {
9211 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9212 						   0, SOCKET_ID_ANY);
9213 
9214 	if (!tte)
9215 		return NULL;
9216 	memcpy(tte, oentry, sizeof(*tte));
9217 	return &tte->hash;
9218 }
9219 
9220 static void
9221 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9222 				       struct mlx5_list_entry *entry)
9223 {
9224 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9225 
9226 	mlx5_free(tte);
9227 }
9228 
9229 static uint32_t
9230 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9231 				const struct mlx5_flow_tunnel *tunnel,
9232 				uint32_t group, uint32_t *table,
9233 				struct rte_flow_error *error)
9234 {
9235 	struct mlx5_list_entry *he;
9236 	struct tunnel_tbl_entry *tte;
9237 	union tunnel_tbl_key key = {
9238 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9239 		.group = group
9240 	};
9241 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9242 	struct mlx5_hlist *group_hash;
9243 	struct mlx5_flow_cb_ctx ctx = {
9244 		.data = &key.val,
9245 	};
9246 
9247 	group_hash = tunnel ? tunnel->groups : thub->groups;
9248 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9249 	if (!he)
9250 		return rte_flow_error_set(error, EINVAL,
9251 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9252 					  NULL,
9253 					  "tunnel group index not supported");
9254 	tte = container_of(he, typeof(*tte), hash);
9255 	*table = tte->flow_table;
9256 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9257 		dev->data->port_id, key.tunnel_id, group, *table);
9258 	return 0;
9259 }
9260 
9261 static void
9262 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9263 		      struct mlx5_flow_tunnel *tunnel)
9264 {
9265 	struct mlx5_priv *priv = dev->data->dev_private;
9266 	struct mlx5_indexed_pool *ipool;
9267 
9268 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9269 		dev->data->port_id, tunnel->tunnel_id);
9270 	LIST_REMOVE(tunnel, chain);
9271 	mlx5_hlist_destroy(tunnel->groups);
9272 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9273 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9274 }
9275 
9276 static bool
9277 mlx5_access_tunnel_offload_db
9278 	(struct rte_eth_dev *dev,
9279 	 bool (*match)(struct rte_eth_dev *,
9280 		       struct mlx5_flow_tunnel *, const void *),
9281 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9282 	 void (*miss)(struct rte_eth_dev *, void *),
9283 	 void *ctx, bool lock_op)
9284 {
9285 	bool verdict = false;
9286 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9287 	struct mlx5_flow_tunnel *tunnel;
9288 
9289 	rte_spinlock_lock(&thub->sl);
9290 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9291 		verdict = match(dev, tunnel, (const void *)ctx);
9292 		if (verdict)
9293 			break;
9294 	}
9295 	if (!lock_op)
9296 		rte_spinlock_unlock(&thub->sl);
9297 	if (verdict && hit)
9298 		hit(dev, tunnel, ctx);
9299 	if (!verdict && miss)
9300 		miss(dev, ctx);
9301 	if (lock_op)
9302 		rte_spinlock_unlock(&thub->sl);
9303 
9304 	return verdict;
9305 }
9306 
9307 struct tunnel_db_find_tunnel_id_ctx {
9308 	uint32_t tunnel_id;
9309 	struct mlx5_flow_tunnel *tunnel;
9310 };
9311 
9312 static bool
9313 find_tunnel_id_match(struct rte_eth_dev *dev,
9314 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9315 {
9316 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9317 
9318 	RTE_SET_USED(dev);
9319 	return tunnel->tunnel_id == ctx->tunnel_id;
9320 }
9321 
9322 static void
9323 find_tunnel_id_hit(struct rte_eth_dev *dev,
9324 		   struct mlx5_flow_tunnel *tunnel, void *x)
9325 {
9326 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9327 	RTE_SET_USED(dev);
9328 	ctx->tunnel = tunnel;
9329 }
9330 
9331 static struct mlx5_flow_tunnel *
9332 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9333 {
9334 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9335 		.tunnel_id = id,
9336 	};
9337 
9338 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9339 				      find_tunnel_id_hit, NULL, &ctx, true);
9340 
9341 	return ctx.tunnel;
9342 }
9343 
9344 static struct mlx5_flow_tunnel *
9345 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9346 			  const struct rte_flow_tunnel *app_tunnel)
9347 {
9348 	struct mlx5_priv *priv = dev->data->dev_private;
9349 	struct mlx5_indexed_pool *ipool;
9350 	struct mlx5_flow_tunnel *tunnel;
9351 	uint32_t id;
9352 
9353 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9354 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9355 	if (!tunnel)
9356 		return NULL;
9357 	if (id >= MLX5_MAX_TUNNELS) {
9358 		mlx5_ipool_free(ipool, id);
9359 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9360 		return NULL;
9361 	}
9362 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9363 					   priv->sh,
9364 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9365 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9366 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9367 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9368 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9369 	if (!tunnel->groups) {
9370 		mlx5_ipool_free(ipool, id);
9371 		return NULL;
9372 	}
9373 	/* initiate new PMD tunnel */
9374 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9375 	tunnel->tunnel_id = id;
9376 	tunnel->action.type = (typeof(tunnel->action.type))
9377 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9378 	tunnel->action.conf = tunnel;
9379 	tunnel->item.type = (typeof(tunnel->item.type))
9380 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9381 	tunnel->item.spec = tunnel;
9382 	tunnel->item.last = NULL;
9383 	tunnel->item.mask = NULL;
9384 
9385 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9386 		dev->data->port_id, tunnel->tunnel_id);
9387 
9388 	return tunnel;
9389 }
9390 
9391 struct tunnel_db_get_tunnel_ctx {
9392 	const struct rte_flow_tunnel *app_tunnel;
9393 	struct mlx5_flow_tunnel *tunnel;
9394 };
9395 
9396 static bool get_tunnel_match(struct rte_eth_dev *dev,
9397 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9398 {
9399 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9400 
9401 	RTE_SET_USED(dev);
9402 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9403 		       sizeof(*ctx->app_tunnel));
9404 }
9405 
9406 static void get_tunnel_hit(struct rte_eth_dev *dev,
9407 			   struct mlx5_flow_tunnel *tunnel, void *x)
9408 {
9409 	/* called under tunnel spinlock protection */
9410 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9411 
9412 	RTE_SET_USED(dev);
9413 	tunnel->refctn++;
9414 	ctx->tunnel = tunnel;
9415 }
9416 
9417 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9418 {
9419 	/* called under tunnel spinlock protection */
9420 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9421 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9422 
9423 	rte_spinlock_unlock(&thub->sl);
9424 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9425 	rte_spinlock_lock(&thub->sl);
9426 	if (ctx->tunnel) {
9427 		ctx->tunnel->refctn = 1;
9428 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9429 	}
9430 }
9431 
9432 
9433 static int
9434 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9435 		     const struct rte_flow_tunnel *app_tunnel,
9436 		     struct mlx5_flow_tunnel **tunnel)
9437 {
9438 	struct tunnel_db_get_tunnel_ctx ctx = {
9439 		.app_tunnel = app_tunnel,
9440 	};
9441 
9442 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9443 				      get_tunnel_miss, &ctx, true);
9444 	*tunnel = ctx.tunnel;
9445 	return ctx.tunnel ? 0 : -ENOMEM;
9446 }
9447 
9448 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9449 {
9450 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9451 
9452 	if (!thub)
9453 		return;
9454 	if (!LIST_EMPTY(&thub->tunnels))
9455 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9456 	mlx5_hlist_destroy(thub->groups);
9457 	mlx5_free(thub);
9458 }
9459 
9460 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9461 {
9462 	int err;
9463 	struct mlx5_flow_tunnel_hub *thub;
9464 
9465 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9466 			   0, SOCKET_ID_ANY);
9467 	if (!thub)
9468 		return -ENOMEM;
9469 	LIST_INIT(&thub->tunnels);
9470 	rte_spinlock_init(&thub->sl);
9471 	thub->groups = mlx5_hlist_create("flow groups", 64,
9472 					 false, true, sh,
9473 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9474 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9475 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9476 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9477 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9478 	if (!thub->groups) {
9479 		err = -rte_errno;
9480 		goto err;
9481 	}
9482 	sh->tunnel_hub = thub;
9483 
9484 	return 0;
9485 
9486 err:
9487 	if (thub->groups)
9488 		mlx5_hlist_destroy(thub->groups);
9489 	if (thub)
9490 		mlx5_free(thub);
9491 	return err;
9492 }
9493 
9494 static inline bool
9495 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9496 			  struct rte_flow_tunnel *tunnel,
9497 			  const char *err_msg)
9498 {
9499 	err_msg = NULL;
9500 	if (!is_tunnel_offload_active(dev)) {
9501 		err_msg = "tunnel offload was not activated";
9502 		goto out;
9503 	} else if (!tunnel) {
9504 		err_msg = "no application tunnel";
9505 		goto out;
9506 	}
9507 
9508 	switch (tunnel->type) {
9509 	default:
9510 		err_msg = "unsupported tunnel type";
9511 		goto out;
9512 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9513 	case RTE_FLOW_ITEM_TYPE_GRE:
9514 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9515 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9516 		break;
9517 	}
9518 
9519 out:
9520 	return !err_msg;
9521 }
9522 
9523 static int
9524 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9525 		    struct rte_flow_tunnel *app_tunnel,
9526 		    struct rte_flow_action **actions,
9527 		    uint32_t *num_of_actions,
9528 		    struct rte_flow_error *error)
9529 {
9530 	int ret;
9531 	struct mlx5_flow_tunnel *tunnel;
9532 	const char *err_msg = NULL;
9533 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9534 
9535 	if (!verdict)
9536 		return rte_flow_error_set(error, EINVAL,
9537 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9538 					  err_msg);
9539 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9540 	if (ret < 0) {
9541 		return rte_flow_error_set(error, ret,
9542 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9543 					  "failed to initialize pmd tunnel");
9544 	}
9545 	*actions = &tunnel->action;
9546 	*num_of_actions = 1;
9547 	return 0;
9548 }
9549 
9550 static int
9551 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9552 		       struct rte_flow_tunnel *app_tunnel,
9553 		       struct rte_flow_item **items,
9554 		       uint32_t *num_of_items,
9555 		       struct rte_flow_error *error)
9556 {
9557 	int ret;
9558 	struct mlx5_flow_tunnel *tunnel;
9559 	const char *err_msg = NULL;
9560 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9561 
9562 	if (!verdict)
9563 		return rte_flow_error_set(error, EINVAL,
9564 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9565 					  err_msg);
9566 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9567 	if (ret < 0) {
9568 		return rte_flow_error_set(error, ret,
9569 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9570 					  "failed to initialize pmd tunnel");
9571 	}
9572 	*items = &tunnel->item;
9573 	*num_of_items = 1;
9574 	return 0;
9575 }
9576 
9577 struct tunnel_db_element_release_ctx {
9578 	struct rte_flow_item *items;
9579 	struct rte_flow_action *actions;
9580 	uint32_t num_elements;
9581 	struct rte_flow_error *error;
9582 	int ret;
9583 };
9584 
9585 static bool
9586 tunnel_element_release_match(struct rte_eth_dev *dev,
9587 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9588 {
9589 	const struct tunnel_db_element_release_ctx *ctx = x;
9590 
9591 	RTE_SET_USED(dev);
9592 	if (ctx->num_elements != 1)
9593 		return false;
9594 	else if (ctx->items)
9595 		return ctx->items == &tunnel->item;
9596 	else if (ctx->actions)
9597 		return ctx->actions == &tunnel->action;
9598 
9599 	return false;
9600 }
9601 
9602 static void
9603 tunnel_element_release_hit(struct rte_eth_dev *dev,
9604 			   struct mlx5_flow_tunnel *tunnel, void *x)
9605 {
9606 	struct tunnel_db_element_release_ctx *ctx = x;
9607 	ctx->ret = 0;
9608 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9609 		mlx5_flow_tunnel_free(dev, tunnel);
9610 }
9611 
9612 static void
9613 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9614 {
9615 	struct tunnel_db_element_release_ctx *ctx = x;
9616 	RTE_SET_USED(dev);
9617 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9618 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9619 				      "invalid argument");
9620 }
9621 
9622 static int
9623 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9624 		       struct rte_flow_item *pmd_items,
9625 		       uint32_t num_items, struct rte_flow_error *err)
9626 {
9627 	struct tunnel_db_element_release_ctx ctx = {
9628 		.items = pmd_items,
9629 		.actions = NULL,
9630 		.num_elements = num_items,
9631 		.error = err,
9632 	};
9633 
9634 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9635 				      tunnel_element_release_hit,
9636 				      tunnel_element_release_miss, &ctx, false);
9637 
9638 	return ctx.ret;
9639 }
9640 
9641 static int
9642 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9643 			 struct rte_flow_action *pmd_actions,
9644 			 uint32_t num_actions, struct rte_flow_error *err)
9645 {
9646 	struct tunnel_db_element_release_ctx ctx = {
9647 		.items = NULL,
9648 		.actions = pmd_actions,
9649 		.num_elements = num_actions,
9650 		.error = err,
9651 	};
9652 
9653 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9654 				      tunnel_element_release_hit,
9655 				      tunnel_element_release_miss, &ctx, false);
9656 
9657 	return ctx.ret;
9658 }
9659 
9660 static int
9661 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9662 				  struct rte_mbuf *m,
9663 				  struct rte_flow_restore_info *info,
9664 				  struct rte_flow_error *err)
9665 {
9666 	uint64_t ol_flags = m->ol_flags;
9667 	const struct mlx5_flow_tbl_data_entry *tble;
9668 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9669 
9670 	if (!is_tunnel_offload_active(dev)) {
9671 		info->flags = 0;
9672 		return 0;
9673 	}
9674 
9675 	if ((ol_flags & mask) != mask)
9676 		goto err;
9677 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9678 	if (!tble) {
9679 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9680 			dev->data->port_id, m->hash.fdir.hi);
9681 		goto err;
9682 	}
9683 	MLX5_ASSERT(tble->tunnel);
9684 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9685 	info->group_id = tble->group_id;
9686 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9687 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9688 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9689 
9690 	return 0;
9691 
9692 err:
9693 	return rte_flow_error_set(err, EINVAL,
9694 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9695 				  "failed to get restore info");
9696 }
9697 
9698 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9699 static int
9700 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9701 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9702 			   __rte_unused struct rte_flow_action **actions,
9703 			   __rte_unused uint32_t *num_of_actions,
9704 			   __rte_unused struct rte_flow_error *error)
9705 {
9706 	return -ENOTSUP;
9707 }
9708 
9709 static int
9710 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9711 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9712 		       __rte_unused struct rte_flow_item **items,
9713 		       __rte_unused uint32_t *num_of_items,
9714 		       __rte_unused struct rte_flow_error *error)
9715 {
9716 	return -ENOTSUP;
9717 }
9718 
9719 static int
9720 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9721 			      __rte_unused struct rte_flow_item *pmd_items,
9722 			      __rte_unused uint32_t num_items,
9723 			      __rte_unused struct rte_flow_error *err)
9724 {
9725 	return -ENOTSUP;
9726 }
9727 
9728 static int
9729 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9730 				__rte_unused struct rte_flow_action *pmd_action,
9731 				__rte_unused uint32_t num_actions,
9732 				__rte_unused struct rte_flow_error *err)
9733 {
9734 	return -ENOTSUP;
9735 }
9736 
9737 static int
9738 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9739 				  __rte_unused struct rte_mbuf *m,
9740 				  __rte_unused struct rte_flow_restore_info *i,
9741 				  __rte_unused struct rte_flow_error *err)
9742 {
9743 	return -ENOTSUP;
9744 }
9745 
9746 static int
9747 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9748 			     __rte_unused struct rte_flow *flow,
9749 			     __rte_unused const struct rte_flow_attr *attr,
9750 			     __rte_unused const struct rte_flow_action *actions,
9751 			     __rte_unused uint32_t flow_idx,
9752 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9753 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9754 			     __rte_unused struct rte_flow_error *error)
9755 {
9756 	return -ENOTSUP;
9757 }
9758 
9759 static struct mlx5_flow_tunnel *
9760 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9761 		    __rte_unused uint32_t id)
9762 {
9763 	return NULL;
9764 }
9765 
9766 static void
9767 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9768 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9769 {
9770 }
9771 
9772 static uint32_t
9773 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9774 				__rte_unused const struct mlx5_flow_tunnel *t,
9775 				__rte_unused uint32_t group,
9776 				__rte_unused uint32_t *table,
9777 				struct rte_flow_error *error)
9778 {
9779 	return rte_flow_error_set(error, ENOTSUP,
9780 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9781 				  "tunnel offload requires DV support");
9782 }
9783 
9784 void
9785 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9786 			__rte_unused  uint16_t port_id)
9787 {
9788 }
9789 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9790 
9791 static void
9792 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9793 {
9794 	int ret;
9795 	struct rte_flow_error error;
9796 
9797 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9798 		char *item_name;
9799 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9800 				    sizeof(item_name),
9801 				    (void *)(uintptr_t)item->type, &error);
9802 		if (ret > 0)
9803 			printf("%s ", item_name);
9804 		else
9805 			printf("%d\n", (int)item->type);
9806 	}
9807 	printf("END\n");
9808 }
9809 
9810 static int
9811 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9812 {
9813 	const struct rte_flow_item_udp *spec = udp_item->spec;
9814 	const struct rte_flow_item_udp *mask = udp_item->mask;
9815 	uint16_t udp_dport = 0;
9816 
9817 	if (spec != NULL) {
9818 		if (!mask)
9819 			mask = &rte_flow_item_udp_mask;
9820 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9821 				mask->hdr.dst_port);
9822 	}
9823 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9824 }
9825 
9826 static const struct mlx5_flow_expand_node *
9827 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9828 		unsigned int item_idx,
9829 		const struct mlx5_flow_expand_node graph[],
9830 		const struct mlx5_flow_expand_node *node)
9831 {
9832 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9833 
9834 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9835 			node != NULL &&
9836 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9837 		/*
9838 		 * The expansion node is VXLAN and it is also the last
9839 		 * expandable item in the pattern, so need to continue
9840 		 * expansion of the inner tunnel.
9841 		 */
9842 		MLX5_ASSERT(item_idx > 0);
9843 		prev_item = pattern + item_idx - 1;
9844 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9845 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9846 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9847 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9848 	}
9849 	return node;
9850 }
9851 
9852 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9853 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9854 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9855 };
9856 
9857 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9858 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9859 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9860 	{ 9, 10, 11 }, { 12, 13, 14 },
9861 };
9862 
9863 /**
9864  * Discover the number of available flow priorities.
9865  *
9866  * @param dev
9867  *   Ethernet device.
9868  *
9869  * @return
9870  *   On success, number of available flow priorities.
9871  *   On failure, a negative errno-style code and rte_errno is set.
9872  */
9873 int
9874 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9875 {
9876 	static const uint16_t vprio[] = {8, 16};
9877 	const struct mlx5_priv *priv = dev->data->dev_private;
9878 	const struct mlx5_flow_driver_ops *fops;
9879 	enum mlx5_flow_drv_type type;
9880 	int ret;
9881 
9882 	type = mlx5_flow_os_get_type();
9883 	if (type == MLX5_FLOW_TYPE_MAX) {
9884 		type = MLX5_FLOW_TYPE_VERBS;
9885 		if (priv->sh->devx && priv->config.dv_flow_en)
9886 			type = MLX5_FLOW_TYPE_DV;
9887 	}
9888 	fops = flow_get_drv_ops(type);
9889 	if (fops->discover_priorities == NULL) {
9890 		DRV_LOG(ERR, "Priority discovery not supported");
9891 		rte_errno = ENOTSUP;
9892 		return -rte_errno;
9893 	}
9894 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9895 	if (ret < 0)
9896 		return ret;
9897 	switch (ret) {
9898 	case 8:
9899 		ret = RTE_DIM(priority_map_3);
9900 		break;
9901 	case 16:
9902 		ret = RTE_DIM(priority_map_5);
9903 		break;
9904 	default:
9905 		rte_errno = ENOTSUP;
9906 		DRV_LOG(ERR,
9907 			"port %u maximum priority: %d expected 8/16",
9908 			dev->data->port_id, ret);
9909 		return -rte_errno;
9910 	}
9911 	DRV_LOG(INFO, "port %u supported flow priorities:"
9912 		" 0-%d for ingress or egress root table,"
9913 		" 0-%d for non-root table or transfer root table.",
9914 		dev->data->port_id, ret - 2,
9915 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9916 	return ret;
9917 }
9918 
9919 /**
9920  * Adjust flow priority based on the highest layer and the request priority.
9921  *
9922  * @param[in] dev
9923  *   Pointer to the Ethernet device structure.
9924  * @param[in] priority
9925  *   The rule base priority.
9926  * @param[in] subpriority
9927  *   The priority based on the items.
9928  *
9929  * @return
9930  *   The new priority.
9931  */
9932 uint32_t
9933 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
9934 			  uint32_t subpriority)
9935 {
9936 	uint32_t res = 0;
9937 	struct mlx5_priv *priv = dev->data->dev_private;
9938 
9939 	switch (priv->sh->flow_max_priority) {
9940 	case RTE_DIM(priority_map_3):
9941 		res = priority_map_3[priority][subpriority];
9942 		break;
9943 	case RTE_DIM(priority_map_5):
9944 		res = priority_map_5[priority][subpriority];
9945 		break;
9946 	}
9947 	return  res;
9948 }
9949