xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 30a1de105a5f40d77b344a891c4a68f79e815c43)
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 /**
168  * Network Service Header (NSH) and its next protocol values
169  * are described in RFC-8393.
170  */
171 static enum rte_flow_item_type
172 mlx5_nsh_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
173 {
174 	enum rte_flow_item_type type;
175 
176 	switch (proto_mask & proto_spec) {
177 	case RTE_VXLAN_GPE_TYPE_IPV4:
178 		type = RTE_FLOW_ITEM_TYPE_IPV4;
179 		break;
180 	case RTE_VXLAN_GPE_TYPE_IPV6:
181 		type = RTE_VXLAN_GPE_TYPE_IPV6;
182 		break;
183 	case RTE_VXLAN_GPE_TYPE_ETH:
184 		type = RTE_FLOW_ITEM_TYPE_ETH;
185 		break;
186 	default:
187 		type = RTE_FLOW_ITEM_TYPE_END;
188 	}
189 	return type;
190 }
191 
192 static enum rte_flow_item_type
193 mlx5_inet_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
194 {
195 	enum rte_flow_item_type type;
196 
197 	switch (proto_mask & proto_spec) {
198 	case IPPROTO_UDP:
199 		type = RTE_FLOW_ITEM_TYPE_UDP;
200 		break;
201 	case IPPROTO_TCP:
202 		type = RTE_FLOW_ITEM_TYPE_TCP;
203 		break;
204 	case IPPROTO_IP:
205 		type = RTE_FLOW_ITEM_TYPE_IPV4;
206 		break;
207 	case IPPROTO_IPV6:
208 		type = RTE_FLOW_ITEM_TYPE_IPV6;
209 		break;
210 	default:
211 		type = RTE_FLOW_ITEM_TYPE_END;
212 	}
213 	return type;
214 }
215 
216 static enum rte_flow_item_type
217 mlx5_ethertype_to_item_type(rte_be16_t type_spec,
218 			    rte_be16_t type_mask, bool is_tunnel)
219 {
220 	enum rte_flow_item_type type;
221 
222 	switch (rte_be_to_cpu_16(type_spec & type_mask)) {
223 	case RTE_ETHER_TYPE_TEB:
224 		type = is_tunnel ?
225 		       RTE_FLOW_ITEM_TYPE_ETH : RTE_FLOW_ITEM_TYPE_END;
226 		break;
227 	case RTE_ETHER_TYPE_VLAN:
228 		type = !is_tunnel ?
229 		       RTE_FLOW_ITEM_TYPE_VLAN : RTE_FLOW_ITEM_TYPE_END;
230 		break;
231 	case RTE_ETHER_TYPE_IPV4:
232 		type = RTE_FLOW_ITEM_TYPE_IPV4;
233 		break;
234 	case RTE_ETHER_TYPE_IPV6:
235 		type = RTE_FLOW_ITEM_TYPE_IPV6;
236 		break;
237 	default:
238 		type = RTE_FLOW_ITEM_TYPE_END;
239 	}
240 	return type;
241 }
242 
243 static enum rte_flow_item_type
244 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
245 {
246 #define MLX5_XSET_ITEM_MASK_SPEC(type, fld)                              \
247 	do {                                                             \
248 		const void *m = item->mask;                              \
249 		const void *s = item->spec;                              \
250 		mask = m ?                                               \
251 			((const struct rte_flow_item_##type *)m)->fld :  \
252 			rte_flow_item_##type##_mask.fld;                 \
253 		spec = ((const struct rte_flow_item_##type *)s)->fld;    \
254 	} while (0)
255 
256 	enum rte_flow_item_type ret;
257 	uint16_t spec, mask;
258 
259 	if (item == NULL || item->spec == NULL)
260 		return RTE_FLOW_ITEM_TYPE_VOID;
261 	switch (item->type) {
262 	case RTE_FLOW_ITEM_TYPE_ETH:
263 		MLX5_XSET_ITEM_MASK_SPEC(eth, type);
264 		if (!mask)
265 			return RTE_FLOW_ITEM_TYPE_VOID;
266 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
267 		break;
268 	case RTE_FLOW_ITEM_TYPE_VLAN:
269 		MLX5_XSET_ITEM_MASK_SPEC(vlan, inner_type);
270 		if (!mask)
271 			return RTE_FLOW_ITEM_TYPE_VOID;
272 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
273 		break;
274 	case RTE_FLOW_ITEM_TYPE_IPV4:
275 		MLX5_XSET_ITEM_MASK_SPEC(ipv4, hdr.next_proto_id);
276 		if (!mask)
277 			return RTE_FLOW_ITEM_TYPE_VOID;
278 		ret = mlx5_inet_proto_to_item_type(spec, mask);
279 		break;
280 	case RTE_FLOW_ITEM_TYPE_IPV6:
281 		MLX5_XSET_ITEM_MASK_SPEC(ipv6, hdr.proto);
282 		if (!mask)
283 			return RTE_FLOW_ITEM_TYPE_VOID;
284 		ret = mlx5_inet_proto_to_item_type(spec, mask);
285 		break;
286 	case RTE_FLOW_ITEM_TYPE_GENEVE:
287 		MLX5_XSET_ITEM_MASK_SPEC(geneve, protocol);
288 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
289 		break;
290 	case RTE_FLOW_ITEM_TYPE_GRE:
291 		MLX5_XSET_ITEM_MASK_SPEC(gre, protocol);
292 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
293 		break;
294 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
295 		MLX5_XSET_ITEM_MASK_SPEC(vxlan_gpe, protocol);
296 		ret = mlx5_nsh_proto_to_item_type(spec, mask);
297 		break;
298 	default:
299 		ret = RTE_FLOW_ITEM_TYPE_VOID;
300 		break;
301 	}
302 	return ret;
303 #undef MLX5_XSET_ITEM_MASK_SPEC
304 }
305 
306 static const int *
307 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
308 		const int *next_node)
309 {
310 	const struct mlx5_flow_expand_node *node = NULL;
311 	const int *next = next_node;
312 
313 	while (next && *next) {
314 		/*
315 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
316 		 * flag set, because they were not found in the flow pattern.
317 		 */
318 		node = &graph[*next];
319 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
320 			break;
321 		next = node->next;
322 	}
323 	return next;
324 }
325 
326 #define MLX5_RSS_EXP_ELT_N 16
327 
328 /**
329  * Expand RSS flows into several possible flows according to the RSS hash
330  * fields requested and the driver capabilities.
331  *
332  * @param[out] buf
333  *   Buffer to store the result expansion.
334  * @param[in] size
335  *   Buffer size in bytes. If 0, @p buf can be NULL.
336  * @param[in] pattern
337  *   User flow pattern.
338  * @param[in] types
339  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
340  * @param[in] graph
341  *   Input graph to expand @p pattern according to @p types.
342  * @param[in] graph_root_index
343  *   Index of root node in @p graph, typically 0.
344  *
345  * @return
346  *   A positive value representing the size of @p buf in bytes regardless of
347  *   @p size on success, a negative errno value otherwise and rte_errno is
348  *   set, the following errors are defined:
349  *
350  *   -E2BIG: graph-depth @p graph is too deep.
351  *   -EINVAL: @p size has not enough space for expanded pattern.
352  */
353 static int
354 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
355 		     const struct rte_flow_item *pattern, uint64_t types,
356 		     const struct mlx5_flow_expand_node graph[],
357 		     int graph_root_index)
358 {
359 	const struct rte_flow_item *item;
360 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
361 	const int *next_node;
362 	const int *stack[MLX5_RSS_EXP_ELT_N];
363 	int stack_pos = 0;
364 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
365 	unsigned int i, item_idx, last_expand_item_idx = 0;
366 	size_t lsize;
367 	size_t user_pattern_size = 0;
368 	void *addr = NULL;
369 	const struct mlx5_flow_expand_node *next = NULL;
370 	struct rte_flow_item missed_item;
371 	int missed = 0;
372 	int elt = 0;
373 	const struct rte_flow_item *last_expand_item = NULL;
374 
375 	memset(&missed_item, 0, sizeof(missed_item));
376 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
377 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
378 	if (lsize > size)
379 		return -EINVAL;
380 	buf->entry[0].priority = 0;
381 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
382 	buf->entries = 0;
383 	addr = buf->entry[0].pattern;
384 	for (item = pattern, item_idx = 0;
385 			item->type != RTE_FLOW_ITEM_TYPE_END;
386 			item++, item_idx++) {
387 		if (!mlx5_flow_is_rss_expandable_item(item)) {
388 			user_pattern_size += sizeof(*item);
389 			continue;
390 		}
391 		last_expand_item = item;
392 		last_expand_item_idx = item_idx;
393 		i = 0;
394 		while (node->next && node->next[i]) {
395 			next = &graph[node->next[i]];
396 			if (next->type == item->type)
397 				break;
398 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
399 				node = next;
400 				i = 0;
401 			} else {
402 				++i;
403 			}
404 		}
405 		if (next)
406 			node = next;
407 		user_pattern_size += sizeof(*item);
408 	}
409 	user_pattern_size += sizeof(*item); /* Handle END item. */
410 	lsize += user_pattern_size;
411 	if (lsize > size)
412 		return -EINVAL;
413 	/* Copy the user pattern in the first entry of the buffer. */
414 	rte_memcpy(addr, pattern, user_pattern_size);
415 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
416 	buf->entries = 1;
417 	/* Start expanding. */
418 	memset(flow_items, 0, sizeof(flow_items));
419 	user_pattern_size -= sizeof(*item);
420 	/*
421 	 * Check if the last valid item has spec set, need complete pattern,
422 	 * and the pattern can be used for expansion.
423 	 */
424 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
425 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
426 		/* Item type END indicates expansion is not required. */
427 		return lsize;
428 	}
429 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
430 		next = NULL;
431 		missed = 1;
432 		i = 0;
433 		while (node->next && node->next[i]) {
434 			next = &graph[node->next[i]];
435 			if (next->type == missed_item.type) {
436 				flow_items[0].type = missed_item.type;
437 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
438 				break;
439 			}
440 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
441 				node = next;
442 				i = 0;
443 			} else {
444 				++i;
445 			}
446 			next = NULL;
447 		}
448 	}
449 	if (next && missed) {
450 		elt = 2; /* missed item + item end. */
451 		node = next;
452 		lsize += elt * sizeof(*item) + user_pattern_size;
453 		if (lsize > size)
454 			return -EINVAL;
455 		if (node->rss_types & types) {
456 			buf->entry[buf->entries].priority = 1;
457 			buf->entry[buf->entries].pattern = addr;
458 			buf->entries++;
459 			rte_memcpy(addr, buf->entry[0].pattern,
460 				   user_pattern_size);
461 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
462 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
463 			addr = (void *)(((uintptr_t)addr) +
464 					elt * sizeof(*item));
465 		}
466 	} else if (last_expand_item != NULL) {
467 		node = mlx5_flow_expand_rss_adjust_node(pattern,
468 				last_expand_item_idx, graph, node);
469 	}
470 	memset(flow_items, 0, sizeof(flow_items));
471 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
472 			node->next);
473 	stack[stack_pos] = next_node;
474 	node = next_node ? &graph[*next_node] : NULL;
475 	while (node) {
476 		flow_items[stack_pos].type = node->type;
477 		if (node->rss_types & types) {
478 			size_t n;
479 			/*
480 			 * compute the number of items to copy from the
481 			 * expansion and copy it.
482 			 * When the stack_pos is 0, there are 1 element in it,
483 			 * plus the addition END item.
484 			 */
485 			elt = stack_pos + 2;
486 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
487 			lsize += elt * sizeof(*item) + user_pattern_size;
488 			if (lsize > size)
489 				return -EINVAL;
490 			n = elt * sizeof(*item);
491 			buf->entry[buf->entries].priority =
492 				stack_pos + 1 + missed;
493 			buf->entry[buf->entries].pattern = addr;
494 			buf->entries++;
495 			rte_memcpy(addr, buf->entry[0].pattern,
496 				   user_pattern_size);
497 			addr = (void *)(((uintptr_t)addr) +
498 					user_pattern_size);
499 			rte_memcpy(addr, &missed_item,
500 				   missed * sizeof(*item));
501 			addr = (void *)(((uintptr_t)addr) +
502 				missed * sizeof(*item));
503 			rte_memcpy(addr, flow_items, n);
504 			addr = (void *)(((uintptr_t)addr) + n);
505 		}
506 		/* Go deeper. */
507 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
508 				node->next) {
509 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
510 					node->next);
511 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
512 				rte_errno = E2BIG;
513 				return -rte_errno;
514 			}
515 			stack[stack_pos] = next_node;
516 		} else if (*(next_node + 1)) {
517 			/* Follow up with the next possibility. */
518 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
519 					++next_node);
520 		} else if (!stack_pos) {
521 			/*
522 			 * Completing the traverse over the different paths.
523 			 * The next_node is advanced to the terminator.
524 			 */
525 			++next_node;
526 		} else {
527 			/* Move to the next path. */
528 			while (stack_pos) {
529 				next_node = stack[--stack_pos];
530 				next_node++;
531 				if (*next_node)
532 					break;
533 			}
534 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
535 					next_node);
536 			stack[stack_pos] = next_node;
537 		}
538 		node = next_node && *next_node ? &graph[*next_node] : NULL;
539 	};
540 	return lsize;
541 }
542 
543 enum mlx5_expansion {
544 	MLX5_EXPANSION_ROOT,
545 	MLX5_EXPANSION_ROOT_OUTER,
546 	MLX5_EXPANSION_OUTER_ETH,
547 	MLX5_EXPANSION_OUTER_VLAN,
548 	MLX5_EXPANSION_OUTER_IPV4,
549 	MLX5_EXPANSION_OUTER_IPV4_UDP,
550 	MLX5_EXPANSION_OUTER_IPV4_TCP,
551 	MLX5_EXPANSION_OUTER_IPV6,
552 	MLX5_EXPANSION_OUTER_IPV6_UDP,
553 	MLX5_EXPANSION_OUTER_IPV6_TCP,
554 	MLX5_EXPANSION_VXLAN,
555 	MLX5_EXPANSION_STD_VXLAN,
556 	MLX5_EXPANSION_L3_VXLAN,
557 	MLX5_EXPANSION_VXLAN_GPE,
558 	MLX5_EXPANSION_GRE,
559 	MLX5_EXPANSION_NVGRE,
560 	MLX5_EXPANSION_GRE_KEY,
561 	MLX5_EXPANSION_MPLS,
562 	MLX5_EXPANSION_ETH,
563 	MLX5_EXPANSION_VLAN,
564 	MLX5_EXPANSION_IPV4,
565 	MLX5_EXPANSION_IPV4_UDP,
566 	MLX5_EXPANSION_IPV4_TCP,
567 	MLX5_EXPANSION_IPV6,
568 	MLX5_EXPANSION_IPV6_UDP,
569 	MLX5_EXPANSION_IPV6_TCP,
570 	MLX5_EXPANSION_IPV6_FRAG_EXT,
571 	MLX5_EXPANSION_GTP,
572 	MLX5_EXPANSION_GENEVE,
573 };
574 
575 /** Supported expansion of items. */
576 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
577 	[MLX5_EXPANSION_ROOT] = {
578 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
579 						  MLX5_EXPANSION_IPV4,
580 						  MLX5_EXPANSION_IPV6),
581 		.type = RTE_FLOW_ITEM_TYPE_END,
582 	},
583 	[MLX5_EXPANSION_ROOT_OUTER] = {
584 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
585 						  MLX5_EXPANSION_OUTER_IPV4,
586 						  MLX5_EXPANSION_OUTER_IPV6),
587 		.type = RTE_FLOW_ITEM_TYPE_END,
588 	},
589 	[MLX5_EXPANSION_OUTER_ETH] = {
590 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
591 		.type = RTE_FLOW_ITEM_TYPE_ETH,
592 		.rss_types = 0,
593 	},
594 	[MLX5_EXPANSION_OUTER_VLAN] = {
595 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
596 						  MLX5_EXPANSION_OUTER_IPV6),
597 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
598 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
599 	},
600 	[MLX5_EXPANSION_OUTER_IPV4] = {
601 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
602 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
603 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
604 			 MLX5_EXPANSION_GRE,
605 			 MLX5_EXPANSION_NVGRE,
606 			 MLX5_EXPANSION_IPV4,
607 			 MLX5_EXPANSION_IPV6),
608 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
609 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
610 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
611 	},
612 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
613 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
614 						  MLX5_EXPANSION_VXLAN_GPE,
615 						  MLX5_EXPANSION_MPLS,
616 						  MLX5_EXPANSION_GENEVE,
617 						  MLX5_EXPANSION_GTP),
618 		.type = RTE_FLOW_ITEM_TYPE_UDP,
619 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
620 	},
621 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
622 		.type = RTE_FLOW_ITEM_TYPE_TCP,
623 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
624 	},
625 	[MLX5_EXPANSION_OUTER_IPV6] = {
626 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
627 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
628 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
629 			 MLX5_EXPANSION_IPV4,
630 			 MLX5_EXPANSION_IPV6,
631 			 MLX5_EXPANSION_GRE,
632 			 MLX5_EXPANSION_NVGRE),
633 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
634 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
635 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
636 	},
637 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
638 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
639 						  MLX5_EXPANSION_VXLAN_GPE,
640 						  MLX5_EXPANSION_MPLS,
641 						  MLX5_EXPANSION_GENEVE,
642 						  MLX5_EXPANSION_GTP),
643 		.type = RTE_FLOW_ITEM_TYPE_UDP,
644 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
645 	},
646 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
647 		.type = RTE_FLOW_ITEM_TYPE_TCP,
648 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
649 	},
650 	[MLX5_EXPANSION_VXLAN] = {
651 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
652 						  MLX5_EXPANSION_IPV4,
653 						  MLX5_EXPANSION_IPV6),
654 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
655 	},
656 	[MLX5_EXPANSION_STD_VXLAN] = {
657 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
658 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
659 	},
660 	[MLX5_EXPANSION_L3_VXLAN] = {
661 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
662 					MLX5_EXPANSION_IPV6),
663 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
664 	},
665 	[MLX5_EXPANSION_VXLAN_GPE] = {
666 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
667 						  MLX5_EXPANSION_IPV4,
668 						  MLX5_EXPANSION_IPV6),
669 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
670 	},
671 	[MLX5_EXPANSION_GRE] = {
672 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
673 						  MLX5_EXPANSION_IPV4,
674 						  MLX5_EXPANSION_IPV6,
675 						  MLX5_EXPANSION_GRE_KEY,
676 						  MLX5_EXPANSION_MPLS),
677 		.type = RTE_FLOW_ITEM_TYPE_GRE,
678 	},
679 	[MLX5_EXPANSION_GRE_KEY] = {
680 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
681 						  MLX5_EXPANSION_IPV6,
682 						  MLX5_EXPANSION_MPLS),
683 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
684 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
685 	},
686 	[MLX5_EXPANSION_NVGRE] = {
687 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
688 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
689 	},
690 	[MLX5_EXPANSION_MPLS] = {
691 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
692 						  MLX5_EXPANSION_IPV6,
693 						  MLX5_EXPANSION_ETH),
694 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
695 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
696 	},
697 	[MLX5_EXPANSION_ETH] = {
698 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
699 		.type = RTE_FLOW_ITEM_TYPE_ETH,
700 	},
701 	[MLX5_EXPANSION_VLAN] = {
702 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
703 						  MLX5_EXPANSION_IPV6),
704 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
705 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
706 	},
707 	[MLX5_EXPANSION_IPV4] = {
708 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
709 						  MLX5_EXPANSION_IPV4_TCP),
710 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
711 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
712 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
713 	},
714 	[MLX5_EXPANSION_IPV4_UDP] = {
715 		.type = RTE_FLOW_ITEM_TYPE_UDP,
716 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
717 	},
718 	[MLX5_EXPANSION_IPV4_TCP] = {
719 		.type = RTE_FLOW_ITEM_TYPE_TCP,
720 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
721 	},
722 	[MLX5_EXPANSION_IPV6] = {
723 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
724 						  MLX5_EXPANSION_IPV6_TCP,
725 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
726 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
727 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
728 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
729 	},
730 	[MLX5_EXPANSION_IPV6_UDP] = {
731 		.type = RTE_FLOW_ITEM_TYPE_UDP,
732 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
733 	},
734 	[MLX5_EXPANSION_IPV6_TCP] = {
735 		.type = RTE_FLOW_ITEM_TYPE_TCP,
736 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
737 	},
738 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
739 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
740 	},
741 	[MLX5_EXPANSION_GTP] = {
742 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
743 						  MLX5_EXPANSION_IPV6),
744 		.type = RTE_FLOW_ITEM_TYPE_GTP,
745 	},
746 	[MLX5_EXPANSION_GENEVE] = {
747 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
748 						  MLX5_EXPANSION_IPV4,
749 						  MLX5_EXPANSION_IPV6),
750 		.type = RTE_FLOW_ITEM_TYPE_GENEVE,
751 	},
752 };
753 
754 static struct rte_flow_action_handle *
755 mlx5_action_handle_create(struct rte_eth_dev *dev,
756 			  const struct rte_flow_indir_action_conf *conf,
757 			  const struct rte_flow_action *action,
758 			  struct rte_flow_error *error);
759 static int mlx5_action_handle_destroy
760 				(struct rte_eth_dev *dev,
761 				 struct rte_flow_action_handle *handle,
762 				 struct rte_flow_error *error);
763 static int mlx5_action_handle_update
764 				(struct rte_eth_dev *dev,
765 				 struct rte_flow_action_handle *handle,
766 				 const void *update,
767 				 struct rte_flow_error *error);
768 static int mlx5_action_handle_query
769 				(struct rte_eth_dev *dev,
770 				 const struct rte_flow_action_handle *handle,
771 				 void *data,
772 				 struct rte_flow_error *error);
773 static int
774 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
775 		    struct rte_flow_tunnel *app_tunnel,
776 		    struct rte_flow_action **actions,
777 		    uint32_t *num_of_actions,
778 		    struct rte_flow_error *error);
779 static int
780 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
781 		       struct rte_flow_tunnel *app_tunnel,
782 		       struct rte_flow_item **items,
783 		       uint32_t *num_of_items,
784 		       struct rte_flow_error *error);
785 static int
786 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
787 			      struct rte_flow_item *pmd_items,
788 			      uint32_t num_items, struct rte_flow_error *err);
789 static int
790 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
791 				struct rte_flow_action *pmd_actions,
792 				uint32_t num_actions,
793 				struct rte_flow_error *err);
794 static int
795 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
796 				  struct rte_mbuf *m,
797 				  struct rte_flow_restore_info *info,
798 				  struct rte_flow_error *err);
799 static struct rte_flow_item_flex_handle *
800 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
801 			   const struct rte_flow_item_flex_conf *conf,
802 			   struct rte_flow_error *error);
803 static int
804 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
805 			    const struct rte_flow_item_flex_handle *handle,
806 			    struct rte_flow_error *error);
807 
808 static const struct rte_flow_ops mlx5_flow_ops = {
809 	.validate = mlx5_flow_validate,
810 	.create = mlx5_flow_create,
811 	.destroy = mlx5_flow_destroy,
812 	.flush = mlx5_flow_flush,
813 	.isolate = mlx5_flow_isolate,
814 	.query = mlx5_flow_query,
815 	.dev_dump = mlx5_flow_dev_dump,
816 	.get_aged_flows = mlx5_flow_get_aged_flows,
817 	.action_handle_create = mlx5_action_handle_create,
818 	.action_handle_destroy = mlx5_action_handle_destroy,
819 	.action_handle_update = mlx5_action_handle_update,
820 	.action_handle_query = mlx5_action_handle_query,
821 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
822 	.tunnel_match = mlx5_flow_tunnel_match,
823 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
824 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
825 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
826 	.flex_item_create = mlx5_flow_flex_item_create,
827 	.flex_item_release = mlx5_flow_flex_item_release,
828 };
829 
830 /* Tunnel information. */
831 struct mlx5_flow_tunnel_info {
832 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
833 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
834 };
835 
836 static struct mlx5_flow_tunnel_info tunnels_info[] = {
837 	{
838 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
839 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
840 	},
841 	{
842 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
843 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
844 	},
845 	{
846 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
847 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
848 	},
849 	{
850 		.tunnel = MLX5_FLOW_LAYER_GRE,
851 		.ptype = RTE_PTYPE_TUNNEL_GRE,
852 	},
853 	{
854 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
855 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
856 	},
857 	{
858 		.tunnel = MLX5_FLOW_LAYER_MPLS,
859 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
860 	},
861 	{
862 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
863 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
864 	},
865 	{
866 		.tunnel = MLX5_FLOW_LAYER_IPIP,
867 		.ptype = RTE_PTYPE_TUNNEL_IP,
868 	},
869 	{
870 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
871 		.ptype = RTE_PTYPE_TUNNEL_IP,
872 	},
873 	{
874 		.tunnel = MLX5_FLOW_LAYER_GTP,
875 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
876 	},
877 };
878 
879 
880 
881 /**
882  * Translate tag ID to register.
883  *
884  * @param[in] dev
885  *   Pointer to the Ethernet device structure.
886  * @param[in] feature
887  *   The feature that request the register.
888  * @param[in] id
889  *   The request register ID.
890  * @param[out] error
891  *   Error description in case of any.
892  *
893  * @return
894  *   The request register on success, a negative errno
895  *   value otherwise and rte_errno is set.
896  */
897 int
898 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
899 		     enum mlx5_feature_name feature,
900 		     uint32_t id,
901 		     struct rte_flow_error *error)
902 {
903 	struct mlx5_priv *priv = dev->data->dev_private;
904 	struct mlx5_dev_config *config = &priv->config;
905 	enum modify_reg start_reg;
906 	bool skip_mtr_reg = false;
907 
908 	switch (feature) {
909 	case MLX5_HAIRPIN_RX:
910 		return REG_B;
911 	case MLX5_HAIRPIN_TX:
912 		return REG_A;
913 	case MLX5_METADATA_RX:
914 		switch (config->dv_xmeta_en) {
915 		case MLX5_XMETA_MODE_LEGACY:
916 			return REG_B;
917 		case MLX5_XMETA_MODE_META16:
918 			return REG_C_0;
919 		case MLX5_XMETA_MODE_META32:
920 			return REG_C_1;
921 		}
922 		break;
923 	case MLX5_METADATA_TX:
924 		return REG_A;
925 	case MLX5_METADATA_FDB:
926 		switch (config->dv_xmeta_en) {
927 		case MLX5_XMETA_MODE_LEGACY:
928 			return REG_NON;
929 		case MLX5_XMETA_MODE_META16:
930 			return REG_C_0;
931 		case MLX5_XMETA_MODE_META32:
932 			return REG_C_1;
933 		}
934 		break;
935 	case MLX5_FLOW_MARK:
936 		switch (config->dv_xmeta_en) {
937 		case MLX5_XMETA_MODE_LEGACY:
938 			return REG_NON;
939 		case MLX5_XMETA_MODE_META16:
940 			return REG_C_1;
941 		case MLX5_XMETA_MODE_META32:
942 			return REG_C_0;
943 		}
944 		break;
945 	case MLX5_MTR_ID:
946 		/*
947 		 * If meter color and meter id share one register, flow match
948 		 * should use the meter color register for match.
949 		 */
950 		if (priv->mtr_reg_share)
951 			return priv->mtr_color_reg;
952 		else
953 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
954 			       REG_C_3;
955 	case MLX5_MTR_COLOR:
956 	case MLX5_ASO_FLOW_HIT:
957 	case MLX5_ASO_CONNTRACK:
958 	case MLX5_SAMPLE_ID:
959 		/* All features use the same REG_C. */
960 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
961 		return priv->mtr_color_reg;
962 	case MLX5_COPY_MARK:
963 		/*
964 		 * Metadata COPY_MARK register using is in meter suffix sub
965 		 * flow while with meter. It's safe to share the same register.
966 		 */
967 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
968 	case MLX5_APP_TAG:
969 		/*
970 		 * If meter is enable, it will engage the register for color
971 		 * match and flow match. If meter color match is not using the
972 		 * REG_C_2, need to skip the REG_C_x be used by meter color
973 		 * match.
974 		 * If meter is disable, free to use all available registers.
975 		 */
976 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
977 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
978 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
979 		if (id > (uint32_t)(REG_C_7 - start_reg))
980 			return rte_flow_error_set(error, EINVAL,
981 						  RTE_FLOW_ERROR_TYPE_ITEM,
982 						  NULL, "invalid tag id");
983 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
984 			return rte_flow_error_set(error, ENOTSUP,
985 						  RTE_FLOW_ERROR_TYPE_ITEM,
986 						  NULL, "unsupported tag id");
987 		/*
988 		 * This case means meter is using the REG_C_x great than 2.
989 		 * Take care not to conflict with meter color REG_C_x.
990 		 * If the available index REG_C_y >= REG_C_x, skip the
991 		 * color register.
992 		 */
993 		if (skip_mtr_reg && priv->sh->flow_mreg_c
994 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
995 			if (id >= (uint32_t)(REG_C_7 - start_reg))
996 				return rte_flow_error_set(error, EINVAL,
997 						       RTE_FLOW_ERROR_TYPE_ITEM,
998 							NULL, "invalid tag id");
999 			if (priv->sh->flow_mreg_c
1000 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
1001 				return priv->sh->flow_mreg_c
1002 					       [id + 1 + start_reg - REG_C_0];
1003 			return rte_flow_error_set(error, ENOTSUP,
1004 						  RTE_FLOW_ERROR_TYPE_ITEM,
1005 						  NULL, "unsupported tag id");
1006 		}
1007 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
1008 	}
1009 	MLX5_ASSERT(false);
1010 	return rte_flow_error_set(error, EINVAL,
1011 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1012 				  NULL, "invalid feature name");
1013 }
1014 
1015 /**
1016  * Check extensive flow metadata register support.
1017  *
1018  * @param dev
1019  *   Pointer to rte_eth_dev structure.
1020  *
1021  * @return
1022  *   True if device supports extensive flow metadata register, otherwise false.
1023  */
1024 bool
1025 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1026 {
1027 	struct mlx5_priv *priv = dev->data->dev_private;
1028 
1029 	/*
1030 	 * Having available reg_c can be regarded inclusively as supporting
1031 	 * extensive flow metadata register, which could mean,
1032 	 * - metadata register copy action by modify header.
1033 	 * - 16 modify header actions is supported.
1034 	 * - reg_c's are preserved across different domain (FDB and NIC) on
1035 	 *   packet loopback by flow lookup miss.
1036 	 */
1037 	return priv->sh->flow_mreg_c[2] != REG_NON;
1038 }
1039 
1040 /**
1041  * Get the lowest priority.
1042  *
1043  * @param[in] dev
1044  *   Pointer to the Ethernet device structure.
1045  * @param[in] attributes
1046  *   Pointer to device flow rule attributes.
1047  *
1048  * @return
1049  *   The value of lowest priority of flow.
1050  */
1051 uint32_t
1052 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1053 			  const struct rte_flow_attr *attr)
1054 {
1055 	struct mlx5_priv *priv = dev->data->dev_private;
1056 
1057 	if (!attr->group && !attr->transfer)
1058 		return priv->sh->flow_max_priority - 2;
1059 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1060 }
1061 
1062 /**
1063  * Calculate matcher priority of the flow.
1064  *
1065  * @param[in] dev
1066  *   Pointer to the Ethernet device structure.
1067  * @param[in] attr
1068  *   Pointer to device flow rule attributes.
1069  * @param[in] subpriority
1070  *   The priority based on the items.
1071  * @param[in] external
1072  *   Flow is user flow.
1073  * @return
1074  *   The matcher priority of the flow.
1075  */
1076 uint16_t
1077 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1078 			  const struct rte_flow_attr *attr,
1079 			  uint32_t subpriority, bool external)
1080 {
1081 	uint16_t priority = (uint16_t)attr->priority;
1082 	struct mlx5_priv *priv = dev->data->dev_private;
1083 
1084 	if (!attr->group && !attr->transfer) {
1085 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1086 			priority = priv->sh->flow_max_priority - 1;
1087 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1088 	} else if (!external && attr->transfer && attr->group == 0 &&
1089 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1090 		return (priv->sh->flow_max_priority - 1) * 3;
1091 	}
1092 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1093 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1094 	return priority * 3 + subpriority;
1095 }
1096 
1097 /**
1098  * Verify the @p item specifications (spec, last, mask) are compatible with the
1099  * NIC capabilities.
1100  *
1101  * @param[in] item
1102  *   Item specification.
1103  * @param[in] mask
1104  *   @p item->mask or flow default bit-masks.
1105  * @param[in] nic_mask
1106  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1107  * @param[in] size
1108  *   Bit-masks size in bytes.
1109  * @param[in] range_accepted
1110  *   True if range of values is accepted for specific fields, false otherwise.
1111  * @param[out] error
1112  *   Pointer to error structure.
1113  *
1114  * @return
1115  *   0 on success, a negative errno value otherwise and rte_errno is set.
1116  */
1117 int
1118 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1119 			  const uint8_t *mask,
1120 			  const uint8_t *nic_mask,
1121 			  unsigned int size,
1122 			  bool range_accepted,
1123 			  struct rte_flow_error *error)
1124 {
1125 	unsigned int i;
1126 
1127 	MLX5_ASSERT(nic_mask);
1128 	for (i = 0; i < size; ++i)
1129 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1130 			return rte_flow_error_set(error, ENOTSUP,
1131 						  RTE_FLOW_ERROR_TYPE_ITEM,
1132 						  item,
1133 						  "mask enables non supported"
1134 						  " bits");
1135 	if (!item->spec && (item->mask || item->last))
1136 		return rte_flow_error_set(error, EINVAL,
1137 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1138 					  "mask/last without a spec is not"
1139 					  " supported");
1140 	if (item->spec && item->last && !range_accepted) {
1141 		uint8_t spec[size];
1142 		uint8_t last[size];
1143 		unsigned int i;
1144 		int ret;
1145 
1146 		for (i = 0; i < size; ++i) {
1147 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1148 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1149 		}
1150 		ret = memcmp(spec, last, size);
1151 		if (ret != 0)
1152 			return rte_flow_error_set(error, EINVAL,
1153 						  RTE_FLOW_ERROR_TYPE_ITEM,
1154 						  item,
1155 						  "range is not valid");
1156 	}
1157 	return 0;
1158 }
1159 
1160 /**
1161  * Adjust the hash fields according to the @p flow information.
1162  *
1163  * @param[in] dev_flow.
1164  *   Pointer to the mlx5_flow.
1165  * @param[in] tunnel
1166  *   1 when the hash field is for a tunnel item.
1167  * @param[in] layer_types
1168  *   RTE_ETH_RSS_* types.
1169  * @param[in] hash_fields
1170  *   Item hash fields.
1171  *
1172  * @return
1173  *   The hash fields that should be used.
1174  */
1175 uint64_t
1176 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1177 			    int tunnel __rte_unused, uint64_t layer_types,
1178 			    uint64_t hash_fields)
1179 {
1180 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1181 	int rss_request_inner = rss_desc->level >= 2;
1182 
1183 	/* Check RSS hash level for tunnel. */
1184 	if (tunnel && rss_request_inner)
1185 		hash_fields |= IBV_RX_HASH_INNER;
1186 	else if (tunnel || rss_request_inner)
1187 		return 0;
1188 #endif
1189 	/* Check if requested layer matches RSS hash fields. */
1190 	if (!(rss_desc->types & layer_types))
1191 		return 0;
1192 	return hash_fields;
1193 }
1194 
1195 /**
1196  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1197  * if several tunnel rules are used on this queue, the tunnel ptype will be
1198  * cleared.
1199  *
1200  * @param rxq_ctrl
1201  *   Rx queue to update.
1202  */
1203 static void
1204 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1205 {
1206 	unsigned int i;
1207 	uint32_t tunnel_ptype = 0;
1208 
1209 	/* Look up for the ptype to use. */
1210 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1211 		if (!rxq_ctrl->flow_tunnels_n[i])
1212 			continue;
1213 		if (!tunnel_ptype) {
1214 			tunnel_ptype = tunnels_info[i].ptype;
1215 		} else {
1216 			tunnel_ptype = 0;
1217 			break;
1218 		}
1219 	}
1220 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1221 }
1222 
1223 /**
1224  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
1225  * flow.
1226  *
1227  * @param[in] dev
1228  *   Pointer to the Ethernet device structure.
1229  * @param[in] dev_handle
1230  *   Pointer to device flow handle structure.
1231  */
1232 void
1233 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1234 		       struct mlx5_flow_handle *dev_handle)
1235 {
1236 	struct mlx5_priv *priv = dev->data->dev_private;
1237 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1238 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1239 	unsigned int i;
1240 
1241 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1242 		struct mlx5_hrxq *hrxq;
1243 
1244 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1245 			      dev_handle->rix_hrxq);
1246 		if (hrxq)
1247 			ind_tbl = hrxq->ind_table;
1248 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1249 		struct mlx5_shared_action_rss *shared_rss;
1250 
1251 		shared_rss = mlx5_ipool_get
1252 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1253 			 dev_handle->rix_srss);
1254 		if (shared_rss)
1255 			ind_tbl = shared_rss->ind_tbl;
1256 	}
1257 	if (!ind_tbl)
1258 		return;
1259 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1260 		int idx = ind_tbl->queues[i];
1261 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1262 
1263 		MLX5_ASSERT(rxq_ctrl != NULL);
1264 		if (rxq_ctrl == NULL)
1265 			continue;
1266 		/*
1267 		 * To support metadata register copy on Tx loopback,
1268 		 * this must be always enabled (metadata may arive
1269 		 * from other port - not from local flows only.
1270 		 */
1271 		if (tunnel) {
1272 			unsigned int j;
1273 
1274 			/* Increase the counter matching the flow. */
1275 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1276 				if ((tunnels_info[j].tunnel &
1277 				     dev_handle->layers) ==
1278 				    tunnels_info[j].tunnel) {
1279 					rxq_ctrl->flow_tunnels_n[j]++;
1280 					break;
1281 				}
1282 			}
1283 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1284 		}
1285 	}
1286 }
1287 
1288 static void
1289 flow_rxq_mark_flag_set(struct rte_eth_dev *dev)
1290 {
1291 	struct mlx5_priv *priv = dev->data->dev_private;
1292 	struct mlx5_rxq_ctrl *rxq_ctrl;
1293 
1294 	if (priv->mark_enabled)
1295 		return;
1296 	LIST_FOREACH(rxq_ctrl, &priv->rxqsctrl, next) {
1297 		rxq_ctrl->rxq.mark = 1;
1298 	}
1299 	priv->mark_enabled = 1;
1300 }
1301 
1302 /**
1303  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1304  *
1305  * @param[in] dev
1306  *   Pointer to the Ethernet device structure.
1307  * @param[in] flow
1308  *   Pointer to flow structure.
1309  */
1310 static void
1311 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1312 {
1313 	struct mlx5_priv *priv = dev->data->dev_private;
1314 	uint32_t handle_idx;
1315 	struct mlx5_flow_handle *dev_handle;
1316 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1317 
1318 	MLX5_ASSERT(wks);
1319 	if (wks->mark)
1320 		flow_rxq_mark_flag_set(dev);
1321 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1322 		       handle_idx, dev_handle, next)
1323 		flow_drv_rxq_flags_set(dev, dev_handle);
1324 }
1325 
1326 /**
1327  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1328  * device flow if no other flow uses it with the same kind of request.
1329  *
1330  * @param dev
1331  *   Pointer to Ethernet device.
1332  * @param[in] dev_handle
1333  *   Pointer to the device flow handle structure.
1334  */
1335 static void
1336 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1337 			struct mlx5_flow_handle *dev_handle)
1338 {
1339 	struct mlx5_priv *priv = dev->data->dev_private;
1340 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1341 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1342 	unsigned int i;
1343 
1344 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1345 		struct mlx5_hrxq *hrxq;
1346 
1347 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1348 			      dev_handle->rix_hrxq);
1349 		if (hrxq)
1350 			ind_tbl = hrxq->ind_table;
1351 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1352 		struct mlx5_shared_action_rss *shared_rss;
1353 
1354 		shared_rss = mlx5_ipool_get
1355 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1356 			 dev_handle->rix_srss);
1357 		if (shared_rss)
1358 			ind_tbl = shared_rss->ind_tbl;
1359 	}
1360 	if (!ind_tbl)
1361 		return;
1362 	MLX5_ASSERT(dev->data->dev_started);
1363 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1364 		int idx = ind_tbl->queues[i];
1365 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1366 
1367 		MLX5_ASSERT(rxq_ctrl != NULL);
1368 		if (rxq_ctrl == NULL)
1369 			continue;
1370 		if (tunnel) {
1371 			unsigned int j;
1372 
1373 			/* Decrease the counter matching the flow. */
1374 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1375 				if ((tunnels_info[j].tunnel &
1376 				     dev_handle->layers) ==
1377 				    tunnels_info[j].tunnel) {
1378 					rxq_ctrl->flow_tunnels_n[j]--;
1379 					break;
1380 				}
1381 			}
1382 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1383 		}
1384 	}
1385 }
1386 
1387 /**
1388  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1389  * @p flow if no other flow uses it with the same kind of request.
1390  *
1391  * @param dev
1392  *   Pointer to Ethernet device.
1393  * @param[in] flow
1394  *   Pointer to the flow.
1395  */
1396 static void
1397 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1398 {
1399 	struct mlx5_priv *priv = dev->data->dev_private;
1400 	uint32_t handle_idx;
1401 	struct mlx5_flow_handle *dev_handle;
1402 
1403 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1404 		       handle_idx, dev_handle, next)
1405 		flow_drv_rxq_flags_trim(dev, dev_handle);
1406 }
1407 
1408 /**
1409  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1410  *
1411  * @param dev
1412  *   Pointer to Ethernet device.
1413  */
1414 static void
1415 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1416 {
1417 	struct mlx5_priv *priv = dev->data->dev_private;
1418 	unsigned int i;
1419 
1420 	for (i = 0; i != priv->rxqs_n; ++i) {
1421 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1422 		unsigned int j;
1423 
1424 		if (rxq == NULL || rxq->ctrl == NULL)
1425 			continue;
1426 		rxq->ctrl->rxq.mark = 0;
1427 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1428 			rxq->ctrl->flow_tunnels_n[j] = 0;
1429 		rxq->ctrl->rxq.tunnel = 0;
1430 	}
1431 	priv->mark_enabled = 0;
1432 }
1433 
1434 /**
1435  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1436  *
1437  * @param[in] dev
1438  *   Pointer to the Ethernet device structure.
1439  */
1440 void
1441 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1442 {
1443 	struct mlx5_priv *priv = dev->data->dev_private;
1444 	unsigned int i;
1445 
1446 	for (i = 0; i != priv->rxqs_n; ++i) {
1447 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1448 		struct mlx5_rxq_data *data;
1449 
1450 		if (rxq == NULL || rxq->ctrl == NULL)
1451 			continue;
1452 		data = &rxq->ctrl->rxq;
1453 		if (!rte_flow_dynf_metadata_avail()) {
1454 			data->dynf_meta = 0;
1455 			data->flow_meta_mask = 0;
1456 			data->flow_meta_offset = -1;
1457 			data->flow_meta_port_mask = 0;
1458 		} else {
1459 			data->dynf_meta = 1;
1460 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1461 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1462 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1463 		}
1464 	}
1465 }
1466 
1467 /*
1468  * return a pointer to the desired action in the list of actions.
1469  *
1470  * @param[in] actions
1471  *   The list of actions to search the action in.
1472  * @param[in] action
1473  *   The action to find.
1474  *
1475  * @return
1476  *   Pointer to the action in the list, if found. NULL otherwise.
1477  */
1478 const struct rte_flow_action *
1479 mlx5_flow_find_action(const struct rte_flow_action *actions,
1480 		      enum rte_flow_action_type action)
1481 {
1482 	if (actions == NULL)
1483 		return NULL;
1484 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1485 		if (actions->type == action)
1486 			return actions;
1487 	return NULL;
1488 }
1489 
1490 /*
1491  * Validate the flag action.
1492  *
1493  * @param[in] action_flags
1494  *   Bit-fields that holds the actions detected until now.
1495  * @param[in] attr
1496  *   Attributes of flow that includes this action.
1497  * @param[out] error
1498  *   Pointer to error structure.
1499  *
1500  * @return
1501  *   0 on success, a negative errno value otherwise and rte_errno is set.
1502  */
1503 int
1504 mlx5_flow_validate_action_flag(uint64_t action_flags,
1505 			       const struct rte_flow_attr *attr,
1506 			       struct rte_flow_error *error)
1507 {
1508 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1509 		return rte_flow_error_set(error, EINVAL,
1510 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1511 					  "can't mark and flag in same flow");
1512 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1513 		return rte_flow_error_set(error, EINVAL,
1514 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1515 					  "can't have 2 flag"
1516 					  " actions in same flow");
1517 	if (attr->egress)
1518 		return rte_flow_error_set(error, ENOTSUP,
1519 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1520 					  "flag action not supported for "
1521 					  "egress");
1522 	return 0;
1523 }
1524 
1525 /*
1526  * Validate the mark action.
1527  *
1528  * @param[in] action
1529  *   Pointer to the queue action.
1530  * @param[in] action_flags
1531  *   Bit-fields that holds the actions detected until now.
1532  * @param[in] attr
1533  *   Attributes of flow that includes this action.
1534  * @param[out] error
1535  *   Pointer to error structure.
1536  *
1537  * @return
1538  *   0 on success, a negative errno value otherwise and rte_errno is set.
1539  */
1540 int
1541 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1542 			       uint64_t action_flags,
1543 			       const struct rte_flow_attr *attr,
1544 			       struct rte_flow_error *error)
1545 {
1546 	const struct rte_flow_action_mark *mark = action->conf;
1547 
1548 	if (!mark)
1549 		return rte_flow_error_set(error, EINVAL,
1550 					  RTE_FLOW_ERROR_TYPE_ACTION,
1551 					  action,
1552 					  "configuration cannot be null");
1553 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1554 		return rte_flow_error_set(error, EINVAL,
1555 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1556 					  &mark->id,
1557 					  "mark id must in 0 <= id < "
1558 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1559 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1560 		return rte_flow_error_set(error, EINVAL,
1561 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1562 					  "can't flag and mark in same flow");
1563 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1564 		return rte_flow_error_set(error, EINVAL,
1565 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1566 					  "can't have 2 mark actions in same"
1567 					  " flow");
1568 	if (attr->egress)
1569 		return rte_flow_error_set(error, ENOTSUP,
1570 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1571 					  "mark action not supported for "
1572 					  "egress");
1573 	return 0;
1574 }
1575 
1576 /*
1577  * Validate the drop action.
1578  *
1579  * @param[in] action_flags
1580  *   Bit-fields that holds the actions detected until now.
1581  * @param[in] attr
1582  *   Attributes of flow that includes this action.
1583  * @param[out] error
1584  *   Pointer to error structure.
1585  *
1586  * @return
1587  *   0 on success, a negative errno value otherwise and rte_errno is set.
1588  */
1589 int
1590 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1591 			       const struct rte_flow_attr *attr,
1592 			       struct rte_flow_error *error)
1593 {
1594 	if (attr->egress)
1595 		return rte_flow_error_set(error, ENOTSUP,
1596 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1597 					  "drop action not supported for "
1598 					  "egress");
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Validate the queue action.
1604  *
1605  * @param[in] action
1606  *   Pointer to the queue action.
1607  * @param[in] action_flags
1608  *   Bit-fields that holds the actions detected until now.
1609  * @param[in] dev
1610  *   Pointer to the Ethernet device structure.
1611  * @param[in] attr
1612  *   Attributes of flow that includes this action.
1613  * @param[out] error
1614  *   Pointer to error structure.
1615  *
1616  * @return
1617  *   0 on success, a negative errno value otherwise and rte_errno is set.
1618  */
1619 int
1620 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1621 				uint64_t action_flags,
1622 				struct rte_eth_dev *dev,
1623 				const struct rte_flow_attr *attr,
1624 				struct rte_flow_error *error)
1625 {
1626 	struct mlx5_priv *priv = dev->data->dev_private;
1627 	const struct rte_flow_action_queue *queue = action->conf;
1628 
1629 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1630 		return rte_flow_error_set(error, EINVAL,
1631 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1632 					  "can't have 2 fate actions in"
1633 					  " same flow");
1634 	if (!priv->rxqs_n)
1635 		return rte_flow_error_set(error, EINVAL,
1636 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1637 					  NULL, "No Rx queues configured");
1638 	if (queue->index >= priv->rxqs_n)
1639 		return rte_flow_error_set(error, EINVAL,
1640 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1641 					  &queue->index,
1642 					  "queue index out of range");
1643 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1644 		return rte_flow_error_set(error, EINVAL,
1645 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1646 					  &queue->index,
1647 					  "queue is not configured");
1648 	if (attr->egress)
1649 		return rte_flow_error_set(error, ENOTSUP,
1650 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1651 					  "queue action not supported for "
1652 					  "egress");
1653 	return 0;
1654 }
1655 
1656 /**
1657  * Validate queue numbers for device RSS.
1658  *
1659  * @param[in] dev
1660  *   Configured device.
1661  * @param[in] queues
1662  *   Array of queue numbers.
1663  * @param[in] queues_n
1664  *   Size of the @p queues array.
1665  * @param[out] error
1666  *   On error, filled with a textual error description.
1667  * @param[out] queue
1668  *   On error, filled with an offending queue index in @p queues array.
1669  *
1670  * @return
1671  *   0 on success, a negative errno code on error.
1672  */
1673 static int
1674 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1675 			 const uint16_t *queues, uint32_t queues_n,
1676 			 const char **error, uint32_t *queue_idx)
1677 {
1678 	const struct mlx5_priv *priv = dev->data->dev_private;
1679 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1680 	uint32_t i;
1681 
1682 	for (i = 0; i != queues_n; ++i) {
1683 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1684 								   queues[i]);
1685 
1686 		if (queues[i] >= priv->rxqs_n) {
1687 			*error = "queue index out of range";
1688 			*queue_idx = i;
1689 			return -EINVAL;
1690 		}
1691 		if (rxq_ctrl == NULL) {
1692 			*error =  "queue is not configured";
1693 			*queue_idx = i;
1694 			return -EINVAL;
1695 		}
1696 		if (i == 0)
1697 			rxq_type = rxq_ctrl->type;
1698 		if (rxq_type != rxq_ctrl->type) {
1699 			*error = "combining hairpin and regular RSS queues is not supported";
1700 			*queue_idx = i;
1701 			return -ENOTSUP;
1702 		}
1703 	}
1704 	return 0;
1705 }
1706 
1707 /*
1708  * Validate the rss action.
1709  *
1710  * @param[in] dev
1711  *   Pointer to the Ethernet device structure.
1712  * @param[in] action
1713  *   Pointer to the queue action.
1714  * @param[out] error
1715  *   Pointer to error structure.
1716  *
1717  * @return
1718  *   0 on success, a negative errno value otherwise and rte_errno is set.
1719  */
1720 int
1721 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1722 			 const struct rte_flow_action *action,
1723 			 struct rte_flow_error *error)
1724 {
1725 	struct mlx5_priv *priv = dev->data->dev_private;
1726 	const struct rte_flow_action_rss *rss = action->conf;
1727 	int ret;
1728 	const char *message;
1729 	uint32_t queue_idx;
1730 
1731 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1732 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1733 		return rte_flow_error_set(error, ENOTSUP,
1734 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1735 					  &rss->func,
1736 					  "RSS hash function not supported");
1737 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1738 	if (rss->level > 2)
1739 #else
1740 	if (rss->level > 1)
1741 #endif
1742 		return rte_flow_error_set(error, ENOTSUP,
1743 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1744 					  &rss->level,
1745 					  "tunnel RSS is not supported");
1746 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1747 	if (rss->key_len == 0 && rss->key != NULL)
1748 		return rte_flow_error_set(error, ENOTSUP,
1749 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1750 					  &rss->key_len,
1751 					  "RSS hash key length 0");
1752 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1753 		return rte_flow_error_set(error, ENOTSUP,
1754 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1755 					  &rss->key_len,
1756 					  "RSS hash key too small");
1757 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1758 		return rte_flow_error_set(error, ENOTSUP,
1759 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1760 					  &rss->key_len,
1761 					  "RSS hash key too large");
1762 	if (rss->queue_num > priv->config.ind_table_max_size)
1763 		return rte_flow_error_set(error, ENOTSUP,
1764 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1765 					  &rss->queue_num,
1766 					  "number of queues too large");
1767 	if (rss->types & MLX5_RSS_HF_MASK)
1768 		return rte_flow_error_set(error, ENOTSUP,
1769 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1770 					  &rss->types,
1771 					  "some RSS protocols are not"
1772 					  " supported");
1773 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1774 	    !(rss->types & RTE_ETH_RSS_IP))
1775 		return rte_flow_error_set(error, EINVAL,
1776 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1777 					  "L3 partial RSS requested but L3 RSS"
1778 					  " type not specified");
1779 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1780 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1781 		return rte_flow_error_set(error, EINVAL,
1782 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1783 					  "L4 partial RSS requested but L4 RSS"
1784 					  " type not specified");
1785 	if (!priv->rxqs_n)
1786 		return rte_flow_error_set(error, EINVAL,
1787 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1788 					  NULL, "No Rx queues configured");
1789 	if (!rss->queue_num)
1790 		return rte_flow_error_set(error, EINVAL,
1791 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1792 					  NULL, "No queues configured");
1793 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1794 				       &message, &queue_idx);
1795 	if (ret != 0) {
1796 		return rte_flow_error_set(error, -ret,
1797 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1798 					  &rss->queue[queue_idx], message);
1799 	}
1800 	return 0;
1801 }
1802 
1803 /*
1804  * Validate the rss action.
1805  *
1806  * @param[in] action
1807  *   Pointer to the queue action.
1808  * @param[in] action_flags
1809  *   Bit-fields that holds the actions detected until now.
1810  * @param[in] dev
1811  *   Pointer to the Ethernet device structure.
1812  * @param[in] attr
1813  *   Attributes of flow that includes this action.
1814  * @param[in] item_flags
1815  *   Items that were detected.
1816  * @param[out] error
1817  *   Pointer to error structure.
1818  *
1819  * @return
1820  *   0 on success, a negative errno value otherwise and rte_errno is set.
1821  */
1822 int
1823 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1824 			      uint64_t action_flags,
1825 			      struct rte_eth_dev *dev,
1826 			      const struct rte_flow_attr *attr,
1827 			      uint64_t item_flags,
1828 			      struct rte_flow_error *error)
1829 {
1830 	const struct rte_flow_action_rss *rss = action->conf;
1831 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1832 	int ret;
1833 
1834 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1835 		return rte_flow_error_set(error, EINVAL,
1836 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1837 					  "can't have 2 fate actions"
1838 					  " in same flow");
1839 	ret = mlx5_validate_action_rss(dev, action, error);
1840 	if (ret)
1841 		return ret;
1842 	if (attr->egress)
1843 		return rte_flow_error_set(error, ENOTSUP,
1844 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1845 					  "rss action not supported for "
1846 					  "egress");
1847 	if (rss->level > 1 && !tunnel)
1848 		return rte_flow_error_set(error, EINVAL,
1849 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1850 					  "inner RSS is not supported for "
1851 					  "non-tunnel flows");
1852 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1853 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1854 		return rte_flow_error_set(error, EINVAL,
1855 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1856 					  "RSS on eCPRI is not supported now");
1857 	}
1858 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1859 	    !(item_flags &
1860 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1861 	    rss->level > 1)
1862 		return rte_flow_error_set(error, EINVAL,
1863 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1864 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1865 	return 0;
1866 }
1867 
1868 /*
1869  * Validate the default miss action.
1870  *
1871  * @param[in] action_flags
1872  *   Bit-fields that holds the actions detected until now.
1873  * @param[out] error
1874  *   Pointer to error structure.
1875  *
1876  * @return
1877  *   0 on success, a negative errno value otherwise and rte_errno is set.
1878  */
1879 int
1880 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1881 				const struct rte_flow_attr *attr,
1882 				struct rte_flow_error *error)
1883 {
1884 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1885 		return rte_flow_error_set(error, EINVAL,
1886 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1887 					  "can't have 2 fate actions in"
1888 					  " same flow");
1889 	if (attr->egress)
1890 		return rte_flow_error_set(error, ENOTSUP,
1891 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1892 					  "default miss action not supported "
1893 					  "for egress");
1894 	if (attr->group)
1895 		return rte_flow_error_set(error, ENOTSUP,
1896 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1897 					  "only group 0 is supported");
1898 	if (attr->transfer)
1899 		return rte_flow_error_set(error, ENOTSUP,
1900 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1901 					  NULL, "transfer is not supported");
1902 	return 0;
1903 }
1904 
1905 /*
1906  * Validate the count action.
1907  *
1908  * @param[in] dev
1909  *   Pointer to the Ethernet device structure.
1910  * @param[in] attr
1911  *   Attributes of flow that includes this action.
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_action_count(struct rte_eth_dev *dev __rte_unused,
1920 				const struct rte_flow_attr *attr,
1921 				struct rte_flow_error *error)
1922 {
1923 	if (attr->egress)
1924 		return rte_flow_error_set(error, ENOTSUP,
1925 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1926 					  "count action not supported for "
1927 					  "egress");
1928 	return 0;
1929 }
1930 
1931 /*
1932  * Validate the ASO CT action.
1933  *
1934  * @param[in] dev
1935  *   Pointer to the Ethernet device structure.
1936  * @param[in] conntrack
1937  *   Pointer to the CT action profile.
1938  * @param[out] error
1939  *   Pointer to error structure.
1940  *
1941  * @return
1942  *   0 on success, a negative errno value otherwise and rte_errno is set.
1943  */
1944 int
1945 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1946 			const struct rte_flow_action_conntrack *conntrack,
1947 			struct rte_flow_error *error)
1948 {
1949 	RTE_SET_USED(dev);
1950 
1951 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1952 		return rte_flow_error_set(error, EINVAL,
1953 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1954 					  "Invalid CT state");
1955 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1956 		return rte_flow_error_set(error, EINVAL,
1957 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1958 					  "Invalid last TCP packet flag");
1959 	return 0;
1960 }
1961 
1962 /**
1963  * Verify the @p attributes will be correctly understood by the NIC and store
1964  * them in the @p flow if everything is correct.
1965  *
1966  * @param[in] dev
1967  *   Pointer to the Ethernet device structure.
1968  * @param[in] attributes
1969  *   Pointer to flow attributes
1970  * @param[out] error
1971  *   Pointer to error structure.
1972  *
1973  * @return
1974  *   0 on success, a negative errno value otherwise and rte_errno is set.
1975  */
1976 int
1977 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1978 			      const struct rte_flow_attr *attributes,
1979 			      struct rte_flow_error *error)
1980 {
1981 	struct mlx5_priv *priv = dev->data->dev_private;
1982 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1983 
1984 	if (attributes->group)
1985 		return rte_flow_error_set(error, ENOTSUP,
1986 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1987 					  NULL, "groups is not supported");
1988 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1989 	    attributes->priority >= priority_max)
1990 		return rte_flow_error_set(error, ENOTSUP,
1991 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1992 					  NULL, "priority out of range");
1993 	if (attributes->egress)
1994 		return rte_flow_error_set(error, ENOTSUP,
1995 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1996 					  "egress is not supported");
1997 	if (attributes->transfer && !priv->config.dv_esw_en)
1998 		return rte_flow_error_set(error, ENOTSUP,
1999 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2000 					  NULL, "transfer is not supported");
2001 	if (!attributes->ingress)
2002 		return rte_flow_error_set(error, EINVAL,
2003 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2004 					  NULL,
2005 					  "ingress attribute is mandatory");
2006 	return 0;
2007 }
2008 
2009 /**
2010  * Validate ICMP6 item.
2011  *
2012  * @param[in] item
2013  *   Item specification.
2014  * @param[in] item_flags
2015  *   Bit-fields that holds the items detected until now.
2016  * @param[in] ext_vlan_sup
2017  *   Whether extended VLAN features are supported or not.
2018  * @param[out] error
2019  *   Pointer to error structure.
2020  *
2021  * @return
2022  *   0 on success, a negative errno value otherwise and rte_errno is set.
2023  */
2024 int
2025 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2026 			       uint64_t item_flags,
2027 			       uint8_t target_protocol,
2028 			       struct rte_flow_error *error)
2029 {
2030 	const struct rte_flow_item_icmp6 *mask = item->mask;
2031 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2032 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2033 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2034 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2035 				      MLX5_FLOW_LAYER_OUTER_L4;
2036 	int ret;
2037 
2038 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2039 		return rte_flow_error_set(error, EINVAL,
2040 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2041 					  "protocol filtering not compatible"
2042 					  " with ICMP6 layer");
2043 	if (!(item_flags & l3m))
2044 		return rte_flow_error_set(error, EINVAL,
2045 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2046 					  "IPv6 is mandatory to filter on"
2047 					  " ICMP6");
2048 	if (item_flags & l4m)
2049 		return rte_flow_error_set(error, EINVAL,
2050 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2051 					  "multiple L4 layers not supported");
2052 	if (!mask)
2053 		mask = &rte_flow_item_icmp6_mask;
2054 	ret = mlx5_flow_item_acceptable
2055 		(item, (const uint8_t *)mask,
2056 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2057 		 sizeof(struct rte_flow_item_icmp6),
2058 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2059 	if (ret < 0)
2060 		return ret;
2061 	return 0;
2062 }
2063 
2064 /**
2065  * Validate ICMP item.
2066  *
2067  * @param[in] item
2068  *   Item specification.
2069  * @param[in] item_flags
2070  *   Bit-fields that holds the items detected until now.
2071  * @param[out] error
2072  *   Pointer to error structure.
2073  *
2074  * @return
2075  *   0 on success, a negative errno value otherwise and rte_errno is set.
2076  */
2077 int
2078 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2079 			     uint64_t item_flags,
2080 			     uint8_t target_protocol,
2081 			     struct rte_flow_error *error)
2082 {
2083 	const struct rte_flow_item_icmp *mask = item->mask;
2084 	const struct rte_flow_item_icmp nic_mask = {
2085 		.hdr.icmp_type = 0xff,
2086 		.hdr.icmp_code = 0xff,
2087 		.hdr.icmp_ident = RTE_BE16(0xffff),
2088 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2089 	};
2090 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2091 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2092 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2093 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2094 				      MLX5_FLOW_LAYER_OUTER_L4;
2095 	int ret;
2096 
2097 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2098 		return rte_flow_error_set(error, EINVAL,
2099 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2100 					  "protocol filtering not compatible"
2101 					  " with ICMP layer");
2102 	if (!(item_flags & l3m))
2103 		return rte_flow_error_set(error, EINVAL,
2104 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2105 					  "IPv4 is mandatory to filter"
2106 					  " on ICMP");
2107 	if (item_flags & l4m)
2108 		return rte_flow_error_set(error, EINVAL,
2109 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2110 					  "multiple L4 layers not supported");
2111 	if (!mask)
2112 		mask = &nic_mask;
2113 	ret = mlx5_flow_item_acceptable
2114 		(item, (const uint8_t *)mask,
2115 		 (const uint8_t *)&nic_mask,
2116 		 sizeof(struct rte_flow_item_icmp),
2117 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2118 	if (ret < 0)
2119 		return ret;
2120 	return 0;
2121 }
2122 
2123 /**
2124  * Validate Ethernet 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[out] error
2131  *   Pointer to error structure.
2132  *
2133  * @return
2134  *   0 on success, a negative errno value otherwise and rte_errno is set.
2135  */
2136 int
2137 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2138 			    uint64_t item_flags, bool ext_vlan_sup,
2139 			    struct rte_flow_error *error)
2140 {
2141 	const struct rte_flow_item_eth *mask = item->mask;
2142 	const struct rte_flow_item_eth nic_mask = {
2143 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2144 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2145 		.type = RTE_BE16(0xffff),
2146 		.has_vlan = ext_vlan_sup ? 1 : 0,
2147 	};
2148 	int ret;
2149 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2150 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2151 				       MLX5_FLOW_LAYER_OUTER_L2;
2152 
2153 	if (item_flags & ethm)
2154 		return rte_flow_error_set(error, ENOTSUP,
2155 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2156 					  "multiple L2 layers not supported");
2157 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2158 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2159 		return rte_flow_error_set(error, EINVAL,
2160 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2161 					  "L2 layer should not follow "
2162 					  "L3 layers");
2163 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2164 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2165 		return rte_flow_error_set(error, EINVAL,
2166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 					  "L2 layer should not follow VLAN");
2168 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2169 		return rte_flow_error_set(error, EINVAL,
2170 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2171 					  "L2 layer should not follow GTP");
2172 	if (!mask)
2173 		mask = &rte_flow_item_eth_mask;
2174 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2175 					(const uint8_t *)&nic_mask,
2176 					sizeof(struct rte_flow_item_eth),
2177 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2178 	return ret;
2179 }
2180 
2181 /**
2182  * Validate VLAN item.
2183  *
2184  * @param[in] item
2185  *   Item specification.
2186  * @param[in] item_flags
2187  *   Bit-fields that holds the items detected until now.
2188  * @param[in] dev
2189  *   Ethernet device flow is being created on.
2190  * @param[out] error
2191  *   Pointer to error structure.
2192  *
2193  * @return
2194  *   0 on success, a negative errno value otherwise and rte_errno is set.
2195  */
2196 int
2197 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2198 			     uint64_t item_flags,
2199 			     struct rte_eth_dev *dev,
2200 			     struct rte_flow_error *error)
2201 {
2202 	const struct rte_flow_item_vlan *spec = item->spec;
2203 	const struct rte_flow_item_vlan *mask = item->mask;
2204 	const struct rte_flow_item_vlan nic_mask = {
2205 		.tci = RTE_BE16(UINT16_MAX),
2206 		.inner_type = RTE_BE16(UINT16_MAX),
2207 	};
2208 	uint16_t vlan_tag = 0;
2209 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2210 	int ret;
2211 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2212 					MLX5_FLOW_LAYER_INNER_L4) :
2213 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2214 					MLX5_FLOW_LAYER_OUTER_L4);
2215 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2216 					MLX5_FLOW_LAYER_OUTER_VLAN;
2217 
2218 	if (item_flags & vlanm)
2219 		return rte_flow_error_set(error, EINVAL,
2220 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2221 					  "multiple VLAN layers not supported");
2222 	else if ((item_flags & l34m) != 0)
2223 		return rte_flow_error_set(error, EINVAL,
2224 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2225 					  "VLAN cannot follow L3/L4 layer");
2226 	if (!mask)
2227 		mask = &rte_flow_item_vlan_mask;
2228 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2229 					(const uint8_t *)&nic_mask,
2230 					sizeof(struct rte_flow_item_vlan),
2231 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2232 	if (ret)
2233 		return ret;
2234 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2235 		struct mlx5_priv *priv = dev->data->dev_private;
2236 
2237 		if (priv->vmwa_context) {
2238 			/*
2239 			 * Non-NULL context means we have a virtual machine
2240 			 * and SR-IOV enabled, we have to create VLAN interface
2241 			 * to make hypervisor to setup E-Switch vport
2242 			 * context correctly. We avoid creating the multiple
2243 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2244 			 */
2245 			return rte_flow_error_set(error, EINVAL,
2246 						  RTE_FLOW_ERROR_TYPE_ITEM,
2247 						  item,
2248 						  "VLAN tag mask is not"
2249 						  " supported in virtual"
2250 						  " environment");
2251 		}
2252 	}
2253 	if (spec) {
2254 		vlan_tag = spec->tci;
2255 		vlan_tag &= mask->tci;
2256 	}
2257 	/*
2258 	 * From verbs perspective an empty VLAN is equivalent
2259 	 * to a packet without VLAN layer.
2260 	 */
2261 	if (!vlan_tag)
2262 		return rte_flow_error_set(error, EINVAL,
2263 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2264 					  item->spec,
2265 					  "VLAN cannot be empty");
2266 	return 0;
2267 }
2268 
2269 /**
2270  * Validate IPV4 item.
2271  *
2272  * @param[in] item
2273  *   Item specification.
2274  * @param[in] item_flags
2275  *   Bit-fields that holds the items detected until now.
2276  * @param[in] last_item
2277  *   Previous validated item in the pattern items.
2278  * @param[in] ether_type
2279  *   Type in the ethernet layer header (including dot1q).
2280  * @param[in] acc_mask
2281  *   Acceptable mask, if NULL default internal default mask
2282  *   will be used to check whether item fields are supported.
2283  * @param[in] range_accepted
2284  *   True if range of values is accepted for specific fields, false otherwise.
2285  * @param[out] error
2286  *   Pointer to error structure.
2287  *
2288  * @return
2289  *   0 on success, a negative errno value otherwise and rte_errno is set.
2290  */
2291 int
2292 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2293 			     uint64_t item_flags,
2294 			     uint64_t last_item,
2295 			     uint16_t ether_type,
2296 			     const struct rte_flow_item_ipv4 *acc_mask,
2297 			     bool range_accepted,
2298 			     struct rte_flow_error *error)
2299 {
2300 	const struct rte_flow_item_ipv4 *mask = item->mask;
2301 	const struct rte_flow_item_ipv4 *spec = item->spec;
2302 	const struct rte_flow_item_ipv4 nic_mask = {
2303 		.hdr = {
2304 			.src_addr = RTE_BE32(0xffffffff),
2305 			.dst_addr = RTE_BE32(0xffffffff),
2306 			.type_of_service = 0xff,
2307 			.next_proto_id = 0xff,
2308 		},
2309 	};
2310 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2311 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2312 				      MLX5_FLOW_LAYER_OUTER_L3;
2313 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2314 				      MLX5_FLOW_LAYER_OUTER_L4;
2315 	int ret;
2316 	uint8_t next_proto = 0xFF;
2317 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2318 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2319 				  MLX5_FLOW_LAYER_INNER_VLAN);
2320 
2321 	if ((last_item & l2_vlan) && ether_type &&
2322 	    ether_type != RTE_ETHER_TYPE_IPV4)
2323 		return rte_flow_error_set(error, EINVAL,
2324 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2325 					  "IPv4 cannot follow L2/VLAN layer "
2326 					  "which ether type is not IPv4");
2327 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2328 		if (mask && spec)
2329 			next_proto = mask->hdr.next_proto_id &
2330 				     spec->hdr.next_proto_id;
2331 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2332 			return rte_flow_error_set(error, EINVAL,
2333 						  RTE_FLOW_ERROR_TYPE_ITEM,
2334 						  item,
2335 						  "multiple tunnel "
2336 						  "not supported");
2337 	}
2338 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2339 		return rte_flow_error_set(error, EINVAL,
2340 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2341 					  "wrong tunnel type - IPv6 specified "
2342 					  "but IPv4 item provided");
2343 	if (item_flags & l3m)
2344 		return rte_flow_error_set(error, ENOTSUP,
2345 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2346 					  "multiple L3 layers not supported");
2347 	else if (item_flags & l4m)
2348 		return rte_flow_error_set(error, EINVAL,
2349 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2350 					  "L3 cannot follow an L4 layer.");
2351 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2352 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2353 		return rte_flow_error_set(error, EINVAL,
2354 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 					  "L3 cannot follow an NVGRE layer.");
2356 	if (!mask)
2357 		mask = &rte_flow_item_ipv4_mask;
2358 	else if (mask->hdr.next_proto_id != 0 &&
2359 		 mask->hdr.next_proto_id != 0xff)
2360 		return rte_flow_error_set(error, EINVAL,
2361 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2362 					  "partial mask is not supported"
2363 					  " for protocol");
2364 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2365 					acc_mask ? (const uint8_t *)acc_mask
2366 						 : (const uint8_t *)&nic_mask,
2367 					sizeof(struct rte_flow_item_ipv4),
2368 					range_accepted, error);
2369 	if (ret < 0)
2370 		return ret;
2371 	return 0;
2372 }
2373 
2374 /**
2375  * Validate IPV6 item.
2376  *
2377  * @param[in] item
2378  *   Item specification.
2379  * @param[in] item_flags
2380  *   Bit-fields that holds the items detected until now.
2381  * @param[in] last_item
2382  *   Previous validated item in the pattern items.
2383  * @param[in] ether_type
2384  *   Type in the ethernet layer header (including dot1q).
2385  * @param[in] acc_mask
2386  *   Acceptable mask, if NULL default internal default mask
2387  *   will be used to check whether item fields are supported.
2388  * @param[out] error
2389  *   Pointer to error structure.
2390  *
2391  * @return
2392  *   0 on success, a negative errno value otherwise and rte_errno is set.
2393  */
2394 int
2395 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2396 			     uint64_t item_flags,
2397 			     uint64_t last_item,
2398 			     uint16_t ether_type,
2399 			     const struct rte_flow_item_ipv6 *acc_mask,
2400 			     struct rte_flow_error *error)
2401 {
2402 	const struct rte_flow_item_ipv6 *mask = item->mask;
2403 	const struct rte_flow_item_ipv6 *spec = item->spec;
2404 	const struct rte_flow_item_ipv6 nic_mask = {
2405 		.hdr = {
2406 			.src_addr =
2407 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2408 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2409 			.dst_addr =
2410 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2411 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2412 			.vtc_flow = RTE_BE32(0xffffffff),
2413 			.proto = 0xff,
2414 		},
2415 	};
2416 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2417 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2418 				      MLX5_FLOW_LAYER_OUTER_L3;
2419 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2420 				      MLX5_FLOW_LAYER_OUTER_L4;
2421 	int ret;
2422 	uint8_t next_proto = 0xFF;
2423 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2424 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2425 				  MLX5_FLOW_LAYER_INNER_VLAN);
2426 
2427 	if ((last_item & l2_vlan) && ether_type &&
2428 	    ether_type != RTE_ETHER_TYPE_IPV6)
2429 		return rte_flow_error_set(error, EINVAL,
2430 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2431 					  "IPv6 cannot follow L2/VLAN layer "
2432 					  "which ether type is not IPv6");
2433 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2434 		next_proto = spec->hdr.proto;
2435 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2436 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2437 			return rte_flow_error_set(error, EINVAL,
2438 						  RTE_FLOW_ERROR_TYPE_ITEM,
2439 						  item,
2440 						  "multiple tunnel "
2441 						  "not supported");
2442 	}
2443 	if (next_proto == IPPROTO_HOPOPTS  ||
2444 	    next_proto == IPPROTO_ROUTING  ||
2445 	    next_proto == IPPROTO_FRAGMENT ||
2446 	    next_proto == IPPROTO_ESP	   ||
2447 	    next_proto == IPPROTO_AH	   ||
2448 	    next_proto == IPPROTO_DSTOPTS)
2449 		return rte_flow_error_set(error, EINVAL,
2450 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2451 					  "IPv6 proto (next header) should "
2452 					  "not be set as extension header");
2453 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2454 		return rte_flow_error_set(error, EINVAL,
2455 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2456 					  "wrong tunnel type - IPv4 specified "
2457 					  "but IPv6 item provided");
2458 	if (item_flags & l3m)
2459 		return rte_flow_error_set(error, ENOTSUP,
2460 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 					  "multiple L3 layers not supported");
2462 	else if (item_flags & l4m)
2463 		return rte_flow_error_set(error, EINVAL,
2464 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2465 					  "L3 cannot follow an L4 layer.");
2466 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2467 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2468 		return rte_flow_error_set(error, EINVAL,
2469 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2470 					  "L3 cannot follow an NVGRE layer.");
2471 	if (!mask)
2472 		mask = &rte_flow_item_ipv6_mask;
2473 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2474 					acc_mask ? (const uint8_t *)acc_mask
2475 						 : (const uint8_t *)&nic_mask,
2476 					sizeof(struct rte_flow_item_ipv6),
2477 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2478 	if (ret < 0)
2479 		return ret;
2480 	return 0;
2481 }
2482 
2483 /**
2484  * Validate UDP item.
2485  *
2486  * @param[in] item
2487  *   Item specification.
2488  * @param[in] item_flags
2489  *   Bit-fields that holds the items detected until now.
2490  * @param[in] target_protocol
2491  *   The next protocol in the previous item.
2492  * @param[in] flow_mask
2493  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2494  * @param[out] error
2495  *   Pointer to error structure.
2496  *
2497  * @return
2498  *   0 on success, a negative errno value otherwise and rte_errno is set.
2499  */
2500 int
2501 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2502 			    uint64_t item_flags,
2503 			    uint8_t target_protocol,
2504 			    struct rte_flow_error *error)
2505 {
2506 	const struct rte_flow_item_udp *mask = item->mask;
2507 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2508 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2509 				      MLX5_FLOW_LAYER_OUTER_L3;
2510 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2511 				      MLX5_FLOW_LAYER_OUTER_L4;
2512 	int ret;
2513 
2514 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2515 		return rte_flow_error_set(error, EINVAL,
2516 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2517 					  "protocol filtering not compatible"
2518 					  " with UDP layer");
2519 	if (!(item_flags & l3m))
2520 		return rte_flow_error_set(error, EINVAL,
2521 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2522 					  "L3 is mandatory to filter on L4");
2523 	if (item_flags & l4m)
2524 		return rte_flow_error_set(error, EINVAL,
2525 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2526 					  "multiple L4 layers not supported");
2527 	if (!mask)
2528 		mask = &rte_flow_item_udp_mask;
2529 	ret = mlx5_flow_item_acceptable
2530 		(item, (const uint8_t *)mask,
2531 		 (const uint8_t *)&rte_flow_item_udp_mask,
2532 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2533 		 error);
2534 	if (ret < 0)
2535 		return ret;
2536 	return 0;
2537 }
2538 
2539 /**
2540  * Validate TCP item.
2541  *
2542  * @param[in] item
2543  *   Item specification.
2544  * @param[in] item_flags
2545  *   Bit-fields that holds the items detected until now.
2546  * @param[in] target_protocol
2547  *   The next protocol in the previous item.
2548  * @param[out] error
2549  *   Pointer to error structure.
2550  *
2551  * @return
2552  *   0 on success, a negative errno value otherwise and rte_errno is set.
2553  */
2554 int
2555 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2556 			    uint64_t item_flags,
2557 			    uint8_t target_protocol,
2558 			    const struct rte_flow_item_tcp *flow_mask,
2559 			    struct rte_flow_error *error)
2560 {
2561 	const struct rte_flow_item_tcp *mask = item->mask;
2562 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2563 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2564 				      MLX5_FLOW_LAYER_OUTER_L3;
2565 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2566 				      MLX5_FLOW_LAYER_OUTER_L4;
2567 	int ret;
2568 
2569 	MLX5_ASSERT(flow_mask);
2570 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2571 		return rte_flow_error_set(error, EINVAL,
2572 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2573 					  "protocol filtering not compatible"
2574 					  " with TCP layer");
2575 	if (!(item_flags & l3m))
2576 		return rte_flow_error_set(error, EINVAL,
2577 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2578 					  "L3 is mandatory to filter on L4");
2579 	if (item_flags & l4m)
2580 		return rte_flow_error_set(error, EINVAL,
2581 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2582 					  "multiple L4 layers not supported");
2583 	if (!mask)
2584 		mask = &rte_flow_item_tcp_mask;
2585 	ret = mlx5_flow_item_acceptable
2586 		(item, (const uint8_t *)mask,
2587 		 (const uint8_t *)flow_mask,
2588 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2589 		 error);
2590 	if (ret < 0)
2591 		return ret;
2592 	return 0;
2593 }
2594 
2595 /**
2596  * Validate VXLAN item.
2597  *
2598  * @param[in] dev
2599  *   Pointer to the Ethernet device structure.
2600  * @param[in] udp_dport
2601  *   UDP destination port
2602  * @param[in] item
2603  *   Item specification.
2604  * @param[in] item_flags
2605  *   Bit-fields that holds the items detected until now.
2606  * @param[in] attr
2607  *   Flow rule attributes.
2608  * @param[out] error
2609  *   Pointer to error structure.
2610  *
2611  * @return
2612  *   0 on success, a negative errno value otherwise and rte_errno is set.
2613  */
2614 int
2615 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2616 			      uint16_t udp_dport,
2617 			      const struct rte_flow_item *item,
2618 			      uint64_t item_flags,
2619 			      const struct rte_flow_attr *attr,
2620 			      struct rte_flow_error *error)
2621 {
2622 	const struct rte_flow_item_vxlan *spec = item->spec;
2623 	const struct rte_flow_item_vxlan *mask = item->mask;
2624 	int ret;
2625 	struct mlx5_priv *priv = dev->data->dev_private;
2626 	union vni {
2627 		uint32_t vlan_id;
2628 		uint8_t vni[4];
2629 	} id = { .vlan_id = 0, };
2630 	const struct rte_flow_item_vxlan nic_mask = {
2631 		.vni = "\xff\xff\xff",
2632 		.rsvd1 = 0xff,
2633 	};
2634 	const struct rte_flow_item_vxlan *valid_mask;
2635 
2636 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2637 		return rte_flow_error_set(error, ENOTSUP,
2638 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2639 					  "multiple tunnel layers not"
2640 					  " supported");
2641 	valid_mask = &rte_flow_item_vxlan_mask;
2642 	/*
2643 	 * Verify only UDPv4 is present as defined in
2644 	 * https://tools.ietf.org/html/rfc7348
2645 	 */
2646 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2647 		return rte_flow_error_set(error, EINVAL,
2648 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2649 					  "no outer UDP layer found");
2650 	if (!mask)
2651 		mask = &rte_flow_item_vxlan_mask;
2652 
2653 	if (priv->sh->steering_format_version !=
2654 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2655 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2656 		/* FDB domain & NIC domain non-zero group */
2657 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2658 			valid_mask = &nic_mask;
2659 		/* Group zero in NIC domain */
2660 		if (!attr->group && !attr->transfer &&
2661 		    priv->sh->tunnel_header_0_1)
2662 			valid_mask = &nic_mask;
2663 	}
2664 	ret = mlx5_flow_item_acceptable
2665 		(item, (const uint8_t *)mask,
2666 		 (const uint8_t *)valid_mask,
2667 		 sizeof(struct rte_flow_item_vxlan),
2668 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2669 	if (ret < 0)
2670 		return ret;
2671 	if (spec) {
2672 		memcpy(&id.vni[1], spec->vni, 3);
2673 		memcpy(&id.vni[1], mask->vni, 3);
2674 	}
2675 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2676 		return rte_flow_error_set(error, ENOTSUP,
2677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2678 					  "VXLAN tunnel must be fully defined");
2679 	return 0;
2680 }
2681 
2682 /**
2683  * Validate VXLAN_GPE item.
2684  *
2685  * @param[in] item
2686  *   Item specification.
2687  * @param[in] item_flags
2688  *   Bit-fields that holds the items detected until now.
2689  * @param[in] priv
2690  *   Pointer to the private data structure.
2691  * @param[in] target_protocol
2692  *   The next protocol in the previous item.
2693  * @param[out] error
2694  *   Pointer to error structure.
2695  *
2696  * @return
2697  *   0 on success, a negative errno value otherwise and rte_errno is set.
2698  */
2699 int
2700 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2701 				  uint64_t item_flags,
2702 				  struct rte_eth_dev *dev,
2703 				  struct rte_flow_error *error)
2704 {
2705 	struct mlx5_priv *priv = dev->data->dev_private;
2706 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2707 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2708 	int ret;
2709 	union vni {
2710 		uint32_t vlan_id;
2711 		uint8_t vni[4];
2712 	} id = { .vlan_id = 0, };
2713 
2714 	if (!priv->config.l3_vxlan_en)
2715 		return rte_flow_error_set(error, ENOTSUP,
2716 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2717 					  "L3 VXLAN is not enabled by device"
2718 					  " parameter and/or not configured in"
2719 					  " firmware");
2720 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2721 		return rte_flow_error_set(error, ENOTSUP,
2722 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2723 					  "multiple tunnel layers not"
2724 					  " supported");
2725 	/*
2726 	 * Verify only UDPv4 is present as defined in
2727 	 * https://tools.ietf.org/html/rfc7348
2728 	 */
2729 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2730 		return rte_flow_error_set(error, EINVAL,
2731 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2732 					  "no outer UDP layer found");
2733 	if (!mask)
2734 		mask = &rte_flow_item_vxlan_gpe_mask;
2735 	ret = mlx5_flow_item_acceptable
2736 		(item, (const uint8_t *)mask,
2737 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2738 		 sizeof(struct rte_flow_item_vxlan_gpe),
2739 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2740 	if (ret < 0)
2741 		return ret;
2742 	if (spec) {
2743 		if (spec->protocol)
2744 			return rte_flow_error_set(error, ENOTSUP,
2745 						  RTE_FLOW_ERROR_TYPE_ITEM,
2746 						  item,
2747 						  "VxLAN-GPE protocol"
2748 						  " not supported");
2749 		memcpy(&id.vni[1], spec->vni, 3);
2750 		memcpy(&id.vni[1], mask->vni, 3);
2751 	}
2752 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2753 		return rte_flow_error_set(error, ENOTSUP,
2754 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2755 					  "VXLAN-GPE tunnel must be fully"
2756 					  " defined");
2757 	return 0;
2758 }
2759 /**
2760  * Validate GRE Key item.
2761  *
2762  * @param[in] item
2763  *   Item specification.
2764  * @param[in] item_flags
2765  *   Bit flags to mark detected items.
2766  * @param[in] gre_item
2767  *   Pointer to gre_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_key(const struct rte_flow_item *item,
2776 				uint64_t item_flags,
2777 				const struct rte_flow_item *gre_item,
2778 				struct rte_flow_error *error)
2779 {
2780 	const rte_be32_t *mask = item->mask;
2781 	int ret = 0;
2782 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2783 	const struct rte_flow_item_gre *gre_spec;
2784 	const struct rte_flow_item_gre *gre_mask;
2785 
2786 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2787 		return rte_flow_error_set(error, ENOTSUP,
2788 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2789 					  "Multiple GRE key not support");
2790 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2791 		return rte_flow_error_set(error, ENOTSUP,
2792 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 					  "No preceding GRE header");
2794 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2795 		return rte_flow_error_set(error, ENOTSUP,
2796 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2797 					  "GRE key following a wrong item");
2798 	gre_mask = gre_item->mask;
2799 	if (!gre_mask)
2800 		gre_mask = &rte_flow_item_gre_mask;
2801 	gre_spec = gre_item->spec;
2802 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2803 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2804 		return rte_flow_error_set(error, EINVAL,
2805 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2806 					  "Key bit must be on");
2807 
2808 	if (!mask)
2809 		mask = &gre_key_default_mask;
2810 	ret = mlx5_flow_item_acceptable
2811 		(item, (const uint8_t *)mask,
2812 		 (const uint8_t *)&gre_key_default_mask,
2813 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2814 	return ret;
2815 }
2816 
2817 /**
2818  * Validate GRE item.
2819  *
2820  * @param[in] item
2821  *   Item specification.
2822  * @param[in] item_flags
2823  *   Bit flags to mark detected items.
2824  * @param[in] target_protocol
2825  *   The next protocol in the previous item.
2826  * @param[out] error
2827  *   Pointer to error structure.
2828  *
2829  * @return
2830  *   0 on success, a negative errno value otherwise and rte_errno is set.
2831  */
2832 int
2833 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2834 			    uint64_t item_flags,
2835 			    uint8_t target_protocol,
2836 			    struct rte_flow_error *error)
2837 {
2838 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2839 	const struct rte_flow_item_gre *mask = item->mask;
2840 	int ret;
2841 	const struct rte_flow_item_gre nic_mask = {
2842 		.c_rsvd0_ver = RTE_BE16(0xB000),
2843 		.protocol = RTE_BE16(UINT16_MAX),
2844 	};
2845 
2846 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2847 		return rte_flow_error_set(error, EINVAL,
2848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2849 					  "protocol filtering not compatible"
2850 					  " with this GRE layer");
2851 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2852 		return rte_flow_error_set(error, ENOTSUP,
2853 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2854 					  "multiple tunnel layers not"
2855 					  " supported");
2856 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2857 		return rte_flow_error_set(error, ENOTSUP,
2858 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 					  "L3 Layer is missing");
2860 	if (!mask)
2861 		mask = &rte_flow_item_gre_mask;
2862 	ret = mlx5_flow_item_acceptable
2863 		(item, (const uint8_t *)mask,
2864 		 (const uint8_t *)&nic_mask,
2865 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2866 		 error);
2867 	if (ret < 0)
2868 		return ret;
2869 #ifndef HAVE_MLX5DV_DR
2870 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2871 	if (spec && (spec->protocol & mask->protocol))
2872 		return rte_flow_error_set(error, ENOTSUP,
2873 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 					  "without MPLS support the"
2875 					  " specification cannot be used for"
2876 					  " filtering");
2877 #endif
2878 #endif
2879 	return 0;
2880 }
2881 
2882 /**
2883  * Validate Geneve item.
2884  *
2885  * @param[in] item
2886  *   Item specification.
2887  * @param[in] itemFlags
2888  *   Bit-fields that holds the items detected until now.
2889  * @param[in] enPriv
2890  *   Pointer to the private data structure.
2891  * @param[out] error
2892  *   Pointer to error structure.
2893  *
2894  * @return
2895  *   0 on success, a negative errno value otherwise and rte_errno is set.
2896  */
2897 
2898 int
2899 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2900 			       uint64_t item_flags,
2901 			       struct rte_eth_dev *dev,
2902 			       struct rte_flow_error *error)
2903 {
2904 	struct mlx5_priv *priv = dev->data->dev_private;
2905 	const struct rte_flow_item_geneve *spec = item->spec;
2906 	const struct rte_flow_item_geneve *mask = item->mask;
2907 	int ret;
2908 	uint16_t gbhdr;
2909 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2910 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2911 	const struct rte_flow_item_geneve nic_mask = {
2912 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2913 		.vni = "\xff\xff\xff",
2914 		.protocol = RTE_BE16(UINT16_MAX),
2915 	};
2916 
2917 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2918 		return rte_flow_error_set(error, ENOTSUP,
2919 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2920 					  "L3 Geneve is not enabled by device"
2921 					  " parameter and/or not configured in"
2922 					  " firmware");
2923 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2924 		return rte_flow_error_set(error, ENOTSUP,
2925 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2926 					  "multiple tunnel layers not"
2927 					  " supported");
2928 	/*
2929 	 * Verify only UDPv4 is present as defined in
2930 	 * https://tools.ietf.org/html/rfc7348
2931 	 */
2932 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2933 		return rte_flow_error_set(error, EINVAL,
2934 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2935 					  "no outer UDP layer found");
2936 	if (!mask)
2937 		mask = &rte_flow_item_geneve_mask;
2938 	ret = mlx5_flow_item_acceptable
2939 				  (item, (const uint8_t *)mask,
2940 				   (const uint8_t *)&nic_mask,
2941 				   sizeof(struct rte_flow_item_geneve),
2942 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2943 	if (ret)
2944 		return ret;
2945 	if (spec) {
2946 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2947 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2948 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2949 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2950 			return rte_flow_error_set(error, ENOTSUP,
2951 						  RTE_FLOW_ERROR_TYPE_ITEM,
2952 						  item,
2953 						  "Geneve protocol unsupported"
2954 						  " fields are being used");
2955 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2956 			return rte_flow_error_set
2957 					(error, ENOTSUP,
2958 					 RTE_FLOW_ERROR_TYPE_ITEM,
2959 					 item,
2960 					 "Unsupported Geneve options length");
2961 	}
2962 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2963 		return rte_flow_error_set
2964 				    (error, ENOTSUP,
2965 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2966 				     "Geneve tunnel must be fully defined");
2967 	return 0;
2968 }
2969 
2970 /**
2971  * Validate Geneve TLV option item.
2972  *
2973  * @param[in] item
2974  *   Item specification.
2975  * @param[in] last_item
2976  *   Previous validated item in the pattern items.
2977  * @param[in] geneve_item
2978  *   Previous GENEVE item specification.
2979  * @param[in] dev
2980  *   Pointer to the rte_eth_dev structure.
2981  * @param[out] error
2982  *   Pointer to error structure.
2983  *
2984  * @return
2985  *   0 on success, a negative errno value otherwise and rte_errno is set.
2986  */
2987 int
2988 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2989 				   uint64_t last_item,
2990 				   const struct rte_flow_item *geneve_item,
2991 				   struct rte_eth_dev *dev,
2992 				   struct rte_flow_error *error)
2993 {
2994 	struct mlx5_priv *priv = dev->data->dev_private;
2995 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2996 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2997 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2998 	uint8_t data_max_supported =
2999 			hca_attr->max_geneve_tlv_option_data_len * 4;
3000 	struct mlx5_dev_config *config = &priv->config;
3001 	const struct rte_flow_item_geneve *geneve_spec;
3002 	const struct rte_flow_item_geneve *geneve_mask;
3003 	const struct rte_flow_item_geneve_opt *spec = item->spec;
3004 	const struct rte_flow_item_geneve_opt *mask = item->mask;
3005 	unsigned int i;
3006 	unsigned int data_len;
3007 	uint8_t tlv_option_len;
3008 	uint16_t optlen_m, optlen_v;
3009 	const struct rte_flow_item_geneve_opt full_mask = {
3010 		.option_class = RTE_BE16(0xffff),
3011 		.option_type = 0xff,
3012 		.option_len = 0x1f,
3013 	};
3014 
3015 	if (!mask)
3016 		mask = &rte_flow_item_geneve_opt_mask;
3017 	if (!spec)
3018 		return rte_flow_error_set
3019 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3020 			"Geneve TLV opt class/type/length must be specified");
3021 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3022 		return rte_flow_error_set
3023 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 			"Geneve TLV opt length exceeds the limit (31)");
3025 	/* Check if class type and length masks are full. */
3026 	if (full_mask.option_class != mask->option_class ||
3027 	    full_mask.option_type != mask->option_type ||
3028 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
3029 		return rte_flow_error_set
3030 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3031 			"Geneve TLV opt class/type/length masks must be full");
3032 	/* Check if length is supported */
3033 	if ((uint32_t)spec->option_len >
3034 			config->hca_attr.max_geneve_tlv_option_data_len)
3035 		return rte_flow_error_set
3036 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 			"Geneve TLV opt length not supported");
3038 	if (config->hca_attr.max_geneve_tlv_options > 1)
3039 		DRV_LOG(DEBUG,
3040 			"max_geneve_tlv_options supports more than 1 option");
3041 	/* Check GENEVE item preceding. */
3042 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3043 		return rte_flow_error_set
3044 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3045 			"Geneve opt item must be preceded with Geneve item");
3046 	geneve_spec = geneve_item->spec;
3047 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3048 					  &rte_flow_item_geneve_mask;
3049 	/* Check if GENEVE TLV option size doesn't exceed option length */
3050 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3051 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3052 		tlv_option_len = spec->option_len & mask->option_len;
3053 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3054 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3055 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3056 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3057 		if ((optlen_v & optlen_m) <= tlv_option_len)
3058 			return rte_flow_error_set
3059 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3060 				 "GENEVE TLV option length exceeds optlen");
3061 	}
3062 	/* Check if length is 0 or data is 0. */
3063 	if (spec->data == NULL || spec->option_len == 0)
3064 		return rte_flow_error_set
3065 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3066 			"Geneve TLV opt with zero data/length not supported");
3067 	/* Check not all data & mask are 0. */
3068 	data_len = spec->option_len * 4;
3069 	if (mask->data == NULL) {
3070 		for (i = 0; i < data_len; i++)
3071 			if (spec->data[i])
3072 				break;
3073 		if (i == data_len)
3074 			return rte_flow_error_set(error, ENOTSUP,
3075 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3076 				"Can't match on Geneve option data 0");
3077 	} else {
3078 		for (i = 0; i < data_len; i++)
3079 			if (spec->data[i] & mask->data[i])
3080 				break;
3081 		if (i == data_len)
3082 			return rte_flow_error_set(error, ENOTSUP,
3083 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3084 				"Can't match on Geneve option data and mask 0");
3085 		/* Check data mask supported. */
3086 		for (i = data_max_supported; i < data_len ; i++)
3087 			if (mask->data[i])
3088 				return rte_flow_error_set(error, ENOTSUP,
3089 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3090 					"Data mask is of unsupported size");
3091 	}
3092 	/* Check GENEVE option is supported in NIC. */
3093 	if (!config->hca_attr.geneve_tlv_opt)
3094 		return rte_flow_error_set
3095 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3096 			"Geneve TLV opt not supported");
3097 	/* Check if we already have geneve option with different type/class. */
3098 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3099 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3100 	if (geneve_opt_resource != NULL)
3101 		if (geneve_opt_resource->option_class != spec->option_class ||
3102 		    geneve_opt_resource->option_type != spec->option_type ||
3103 		    geneve_opt_resource->length != spec->option_len) {
3104 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3105 			return rte_flow_error_set(error, ENOTSUP,
3106 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 				"Only one Geneve TLV option supported");
3108 		}
3109 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3110 	return 0;
3111 }
3112 
3113 /**
3114  * Validate MPLS item.
3115  *
3116  * @param[in] dev
3117  *   Pointer to the rte_eth_dev structure.
3118  * @param[in] item
3119  *   Item specification.
3120  * @param[in] item_flags
3121  *   Bit-fields that holds the items detected until now.
3122  * @param[in] prev_layer
3123  *   The protocol layer indicated in previous item.
3124  * @param[out] error
3125  *   Pointer to error structure.
3126  *
3127  * @return
3128  *   0 on success, a negative errno value otherwise and rte_errno is set.
3129  */
3130 int
3131 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3132 			     const struct rte_flow_item *item __rte_unused,
3133 			     uint64_t item_flags __rte_unused,
3134 			     uint64_t prev_layer __rte_unused,
3135 			     struct rte_flow_error *error)
3136 {
3137 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3138 	const struct rte_flow_item_mpls *mask = item->mask;
3139 	struct mlx5_priv *priv = dev->data->dev_private;
3140 	int ret;
3141 
3142 	if (!priv->config.mpls_en)
3143 		return rte_flow_error_set(error, ENOTSUP,
3144 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3145 					  "MPLS not supported or"
3146 					  " disabled in firmware"
3147 					  " configuration.");
3148 	/* MPLS over UDP, GRE is allowed */
3149 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3150 			    MLX5_FLOW_LAYER_GRE |
3151 			    MLX5_FLOW_LAYER_GRE_KEY)))
3152 		return rte_flow_error_set(error, EINVAL,
3153 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3154 					  "protocol filtering not compatible"
3155 					  " with MPLS layer");
3156 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3157 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3158 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3159 		return rte_flow_error_set(error, ENOTSUP,
3160 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3161 					  "multiple tunnel layers not"
3162 					  " supported");
3163 	if (!mask)
3164 		mask = &rte_flow_item_mpls_mask;
3165 	ret = mlx5_flow_item_acceptable
3166 		(item, (const uint8_t *)mask,
3167 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3168 		 sizeof(struct rte_flow_item_mpls),
3169 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3170 	if (ret < 0)
3171 		return ret;
3172 	return 0;
3173 #else
3174 	return rte_flow_error_set(error, ENOTSUP,
3175 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3176 				  "MPLS is not supported by Verbs, please"
3177 				  " update.");
3178 #endif
3179 }
3180 
3181 /**
3182  * Validate NVGRE item.
3183  *
3184  * @param[in] item
3185  *   Item specification.
3186  * @param[in] item_flags
3187  *   Bit flags to mark detected items.
3188  * @param[in] target_protocol
3189  *   The next protocol in the previous item.
3190  * @param[out] error
3191  *   Pointer to error structure.
3192  *
3193  * @return
3194  *   0 on success, a negative errno value otherwise and rte_errno is set.
3195  */
3196 int
3197 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3198 			      uint64_t item_flags,
3199 			      uint8_t target_protocol,
3200 			      struct rte_flow_error *error)
3201 {
3202 	const struct rte_flow_item_nvgre *mask = item->mask;
3203 	int ret;
3204 
3205 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3206 		return rte_flow_error_set(error, EINVAL,
3207 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3208 					  "protocol filtering not compatible"
3209 					  " with this GRE layer");
3210 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3211 		return rte_flow_error_set(error, ENOTSUP,
3212 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3213 					  "multiple tunnel layers not"
3214 					  " supported");
3215 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3216 		return rte_flow_error_set(error, ENOTSUP,
3217 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3218 					  "L3 Layer is missing");
3219 	if (!mask)
3220 		mask = &rte_flow_item_nvgre_mask;
3221 	ret = mlx5_flow_item_acceptable
3222 		(item, (const uint8_t *)mask,
3223 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3224 		 sizeof(struct rte_flow_item_nvgre),
3225 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3226 	if (ret < 0)
3227 		return ret;
3228 	return 0;
3229 }
3230 
3231 /**
3232  * Validate eCPRI item.
3233  *
3234  * @param[in] item
3235  *   Item specification.
3236  * @param[in] item_flags
3237  *   Bit-fields that holds the items detected until now.
3238  * @param[in] last_item
3239  *   Previous validated item in the pattern items.
3240  * @param[in] ether_type
3241  *   Type in the ethernet layer header (including dot1q).
3242  * @param[in] acc_mask
3243  *   Acceptable mask, if NULL default internal default mask
3244  *   will be used to check whether item fields are supported.
3245  * @param[out] error
3246  *   Pointer to error structure.
3247  *
3248  * @return
3249  *   0 on success, a negative errno value otherwise and rte_errno is set.
3250  */
3251 int
3252 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3253 			      uint64_t item_flags,
3254 			      uint64_t last_item,
3255 			      uint16_t ether_type,
3256 			      const struct rte_flow_item_ecpri *acc_mask,
3257 			      struct rte_flow_error *error)
3258 {
3259 	const struct rte_flow_item_ecpri *mask = item->mask;
3260 	const struct rte_flow_item_ecpri nic_mask = {
3261 		.hdr = {
3262 			.common = {
3263 				.u32 =
3264 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3265 					.type = 0xFF,
3266 					}).u32),
3267 			},
3268 			.dummy[0] = 0xFFFFFFFF,
3269 		},
3270 	};
3271 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3272 					MLX5_FLOW_LAYER_OUTER_VLAN);
3273 	struct rte_flow_item_ecpri mask_lo;
3274 
3275 	if (!(last_item & outer_l2_vlan) &&
3276 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3277 		return rte_flow_error_set(error, EINVAL,
3278 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3279 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3280 	if ((last_item & outer_l2_vlan) && ether_type &&
3281 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3282 		return rte_flow_error_set(error, EINVAL,
3283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3284 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3285 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3286 		return rte_flow_error_set(error, EINVAL,
3287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3288 					  "eCPRI with tunnel is not supported right now");
3289 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3290 		return rte_flow_error_set(error, ENOTSUP,
3291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3292 					  "multiple L3 layers not supported");
3293 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3294 		return rte_flow_error_set(error, EINVAL,
3295 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3296 					  "eCPRI cannot coexist with a TCP layer");
3297 	/* In specification, eCPRI could be over UDP layer. */
3298 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3299 		return rte_flow_error_set(error, EINVAL,
3300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3301 					  "eCPRI over UDP layer is not yet supported right now");
3302 	/* Mask for type field in common header could be zero. */
3303 	if (!mask)
3304 		mask = &rte_flow_item_ecpri_mask;
3305 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3306 	/* Input mask is in big-endian format. */
3307 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3308 		return rte_flow_error_set(error, EINVAL,
3309 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3310 					  "partial mask is not supported for protocol");
3311 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3312 		return rte_flow_error_set(error, EINVAL,
3313 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3314 					  "message header mask must be after a type mask");
3315 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3316 					 acc_mask ? (const uint8_t *)acc_mask
3317 						  : (const uint8_t *)&nic_mask,
3318 					 sizeof(struct rte_flow_item_ecpri),
3319 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3320 }
3321 
3322 static int
3323 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3324 		   const struct rte_flow_attr *attr __rte_unused,
3325 		   const struct rte_flow_item items[] __rte_unused,
3326 		   const struct rte_flow_action actions[] __rte_unused,
3327 		   bool external __rte_unused,
3328 		   int hairpin __rte_unused,
3329 		   struct rte_flow_error *error)
3330 {
3331 	return rte_flow_error_set(error, ENOTSUP,
3332 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3333 }
3334 
3335 static struct mlx5_flow *
3336 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3337 		  const struct rte_flow_attr *attr __rte_unused,
3338 		  const struct rte_flow_item items[] __rte_unused,
3339 		  const struct rte_flow_action actions[] __rte_unused,
3340 		  struct rte_flow_error *error)
3341 {
3342 	rte_flow_error_set(error, ENOTSUP,
3343 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3344 	return NULL;
3345 }
3346 
3347 static int
3348 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3349 		    struct mlx5_flow *dev_flow __rte_unused,
3350 		    const struct rte_flow_attr *attr __rte_unused,
3351 		    const struct rte_flow_item items[] __rte_unused,
3352 		    const struct rte_flow_action actions[] __rte_unused,
3353 		    struct rte_flow_error *error)
3354 {
3355 	return rte_flow_error_set(error, ENOTSUP,
3356 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3357 }
3358 
3359 static int
3360 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3361 		struct rte_flow *flow __rte_unused,
3362 		struct rte_flow_error *error)
3363 {
3364 	return rte_flow_error_set(error, ENOTSUP,
3365 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3366 }
3367 
3368 static void
3369 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3370 		 struct rte_flow *flow __rte_unused)
3371 {
3372 }
3373 
3374 static void
3375 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3376 		  struct rte_flow *flow __rte_unused)
3377 {
3378 }
3379 
3380 static int
3381 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3382 		struct rte_flow *flow __rte_unused,
3383 		const struct rte_flow_action *actions __rte_unused,
3384 		void *data __rte_unused,
3385 		struct rte_flow_error *error)
3386 {
3387 	return rte_flow_error_set(error, ENOTSUP,
3388 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3389 }
3390 
3391 static int
3392 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3393 		      uint32_t domains __rte_unused,
3394 		      uint32_t flags __rte_unused)
3395 {
3396 	return 0;
3397 }
3398 
3399 /* Void driver to protect from null pointer reference. */
3400 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3401 	.validate = flow_null_validate,
3402 	.prepare = flow_null_prepare,
3403 	.translate = flow_null_translate,
3404 	.apply = flow_null_apply,
3405 	.remove = flow_null_remove,
3406 	.destroy = flow_null_destroy,
3407 	.query = flow_null_query,
3408 	.sync_domain = flow_null_sync_domain,
3409 };
3410 
3411 /**
3412  * Select flow driver type according to flow attributes and device
3413  * configuration.
3414  *
3415  * @param[in] dev
3416  *   Pointer to the dev structure.
3417  * @param[in] attr
3418  *   Pointer to the flow attributes.
3419  *
3420  * @return
3421  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3422  */
3423 static enum mlx5_flow_drv_type
3424 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3425 {
3426 	struct mlx5_priv *priv = dev->data->dev_private;
3427 	/* The OS can determine first a specific flow type (DV, VERBS) */
3428 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3429 
3430 	if (type != MLX5_FLOW_TYPE_MAX)
3431 		return type;
3432 	/* If no OS specific type - continue with DV/VERBS selection */
3433 	if (attr->transfer && priv->config.dv_esw_en)
3434 		type = MLX5_FLOW_TYPE_DV;
3435 	if (!attr->transfer)
3436 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3437 						 MLX5_FLOW_TYPE_VERBS;
3438 	return type;
3439 }
3440 
3441 #define flow_get_drv_ops(type) flow_drv_ops[type]
3442 
3443 /**
3444  * Flow driver validation API. This abstracts calling driver specific functions.
3445  * The type of flow driver is determined according to flow attributes.
3446  *
3447  * @param[in] dev
3448  *   Pointer to the dev structure.
3449  * @param[in] attr
3450  *   Pointer to the flow attributes.
3451  * @param[in] items
3452  *   Pointer to the list of items.
3453  * @param[in] actions
3454  *   Pointer to the list of actions.
3455  * @param[in] external
3456  *   This flow rule is created by request external to PMD.
3457  * @param[in] hairpin
3458  *   Number of hairpin TX actions, 0 means classic flow.
3459  * @param[out] error
3460  *   Pointer to the error structure.
3461  *
3462  * @return
3463  *   0 on success, a negative errno value otherwise and rte_errno is set.
3464  */
3465 static inline int
3466 flow_drv_validate(struct rte_eth_dev *dev,
3467 		  const struct rte_flow_attr *attr,
3468 		  const struct rte_flow_item items[],
3469 		  const struct rte_flow_action actions[],
3470 		  bool external, int hairpin, struct rte_flow_error *error)
3471 {
3472 	const struct mlx5_flow_driver_ops *fops;
3473 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3474 
3475 	fops = flow_get_drv_ops(type);
3476 	return fops->validate(dev, attr, items, actions, external,
3477 			      hairpin, error);
3478 }
3479 
3480 /**
3481  * Flow driver preparation API. This abstracts calling driver specific
3482  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3483  * calculates the size of memory required for device flow, allocates the memory,
3484  * initializes the device flow and returns the pointer.
3485  *
3486  * @note
3487  *   This function initializes device flow structure such as dv or verbs in
3488  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3489  *   rest. For example, adding returning device flow to flow->dev_flow list and
3490  *   setting backward reference to the flow should be done out of this function.
3491  *   layers field is not filled either.
3492  *
3493  * @param[in] dev
3494  *   Pointer to the dev structure.
3495  * @param[in] attr
3496  *   Pointer to the flow attributes.
3497  * @param[in] items
3498  *   Pointer to the list of items.
3499  * @param[in] actions
3500  *   Pointer to the list of actions.
3501  * @param[in] flow_idx
3502  *   This memory pool index to the flow.
3503  * @param[out] error
3504  *   Pointer to the error structure.
3505  *
3506  * @return
3507  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3508  */
3509 static inline struct mlx5_flow *
3510 flow_drv_prepare(struct rte_eth_dev *dev,
3511 		 const struct rte_flow *flow,
3512 		 const struct rte_flow_attr *attr,
3513 		 const struct rte_flow_item items[],
3514 		 const struct rte_flow_action actions[],
3515 		 uint32_t flow_idx,
3516 		 struct rte_flow_error *error)
3517 {
3518 	const struct mlx5_flow_driver_ops *fops;
3519 	enum mlx5_flow_drv_type type = flow->drv_type;
3520 	struct mlx5_flow *mlx5_flow = NULL;
3521 
3522 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3523 	fops = flow_get_drv_ops(type);
3524 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3525 	if (mlx5_flow)
3526 		mlx5_flow->flow_idx = flow_idx;
3527 	return mlx5_flow;
3528 }
3529 
3530 /**
3531  * Flow driver translation API. This abstracts calling driver specific
3532  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3533  * translates a generic flow into a driver flow. flow_drv_prepare() must
3534  * precede.
3535  *
3536  * @note
3537  *   dev_flow->layers could be filled as a result of parsing during translation
3538  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3539  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3540  *   flow->actions could be overwritten even though all the expanded dev_flows
3541  *   have the same actions.
3542  *
3543  * @param[in] dev
3544  *   Pointer to the rte dev structure.
3545  * @param[in, out] dev_flow
3546  *   Pointer to the mlx5 flow.
3547  * @param[in] attr
3548  *   Pointer to the flow attributes.
3549  * @param[in] items
3550  *   Pointer to the list of items.
3551  * @param[in] actions
3552  *   Pointer to the list of actions.
3553  * @param[out] error
3554  *   Pointer to the error structure.
3555  *
3556  * @return
3557  *   0 on success, a negative errno value otherwise and rte_errno is set.
3558  */
3559 static inline int
3560 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3561 		   const struct rte_flow_attr *attr,
3562 		   const struct rte_flow_item items[],
3563 		   const struct rte_flow_action actions[],
3564 		   struct rte_flow_error *error)
3565 {
3566 	const struct mlx5_flow_driver_ops *fops;
3567 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3568 
3569 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3570 	fops = flow_get_drv_ops(type);
3571 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3572 }
3573 
3574 /**
3575  * Flow driver apply API. This abstracts calling driver specific functions.
3576  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3577  * translated driver flows on to device. flow_drv_translate() must precede.
3578  *
3579  * @param[in] dev
3580  *   Pointer to Ethernet device structure.
3581  * @param[in, out] flow
3582  *   Pointer to flow structure.
3583  * @param[out] error
3584  *   Pointer to error structure.
3585  *
3586  * @return
3587  *   0 on success, a negative errno value otherwise and rte_errno is set.
3588  */
3589 static inline int
3590 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3591 	       struct rte_flow_error *error)
3592 {
3593 	const struct mlx5_flow_driver_ops *fops;
3594 	enum mlx5_flow_drv_type type = flow->drv_type;
3595 
3596 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3597 	fops = flow_get_drv_ops(type);
3598 	return fops->apply(dev, flow, error);
3599 }
3600 
3601 /**
3602  * Flow driver destroy API. This abstracts calling driver specific functions.
3603  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3604  * on device and releases resources of the flow.
3605  *
3606  * @param[in] dev
3607  *   Pointer to Ethernet device.
3608  * @param[in, out] flow
3609  *   Pointer to flow structure.
3610  */
3611 static inline void
3612 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3613 {
3614 	const struct mlx5_flow_driver_ops *fops;
3615 	enum mlx5_flow_drv_type type = flow->drv_type;
3616 
3617 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3618 	fops = flow_get_drv_ops(type);
3619 	fops->destroy(dev, flow);
3620 }
3621 
3622 /**
3623  * Flow driver find RSS policy tbl API. This abstracts calling driver
3624  * specific functions. Parent flow (rte_flow) should have driver
3625  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3626  *
3627  * @param[in] dev
3628  *   Pointer to Ethernet device.
3629  * @param[in, out] flow
3630  *   Pointer to flow structure.
3631  * @param[in] policy
3632  *   Pointer to meter policy table.
3633  * @param[in] rss_desc
3634  *   Pointer to rss_desc
3635  */
3636 static struct mlx5_flow_meter_sub_policy *
3637 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3638 		struct rte_flow *flow,
3639 		struct mlx5_flow_meter_policy *policy,
3640 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3641 {
3642 	const struct mlx5_flow_driver_ops *fops;
3643 	enum mlx5_flow_drv_type type = flow->drv_type;
3644 
3645 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3646 	fops = flow_get_drv_ops(type);
3647 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3648 }
3649 
3650 /**
3651  * Flow driver color tag rule API. This abstracts calling driver
3652  * specific functions. Parent flow (rte_flow) should have driver
3653  * type (drv_type). It will create the color tag rules in hierarchy meter.
3654  *
3655  * @param[in] dev
3656  *   Pointer to Ethernet device.
3657  * @param[in, out] flow
3658  *   Pointer to flow structure.
3659  * @param[in] fm
3660  *   Pointer to flow meter structure.
3661  * @param[in] src_port
3662  *   The src port this extra rule should use.
3663  * @param[in] item
3664  *   The src port id match item.
3665  * @param[out] error
3666  *   Pointer to error structure.
3667  */
3668 static int
3669 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3670 		struct rte_flow *flow,
3671 		struct mlx5_flow_meter_info *fm,
3672 		int32_t src_port,
3673 		const struct rte_flow_item *item,
3674 		struct rte_flow_error *error)
3675 {
3676 	const struct mlx5_flow_driver_ops *fops;
3677 	enum mlx5_flow_drv_type type = flow->drv_type;
3678 
3679 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3680 	fops = flow_get_drv_ops(type);
3681 	return fops->meter_hierarchy_rule_create(dev, fm,
3682 						src_port, item, error);
3683 }
3684 
3685 /**
3686  * Get RSS action from the action list.
3687  *
3688  * @param[in] dev
3689  *   Pointer to Ethernet device.
3690  * @param[in] actions
3691  *   Pointer to the list of actions.
3692  * @param[in] flow
3693  *   Parent flow structure pointer.
3694  *
3695  * @return
3696  *   Pointer to the RSS action if exist, else return NULL.
3697  */
3698 static const struct rte_flow_action_rss*
3699 flow_get_rss_action(struct rte_eth_dev *dev,
3700 		    const struct rte_flow_action actions[])
3701 {
3702 	struct mlx5_priv *priv = dev->data->dev_private;
3703 	const struct rte_flow_action_rss *rss = NULL;
3704 	struct mlx5_meter_policy_action_container *acg;
3705 	struct mlx5_meter_policy_action_container *acy;
3706 
3707 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3708 		switch (actions->type) {
3709 		case RTE_FLOW_ACTION_TYPE_RSS:
3710 			rss = actions->conf;
3711 			break;
3712 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3713 		{
3714 			const struct rte_flow_action_sample *sample =
3715 								actions->conf;
3716 			const struct rte_flow_action *act = sample->actions;
3717 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3718 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3719 					rss = act->conf;
3720 			break;
3721 		}
3722 		case RTE_FLOW_ACTION_TYPE_METER:
3723 		{
3724 			uint32_t mtr_idx;
3725 			struct mlx5_flow_meter_info *fm;
3726 			struct mlx5_flow_meter_policy *policy;
3727 			const struct rte_flow_action_meter *mtr = actions->conf;
3728 
3729 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3730 			if (fm && !fm->def_policy) {
3731 				policy = mlx5_flow_meter_policy_find(dev,
3732 						fm->policy_id, NULL);
3733 				MLX5_ASSERT(policy);
3734 				if (policy->is_hierarchy) {
3735 					policy =
3736 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3737 									policy);
3738 					if (!policy)
3739 						return NULL;
3740 				}
3741 				if (policy->is_rss) {
3742 					acg =
3743 					&policy->act_cnt[RTE_COLOR_GREEN];
3744 					acy =
3745 					&policy->act_cnt[RTE_COLOR_YELLOW];
3746 					if (acg->fate_action ==
3747 					    MLX5_FLOW_FATE_SHARED_RSS)
3748 						rss = acg->rss->conf;
3749 					else if (acy->fate_action ==
3750 						 MLX5_FLOW_FATE_SHARED_RSS)
3751 						rss = acy->rss->conf;
3752 				}
3753 			}
3754 			break;
3755 		}
3756 		default:
3757 			break;
3758 		}
3759 	}
3760 	return rss;
3761 }
3762 
3763 /**
3764  * Get ASO age action by index.
3765  *
3766  * @param[in] dev
3767  *   Pointer to the Ethernet device structure.
3768  * @param[in] age_idx
3769  *   Index to the ASO age action.
3770  *
3771  * @return
3772  *   The specified ASO age action.
3773  */
3774 struct mlx5_aso_age_action*
3775 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3776 {
3777 	uint16_t pool_idx = age_idx & UINT16_MAX;
3778 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3779 	struct mlx5_priv *priv = dev->data->dev_private;
3780 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3781 	struct mlx5_aso_age_pool *pool;
3782 
3783 	rte_rwlock_read_lock(&mng->resize_rwl);
3784 	pool = mng->pools[pool_idx];
3785 	rte_rwlock_read_unlock(&mng->resize_rwl);
3786 	return &pool->actions[offset - 1];
3787 }
3788 
3789 /* maps indirect action to translated direct in some actions array */
3790 struct mlx5_translated_action_handle {
3791 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3792 	int index; /**< Index in related array of rte_flow_action. */
3793 };
3794 
3795 /**
3796  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3797  * direct action if translation possible.
3798  * This functionality used to run same execution path for both direct and
3799  * indirect actions on flow create. All necessary preparations for indirect
3800  * action handling should be performed on *handle* actions list returned
3801  * from this call.
3802  *
3803  * @param[in] dev
3804  *   Pointer to Ethernet device.
3805  * @param[in] actions
3806  *   List of actions to translate.
3807  * @param[out] handle
3808  *   List to store translated indirect action object handles.
3809  * @param[in, out] indir_n
3810  *   Size of *handle* array. On return should be updated with number of
3811  *   indirect actions retrieved from the *actions* list.
3812  * @param[out] translated_actions
3813  *   List of actions where all indirect actions were translated to direct
3814  *   if possible. NULL if no translation took place.
3815  * @param[out] error
3816  *   Pointer to the error structure.
3817  *
3818  * @return
3819  *   0 on success, a negative errno value otherwise and rte_errno is set.
3820  */
3821 static int
3822 flow_action_handles_translate(struct rte_eth_dev *dev,
3823 			      const struct rte_flow_action actions[],
3824 			      struct mlx5_translated_action_handle *handle,
3825 			      int *indir_n,
3826 			      struct rte_flow_action **translated_actions,
3827 			      struct rte_flow_error *error)
3828 {
3829 	struct mlx5_priv *priv = dev->data->dev_private;
3830 	struct rte_flow_action *translated = NULL;
3831 	size_t actions_size;
3832 	int n;
3833 	int copied_n = 0;
3834 	struct mlx5_translated_action_handle *handle_end = NULL;
3835 
3836 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3837 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3838 			continue;
3839 		if (copied_n == *indir_n) {
3840 			return rte_flow_error_set
3841 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3842 				 NULL, "too many shared actions");
3843 		}
3844 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3845 			   sizeof(actions[n].conf));
3846 		handle[copied_n].index = n;
3847 		copied_n++;
3848 	}
3849 	n++;
3850 	*indir_n = copied_n;
3851 	if (!copied_n)
3852 		return 0;
3853 	actions_size = sizeof(struct rte_flow_action) * n;
3854 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3855 	if (!translated) {
3856 		rte_errno = ENOMEM;
3857 		return -ENOMEM;
3858 	}
3859 	memcpy(translated, actions, actions_size);
3860 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3861 		struct mlx5_shared_action_rss *shared_rss;
3862 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3863 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3864 		uint32_t idx = act_idx &
3865 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3866 
3867 		switch (type) {
3868 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3869 			shared_rss = mlx5_ipool_get
3870 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3871 			translated[handle->index].type =
3872 				RTE_FLOW_ACTION_TYPE_RSS;
3873 			translated[handle->index].conf =
3874 				&shared_rss->origin;
3875 			break;
3876 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3877 			translated[handle->index].type =
3878 						(enum rte_flow_action_type)
3879 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3880 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3881 			break;
3882 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3883 			if (priv->sh->flow_hit_aso_en) {
3884 				translated[handle->index].type =
3885 					(enum rte_flow_action_type)
3886 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3887 				translated[handle->index].conf =
3888 							 (void *)(uintptr_t)idx;
3889 				break;
3890 			}
3891 			/* Fall-through */
3892 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3893 			if (priv->sh->ct_aso_en) {
3894 				translated[handle->index].type =
3895 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3896 				translated[handle->index].conf =
3897 							 (void *)(uintptr_t)idx;
3898 				break;
3899 			}
3900 			/* Fall-through */
3901 		default:
3902 			mlx5_free(translated);
3903 			return rte_flow_error_set
3904 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3905 				 NULL, "invalid indirect action type");
3906 		}
3907 	}
3908 	*translated_actions = translated;
3909 	return 0;
3910 }
3911 
3912 /**
3913  * Get Shared RSS action from the action list.
3914  *
3915  * @param[in] dev
3916  *   Pointer to Ethernet device.
3917  * @param[in] shared
3918  *   Pointer to the list of actions.
3919  * @param[in] shared_n
3920  *   Actions list length.
3921  *
3922  * @return
3923  *   The MLX5 RSS action ID if exists, otherwise return 0.
3924  */
3925 static uint32_t
3926 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3927 			   struct mlx5_translated_action_handle *handle,
3928 			   int shared_n)
3929 {
3930 	struct mlx5_translated_action_handle *handle_end;
3931 	struct mlx5_priv *priv = dev->data->dev_private;
3932 	struct mlx5_shared_action_rss *shared_rss;
3933 
3934 
3935 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3936 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3937 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3938 		uint32_t idx = act_idx &
3939 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3940 		switch (type) {
3941 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3942 			shared_rss = mlx5_ipool_get
3943 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3944 									   idx);
3945 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3946 					   __ATOMIC_RELAXED);
3947 			return idx;
3948 		default:
3949 			break;
3950 		}
3951 	}
3952 	return 0;
3953 }
3954 
3955 static unsigned int
3956 find_graph_root(uint32_t rss_level)
3957 {
3958 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3959 			       MLX5_EXPANSION_ROOT_OUTER;
3960 }
3961 
3962 /**
3963  *  Get layer flags from the prefix flow.
3964  *
3965  *  Some flows may be split to several subflows, the prefix subflow gets the
3966  *  match items and the suffix sub flow gets the actions.
3967  *  Some actions need the user defined match item flags to get the detail for
3968  *  the action.
3969  *  This function helps the suffix flow to get the item layer flags from prefix
3970  *  subflow.
3971  *
3972  * @param[in] dev_flow
3973  *   Pointer the created prefix subflow.
3974  *
3975  * @return
3976  *   The layers get from prefix subflow.
3977  */
3978 static inline uint64_t
3979 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3980 {
3981 	uint64_t layers = 0;
3982 
3983 	/*
3984 	 * Layers bits could be localization, but usually the compiler will
3985 	 * help to do the optimization work for source code.
3986 	 * If no decap actions, use the layers directly.
3987 	 */
3988 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3989 		return dev_flow->handle->layers;
3990 	/* Convert L3 layers with decap action. */
3991 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3992 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3993 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3994 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3995 	/* Convert L4 layers with decap action.  */
3996 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3997 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3998 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3999 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
4000 	return layers;
4001 }
4002 
4003 /**
4004  * Get metadata split action information.
4005  *
4006  * @param[in] actions
4007  *   Pointer to the list of actions.
4008  * @param[out] qrss
4009  *   Pointer to the return pointer.
4010  * @param[out] qrss_type
4011  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
4012  *   if no QUEUE/RSS is found.
4013  * @param[out] encap_idx
4014  *   Pointer to the index of the encap action if exists, otherwise the last
4015  *   action index.
4016  *
4017  * @return
4018  *   Total number of actions.
4019  */
4020 static int
4021 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4022 				       const struct rte_flow_action **qrss,
4023 				       int *encap_idx)
4024 {
4025 	const struct rte_flow_action_raw_encap *raw_encap;
4026 	int actions_n = 0;
4027 	int raw_decap_idx = -1;
4028 
4029 	*encap_idx = -1;
4030 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4031 		switch (actions->type) {
4032 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4033 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4034 			*encap_idx = actions_n;
4035 			break;
4036 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4037 			raw_decap_idx = actions_n;
4038 			break;
4039 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4040 			raw_encap = actions->conf;
4041 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4042 				*encap_idx = raw_decap_idx != -1 ?
4043 						      raw_decap_idx : actions_n;
4044 			break;
4045 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4046 		case RTE_FLOW_ACTION_TYPE_RSS:
4047 			*qrss = actions;
4048 			break;
4049 		default:
4050 			break;
4051 		}
4052 		actions_n++;
4053 	}
4054 	if (*encap_idx == -1)
4055 		*encap_idx = actions_n;
4056 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4057 	return actions_n + 1;
4058 }
4059 
4060 /**
4061  * Check if the action will change packet.
4062  *
4063  * @param dev
4064  *   Pointer to Ethernet device.
4065  * @param[in] type
4066  *   action type.
4067  *
4068  * @return
4069  *   true if action will change packet, false otherwise.
4070  */
4071 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4072 					  enum rte_flow_action_type type)
4073 {
4074 	struct mlx5_priv *priv = dev->data->dev_private;
4075 
4076 	switch (type) {
4077 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4078 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4079 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4080 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4081 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4082 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4083 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4084 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4085 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4086 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4087 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4088 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4089 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4090 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4091 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4092 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4093 	case RTE_FLOW_ACTION_TYPE_SET_META:
4094 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4095 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4096 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4097 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4098 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4099 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4100 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4101 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4102 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4103 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4104 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4105 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4106 		return true;
4107 	case RTE_FLOW_ACTION_TYPE_FLAG:
4108 	case RTE_FLOW_ACTION_TYPE_MARK:
4109 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4110 			return true;
4111 		else
4112 			return false;
4113 	default:
4114 		return false;
4115 	}
4116 }
4117 
4118 /**
4119  * Check meter action from the action list.
4120  *
4121  * @param dev
4122  *   Pointer to Ethernet device.
4123  * @param[in] actions
4124  *   Pointer to the list of actions.
4125  * @param[out] has_mtr
4126  *   Pointer to the meter exist flag.
4127  * @param[out] has_modify
4128  *   Pointer to the flag showing there's packet change action.
4129  * @param[out] meter_id
4130  *   Pointer to the meter id.
4131  *
4132  * @return
4133  *   Total number of actions.
4134  */
4135 static int
4136 flow_check_meter_action(struct rte_eth_dev *dev,
4137 			const struct rte_flow_action actions[],
4138 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4139 {
4140 	const struct rte_flow_action_meter *mtr = NULL;
4141 	int actions_n = 0;
4142 
4143 	MLX5_ASSERT(has_mtr);
4144 	*has_mtr = false;
4145 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4146 		switch (actions->type) {
4147 		case RTE_FLOW_ACTION_TYPE_METER:
4148 			mtr = actions->conf;
4149 			*meter_id = mtr->mtr_id;
4150 			*has_mtr = true;
4151 			break;
4152 		default:
4153 			break;
4154 		}
4155 		if (!*has_mtr)
4156 			*has_modify |= flow_check_modify_action_type(dev,
4157 								actions->type);
4158 		actions_n++;
4159 	}
4160 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4161 	return actions_n + 1;
4162 }
4163 
4164 /**
4165  * Check if the flow should be split due to hairpin.
4166  * The reason for the split is that in current HW we can't
4167  * support encap and push-vlan on Rx, so if a flow contains
4168  * these actions we move it to Tx.
4169  *
4170  * @param dev
4171  *   Pointer to Ethernet device.
4172  * @param[in] attr
4173  *   Flow rule attributes.
4174  * @param[in] actions
4175  *   Associated actions (list terminated by the END action).
4176  *
4177  * @return
4178  *   > 0 the number of actions and the flow should be split,
4179  *   0 when no split required.
4180  */
4181 static int
4182 flow_check_hairpin_split(struct rte_eth_dev *dev,
4183 			 const struct rte_flow_attr *attr,
4184 			 const struct rte_flow_action actions[])
4185 {
4186 	int queue_action = 0;
4187 	int action_n = 0;
4188 	int split = 0;
4189 	const struct rte_flow_action_queue *queue;
4190 	const struct rte_flow_action_rss *rss;
4191 	const struct rte_flow_action_raw_encap *raw_encap;
4192 	const struct rte_eth_hairpin_conf *conf;
4193 
4194 	if (!attr->ingress)
4195 		return 0;
4196 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4197 		switch (actions->type) {
4198 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4199 			queue = actions->conf;
4200 			if (queue == NULL)
4201 				return 0;
4202 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4203 			if (conf == NULL || conf->tx_explicit != 0)
4204 				return 0;
4205 			queue_action = 1;
4206 			action_n++;
4207 			break;
4208 		case RTE_FLOW_ACTION_TYPE_RSS:
4209 			rss = actions->conf;
4210 			if (rss == NULL || rss->queue_num == 0)
4211 				return 0;
4212 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4213 			if (conf == NULL || conf->tx_explicit != 0)
4214 				return 0;
4215 			queue_action = 1;
4216 			action_n++;
4217 			break;
4218 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4219 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4220 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4221 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4222 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4223 			split++;
4224 			action_n++;
4225 			break;
4226 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4227 			raw_encap = actions->conf;
4228 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4229 				split++;
4230 			action_n++;
4231 			break;
4232 		default:
4233 			action_n++;
4234 			break;
4235 		}
4236 	}
4237 	if (split && queue_action)
4238 		return action_n;
4239 	return 0;
4240 }
4241 
4242 /* Declare flow create/destroy prototype in advance. */
4243 static uint32_t
4244 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4245 		 const struct rte_flow_attr *attr,
4246 		 const struct rte_flow_item items[],
4247 		 const struct rte_flow_action actions[],
4248 		 bool external, struct rte_flow_error *error);
4249 
4250 static void
4251 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4252 		  uint32_t flow_idx);
4253 
4254 int
4255 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4256 		      struct mlx5_list_entry *entry, void *cb_ctx)
4257 {
4258 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4259 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4260 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4261 
4262 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4263 }
4264 
4265 struct mlx5_list_entry *
4266 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4267 {
4268 	struct rte_eth_dev *dev = tool_ctx;
4269 	struct mlx5_priv *priv = dev->data->dev_private;
4270 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4271 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4272 	struct rte_flow_error *error = ctx->error;
4273 	uint32_t idx = 0;
4274 	int ret;
4275 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4276 	struct rte_flow_attr attr = {
4277 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4278 		.ingress = 1,
4279 	};
4280 	struct mlx5_rte_flow_item_tag tag_spec = {
4281 		.data = mark_id,
4282 	};
4283 	struct rte_flow_item items[] = {
4284 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4285 	};
4286 	struct rte_flow_action_mark ftag = {
4287 		.id = mark_id,
4288 	};
4289 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4290 		.dst = REG_B,
4291 		.src = REG_NON,
4292 	};
4293 	struct rte_flow_action_jump jump = {
4294 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4295 	};
4296 	struct rte_flow_action actions[] = {
4297 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4298 	};
4299 
4300 	/* Fill the register fields in the flow. */
4301 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4302 	if (ret < 0)
4303 		return NULL;
4304 	tag_spec.id = ret;
4305 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4306 	if (ret < 0)
4307 		return NULL;
4308 	cp_mreg.src = ret;
4309 	/* Provide the full width of FLAG specific value. */
4310 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4311 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4312 	/* Build a new flow. */
4313 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4314 		items[0] = (struct rte_flow_item){
4315 			.type = (enum rte_flow_item_type)
4316 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4317 			.spec = &tag_spec,
4318 		};
4319 		items[1] = (struct rte_flow_item){
4320 			.type = RTE_FLOW_ITEM_TYPE_END,
4321 		};
4322 		actions[0] = (struct rte_flow_action){
4323 			.type = (enum rte_flow_action_type)
4324 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4325 			.conf = &ftag,
4326 		};
4327 		actions[1] = (struct rte_flow_action){
4328 			.type = (enum rte_flow_action_type)
4329 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4330 			.conf = &cp_mreg,
4331 		};
4332 		actions[2] = (struct rte_flow_action){
4333 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4334 			.conf = &jump,
4335 		};
4336 		actions[3] = (struct rte_flow_action){
4337 			.type = RTE_FLOW_ACTION_TYPE_END,
4338 		};
4339 	} else {
4340 		/* Default rule, wildcard match. */
4341 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4342 		items[0] = (struct rte_flow_item){
4343 			.type = RTE_FLOW_ITEM_TYPE_END,
4344 		};
4345 		actions[0] = (struct rte_flow_action){
4346 			.type = (enum rte_flow_action_type)
4347 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4348 			.conf = &cp_mreg,
4349 		};
4350 		actions[1] = (struct rte_flow_action){
4351 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4352 			.conf = &jump,
4353 		};
4354 		actions[2] = (struct rte_flow_action){
4355 			.type = RTE_FLOW_ACTION_TYPE_END,
4356 		};
4357 	}
4358 	/* Build a new entry. */
4359 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4360 	if (!mcp_res) {
4361 		rte_errno = ENOMEM;
4362 		return NULL;
4363 	}
4364 	mcp_res->idx = idx;
4365 	mcp_res->mark_id = mark_id;
4366 	/*
4367 	 * The copy Flows are not included in any list. There
4368 	 * ones are referenced from other Flows and can not
4369 	 * be applied, removed, deleted in arbitrary order
4370 	 * by list traversing.
4371 	 */
4372 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4373 					&attr, items, actions, false, error);
4374 	if (!mcp_res->rix_flow) {
4375 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4376 		return NULL;
4377 	}
4378 	return &mcp_res->hlist_ent;
4379 }
4380 
4381 struct mlx5_list_entry *
4382 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4383 		      void *cb_ctx __rte_unused)
4384 {
4385 	struct rte_eth_dev *dev = tool_ctx;
4386 	struct mlx5_priv *priv = dev->data->dev_private;
4387 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4388 	uint32_t idx = 0;
4389 
4390 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4391 	if (!mcp_res) {
4392 		rte_errno = ENOMEM;
4393 		return NULL;
4394 	}
4395 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4396 	mcp_res->idx = idx;
4397 	return &mcp_res->hlist_ent;
4398 }
4399 
4400 void
4401 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4402 {
4403 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4404 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4405 	struct rte_eth_dev *dev = tool_ctx;
4406 	struct mlx5_priv *priv = dev->data->dev_private;
4407 
4408 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4409 }
4410 
4411 /**
4412  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4413  *
4414  * As mark_id is unique, if there's already a registered flow for the mark_id,
4415  * return by increasing the reference counter of the resource. Otherwise, create
4416  * the resource (mcp_res) and flow.
4417  *
4418  * Flow looks like,
4419  *   - If ingress port is ANY and reg_c[1] is mark_id,
4420  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4421  *
4422  * For default flow (zero mark_id), flow is like,
4423  *   - If ingress port is ANY,
4424  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4425  *
4426  * @param dev
4427  *   Pointer to Ethernet device.
4428  * @param mark_id
4429  *   ID of MARK action, zero means default flow for META.
4430  * @param[out] error
4431  *   Perform verbose error reporting if not NULL.
4432  *
4433  * @return
4434  *   Associated resource on success, NULL otherwise and rte_errno is set.
4435  */
4436 static struct mlx5_flow_mreg_copy_resource *
4437 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4438 			  struct rte_flow_error *error)
4439 {
4440 	struct mlx5_priv *priv = dev->data->dev_private;
4441 	struct mlx5_list_entry *entry;
4442 	struct mlx5_flow_cb_ctx ctx = {
4443 		.dev = dev,
4444 		.error = error,
4445 		.data = &mark_id,
4446 	};
4447 
4448 	/* Check if already registered. */
4449 	MLX5_ASSERT(priv->mreg_cp_tbl);
4450 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4451 	if (!entry)
4452 		return NULL;
4453 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4454 			    hlist_ent);
4455 }
4456 
4457 void
4458 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4459 {
4460 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4461 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4462 	struct rte_eth_dev *dev = tool_ctx;
4463 	struct mlx5_priv *priv = dev->data->dev_private;
4464 
4465 	MLX5_ASSERT(mcp_res->rix_flow);
4466 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4467 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4468 }
4469 
4470 /**
4471  * Release flow in RX_CP_TBL.
4472  *
4473  * @param dev
4474  *   Pointer to Ethernet device.
4475  * @flow
4476  *   Parent flow for wich copying is provided.
4477  */
4478 static void
4479 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4480 			  struct rte_flow *flow)
4481 {
4482 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4483 	struct mlx5_priv *priv = dev->data->dev_private;
4484 
4485 	if (!flow->rix_mreg_copy)
4486 		return;
4487 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4488 				 flow->rix_mreg_copy);
4489 	if (!mcp_res || !priv->mreg_cp_tbl)
4490 		return;
4491 	MLX5_ASSERT(mcp_res->rix_flow);
4492 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4493 	flow->rix_mreg_copy = 0;
4494 }
4495 
4496 /**
4497  * Remove the default copy action from RX_CP_TBL.
4498  *
4499  * This functions is called in the mlx5_dev_start(). No thread safe
4500  * is guaranteed.
4501  *
4502  * @param dev
4503  *   Pointer to Ethernet device.
4504  */
4505 static void
4506 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4507 {
4508 	struct mlx5_list_entry *entry;
4509 	struct mlx5_priv *priv = dev->data->dev_private;
4510 	struct mlx5_flow_cb_ctx ctx;
4511 	uint32_t mark_id;
4512 
4513 	/* Check if default flow is registered. */
4514 	if (!priv->mreg_cp_tbl)
4515 		return;
4516 	mark_id = MLX5_DEFAULT_COPY_ID;
4517 	ctx.data = &mark_id;
4518 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4519 	if (!entry)
4520 		return;
4521 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4522 }
4523 
4524 /**
4525  * Add the default copy action in in RX_CP_TBL.
4526  *
4527  * This functions is called in the mlx5_dev_start(). No thread safe
4528  * is guaranteed.
4529  *
4530  * @param dev
4531  *   Pointer to Ethernet device.
4532  * @param[out] error
4533  *   Perform verbose error reporting if not NULL.
4534  *
4535  * @return
4536  *   0 for success, negative value otherwise and rte_errno is set.
4537  */
4538 static int
4539 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4540 				  struct rte_flow_error *error)
4541 {
4542 	struct mlx5_priv *priv = dev->data->dev_private;
4543 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4544 	struct mlx5_flow_cb_ctx ctx;
4545 	uint32_t mark_id;
4546 
4547 	/* Check whether extensive metadata feature is engaged. */
4548 	if (!priv->config.dv_flow_en ||
4549 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4550 	    !mlx5_flow_ext_mreg_supported(dev) ||
4551 	    !priv->sh->dv_regc0_mask)
4552 		return 0;
4553 	/*
4554 	 * Add default mreg copy flow may be called multiple time, but
4555 	 * only be called once in stop. Avoid register it twice.
4556 	 */
4557 	mark_id = MLX5_DEFAULT_COPY_ID;
4558 	ctx.data = &mark_id;
4559 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4560 		return 0;
4561 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4562 	if (!mcp_res)
4563 		return -rte_errno;
4564 	return 0;
4565 }
4566 
4567 /**
4568  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4569  *
4570  * All the flow having Q/RSS action should be split by
4571  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4572  * performs the following,
4573  *   - CQE->flow_tag := reg_c[1] (MARK)
4574  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4575  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4576  * but there should be a flow per each MARK ID set by MARK action.
4577  *
4578  * For the aforementioned reason, if there's a MARK action in flow's action
4579  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4580  * the MARK ID to CQE's flow_tag like,
4581  *   - If reg_c[1] is mark_id,
4582  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4583  *
4584  * For SET_META action which stores value in reg_c[0], as the destination is
4585  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4586  * MARK ID means the default flow. The default flow looks like,
4587  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4588  *
4589  * @param dev
4590  *   Pointer to Ethernet device.
4591  * @param flow
4592  *   Pointer to flow structure.
4593  * @param[in] actions
4594  *   Pointer to the list of actions.
4595  * @param[out] error
4596  *   Perform verbose error reporting if not NULL.
4597  *
4598  * @return
4599  *   0 on success, negative value otherwise and rte_errno is set.
4600  */
4601 static int
4602 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4603 			    struct rte_flow *flow,
4604 			    const struct rte_flow_action *actions,
4605 			    struct rte_flow_error *error)
4606 {
4607 	struct mlx5_priv *priv = dev->data->dev_private;
4608 	struct mlx5_dev_config *config = &priv->config;
4609 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4610 	const struct rte_flow_action_mark *mark;
4611 
4612 	/* Check whether extensive metadata feature is engaged. */
4613 	if (!config->dv_flow_en ||
4614 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4615 	    !mlx5_flow_ext_mreg_supported(dev) ||
4616 	    !priv->sh->dv_regc0_mask)
4617 		return 0;
4618 	/* Find MARK action. */
4619 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4620 		switch (actions->type) {
4621 		case RTE_FLOW_ACTION_TYPE_FLAG:
4622 			mcp_res = flow_mreg_add_copy_action
4623 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4624 			if (!mcp_res)
4625 				return -rte_errno;
4626 			flow->rix_mreg_copy = mcp_res->idx;
4627 			return 0;
4628 		case RTE_FLOW_ACTION_TYPE_MARK:
4629 			mark = (const struct rte_flow_action_mark *)
4630 				actions->conf;
4631 			mcp_res =
4632 				flow_mreg_add_copy_action(dev, mark->id, error);
4633 			if (!mcp_res)
4634 				return -rte_errno;
4635 			flow->rix_mreg_copy = mcp_res->idx;
4636 			return 0;
4637 		default:
4638 			break;
4639 		}
4640 	}
4641 	return 0;
4642 }
4643 
4644 #define MLX5_MAX_SPLIT_ACTIONS 24
4645 #define MLX5_MAX_SPLIT_ITEMS 24
4646 
4647 /**
4648  * Split the hairpin flow.
4649  * Since HW can't support encap and push-vlan on Rx, we move these
4650  * actions to Tx.
4651  * If the count action is after the encap then we also
4652  * move the count action. in this case the count will also measure
4653  * the outer bytes.
4654  *
4655  * @param dev
4656  *   Pointer to Ethernet device.
4657  * @param[in] actions
4658  *   Associated actions (list terminated by the END action).
4659  * @param[out] actions_rx
4660  *   Rx flow actions.
4661  * @param[out] actions_tx
4662  *   Tx flow actions..
4663  * @param[out] pattern_tx
4664  *   The pattern items for the Tx flow.
4665  * @param[out] flow_id
4666  *   The flow ID connected to this flow.
4667  *
4668  * @return
4669  *   0 on success.
4670  */
4671 static int
4672 flow_hairpin_split(struct rte_eth_dev *dev,
4673 		   const struct rte_flow_action actions[],
4674 		   struct rte_flow_action actions_rx[],
4675 		   struct rte_flow_action actions_tx[],
4676 		   struct rte_flow_item pattern_tx[],
4677 		   uint32_t flow_id)
4678 {
4679 	const struct rte_flow_action_raw_encap *raw_encap;
4680 	const struct rte_flow_action_raw_decap *raw_decap;
4681 	struct mlx5_rte_flow_action_set_tag *set_tag;
4682 	struct rte_flow_action *tag_action;
4683 	struct mlx5_rte_flow_item_tag *tag_item;
4684 	struct rte_flow_item *item;
4685 	char *addr;
4686 	int encap = 0;
4687 
4688 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4689 		switch (actions->type) {
4690 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4691 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4692 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4693 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4694 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4695 			rte_memcpy(actions_tx, actions,
4696 			       sizeof(struct rte_flow_action));
4697 			actions_tx++;
4698 			break;
4699 		case RTE_FLOW_ACTION_TYPE_COUNT:
4700 			if (encap) {
4701 				rte_memcpy(actions_tx, actions,
4702 					   sizeof(struct rte_flow_action));
4703 				actions_tx++;
4704 			} else {
4705 				rte_memcpy(actions_rx, actions,
4706 					   sizeof(struct rte_flow_action));
4707 				actions_rx++;
4708 			}
4709 			break;
4710 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4711 			raw_encap = actions->conf;
4712 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4713 				memcpy(actions_tx, actions,
4714 				       sizeof(struct rte_flow_action));
4715 				actions_tx++;
4716 				encap = 1;
4717 			} else {
4718 				rte_memcpy(actions_rx, actions,
4719 					   sizeof(struct rte_flow_action));
4720 				actions_rx++;
4721 			}
4722 			break;
4723 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4724 			raw_decap = actions->conf;
4725 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4726 				memcpy(actions_tx, actions,
4727 				       sizeof(struct rte_flow_action));
4728 				actions_tx++;
4729 			} else {
4730 				rte_memcpy(actions_rx, actions,
4731 					   sizeof(struct rte_flow_action));
4732 				actions_rx++;
4733 			}
4734 			break;
4735 		default:
4736 			rte_memcpy(actions_rx, actions,
4737 				   sizeof(struct rte_flow_action));
4738 			actions_rx++;
4739 			break;
4740 		}
4741 	}
4742 	/* Add set meta action and end action for the Rx flow. */
4743 	tag_action = actions_rx;
4744 	tag_action->type = (enum rte_flow_action_type)
4745 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4746 	actions_rx++;
4747 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4748 	actions_rx++;
4749 	set_tag = (void *)actions_rx;
4750 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4751 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4752 		.data = flow_id,
4753 	};
4754 	MLX5_ASSERT(set_tag->id > REG_NON);
4755 	tag_action->conf = set_tag;
4756 	/* Create Tx item list. */
4757 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4758 	addr = (void *)&pattern_tx[2];
4759 	item = pattern_tx;
4760 	item->type = (enum rte_flow_item_type)
4761 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4762 	tag_item = (void *)addr;
4763 	tag_item->data = flow_id;
4764 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4765 	MLX5_ASSERT(set_tag->id > REG_NON);
4766 	item->spec = tag_item;
4767 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4768 	tag_item = (void *)addr;
4769 	tag_item->data = UINT32_MAX;
4770 	tag_item->id = UINT16_MAX;
4771 	item->mask = tag_item;
4772 	item->last = NULL;
4773 	item++;
4774 	item->type = RTE_FLOW_ITEM_TYPE_END;
4775 	return 0;
4776 }
4777 
4778 /**
4779  * The last stage of splitting chain, just creates the subflow
4780  * without any modification.
4781  *
4782  * @param[in] dev
4783  *   Pointer to Ethernet device.
4784  * @param[in] flow
4785  *   Parent flow structure pointer.
4786  * @param[in, out] sub_flow
4787  *   Pointer to return the created subflow, may be NULL.
4788  * @param[in] attr
4789  *   Flow rule attributes.
4790  * @param[in] items
4791  *   Pattern specification (list terminated by the END pattern item).
4792  * @param[in] actions
4793  *   Associated actions (list terminated by the END action).
4794  * @param[in] flow_split_info
4795  *   Pointer to flow split info structure.
4796  * @param[out] error
4797  *   Perform verbose error reporting if not NULL.
4798  * @return
4799  *   0 on success, negative value otherwise
4800  */
4801 static int
4802 flow_create_split_inner(struct rte_eth_dev *dev,
4803 			struct rte_flow *flow,
4804 			struct mlx5_flow **sub_flow,
4805 			const struct rte_flow_attr *attr,
4806 			const struct rte_flow_item items[],
4807 			const struct rte_flow_action actions[],
4808 			struct mlx5_flow_split_info *flow_split_info,
4809 			struct rte_flow_error *error)
4810 {
4811 	struct mlx5_flow *dev_flow;
4812 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
4813 
4814 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4815 				    flow_split_info->flow_idx, error);
4816 	if (!dev_flow)
4817 		return -rte_errno;
4818 	dev_flow->flow = flow;
4819 	dev_flow->external = flow_split_info->external;
4820 	dev_flow->skip_scale = flow_split_info->skip_scale;
4821 	/* Subflow object was created, we must include one in the list. */
4822 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4823 		      dev_flow->handle, next);
4824 	/*
4825 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4826 	 * flow may need some user defined item layer flags, and pass the
4827 	 * Metadata rxq mark flag to suffix flow as well.
4828 	 */
4829 	if (flow_split_info->prefix_layers)
4830 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4831 	if (flow_split_info->prefix_mark) {
4832 		MLX5_ASSERT(wks);
4833 		wks->mark = 1;
4834 	}
4835 	if (sub_flow)
4836 		*sub_flow = dev_flow;
4837 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4838 	dev_flow->dv.table_id = flow_split_info->table_id;
4839 #endif
4840 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4841 }
4842 
4843 /**
4844  * Get the sub policy of a meter.
4845  *
4846  * @param[in] dev
4847  *   Pointer to Ethernet device.
4848  * @param[in] flow
4849  *   Parent flow structure pointer.
4850  * @param wks
4851  *   Pointer to thread flow work space.
4852  * @param[in] attr
4853  *   Flow rule attributes.
4854  * @param[in] items
4855  *   Pattern specification (list terminated by the END pattern item).
4856  * @param[out] error
4857  *   Perform verbose error reporting if not NULL.
4858  *
4859  * @return
4860  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4861  */
4862 static struct mlx5_flow_meter_sub_policy *
4863 get_meter_sub_policy(struct rte_eth_dev *dev,
4864 		     struct rte_flow *flow,
4865 		     struct mlx5_flow_workspace *wks,
4866 		     const struct rte_flow_attr *attr,
4867 		     const struct rte_flow_item items[],
4868 		     struct rte_flow_error *error)
4869 {
4870 	struct mlx5_flow_meter_policy *policy;
4871 	struct mlx5_flow_meter_policy *final_policy;
4872 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4873 
4874 	policy = wks->policy;
4875 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4876 	if (final_policy->is_rss || final_policy->is_queue) {
4877 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4878 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4879 		uint32_t i;
4880 
4881 		/*
4882 		 * This is a tmp dev_flow,
4883 		 * no need to register any matcher for it in translate.
4884 		 */
4885 		wks->skip_matcher_reg = 1;
4886 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4887 			struct mlx5_flow dev_flow = {0};
4888 			struct mlx5_flow_handle dev_handle = { {0} };
4889 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4890 
4891 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4892 				const struct rte_flow_action_rss *rss_act =
4893 					final_policy->act_cnt[i].rss->conf;
4894 				struct rte_flow_action rss_actions[2] = {
4895 					[0] = {
4896 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4897 					.conf = rss_act,
4898 					},
4899 					[1] = {
4900 					.type = RTE_FLOW_ACTION_TYPE_END,
4901 					.conf = NULL,
4902 					}
4903 				};
4904 
4905 				dev_flow.handle = &dev_handle;
4906 				dev_flow.ingress = attr->ingress;
4907 				dev_flow.flow = flow;
4908 				dev_flow.external = 0;
4909 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4910 				dev_flow.dv.transfer = attr->transfer;
4911 #endif
4912 				/**
4913 				 * Translate RSS action to get rss hash fields.
4914 				 */
4915 				if (flow_drv_translate(dev, &dev_flow, attr,
4916 						items, rss_actions, error))
4917 					goto exit;
4918 				rss_desc_v[i] = wks->rss_desc;
4919 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4920 				rss_desc_v[i].hash_fields =
4921 						dev_flow.hash_fields;
4922 				rss_desc_v[i].queue_num =
4923 						rss_desc_v[i].hash_fields ?
4924 						rss_desc_v[i].queue_num : 1;
4925 				rss_desc_v[i].tunnel =
4926 						!!(dev_flow.handle->layers &
4927 						   MLX5_FLOW_LAYER_TUNNEL);
4928 				/* Use the RSS queues in the containers. */
4929 				rss_desc_v[i].queue =
4930 					(uint16_t *)(uintptr_t)rss_act->queue;
4931 				rss_desc[i] = &rss_desc_v[i];
4932 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4933 				/* This is queue action. */
4934 				rss_desc_v[i] = wks->rss_desc;
4935 				rss_desc_v[i].key_len = 0;
4936 				rss_desc_v[i].hash_fields = 0;
4937 				rss_desc_v[i].queue =
4938 					&final_policy->act_cnt[i].queue;
4939 				rss_desc_v[i].queue_num = 1;
4940 				rss_desc[i] = &rss_desc_v[i];
4941 			} else {
4942 				rss_desc[i] = NULL;
4943 			}
4944 		}
4945 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4946 						flow, policy, rss_desc);
4947 	} else {
4948 		enum mlx5_meter_domain mtr_domain =
4949 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4950 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4951 						MLX5_MTR_DOMAIN_INGRESS);
4952 		sub_policy = policy->sub_policys[mtr_domain][0];
4953 	}
4954 	if (!sub_policy)
4955 		rte_flow_error_set(error, EINVAL,
4956 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4957 				   "Failed to get meter sub-policy.");
4958 exit:
4959 	return sub_policy;
4960 }
4961 
4962 /**
4963  * Split the meter flow.
4964  *
4965  * As meter flow will split to three sub flow, other than meter
4966  * action, the other actions make sense to only meter accepts
4967  * the packet. If it need to be dropped, no other additional
4968  * actions should be take.
4969  *
4970  * One kind of special action which decapsulates the L3 tunnel
4971  * header will be in the prefix sub flow, as not to take the
4972  * L3 tunnel header into account.
4973  *
4974  * @param[in] dev
4975  *   Pointer to Ethernet device.
4976  * @param[in] flow
4977  *   Parent flow structure pointer.
4978  * @param wks
4979  *   Pointer to thread flow work space.
4980  * @param[in] attr
4981  *   Flow rule attributes.
4982  * @param[in] items
4983  *   Pattern specification (list terminated by the END pattern item).
4984  * @param[out] sfx_items
4985  *   Suffix flow match items (list terminated by the END pattern item).
4986  * @param[in] actions
4987  *   Associated actions (list terminated by the END action).
4988  * @param[out] actions_sfx
4989  *   Suffix flow actions.
4990  * @param[out] actions_pre
4991  *   Prefix flow actions.
4992  * @param[out] mtr_flow_id
4993  *   Pointer to meter flow id.
4994  * @param[out] error
4995  *   Perform verbose error reporting if not NULL.
4996  *
4997  * @return
4998  *   0 on success, a negative errno value otherwise and rte_errno is set.
4999  */
5000 static int
5001 flow_meter_split_prep(struct rte_eth_dev *dev,
5002 		      struct rte_flow *flow,
5003 		      struct mlx5_flow_workspace *wks,
5004 		      const struct rte_flow_attr *attr,
5005 		      const struct rte_flow_item items[],
5006 		      struct rte_flow_item sfx_items[],
5007 		      const struct rte_flow_action actions[],
5008 		      struct rte_flow_action actions_sfx[],
5009 		      struct rte_flow_action actions_pre[],
5010 		      uint32_t *mtr_flow_id,
5011 		      struct rte_flow_error *error)
5012 {
5013 	struct mlx5_priv *priv = dev->data->dev_private;
5014 	struct mlx5_flow_meter_info *fm = wks->fm;
5015 	struct rte_flow_action *tag_action = NULL;
5016 	struct rte_flow_item *tag_item;
5017 	struct mlx5_rte_flow_action_set_tag *set_tag;
5018 	const struct rte_flow_action_raw_encap *raw_encap;
5019 	const struct rte_flow_action_raw_decap *raw_decap;
5020 	struct mlx5_rte_flow_item_tag *tag_item_spec;
5021 	struct mlx5_rte_flow_item_tag *tag_item_mask;
5022 	uint32_t tag_id = 0;
5023 	struct rte_flow_item *vlan_item_dst = NULL;
5024 	const struct rte_flow_item *vlan_item_src = NULL;
5025 	struct rte_flow_action *hw_mtr_action;
5026 	struct rte_flow_action *action_pre_head = NULL;
5027 	int32_t flow_src_port = priv->representor_id;
5028 	bool mtr_first;
5029 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5030 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5031 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5032 	uint32_t flow_id = 0;
5033 	uint32_t flow_id_reversed = 0;
5034 	uint8_t flow_id_bits = 0;
5035 	int shift;
5036 
5037 	/* Prepare the suffix subflow items. */
5038 	tag_item = sfx_items++;
5039 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5040 		struct mlx5_priv *port_priv;
5041 		const struct rte_flow_item_port_id *pid_v;
5042 		int item_type = items->type;
5043 
5044 		switch (item_type) {
5045 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5046 			pid_v = items->spec;
5047 			MLX5_ASSERT(pid_v);
5048 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5049 			if (!port_priv)
5050 				return rte_flow_error_set(error,
5051 						rte_errno,
5052 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5053 						pid_v,
5054 						"Failed to get port info.");
5055 			flow_src_port = port_priv->representor_id;
5056 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5057 			    flow_src_port != priv->representor_id) {
5058 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5059 								flow, fm,
5060 								flow_src_port,
5061 								items,
5062 								error))
5063 					return -rte_errno;
5064 			}
5065 			memcpy(sfx_items, items, sizeof(*sfx_items));
5066 			sfx_items++;
5067 			break;
5068 		case RTE_FLOW_ITEM_TYPE_VLAN:
5069 			/* Determine if copy vlan item below. */
5070 			vlan_item_src = items;
5071 			vlan_item_dst = sfx_items++;
5072 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5073 			break;
5074 		default:
5075 			break;
5076 		}
5077 	}
5078 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5079 	sfx_items++;
5080 	mtr_first = priv->sh->meter_aso_en &&
5081 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5082 	/* For ASO meter, meter must be before tag in TX direction. */
5083 	if (mtr_first) {
5084 		action_pre_head = actions_pre++;
5085 		/* Leave space for tag action. */
5086 		tag_action = actions_pre++;
5087 	}
5088 	/* Prepare the actions for prefix and suffix flow. */
5089 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5090 		struct rte_flow_action *action_cur = NULL;
5091 
5092 		switch (actions->type) {
5093 		case RTE_FLOW_ACTION_TYPE_METER:
5094 			if (mtr_first) {
5095 				action_cur = action_pre_head;
5096 			} else {
5097 				/* Leave space for tag action. */
5098 				tag_action = actions_pre++;
5099 				action_cur = actions_pre++;
5100 			}
5101 			break;
5102 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5103 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5104 			action_cur = actions_pre++;
5105 			break;
5106 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5107 			raw_encap = actions->conf;
5108 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5109 				action_cur = actions_pre++;
5110 			break;
5111 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5112 			raw_decap = actions->conf;
5113 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5114 				action_cur = actions_pre++;
5115 			break;
5116 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5117 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5118 			if (vlan_item_dst && vlan_item_src) {
5119 				memcpy(vlan_item_dst, vlan_item_src,
5120 					sizeof(*vlan_item_dst));
5121 				/*
5122 				 * Convert to internal match item, it is used
5123 				 * for vlan push and set vid.
5124 				 */
5125 				vlan_item_dst->type = (enum rte_flow_item_type)
5126 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5127 			}
5128 			break;
5129 		default:
5130 			break;
5131 		}
5132 		if (!action_cur)
5133 			action_cur = (fm->def_policy) ?
5134 					actions_sfx++ : actions_pre++;
5135 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5136 	}
5137 	/* Add end action to the actions. */
5138 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5139 	if (priv->sh->meter_aso_en) {
5140 		/**
5141 		 * For ASO meter, need to add an extra jump action explicitly,
5142 		 * to jump from meter to policer table.
5143 		 */
5144 		struct mlx5_flow_meter_sub_policy *sub_policy;
5145 		struct mlx5_flow_tbl_data_entry *tbl_data;
5146 
5147 		if (!fm->def_policy) {
5148 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5149 							  attr, items, error);
5150 			if (!sub_policy)
5151 				return -rte_errno;
5152 		} else {
5153 			enum mlx5_meter_domain mtr_domain =
5154 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5155 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5156 						MLX5_MTR_DOMAIN_INGRESS);
5157 
5158 			sub_policy =
5159 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5160 		}
5161 		tbl_data = container_of(sub_policy->tbl_rsc,
5162 					struct mlx5_flow_tbl_data_entry, tbl);
5163 		hw_mtr_action = actions_pre++;
5164 		hw_mtr_action->type = (enum rte_flow_action_type)
5165 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5166 		hw_mtr_action->conf = tbl_data->jump.action;
5167 	}
5168 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5169 	actions_pre++;
5170 	if (!tag_action)
5171 		return rte_flow_error_set(error, ENOMEM,
5172 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5173 					  NULL, "No tag action space.");
5174 	if (!mtr_flow_id) {
5175 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5176 		goto exit;
5177 	}
5178 	/* Only default-policy Meter creates mtr flow id. */
5179 	if (fm->def_policy) {
5180 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5181 		if (!tag_id)
5182 			return rte_flow_error_set(error, ENOMEM,
5183 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5184 					"Failed to allocate meter flow id.");
5185 		flow_id = tag_id - 1;
5186 		flow_id_bits = (!flow_id) ? 1 :
5187 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5188 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5189 		    mtr_reg_bits) {
5190 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5191 			return rte_flow_error_set(error, EINVAL,
5192 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5193 					"Meter flow id exceeds max limit.");
5194 		}
5195 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5196 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5197 	}
5198 	/* Build tag actions and items for meter_id/meter flow_id. */
5199 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5200 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5201 	tag_item_mask = tag_item_spec + 1;
5202 	/* Both flow_id and meter_id share the same register. */
5203 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5204 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5205 							    0, error),
5206 		.offset = mtr_id_offset,
5207 		.length = mtr_reg_bits,
5208 		.data = flow->meter,
5209 	};
5210 	/*
5211 	 * The color Reg bits used by flow_id are growing from
5212 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5213 	 */
5214 	for (shift = 0; shift < flow_id_bits; shift++)
5215 		flow_id_reversed = (flow_id_reversed << 1) |
5216 				((flow_id >> shift) & 0x1);
5217 	set_tag->data |=
5218 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5219 	tag_item_spec->id = set_tag->id;
5220 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5221 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5222 	tag_action->type = (enum rte_flow_action_type)
5223 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5224 	tag_action->conf = set_tag;
5225 	tag_item->type = (enum rte_flow_item_type)
5226 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5227 	tag_item->spec = tag_item_spec;
5228 	tag_item->last = NULL;
5229 	tag_item->mask = tag_item_mask;
5230 exit:
5231 	if (mtr_flow_id)
5232 		*mtr_flow_id = tag_id;
5233 	return 0;
5234 }
5235 
5236 /**
5237  * Split action list having QUEUE/RSS for metadata register copy.
5238  *
5239  * Once Q/RSS action is detected in user's action list, the flow action
5240  * should be split in order to copy metadata registers, which will happen in
5241  * RX_CP_TBL like,
5242  *   - CQE->flow_tag := reg_c[1] (MARK)
5243  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5244  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5245  * This is because the last action of each flow must be a terminal action
5246  * (QUEUE, RSS or DROP).
5247  *
5248  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5249  * stored and kept in the mlx5_flow structure per each sub_flow.
5250  *
5251  * The Q/RSS action is replaced with,
5252  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5253  * And the following JUMP action is added at the end,
5254  *   - JUMP, to RX_CP_TBL.
5255  *
5256  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5257  * flow_create_split_metadata() routine. The flow will look like,
5258  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5259  *
5260  * @param dev
5261  *   Pointer to Ethernet device.
5262  * @param[out] split_actions
5263  *   Pointer to store split actions to jump to CP_TBL.
5264  * @param[in] actions
5265  *   Pointer to the list of original flow actions.
5266  * @param[in] qrss
5267  *   Pointer to the Q/RSS action.
5268  * @param[in] actions_n
5269  *   Number of original actions.
5270  * @param[in] mtr_sfx
5271  *   Check if it is in meter suffix table.
5272  * @param[out] error
5273  *   Perform verbose error reporting if not NULL.
5274  *
5275  * @return
5276  *   non-zero unique flow_id on success, otherwise 0 and
5277  *   error/rte_error are set.
5278  */
5279 static uint32_t
5280 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5281 			  struct rte_flow_action *split_actions,
5282 			  const struct rte_flow_action *actions,
5283 			  const struct rte_flow_action *qrss,
5284 			  int actions_n, int mtr_sfx,
5285 			  struct rte_flow_error *error)
5286 {
5287 	struct mlx5_priv *priv = dev->data->dev_private;
5288 	struct mlx5_rte_flow_action_set_tag *set_tag;
5289 	struct rte_flow_action_jump *jump;
5290 	const int qrss_idx = qrss - actions;
5291 	uint32_t flow_id = 0;
5292 	int ret = 0;
5293 
5294 	/*
5295 	 * Given actions will be split
5296 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5297 	 * - Add jump to mreg CP_TBL.
5298 	 * As a result, there will be one more action.
5299 	 */
5300 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5301 	/* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5302 	++actions_n;
5303 	set_tag = (void *)(split_actions + actions_n);
5304 	/*
5305 	 * If we are not the meter suffix flow, add the tag action.
5306 	 * Since meter suffix flow already has the tag added.
5307 	 */
5308 	if (!mtr_sfx) {
5309 		/*
5310 		 * Allocate the new subflow ID. This one is unique within
5311 		 * device and not shared with representors. Otherwise,
5312 		 * we would have to resolve multi-thread access synch
5313 		 * issue. Each flow on the shared device is appended
5314 		 * with source vport identifier, so the resulting
5315 		 * flows will be unique in the shared (by master and
5316 		 * representors) domain even if they have coinciding
5317 		 * IDs.
5318 		 */
5319 		mlx5_ipool_malloc(priv->sh->ipool
5320 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5321 		if (!flow_id)
5322 			return rte_flow_error_set(error, ENOMEM,
5323 						  RTE_FLOW_ERROR_TYPE_ACTION,
5324 						  NULL, "can't allocate id "
5325 						  "for split Q/RSS subflow");
5326 		/* Internal SET_TAG action to set flow ID. */
5327 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5328 			.data = flow_id,
5329 		};
5330 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5331 		if (ret < 0)
5332 			return ret;
5333 		set_tag->id = ret;
5334 		/* Construct new actions array. */
5335 		/* Replace QUEUE/RSS action. */
5336 		split_actions[qrss_idx] = (struct rte_flow_action){
5337 			.type = (enum rte_flow_action_type)
5338 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5339 			.conf = set_tag,
5340 		};
5341 	} else {
5342 		/*
5343 		 * If we are the suffix flow of meter, tag already exist.
5344 		 * Set the QUEUE/RSS action to void.
5345 		 */
5346 		split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5347 	}
5348 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5349 	jump = (void *)(set_tag + 1);
5350 	*jump = (struct rte_flow_action_jump){
5351 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5352 	};
5353 	split_actions[actions_n - 2] = (struct rte_flow_action){
5354 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5355 		.conf = jump,
5356 	};
5357 	split_actions[actions_n - 1] = (struct rte_flow_action){
5358 		.type = RTE_FLOW_ACTION_TYPE_END,
5359 	};
5360 	return flow_id;
5361 }
5362 
5363 /**
5364  * Extend the given action list for Tx metadata copy.
5365  *
5366  * Copy the given action list to the ext_actions and add flow metadata register
5367  * copy action in order to copy reg_a set by WQE to reg_c[0].
5368  *
5369  * @param[out] ext_actions
5370  *   Pointer to the extended action list.
5371  * @param[in] actions
5372  *   Pointer to the list of actions.
5373  * @param[in] actions_n
5374  *   Number of actions in the list.
5375  * @param[out] error
5376  *   Perform verbose error reporting if not NULL.
5377  * @param[in] encap_idx
5378  *   The encap action index.
5379  *
5380  * @return
5381  *   0 on success, negative value otherwise
5382  */
5383 static int
5384 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5385 		       struct rte_flow_action *ext_actions,
5386 		       const struct rte_flow_action *actions,
5387 		       int actions_n, struct rte_flow_error *error,
5388 		       int encap_idx)
5389 {
5390 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5391 		(struct mlx5_flow_action_copy_mreg *)
5392 			(ext_actions + actions_n + 1);
5393 	int ret;
5394 
5395 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5396 	if (ret < 0)
5397 		return ret;
5398 	cp_mreg->dst = ret;
5399 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5400 	if (ret < 0)
5401 		return ret;
5402 	cp_mreg->src = ret;
5403 	if (encap_idx != 0)
5404 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5405 	if (encap_idx == actions_n - 1) {
5406 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5407 			.type = (enum rte_flow_action_type)
5408 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5409 			.conf = cp_mreg,
5410 		};
5411 		ext_actions[actions_n] = (struct rte_flow_action){
5412 			.type = RTE_FLOW_ACTION_TYPE_END,
5413 		};
5414 	} else {
5415 		ext_actions[encap_idx] = (struct rte_flow_action){
5416 			.type = (enum rte_flow_action_type)
5417 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5418 			.conf = cp_mreg,
5419 		};
5420 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5421 				sizeof(*ext_actions) * (actions_n - encap_idx));
5422 	}
5423 	return 0;
5424 }
5425 
5426 /**
5427  * Check the match action from the action list.
5428  *
5429  * @param[in] actions
5430  *   Pointer to the list of actions.
5431  * @param[in] attr
5432  *   Flow rule attributes.
5433  * @param[in] action
5434  *   The action to be check if exist.
5435  * @param[out] match_action_pos
5436  *   Pointer to the position of the matched action if exists, otherwise is -1.
5437  * @param[out] qrss_action_pos
5438  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5439  * @param[out] modify_after_mirror
5440  *   Pointer to the flag of modify action after FDB mirroring.
5441  *
5442  * @return
5443  *   > 0 the total number of actions.
5444  *   0 if not found match action in action list.
5445  */
5446 static int
5447 flow_check_match_action(const struct rte_flow_action actions[],
5448 			const struct rte_flow_attr *attr,
5449 			enum rte_flow_action_type action,
5450 			int *match_action_pos, int *qrss_action_pos,
5451 			int *modify_after_mirror)
5452 {
5453 	const struct rte_flow_action_sample *sample;
5454 	const struct rte_flow_action_raw_decap *decap;
5455 	int actions_n = 0;
5456 	uint32_t ratio = 0;
5457 	int sub_type = 0;
5458 	int flag = 0;
5459 	int fdb_mirror = 0;
5460 
5461 	*match_action_pos = -1;
5462 	*qrss_action_pos = -1;
5463 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5464 		if (actions->type == action) {
5465 			flag = 1;
5466 			*match_action_pos = actions_n;
5467 		}
5468 		switch (actions->type) {
5469 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5470 		case RTE_FLOW_ACTION_TYPE_RSS:
5471 			*qrss_action_pos = actions_n;
5472 			break;
5473 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5474 			sample = actions->conf;
5475 			ratio = sample->ratio;
5476 			sub_type = ((const struct rte_flow_action *)
5477 					(sample->actions))->type;
5478 			if (ratio == 1 && attr->transfer)
5479 				fdb_mirror = 1;
5480 			break;
5481 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5482 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5483 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5484 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5485 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5486 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5487 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5488 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5489 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5490 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5491 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5492 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5493 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5494 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5495 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5496 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5497 		case RTE_FLOW_ACTION_TYPE_FLAG:
5498 		case RTE_FLOW_ACTION_TYPE_MARK:
5499 		case RTE_FLOW_ACTION_TYPE_SET_META:
5500 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5501 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5502 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5503 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5504 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5505 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5506 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5507 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5508 		case RTE_FLOW_ACTION_TYPE_METER:
5509 			if (fdb_mirror)
5510 				*modify_after_mirror = 1;
5511 			break;
5512 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5513 			decap = actions->conf;
5514 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5515 				;
5516 			actions_n++;
5517 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5518 				const struct rte_flow_action_raw_encap *encap =
5519 								actions->conf;
5520 				if (decap->size <=
5521 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5522 				    encap->size >
5523 					MLX5_ENCAPSULATION_DECISION_SIZE)
5524 					/* L3 encap. */
5525 					break;
5526 			}
5527 			if (fdb_mirror)
5528 				*modify_after_mirror = 1;
5529 			break;
5530 		default:
5531 			break;
5532 		}
5533 		actions_n++;
5534 	}
5535 	if (flag && fdb_mirror && !*modify_after_mirror) {
5536 		/* FDB mirroring uses the destination array to implement
5537 		 * instead of FLOW_SAMPLER object.
5538 		 */
5539 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5540 			flag = 0;
5541 	}
5542 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5543 	return flag ? actions_n + 1 : 0;
5544 }
5545 
5546 #define SAMPLE_SUFFIX_ITEM 2
5547 
5548 /**
5549  * Split the sample flow.
5550  *
5551  * As sample flow will split to two sub flow, sample flow with
5552  * sample action, the other actions will move to new suffix flow.
5553  *
5554  * Also add unique tag id with tag action in the sample flow,
5555  * the same tag id will be as match in the suffix flow.
5556  *
5557  * @param dev
5558  *   Pointer to Ethernet device.
5559  * @param[in] add_tag
5560  *   Add extra tag action flag.
5561  * @param[out] sfx_items
5562  *   Suffix flow match items (list terminated by the END pattern item).
5563  * @param[in] actions
5564  *   Associated actions (list terminated by the END action).
5565  * @param[out] actions_sfx
5566  *   Suffix flow actions.
5567  * @param[out] actions_pre
5568  *   Prefix flow actions.
5569  * @param[in] actions_n
5570  *  The total number of actions.
5571  * @param[in] sample_action_pos
5572  *   The sample action position.
5573  * @param[in] qrss_action_pos
5574  *   The Queue/RSS action position.
5575  * @param[in] jump_table
5576  *   Add extra jump action flag.
5577  * @param[out] error
5578  *   Perform verbose error reporting if not NULL.
5579  *
5580  * @return
5581  *   0 on success, or unique flow_id, a negative errno value
5582  *   otherwise and rte_errno is set.
5583  */
5584 static int
5585 flow_sample_split_prep(struct rte_eth_dev *dev,
5586 		       int add_tag,
5587 		       struct rte_flow_item sfx_items[],
5588 		       const struct rte_flow_action actions[],
5589 		       struct rte_flow_action actions_sfx[],
5590 		       struct rte_flow_action actions_pre[],
5591 		       int actions_n,
5592 		       int sample_action_pos,
5593 		       int qrss_action_pos,
5594 		       int jump_table,
5595 		       struct rte_flow_error *error)
5596 {
5597 	struct mlx5_priv *priv = dev->data->dev_private;
5598 	struct mlx5_rte_flow_action_set_tag *set_tag;
5599 	struct mlx5_rte_flow_item_tag *tag_spec;
5600 	struct mlx5_rte_flow_item_tag *tag_mask;
5601 	struct rte_flow_action_jump *jump_action;
5602 	uint32_t tag_id = 0;
5603 	int index;
5604 	int append_index = 0;
5605 	int ret;
5606 
5607 	if (sample_action_pos < 0)
5608 		return rte_flow_error_set(error, EINVAL,
5609 					  RTE_FLOW_ERROR_TYPE_ACTION,
5610 					  NULL, "invalid position of sample "
5611 					  "action in list");
5612 	/* Prepare the actions for prefix and suffix flow. */
5613 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5614 		index = qrss_action_pos;
5615 		/* Put the preceding the Queue/RSS action into prefix flow. */
5616 		if (index != 0)
5617 			memcpy(actions_pre, actions,
5618 			       sizeof(struct rte_flow_action) * index);
5619 		/* Put others preceding the sample action into prefix flow. */
5620 		if (sample_action_pos > index + 1)
5621 			memcpy(actions_pre + index, actions + index + 1,
5622 			       sizeof(struct rte_flow_action) *
5623 			       (sample_action_pos - index - 1));
5624 		index = sample_action_pos - 1;
5625 		/* Put Queue/RSS action into Suffix flow. */
5626 		memcpy(actions_sfx, actions + qrss_action_pos,
5627 		       sizeof(struct rte_flow_action));
5628 		actions_sfx++;
5629 	} else {
5630 		index = sample_action_pos;
5631 		if (index != 0)
5632 			memcpy(actions_pre, actions,
5633 			       sizeof(struct rte_flow_action) * index);
5634 	}
5635 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5636 	 * For CX6DX and above, metadata registers Cx preserve their value,
5637 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5638 	 */
5639 	if (add_tag) {
5640 		/* Prepare the prefix tag action. */
5641 		append_index++;
5642 		set_tag = (void *)(actions_pre + actions_n + append_index);
5643 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5644 		if (ret < 0)
5645 			return ret;
5646 		mlx5_ipool_malloc(priv->sh->ipool
5647 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5648 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5649 			.id = ret,
5650 			.data = tag_id,
5651 		};
5652 		/* Prepare the suffix subflow items. */
5653 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5654 		tag_spec->data = tag_id;
5655 		tag_spec->id = set_tag->id;
5656 		tag_mask = tag_spec + 1;
5657 		tag_mask->data = UINT32_MAX;
5658 		sfx_items[0] = (struct rte_flow_item){
5659 			.type = (enum rte_flow_item_type)
5660 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5661 			.spec = tag_spec,
5662 			.last = NULL,
5663 			.mask = tag_mask,
5664 		};
5665 		sfx_items[1] = (struct rte_flow_item){
5666 			.type = (enum rte_flow_item_type)
5667 				RTE_FLOW_ITEM_TYPE_END,
5668 		};
5669 		/* Prepare the tag action in prefix subflow. */
5670 		actions_pre[index++] =
5671 			(struct rte_flow_action){
5672 			.type = (enum rte_flow_action_type)
5673 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5674 			.conf = set_tag,
5675 		};
5676 	}
5677 	memcpy(actions_pre + index, actions + sample_action_pos,
5678 	       sizeof(struct rte_flow_action));
5679 	index += 1;
5680 	/* For the modify action after the sample action in E-Switch mirroring,
5681 	 * Add the extra jump action in prefix subflow and jump into the next
5682 	 * table, then do the modify action in the new table.
5683 	 */
5684 	if (jump_table) {
5685 		/* Prepare the prefix jump action. */
5686 		append_index++;
5687 		jump_action = (void *)(actions_pre + actions_n + append_index);
5688 		jump_action->group = jump_table;
5689 		actions_pre[index++] =
5690 			(struct rte_flow_action){
5691 			.type = (enum rte_flow_action_type)
5692 				RTE_FLOW_ACTION_TYPE_JUMP,
5693 			.conf = jump_action,
5694 		};
5695 	}
5696 	actions_pre[index] = (struct rte_flow_action){
5697 		.type = (enum rte_flow_action_type)
5698 			RTE_FLOW_ACTION_TYPE_END,
5699 	};
5700 	/* Put the actions after sample into Suffix flow. */
5701 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5702 	       sizeof(struct rte_flow_action) *
5703 	       (actions_n - sample_action_pos - 1));
5704 	return tag_id;
5705 }
5706 
5707 /**
5708  * The splitting for metadata feature.
5709  *
5710  * - Q/RSS action on NIC Rx should be split in order to pass by
5711  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5712  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5713  *
5714  * - All the actions on NIC Tx should have a mreg copy action to
5715  *   copy reg_a from WQE to reg_c[0].
5716  *
5717  * @param dev
5718  *   Pointer to Ethernet device.
5719  * @param[in] flow
5720  *   Parent flow structure pointer.
5721  * @param[in] attr
5722  *   Flow rule attributes.
5723  * @param[in] items
5724  *   Pattern specification (list terminated by the END pattern item).
5725  * @param[in] actions
5726  *   Associated actions (list terminated by the END action).
5727  * @param[in] flow_split_info
5728  *   Pointer to flow split info structure.
5729  * @param[out] error
5730  *   Perform verbose error reporting if not NULL.
5731  * @return
5732  *   0 on success, negative value otherwise
5733  */
5734 static int
5735 flow_create_split_metadata(struct rte_eth_dev *dev,
5736 			   struct rte_flow *flow,
5737 			   const struct rte_flow_attr *attr,
5738 			   const struct rte_flow_item items[],
5739 			   const struct rte_flow_action actions[],
5740 			   struct mlx5_flow_split_info *flow_split_info,
5741 			   struct rte_flow_error *error)
5742 {
5743 	struct mlx5_priv *priv = dev->data->dev_private;
5744 	struct mlx5_dev_config *config = &priv->config;
5745 	const struct rte_flow_action *qrss = NULL;
5746 	struct rte_flow_action *ext_actions = NULL;
5747 	struct mlx5_flow *dev_flow = NULL;
5748 	uint32_t qrss_id = 0;
5749 	int mtr_sfx = 0;
5750 	size_t act_size;
5751 	int actions_n;
5752 	int encap_idx;
5753 	int ret;
5754 
5755 	/* Check whether extensive metadata feature is engaged. */
5756 	if (!config->dv_flow_en ||
5757 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5758 	    !mlx5_flow_ext_mreg_supported(dev))
5759 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5760 					       actions, flow_split_info, error);
5761 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5762 							   &encap_idx);
5763 	if (qrss) {
5764 		/* Exclude hairpin flows from splitting. */
5765 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5766 			const struct rte_flow_action_queue *queue;
5767 
5768 			queue = qrss->conf;
5769 			if (mlx5_rxq_get_type(dev, queue->index) ==
5770 			    MLX5_RXQ_TYPE_HAIRPIN)
5771 				qrss = NULL;
5772 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5773 			const struct rte_flow_action_rss *rss;
5774 
5775 			rss = qrss->conf;
5776 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5777 			    MLX5_RXQ_TYPE_HAIRPIN)
5778 				qrss = NULL;
5779 		}
5780 	}
5781 	if (qrss) {
5782 		/* Check if it is in meter suffix table. */
5783 		mtr_sfx = attr->group == (attr->transfer ?
5784 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5785 			  MLX5_FLOW_TABLE_LEVEL_METER);
5786 		/*
5787 		 * Q/RSS action on NIC Rx should be split in order to pass by
5788 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5789 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5790 		 */
5791 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5792 			   sizeof(struct rte_flow_action_set_tag) +
5793 			   sizeof(struct rte_flow_action_jump);
5794 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5795 					  SOCKET_ID_ANY);
5796 		if (!ext_actions)
5797 			return rte_flow_error_set(error, ENOMEM,
5798 						  RTE_FLOW_ERROR_TYPE_ACTION,
5799 						  NULL, "no memory to split "
5800 						  "metadata flow");
5801 		/*
5802 		 * Create the new actions list with removed Q/RSS action
5803 		 * and appended set tag and jump to register copy table
5804 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5805 		 * in advance, because it is needed for set tag action.
5806 		 */
5807 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5808 						    qrss, actions_n,
5809 						    mtr_sfx, error);
5810 		if (!mtr_sfx && !qrss_id) {
5811 			ret = -rte_errno;
5812 			goto exit;
5813 		}
5814 	} else if (attr->egress && !attr->transfer) {
5815 		/*
5816 		 * All the actions on NIC Tx should have a metadata register
5817 		 * copy action to copy reg_a from WQE to reg_c[meta]
5818 		 */
5819 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5820 			   sizeof(struct mlx5_flow_action_copy_mreg);
5821 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5822 					  SOCKET_ID_ANY);
5823 		if (!ext_actions)
5824 			return rte_flow_error_set(error, ENOMEM,
5825 						  RTE_FLOW_ERROR_TYPE_ACTION,
5826 						  NULL, "no memory to split "
5827 						  "metadata flow");
5828 		/* Create the action list appended with copy register. */
5829 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5830 					     actions_n, error, encap_idx);
5831 		if (ret < 0)
5832 			goto exit;
5833 	}
5834 	/* Add the unmodified original or prefix subflow. */
5835 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5836 				      items, ext_actions ? ext_actions :
5837 				      actions, flow_split_info, error);
5838 	if (ret < 0)
5839 		goto exit;
5840 	MLX5_ASSERT(dev_flow);
5841 	if (qrss) {
5842 		const struct rte_flow_attr q_attr = {
5843 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5844 			.ingress = 1,
5845 		};
5846 		/* Internal PMD action to set register. */
5847 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5848 			.data = qrss_id,
5849 			.id = REG_NON,
5850 		};
5851 		struct rte_flow_item q_items[] = {
5852 			{
5853 				.type = (enum rte_flow_item_type)
5854 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5855 				.spec = &q_tag_spec,
5856 				.last = NULL,
5857 				.mask = NULL,
5858 			},
5859 			{
5860 				.type = RTE_FLOW_ITEM_TYPE_END,
5861 			},
5862 		};
5863 		struct rte_flow_action q_actions[] = {
5864 			{
5865 				.type = qrss->type,
5866 				.conf = qrss->conf,
5867 			},
5868 			{
5869 				.type = RTE_FLOW_ACTION_TYPE_END,
5870 			},
5871 		};
5872 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5873 
5874 		/*
5875 		 * Configure the tag item only if there is no meter subflow.
5876 		 * Since tag is already marked in the meter suffix subflow
5877 		 * we can just use the meter suffix items as is.
5878 		 */
5879 		if (qrss_id) {
5880 			/* Not meter subflow. */
5881 			MLX5_ASSERT(!mtr_sfx);
5882 			/*
5883 			 * Put unique id in prefix flow due to it is destroyed
5884 			 * after suffix flow and id will be freed after there
5885 			 * is no actual flows with this id and identifier
5886 			 * reallocation becomes possible (for example, for
5887 			 * other flows in other threads).
5888 			 */
5889 			dev_flow->handle->split_flow_id = qrss_id;
5890 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5891 						   error);
5892 			if (ret < 0)
5893 				goto exit;
5894 			q_tag_spec.id = ret;
5895 		}
5896 		dev_flow = NULL;
5897 		/* Add suffix subflow to execute Q/RSS. */
5898 		flow_split_info->prefix_layers = layers;
5899 		flow_split_info->prefix_mark = 0;
5900 		flow_split_info->table_id = 0;
5901 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5902 					      &q_attr, mtr_sfx ? items :
5903 					      q_items, q_actions,
5904 					      flow_split_info, error);
5905 		if (ret < 0)
5906 			goto exit;
5907 		/* qrss ID should be freed if failed. */
5908 		qrss_id = 0;
5909 		MLX5_ASSERT(dev_flow);
5910 	}
5911 
5912 exit:
5913 	/*
5914 	 * We do not destroy the partially created sub_flows in case of error.
5915 	 * These ones are included into parent flow list and will be destroyed
5916 	 * by flow_drv_destroy.
5917 	 */
5918 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5919 			qrss_id);
5920 	mlx5_free(ext_actions);
5921 	return ret;
5922 }
5923 
5924 /**
5925  * Create meter internal drop flow with the original pattern.
5926  *
5927  * @param dev
5928  *   Pointer to Ethernet device.
5929  * @param[in] flow
5930  *   Parent flow structure pointer.
5931  * @param[in] attr
5932  *   Flow rule attributes.
5933  * @param[in] items
5934  *   Pattern specification (list terminated by the END pattern item).
5935  * @param[in] flow_split_info
5936  *   Pointer to flow split info structure.
5937  * @param[in] fm
5938  *   Pointer to flow meter structure.
5939  * @param[out] error
5940  *   Perform verbose error reporting if not NULL.
5941  * @return
5942  *   0 on success, negative value otherwise
5943  */
5944 static uint32_t
5945 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5946 			struct rte_flow *flow,
5947 			const struct rte_flow_attr *attr,
5948 			const struct rte_flow_item items[],
5949 			struct mlx5_flow_split_info *flow_split_info,
5950 			struct mlx5_flow_meter_info *fm,
5951 			struct rte_flow_error *error)
5952 {
5953 	struct mlx5_flow *dev_flow = NULL;
5954 	struct rte_flow_attr drop_attr = *attr;
5955 	struct rte_flow_action drop_actions[3];
5956 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5957 
5958 	MLX5_ASSERT(fm->drop_cnt);
5959 	drop_actions[0].type =
5960 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5961 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5962 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5963 	drop_actions[1].conf = NULL;
5964 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5965 	drop_actions[2].conf = NULL;
5966 	drop_split_info.external = false;
5967 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5968 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5969 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5970 	return flow_create_split_inner(dev, flow, &dev_flow,
5971 				&drop_attr, items, drop_actions,
5972 				&drop_split_info, error);
5973 }
5974 
5975 /**
5976  * The splitting for meter feature.
5977  *
5978  * - The meter flow will be split to two flows as prefix and
5979  *   suffix flow. The packets make sense only it pass the prefix
5980  *   meter action.
5981  *
5982  * - Reg_C_5 is used for the packet to match betweend prefix and
5983  *   suffix flow.
5984  *
5985  * @param dev
5986  *   Pointer to Ethernet device.
5987  * @param[in] flow
5988  *   Parent flow structure pointer.
5989  * @param[in] attr
5990  *   Flow rule attributes.
5991  * @param[in] items
5992  *   Pattern specification (list terminated by the END pattern item).
5993  * @param[in] actions
5994  *   Associated actions (list terminated by the END action).
5995  * @param[in] flow_split_info
5996  *   Pointer to flow split info structure.
5997  * @param[out] error
5998  *   Perform verbose error reporting if not NULL.
5999  * @return
6000  *   0 on success, negative value otherwise
6001  */
6002 static int
6003 flow_create_split_meter(struct rte_eth_dev *dev,
6004 			struct rte_flow *flow,
6005 			const struct rte_flow_attr *attr,
6006 			const struct rte_flow_item items[],
6007 			const struct rte_flow_action actions[],
6008 			struct mlx5_flow_split_info *flow_split_info,
6009 			struct rte_flow_error *error)
6010 {
6011 	struct mlx5_priv *priv = dev->data->dev_private;
6012 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6013 	struct rte_flow_action *sfx_actions = NULL;
6014 	struct rte_flow_action *pre_actions = NULL;
6015 	struct rte_flow_item *sfx_items = NULL;
6016 	struct mlx5_flow *dev_flow = NULL;
6017 	struct rte_flow_attr sfx_attr = *attr;
6018 	struct mlx5_flow_meter_info *fm = NULL;
6019 	uint8_t skip_scale_restore;
6020 	bool has_mtr = false;
6021 	bool has_modify = false;
6022 	bool set_mtr_reg = true;
6023 	bool is_mtr_hierarchy = false;
6024 	uint32_t meter_id = 0;
6025 	uint32_t mtr_idx = 0;
6026 	uint32_t mtr_flow_id = 0;
6027 	size_t act_size;
6028 	size_t item_size;
6029 	int actions_n = 0;
6030 	int ret = 0;
6031 
6032 	if (priv->mtr_en)
6033 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6034 						    &has_modify, &meter_id);
6035 	if (has_mtr) {
6036 		if (flow->meter) {
6037 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6038 			if (!fm)
6039 				return rte_flow_error_set(error, EINVAL,
6040 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6041 						NULL, "Meter not found.");
6042 		} else {
6043 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6044 			if (!fm)
6045 				return rte_flow_error_set(error, EINVAL,
6046 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6047 						NULL, "Meter not found.");
6048 			ret = mlx5_flow_meter_attach(priv, fm,
6049 						     &sfx_attr, error);
6050 			if (ret)
6051 				return -rte_errno;
6052 			flow->meter = mtr_idx;
6053 		}
6054 		MLX5_ASSERT(wks);
6055 		wks->fm = fm;
6056 		if (!fm->def_policy) {
6057 			wks->policy = mlx5_flow_meter_policy_find(dev,
6058 								  fm->policy_id,
6059 								  NULL);
6060 			MLX5_ASSERT(wks->policy);
6061 			if (wks->policy->is_hierarchy) {
6062 				wks->final_policy =
6063 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6064 								wks->policy);
6065 				if (!wks->final_policy)
6066 					return rte_flow_error_set(error,
6067 					EINVAL,
6068 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6069 				"Failed to find terminal policy of hierarchy.");
6070 				is_mtr_hierarchy = true;
6071 			}
6072 		}
6073 		/*
6074 		 * If it isn't default-policy Meter, and
6075 		 * 1. There's no action in flow to change
6076 		 *    packet (modify/encap/decap etc.), OR
6077 		 * 2. No drop count needed for this meter.
6078 		 * 3. It's not meter hierarchy.
6079 		 * Then no need to use regC to save meter id anymore.
6080 		 */
6081 		if (!fm->def_policy && !is_mtr_hierarchy &&
6082 		    (!has_modify || !fm->drop_cnt))
6083 			set_mtr_reg = false;
6084 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6085 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6086 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6087 		/* Suffix items: tag, vlan, port id, end. */
6088 #define METER_SUFFIX_ITEM 4
6089 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6090 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6091 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6092 					  0, SOCKET_ID_ANY);
6093 		if (!sfx_actions)
6094 			return rte_flow_error_set(error, ENOMEM,
6095 						  RTE_FLOW_ERROR_TYPE_ACTION,
6096 						  NULL, "no memory to split "
6097 						  "meter flow");
6098 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6099 			     act_size);
6100 		/* There's no suffix flow for meter of non-default policy. */
6101 		if (!fm->def_policy)
6102 			pre_actions = sfx_actions + 1;
6103 		else
6104 			pre_actions = sfx_actions + actions_n;
6105 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6106 					    items, sfx_items, actions,
6107 					    sfx_actions, pre_actions,
6108 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6109 					    error);
6110 		if (ret) {
6111 			ret = -rte_errno;
6112 			goto exit;
6113 		}
6114 		/* Add the prefix subflow. */
6115 		skip_scale_restore = flow_split_info->skip_scale;
6116 		flow_split_info->skip_scale |=
6117 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6118 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6119 					      attr, items, pre_actions,
6120 					      flow_split_info, error);
6121 		flow_split_info->skip_scale = skip_scale_restore;
6122 		if (ret) {
6123 			if (mtr_flow_id)
6124 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6125 			ret = -rte_errno;
6126 			goto exit;
6127 		}
6128 		if (mtr_flow_id) {
6129 			dev_flow->handle->split_flow_id = mtr_flow_id;
6130 			dev_flow->handle->is_meter_flow_id = 1;
6131 		}
6132 		if (!fm->def_policy) {
6133 			if (!set_mtr_reg && fm->drop_cnt)
6134 				ret =
6135 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6136 							&sfx_attr, items,
6137 							flow_split_info,
6138 							fm, error);
6139 			goto exit;
6140 		}
6141 		/* Setting the sfx group atrr. */
6142 		sfx_attr.group = sfx_attr.transfer ?
6143 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6144 				 MLX5_FLOW_TABLE_LEVEL_METER;
6145 		flow_split_info->prefix_layers =
6146 				flow_get_prefix_layer_flags(dev_flow);
6147 		flow_split_info->prefix_mark |= wks->mark;
6148 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6149 	}
6150 	/* Add the prefix subflow. */
6151 	ret = flow_create_split_metadata(dev, flow,
6152 					 &sfx_attr, sfx_items ?
6153 					 sfx_items : items,
6154 					 sfx_actions ? sfx_actions : actions,
6155 					 flow_split_info, error);
6156 exit:
6157 	if (sfx_actions)
6158 		mlx5_free(sfx_actions);
6159 	return ret;
6160 }
6161 
6162 /**
6163  * The splitting for sample feature.
6164  *
6165  * Once Sample action is detected in the action list, the flow actions should
6166  * be split into prefix sub flow and suffix sub flow.
6167  *
6168  * The original items remain in the prefix sub flow, all actions preceding the
6169  * sample action and the sample action itself will be copied to the prefix
6170  * sub flow, the actions following the sample action will be copied to the
6171  * suffix sub flow, Queue action always be located in the suffix sub flow.
6172  *
6173  * In order to make the packet from prefix sub flow matches with suffix sub
6174  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6175  * flow uses tag item with the unique flow id.
6176  *
6177  * @param dev
6178  *   Pointer to Ethernet device.
6179  * @param[in] flow
6180  *   Parent flow structure pointer.
6181  * @param[in] attr
6182  *   Flow rule attributes.
6183  * @param[in] items
6184  *   Pattern specification (list terminated by the END pattern item).
6185  * @param[in] actions
6186  *   Associated actions (list terminated by the END action).
6187  * @param[in] flow_split_info
6188  *   Pointer to flow split info structure.
6189  * @param[out] error
6190  *   Perform verbose error reporting if not NULL.
6191  * @return
6192  *   0 on success, negative value otherwise
6193  */
6194 static int
6195 flow_create_split_sample(struct rte_eth_dev *dev,
6196 			 struct rte_flow *flow,
6197 			 const struct rte_flow_attr *attr,
6198 			 const struct rte_flow_item items[],
6199 			 const struct rte_flow_action actions[],
6200 			 struct mlx5_flow_split_info *flow_split_info,
6201 			 struct rte_flow_error *error)
6202 {
6203 	struct mlx5_priv *priv = dev->data->dev_private;
6204 	struct rte_flow_action *sfx_actions = NULL;
6205 	struct rte_flow_action *pre_actions = NULL;
6206 	struct rte_flow_item *sfx_items = NULL;
6207 	struct mlx5_flow *dev_flow = NULL;
6208 	struct rte_flow_attr sfx_attr = *attr;
6209 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6210 	struct mlx5_flow_dv_sample_resource *sample_res;
6211 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6212 	struct mlx5_flow_tbl_resource *sfx_tbl;
6213 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6214 #endif
6215 	size_t act_size;
6216 	size_t item_size;
6217 	uint32_t fdb_tx = 0;
6218 	int32_t tag_id = 0;
6219 	int actions_n = 0;
6220 	int sample_action_pos;
6221 	int qrss_action_pos;
6222 	int add_tag = 0;
6223 	int modify_after_mirror = 0;
6224 	uint16_t jump_table = 0;
6225 	const uint32_t next_ft_step = 1;
6226 	int ret = 0;
6227 
6228 	if (priv->sampler_en)
6229 		actions_n = flow_check_match_action(actions, attr,
6230 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6231 					&sample_action_pos, &qrss_action_pos,
6232 					&modify_after_mirror);
6233 	if (actions_n) {
6234 		/* The prefix actions must includes sample, tag, end. */
6235 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6236 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6237 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6238 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6239 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6240 					  item_size), 0, SOCKET_ID_ANY);
6241 		if (!sfx_actions)
6242 			return rte_flow_error_set(error, ENOMEM,
6243 						  RTE_FLOW_ERROR_TYPE_ACTION,
6244 						  NULL, "no memory to split "
6245 						  "sample flow");
6246 		/* The representor_id is UINT16_MAX for uplink. */
6247 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6248 		/*
6249 		 * When reg_c_preserve is set, metadata registers Cx preserve
6250 		 * their value even through packet duplication.
6251 		 */
6252 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6253 		if (add_tag)
6254 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6255 					+ act_size);
6256 		if (modify_after_mirror)
6257 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6258 				     next_ft_step;
6259 		pre_actions = sfx_actions + actions_n;
6260 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6261 						actions, sfx_actions,
6262 						pre_actions, actions_n,
6263 						sample_action_pos,
6264 						qrss_action_pos, jump_table,
6265 						error);
6266 		if (tag_id < 0 || (add_tag && !tag_id)) {
6267 			ret = -rte_errno;
6268 			goto exit;
6269 		}
6270 		if (modify_after_mirror)
6271 			flow_split_info->skip_scale =
6272 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6273 		/* Add the prefix subflow. */
6274 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6275 					      items, pre_actions,
6276 					      flow_split_info, error);
6277 		if (ret) {
6278 			ret = -rte_errno;
6279 			goto exit;
6280 		}
6281 		dev_flow->handle->split_flow_id = tag_id;
6282 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6283 		if (!modify_after_mirror) {
6284 			/* Set the sfx group attr. */
6285 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6286 						dev_flow->dv.sample_res;
6287 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6288 						sample_res->normal_path_tbl;
6289 			sfx_tbl_data = container_of(sfx_tbl,
6290 						struct mlx5_flow_tbl_data_entry,
6291 						tbl);
6292 			sfx_attr.group = sfx_attr.transfer ?
6293 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6294 		} else {
6295 			MLX5_ASSERT(attr->transfer);
6296 			sfx_attr.group = jump_table;
6297 		}
6298 		flow_split_info->prefix_layers =
6299 				flow_get_prefix_layer_flags(dev_flow);
6300 		MLX5_ASSERT(wks);
6301 		flow_split_info->prefix_mark |= wks->mark;
6302 		/* Suffix group level already be scaled with factor, set
6303 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6304 		 * again in translation.
6305 		 */
6306 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6307 #endif
6308 	}
6309 	/* Add the suffix subflow. */
6310 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6311 				      sfx_items ? sfx_items : items,
6312 				      sfx_actions ? sfx_actions : actions,
6313 				      flow_split_info, error);
6314 exit:
6315 	if (sfx_actions)
6316 		mlx5_free(sfx_actions);
6317 	return ret;
6318 }
6319 
6320 /**
6321  * Split the flow to subflow set. The splitters might be linked
6322  * in the chain, like this:
6323  * flow_create_split_outer() calls:
6324  *   flow_create_split_meter() calls:
6325  *     flow_create_split_metadata(meter_subflow_0) calls:
6326  *       flow_create_split_inner(metadata_subflow_0)
6327  *       flow_create_split_inner(metadata_subflow_1)
6328  *       flow_create_split_inner(metadata_subflow_2)
6329  *     flow_create_split_metadata(meter_subflow_1) calls:
6330  *       flow_create_split_inner(metadata_subflow_0)
6331  *       flow_create_split_inner(metadata_subflow_1)
6332  *       flow_create_split_inner(metadata_subflow_2)
6333  *
6334  * This provide flexible way to add new levels of flow splitting.
6335  * The all of successfully created subflows are included to the
6336  * parent flow dev_flow list.
6337  *
6338  * @param dev
6339  *   Pointer to Ethernet device.
6340  * @param[in] flow
6341  *   Parent flow structure pointer.
6342  * @param[in] attr
6343  *   Flow rule attributes.
6344  * @param[in] items
6345  *   Pattern specification (list terminated by the END pattern item).
6346  * @param[in] actions
6347  *   Associated actions (list terminated by the END action).
6348  * @param[in] flow_split_info
6349  *   Pointer to flow split info structure.
6350  * @param[out] error
6351  *   Perform verbose error reporting if not NULL.
6352  * @return
6353  *   0 on success, negative value otherwise
6354  */
6355 static int
6356 flow_create_split_outer(struct rte_eth_dev *dev,
6357 			struct rte_flow *flow,
6358 			const struct rte_flow_attr *attr,
6359 			const struct rte_flow_item items[],
6360 			const struct rte_flow_action actions[],
6361 			struct mlx5_flow_split_info *flow_split_info,
6362 			struct rte_flow_error *error)
6363 {
6364 	int ret;
6365 
6366 	ret = flow_create_split_sample(dev, flow, attr, items,
6367 				       actions, flow_split_info, error);
6368 	MLX5_ASSERT(ret <= 0);
6369 	return ret;
6370 }
6371 
6372 static inline struct mlx5_flow_tunnel *
6373 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6374 {
6375 	struct mlx5_flow_tunnel *tunnel;
6376 
6377 #pragma GCC diagnostic push
6378 #pragma GCC diagnostic ignored "-Wcast-qual"
6379 	tunnel = (typeof(tunnel))flow->tunnel;
6380 #pragma GCC diagnostic pop
6381 
6382 	return tunnel;
6383 }
6384 
6385 /**
6386  * Adjust flow RSS workspace if needed.
6387  *
6388  * @param wks
6389  *   Pointer to thread flow work space.
6390  * @param rss_desc
6391  *   Pointer to RSS descriptor.
6392  * @param[in] nrssq_num
6393  *   New RSS queue number.
6394  *
6395  * @return
6396  *   0 on success, -1 otherwise and rte_errno is set.
6397  */
6398 static int
6399 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6400 			  struct mlx5_flow_rss_desc *rss_desc,
6401 			  uint32_t nrssq_num)
6402 {
6403 	if (likely(nrssq_num <= wks->rssq_num))
6404 		return 0;
6405 	rss_desc->queue = realloc(rss_desc->queue,
6406 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6407 	if (!rss_desc->queue) {
6408 		rte_errno = ENOMEM;
6409 		return -1;
6410 	}
6411 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6412 	return 0;
6413 }
6414 
6415 /**
6416  * Create a flow and add it to @p list.
6417  *
6418  * @param dev
6419  *   Pointer to Ethernet device.
6420  * @param list
6421  *   Pointer to a TAILQ flow list. If this parameter NULL,
6422  *   no list insertion occurred, flow is just created,
6423  *   this is caller's responsibility to track the
6424  *   created flow.
6425  * @param[in] attr
6426  *   Flow rule attributes.
6427  * @param[in] items
6428  *   Pattern specification (list terminated by the END pattern item).
6429  * @param[in] actions
6430  *   Associated actions (list terminated by the END action).
6431  * @param[in] external
6432  *   This flow rule is created by request external to PMD.
6433  * @param[out] error
6434  *   Perform verbose error reporting if not NULL.
6435  *
6436  * @return
6437  *   A flow index on success, 0 otherwise and rte_errno is set.
6438  */
6439 static uint32_t
6440 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6441 		 const struct rte_flow_attr *attr,
6442 		 const struct rte_flow_item items[],
6443 		 const struct rte_flow_action original_actions[],
6444 		 bool external, struct rte_flow_error *error)
6445 {
6446 	struct mlx5_priv *priv = dev->data->dev_private;
6447 	struct rte_flow *flow = NULL;
6448 	struct mlx5_flow *dev_flow;
6449 	const struct rte_flow_action_rss *rss = NULL;
6450 	struct mlx5_translated_action_handle
6451 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6452 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6453 	union {
6454 		struct mlx5_flow_expand_rss buf;
6455 		uint8_t buffer[4096];
6456 	} expand_buffer;
6457 	union {
6458 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6459 		uint8_t buffer[2048];
6460 	} actions_rx;
6461 	union {
6462 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6463 		uint8_t buffer[2048];
6464 	} actions_hairpin_tx;
6465 	union {
6466 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6467 		uint8_t buffer[2048];
6468 	} items_tx;
6469 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6470 	struct mlx5_flow_rss_desc *rss_desc;
6471 	const struct rte_flow_action *p_actions_rx;
6472 	uint32_t i;
6473 	uint32_t idx = 0;
6474 	int hairpin_flow;
6475 	struct rte_flow_attr attr_tx = { .priority = 0 };
6476 	const struct rte_flow_action *actions;
6477 	struct rte_flow_action *translated_actions = NULL;
6478 	struct mlx5_flow_tunnel *tunnel;
6479 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6480 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6481 	struct mlx5_flow_split_info flow_split_info = {
6482 		.external = !!external,
6483 		.skip_scale = 0,
6484 		.flow_idx = 0,
6485 		.prefix_mark = 0,
6486 		.prefix_layers = 0,
6487 		.table_id = 0
6488 	};
6489 	int ret;
6490 
6491 	MLX5_ASSERT(wks);
6492 	rss_desc = &wks->rss_desc;
6493 	ret = flow_action_handles_translate(dev, original_actions,
6494 					    indir_actions,
6495 					    &indir_actions_n,
6496 					    &translated_actions, error);
6497 	if (ret < 0) {
6498 		MLX5_ASSERT(translated_actions == NULL);
6499 		return 0;
6500 	}
6501 	actions = translated_actions ? translated_actions : original_actions;
6502 	p_actions_rx = actions;
6503 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6504 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6505 				external, hairpin_flow, error);
6506 	if (ret < 0)
6507 		goto error_before_hairpin_split;
6508 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6509 	if (!flow) {
6510 		rte_errno = ENOMEM;
6511 		goto error_before_hairpin_split;
6512 	}
6513 	if (hairpin_flow > 0) {
6514 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6515 			rte_errno = EINVAL;
6516 			goto error_before_hairpin_split;
6517 		}
6518 		flow_hairpin_split(dev, actions, actions_rx.actions,
6519 				   actions_hairpin_tx.actions, items_tx.items,
6520 				   idx);
6521 		p_actions_rx = actions_rx.actions;
6522 	}
6523 	flow_split_info.flow_idx = idx;
6524 	flow->drv_type = flow_get_drv_type(dev, attr);
6525 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6526 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6527 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6528 	/* RSS Action only works on NIC RX domain */
6529 	if (attr->ingress && !attr->transfer)
6530 		rss = flow_get_rss_action(dev, p_actions_rx);
6531 	if (rss) {
6532 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6533 			return 0;
6534 		/*
6535 		 * The following information is required by
6536 		 * mlx5_flow_hashfields_adjust() in advance.
6537 		 */
6538 		rss_desc->level = rss->level;
6539 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6540 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6541 	}
6542 	flow->dev_handles = 0;
6543 	if (rss && rss->types) {
6544 		unsigned int graph_root;
6545 
6546 		graph_root = find_graph_root(rss->level);
6547 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6548 					   items, rss->types,
6549 					   mlx5_support_expansion, graph_root);
6550 		MLX5_ASSERT(ret > 0 &&
6551 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6552 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6553 			for (i = 0; i < buf->entries; ++i)
6554 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6555 		}
6556 	} else {
6557 		buf->entries = 1;
6558 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6559 	}
6560 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6561 						      indir_actions_n);
6562 	for (i = 0; i < buf->entries; ++i) {
6563 		/* Initialize flow split data. */
6564 		flow_split_info.prefix_layers = 0;
6565 		flow_split_info.prefix_mark = 0;
6566 		flow_split_info.skip_scale = 0;
6567 		/*
6568 		 * The splitter may create multiple dev_flows,
6569 		 * depending on configuration. In the simplest
6570 		 * case it just creates unmodified original flow.
6571 		 */
6572 		ret = flow_create_split_outer(dev, flow, attr,
6573 					      buf->entry[i].pattern,
6574 					      p_actions_rx, &flow_split_info,
6575 					      error);
6576 		if (ret < 0)
6577 			goto error;
6578 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6579 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6580 							   p_actions_rx,
6581 							   idx,
6582 							   wks->flows[0].tunnel,
6583 							   &default_miss_ctx,
6584 							   error);
6585 			if (ret < 0) {
6586 				mlx5_free(default_miss_ctx.queue);
6587 				goto error;
6588 			}
6589 		}
6590 	}
6591 	/* Create the tx flow. */
6592 	if (hairpin_flow) {
6593 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6594 		attr_tx.ingress = 0;
6595 		attr_tx.egress = 1;
6596 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6597 					 actions_hairpin_tx.actions,
6598 					 idx, error);
6599 		if (!dev_flow)
6600 			goto error;
6601 		dev_flow->flow = flow;
6602 		dev_flow->external = 0;
6603 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6604 			      dev_flow->handle, next);
6605 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6606 					 items_tx.items,
6607 					 actions_hairpin_tx.actions, error);
6608 		if (ret < 0)
6609 			goto error;
6610 	}
6611 	/*
6612 	 * Update the metadata register copy table. If extensive
6613 	 * metadata feature is enabled and registers are supported
6614 	 * we might create the extra rte_flow for each unique
6615 	 * MARK/FLAG action ID.
6616 	 *
6617 	 * The table is updated for ingress Flows only, because
6618 	 * the egress Flows belong to the different device and
6619 	 * copy table should be updated in peer NIC Rx domain.
6620 	 */
6621 	if (attr->ingress &&
6622 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6623 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6624 		if (ret)
6625 			goto error;
6626 	}
6627 	/*
6628 	 * If the flow is external (from application) OR device is started,
6629 	 * OR mreg discover, then apply immediately.
6630 	 */
6631 	if (external || dev->data->dev_started ||
6632 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6633 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6634 		ret = flow_drv_apply(dev, flow, error);
6635 		if (ret < 0)
6636 			goto error;
6637 	}
6638 	flow->type = type;
6639 	flow_rxq_flags_set(dev, flow);
6640 	rte_free(translated_actions);
6641 	tunnel = flow_tunnel_from_rule(wks->flows);
6642 	if (tunnel) {
6643 		flow->tunnel = 1;
6644 		flow->tunnel_id = tunnel->tunnel_id;
6645 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6646 		mlx5_free(default_miss_ctx.queue);
6647 	}
6648 	mlx5_flow_pop_thread_workspace();
6649 	return idx;
6650 error:
6651 	MLX5_ASSERT(flow);
6652 	ret = rte_errno; /* Save rte_errno before cleanup. */
6653 	flow_mreg_del_copy_action(dev, flow);
6654 	flow_drv_destroy(dev, flow);
6655 	if (rss_desc->shared_rss)
6656 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6657 			mlx5_ipool_get
6658 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6659 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6660 	mlx5_ipool_free(priv->flows[type], idx);
6661 	rte_errno = ret; /* Restore rte_errno. */
6662 	ret = rte_errno;
6663 	rte_errno = ret;
6664 	mlx5_flow_pop_thread_workspace();
6665 error_before_hairpin_split:
6666 	rte_free(translated_actions);
6667 	return 0;
6668 }
6669 
6670 /**
6671  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6672  * incoming packets to table 1.
6673  *
6674  * Other flow rules, requested for group n, will be created in
6675  * e-switch table n+1.
6676  * Jump action to e-switch group n will be created to group n+1.
6677  *
6678  * Used when working in switchdev mode, to utilise advantages of table 1
6679  * and above.
6680  *
6681  * @param dev
6682  *   Pointer to Ethernet device.
6683  *
6684  * @return
6685  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6686  */
6687 struct rte_flow *
6688 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6689 {
6690 	const struct rte_flow_attr attr = {
6691 		.group = 0,
6692 		.priority = 0,
6693 		.ingress = 1,
6694 		.egress = 0,
6695 		.transfer = 1,
6696 	};
6697 	const struct rte_flow_item pattern = {
6698 		.type = RTE_FLOW_ITEM_TYPE_END,
6699 	};
6700 	struct rte_flow_action_jump jump = {
6701 		.group = 1,
6702 	};
6703 	const struct rte_flow_action actions[] = {
6704 		{
6705 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6706 			.conf = &jump,
6707 		},
6708 		{
6709 			.type = RTE_FLOW_ACTION_TYPE_END,
6710 		},
6711 	};
6712 	struct rte_flow_error error;
6713 
6714 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6715 						   &attr, &pattern,
6716 						   actions, false, &error);
6717 }
6718 
6719 /**
6720  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6721  * and sq number, directs all packets to peer vport.
6722  *
6723  * @param dev
6724  *   Pointer to Ethernet device.
6725  * @param txq
6726  *   Txq index.
6727  *
6728  * @return
6729  *   Flow ID on success, 0 otherwise and rte_errno is set.
6730  */
6731 uint32_t
6732 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6733 {
6734 	struct rte_flow_attr attr = {
6735 		.group = 0,
6736 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6737 		.ingress = 1,
6738 		.egress = 0,
6739 		.transfer = 1,
6740 	};
6741 	struct rte_flow_item_port_id port_spec = {
6742 		.id = MLX5_PORT_ESW_MGR,
6743 	};
6744 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6745 		.queue = txq,
6746 	};
6747 	struct rte_flow_item pattern[] = {
6748 		{
6749 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6750 			.spec = &port_spec,
6751 		},
6752 		{
6753 			.type = (enum rte_flow_item_type)
6754 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6755 			.spec = &txq_spec,
6756 		},
6757 		{
6758 			.type = RTE_FLOW_ITEM_TYPE_END,
6759 		},
6760 	};
6761 	struct rte_flow_action_jump jump = {
6762 		.group = 1,
6763 	};
6764 	struct rte_flow_action_port_id port = {
6765 		.id = dev->data->port_id,
6766 	};
6767 	struct rte_flow_action actions[] = {
6768 		{
6769 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6770 			.conf = &jump,
6771 		},
6772 		{
6773 			.type = RTE_FLOW_ACTION_TYPE_END,
6774 		},
6775 	};
6776 	struct rte_flow_error error;
6777 
6778 	/*
6779 	 * Creates group 0, highest priority jump flow.
6780 	 * Matches txq to bypass kernel packets.
6781 	 */
6782 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6783 			     false, &error) == 0)
6784 		return 0;
6785 	/* Create group 1, lowest priority redirect flow for txq. */
6786 	attr.group = 1;
6787 	actions[0].conf = &port;
6788 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6789 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6790 				actions, false, &error);
6791 }
6792 
6793 /**
6794  * Validate a flow supported by the NIC.
6795  *
6796  * @see rte_flow_validate()
6797  * @see rte_flow_ops
6798  */
6799 int
6800 mlx5_flow_validate(struct rte_eth_dev *dev,
6801 		   const struct rte_flow_attr *attr,
6802 		   const struct rte_flow_item items[],
6803 		   const struct rte_flow_action original_actions[],
6804 		   struct rte_flow_error *error)
6805 {
6806 	int hairpin_flow;
6807 	struct mlx5_translated_action_handle
6808 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6809 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6810 	const struct rte_flow_action *actions;
6811 	struct rte_flow_action *translated_actions = NULL;
6812 	int ret = flow_action_handles_translate(dev, original_actions,
6813 						indir_actions,
6814 						&indir_actions_n,
6815 						&translated_actions, error);
6816 
6817 	if (ret)
6818 		return ret;
6819 	actions = translated_actions ? translated_actions : original_actions;
6820 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6821 	ret = flow_drv_validate(dev, attr, items, actions,
6822 				true, hairpin_flow, error);
6823 	rte_free(translated_actions);
6824 	return ret;
6825 }
6826 
6827 /**
6828  * Create a flow.
6829  *
6830  * @see rte_flow_create()
6831  * @see rte_flow_ops
6832  */
6833 struct rte_flow *
6834 mlx5_flow_create(struct rte_eth_dev *dev,
6835 		 const struct rte_flow_attr *attr,
6836 		 const struct rte_flow_item items[],
6837 		 const struct rte_flow_action actions[],
6838 		 struct rte_flow_error *error)
6839 {
6840 	/*
6841 	 * If the device is not started yet, it is not allowed to created a
6842 	 * flow from application. PMD default flows and traffic control flows
6843 	 * are not affected.
6844 	 */
6845 	if (unlikely(!dev->data->dev_started)) {
6846 		DRV_LOG(DEBUG, "port %u is not started when "
6847 			"inserting a flow", dev->data->port_id);
6848 		rte_flow_error_set(error, ENODEV,
6849 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6850 				   NULL,
6851 				   "port not started");
6852 		return NULL;
6853 	}
6854 
6855 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6856 						   attr, items, actions,
6857 						   true, error);
6858 }
6859 
6860 /**
6861  * Destroy a flow in a list.
6862  *
6863  * @param dev
6864  *   Pointer to Ethernet device.
6865  * @param[in] flow_idx
6866  *   Index of flow to destroy.
6867  */
6868 static void
6869 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6870 		  uint32_t flow_idx)
6871 {
6872 	struct mlx5_priv *priv = dev->data->dev_private;
6873 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6874 
6875 	if (!flow)
6876 		return;
6877 	MLX5_ASSERT(flow->type == type);
6878 	/*
6879 	 * Update RX queue flags only if port is started, otherwise it is
6880 	 * already clean.
6881 	 */
6882 	if (dev->data->dev_started)
6883 		flow_rxq_flags_trim(dev, flow);
6884 	flow_drv_destroy(dev, flow);
6885 	if (flow->tunnel) {
6886 		struct mlx5_flow_tunnel *tunnel;
6887 
6888 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6889 		RTE_VERIFY(tunnel);
6890 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6891 			mlx5_flow_tunnel_free(dev, tunnel);
6892 	}
6893 	flow_mreg_del_copy_action(dev, flow);
6894 	mlx5_ipool_free(priv->flows[type], flow_idx);
6895 }
6896 
6897 /**
6898  * Destroy all flows.
6899  *
6900  * @param dev
6901  *   Pointer to Ethernet device.
6902  * @param type
6903  *   Flow type to be flushed.
6904  * @param active
6905  *   If flushing is called actively.
6906  */
6907 void
6908 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6909 		     bool active)
6910 {
6911 	struct mlx5_priv *priv = dev->data->dev_private;
6912 	uint32_t num_flushed = 0, fidx = 1;
6913 	struct rte_flow *flow;
6914 
6915 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6916 		flow_list_destroy(dev, type, fidx);
6917 		num_flushed++;
6918 	}
6919 	if (active) {
6920 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6921 			dev->data->port_id, num_flushed);
6922 	}
6923 }
6924 
6925 /**
6926  * Stop all default actions for flows.
6927  *
6928  * @param dev
6929  *   Pointer to Ethernet device.
6930  */
6931 void
6932 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6933 {
6934 	flow_mreg_del_default_copy_action(dev);
6935 	flow_rxq_flags_clear(dev);
6936 }
6937 
6938 /**
6939  * Start all default actions for flows.
6940  *
6941  * @param dev
6942  *   Pointer to Ethernet device.
6943  * @return
6944  *   0 on success, a negative errno value otherwise and rte_errno is set.
6945  */
6946 int
6947 mlx5_flow_start_default(struct rte_eth_dev *dev)
6948 {
6949 	struct rte_flow_error error;
6950 
6951 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6952 	return flow_mreg_add_default_copy_action(dev, &error);
6953 }
6954 
6955 /**
6956  * Release key of thread specific flow workspace data.
6957  */
6958 void
6959 flow_release_workspace(void *data)
6960 {
6961 	struct mlx5_flow_workspace *wks = data;
6962 	struct mlx5_flow_workspace *next;
6963 
6964 	while (wks) {
6965 		next = wks->next;
6966 		free(wks->rss_desc.queue);
6967 		free(wks);
6968 		wks = next;
6969 	}
6970 }
6971 
6972 /**
6973  * Get thread specific current flow workspace.
6974  *
6975  * @return pointer to thread specific flow workspace data, NULL on error.
6976  */
6977 struct mlx5_flow_workspace*
6978 mlx5_flow_get_thread_workspace(void)
6979 {
6980 	struct mlx5_flow_workspace *data;
6981 
6982 	data = mlx5_flow_os_get_specific_workspace();
6983 	MLX5_ASSERT(data && data->inuse);
6984 	if (!data || !data->inuse)
6985 		DRV_LOG(ERR, "flow workspace not initialized.");
6986 	return data;
6987 }
6988 
6989 /**
6990  * Allocate and init new flow workspace.
6991  *
6992  * @return pointer to flow workspace data, NULL on error.
6993  */
6994 static struct mlx5_flow_workspace*
6995 flow_alloc_thread_workspace(void)
6996 {
6997 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6998 
6999 	if (!data) {
7000 		DRV_LOG(ERR, "Failed to allocate flow workspace "
7001 			"memory.");
7002 		return NULL;
7003 	}
7004 	data->rss_desc.queue = calloc(1,
7005 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
7006 	if (!data->rss_desc.queue)
7007 		goto err;
7008 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
7009 	return data;
7010 err:
7011 	free(data->rss_desc.queue);
7012 	free(data);
7013 	return NULL;
7014 }
7015 
7016 /**
7017  * Get new thread specific flow workspace.
7018  *
7019  * If current workspace inuse, create new one and set as current.
7020  *
7021  * @return pointer to thread specific flow workspace data, NULL on error.
7022  */
7023 static struct mlx5_flow_workspace*
7024 mlx5_flow_push_thread_workspace(void)
7025 {
7026 	struct mlx5_flow_workspace *curr;
7027 	struct mlx5_flow_workspace *data;
7028 
7029 	curr = mlx5_flow_os_get_specific_workspace();
7030 	if (!curr) {
7031 		data = flow_alloc_thread_workspace();
7032 		if (!data)
7033 			return NULL;
7034 	} else if (!curr->inuse) {
7035 		data = curr;
7036 	} else if (curr->next) {
7037 		data = curr->next;
7038 	} else {
7039 		data = flow_alloc_thread_workspace();
7040 		if (!data)
7041 			return NULL;
7042 		curr->next = data;
7043 		data->prev = curr;
7044 	}
7045 	data->inuse = 1;
7046 	data->flow_idx = 0;
7047 	/* Set as current workspace */
7048 	if (mlx5_flow_os_set_specific_workspace(data))
7049 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7050 	return data;
7051 }
7052 
7053 /**
7054  * Close current thread specific flow workspace.
7055  *
7056  * If previous workspace available, set it as current.
7057  *
7058  * @return pointer to thread specific flow workspace data, NULL on error.
7059  */
7060 static void
7061 mlx5_flow_pop_thread_workspace(void)
7062 {
7063 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7064 
7065 	if (!data)
7066 		return;
7067 	if (!data->inuse) {
7068 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7069 		return;
7070 	}
7071 	data->inuse = 0;
7072 	if (!data->prev)
7073 		return;
7074 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7075 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7076 }
7077 
7078 /**
7079  * Verify the flow list is empty
7080  *
7081  * @param dev
7082  *  Pointer to Ethernet device.
7083  *
7084  * @return the number of flows not released.
7085  */
7086 int
7087 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7088 {
7089 	struct mlx5_priv *priv = dev->data->dev_private;
7090 	struct rte_flow *flow;
7091 	uint32_t idx = 0;
7092 	int ret = 0, i;
7093 
7094 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7095 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7096 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7097 				dev->data->port_id, (void *)flow);
7098 			ret++;
7099 		}
7100 	}
7101 	return ret;
7102 }
7103 
7104 /**
7105  * Enable default hairpin egress flow.
7106  *
7107  * @param dev
7108  *   Pointer to Ethernet device.
7109  * @param queue
7110  *   The queue index.
7111  *
7112  * @return
7113  *   0 on success, a negative errno value otherwise and rte_errno is set.
7114  */
7115 int
7116 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7117 			    uint32_t queue)
7118 {
7119 	const struct rte_flow_attr attr = {
7120 		.egress = 1,
7121 		.priority = 0,
7122 	};
7123 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7124 		.queue = queue,
7125 	};
7126 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7127 		.queue = UINT32_MAX,
7128 	};
7129 	struct rte_flow_item items[] = {
7130 		{
7131 			.type = (enum rte_flow_item_type)
7132 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7133 			.spec = &queue_spec,
7134 			.last = NULL,
7135 			.mask = &queue_mask,
7136 		},
7137 		{
7138 			.type = RTE_FLOW_ITEM_TYPE_END,
7139 		},
7140 	};
7141 	struct rte_flow_action_jump jump = {
7142 		.group = MLX5_HAIRPIN_TX_TABLE,
7143 	};
7144 	struct rte_flow_action actions[2];
7145 	uint32_t flow_idx;
7146 	struct rte_flow_error error;
7147 
7148 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7149 	actions[0].conf = &jump;
7150 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7151 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7152 				    &attr, items, actions, false, &error);
7153 	if (!flow_idx) {
7154 		DRV_LOG(DEBUG,
7155 			"Failed to create ctrl flow: rte_errno(%d),"
7156 			" type(%d), message(%s)",
7157 			rte_errno, error.type,
7158 			error.message ? error.message : " (no stated reason)");
7159 		return -rte_errno;
7160 	}
7161 	return 0;
7162 }
7163 
7164 /**
7165  * Enable a control flow configured from the control plane.
7166  *
7167  * @param dev
7168  *   Pointer to Ethernet device.
7169  * @param eth_spec
7170  *   An Ethernet flow spec to apply.
7171  * @param eth_mask
7172  *   An Ethernet flow mask to apply.
7173  * @param vlan_spec
7174  *   A VLAN flow spec to apply.
7175  * @param vlan_mask
7176  *   A VLAN flow mask to apply.
7177  *
7178  * @return
7179  *   0 on success, a negative errno value otherwise and rte_errno is set.
7180  */
7181 int
7182 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7183 		    struct rte_flow_item_eth *eth_spec,
7184 		    struct rte_flow_item_eth *eth_mask,
7185 		    struct rte_flow_item_vlan *vlan_spec,
7186 		    struct rte_flow_item_vlan *vlan_mask)
7187 {
7188 	struct mlx5_priv *priv = dev->data->dev_private;
7189 	const struct rte_flow_attr attr = {
7190 		.ingress = 1,
7191 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7192 	};
7193 	struct rte_flow_item items[] = {
7194 		{
7195 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7196 			.spec = eth_spec,
7197 			.last = NULL,
7198 			.mask = eth_mask,
7199 		},
7200 		{
7201 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7202 					      RTE_FLOW_ITEM_TYPE_END,
7203 			.spec = vlan_spec,
7204 			.last = NULL,
7205 			.mask = vlan_mask,
7206 		},
7207 		{
7208 			.type = RTE_FLOW_ITEM_TYPE_END,
7209 		},
7210 	};
7211 	uint16_t queue[priv->reta_idx_n];
7212 	struct rte_flow_action_rss action_rss = {
7213 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7214 		.level = 0,
7215 		.types = priv->rss_conf.rss_hf,
7216 		.key_len = priv->rss_conf.rss_key_len,
7217 		.queue_num = priv->reta_idx_n,
7218 		.key = priv->rss_conf.rss_key,
7219 		.queue = queue,
7220 	};
7221 	struct rte_flow_action actions[] = {
7222 		{
7223 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7224 			.conf = &action_rss,
7225 		},
7226 		{
7227 			.type = RTE_FLOW_ACTION_TYPE_END,
7228 		},
7229 	};
7230 	uint32_t flow_idx;
7231 	struct rte_flow_error error;
7232 	unsigned int i;
7233 
7234 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7235 		return 0;
7236 	}
7237 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7238 		action_rss.types = 0;
7239 	for (i = 0; i != priv->reta_idx_n; ++i)
7240 		queue[i] = (*priv->reta_idx)[i];
7241 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7242 				    &attr, items, actions, false, &error);
7243 	if (!flow_idx)
7244 		return -rte_errno;
7245 	return 0;
7246 }
7247 
7248 /**
7249  * Enable a flow control configured from the control plane.
7250  *
7251  * @param dev
7252  *   Pointer to Ethernet device.
7253  * @param eth_spec
7254  *   An Ethernet flow spec to apply.
7255  * @param eth_mask
7256  *   An Ethernet flow mask to apply.
7257  *
7258  * @return
7259  *   0 on success, a negative errno value otherwise and rte_errno is set.
7260  */
7261 int
7262 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7263 	       struct rte_flow_item_eth *eth_spec,
7264 	       struct rte_flow_item_eth *eth_mask)
7265 {
7266 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7267 }
7268 
7269 /**
7270  * Create default miss flow rule matching lacp traffic
7271  *
7272  * @param dev
7273  *   Pointer to Ethernet device.
7274  * @param eth_spec
7275  *   An Ethernet flow spec to apply.
7276  *
7277  * @return
7278  *   0 on success, a negative errno value otherwise and rte_errno is set.
7279  */
7280 int
7281 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7282 {
7283 	/*
7284 	 * The LACP matching is done by only using ether type since using
7285 	 * a multicast dst mac causes kernel to give low priority to this flow.
7286 	 */
7287 	static const struct rte_flow_item_eth lacp_spec = {
7288 		.type = RTE_BE16(0x8809),
7289 	};
7290 	static const struct rte_flow_item_eth lacp_mask = {
7291 		.type = 0xffff,
7292 	};
7293 	const struct rte_flow_attr attr = {
7294 		.ingress = 1,
7295 	};
7296 	struct rte_flow_item items[] = {
7297 		{
7298 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7299 			.spec = &lacp_spec,
7300 			.mask = &lacp_mask,
7301 		},
7302 		{
7303 			.type = RTE_FLOW_ITEM_TYPE_END,
7304 		},
7305 	};
7306 	struct rte_flow_action actions[] = {
7307 		{
7308 			.type = (enum rte_flow_action_type)
7309 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7310 		},
7311 		{
7312 			.type = RTE_FLOW_ACTION_TYPE_END,
7313 		},
7314 	};
7315 	struct rte_flow_error error;
7316 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7317 					&attr, items, actions,
7318 					false, &error);
7319 
7320 	if (!flow_idx)
7321 		return -rte_errno;
7322 	return 0;
7323 }
7324 
7325 /**
7326  * Destroy a flow.
7327  *
7328  * @see rte_flow_destroy()
7329  * @see rte_flow_ops
7330  */
7331 int
7332 mlx5_flow_destroy(struct rte_eth_dev *dev,
7333 		  struct rte_flow *flow,
7334 		  struct rte_flow_error *error __rte_unused)
7335 {
7336 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7337 				(uintptr_t)(void *)flow);
7338 	return 0;
7339 }
7340 
7341 /**
7342  * Destroy all flows.
7343  *
7344  * @see rte_flow_flush()
7345  * @see rte_flow_ops
7346  */
7347 int
7348 mlx5_flow_flush(struct rte_eth_dev *dev,
7349 		struct rte_flow_error *error __rte_unused)
7350 {
7351 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7352 	return 0;
7353 }
7354 
7355 /**
7356  * Isolated mode.
7357  *
7358  * @see rte_flow_isolate()
7359  * @see rte_flow_ops
7360  */
7361 int
7362 mlx5_flow_isolate(struct rte_eth_dev *dev,
7363 		  int enable,
7364 		  struct rte_flow_error *error)
7365 {
7366 	struct mlx5_priv *priv = dev->data->dev_private;
7367 
7368 	if (dev->data->dev_started) {
7369 		rte_flow_error_set(error, EBUSY,
7370 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7371 				   NULL,
7372 				   "port must be stopped first");
7373 		return -rte_errno;
7374 	}
7375 	priv->isolated = !!enable;
7376 	if (enable)
7377 		dev->dev_ops = &mlx5_dev_ops_isolate;
7378 	else
7379 		dev->dev_ops = &mlx5_dev_ops;
7380 
7381 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7382 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7383 
7384 	return 0;
7385 }
7386 
7387 /**
7388  * Query a flow.
7389  *
7390  * @see rte_flow_query()
7391  * @see rte_flow_ops
7392  */
7393 static int
7394 flow_drv_query(struct rte_eth_dev *dev,
7395 	       uint32_t flow_idx,
7396 	       const struct rte_flow_action *actions,
7397 	       void *data,
7398 	       struct rte_flow_error *error)
7399 {
7400 	struct mlx5_priv *priv = dev->data->dev_private;
7401 	const struct mlx5_flow_driver_ops *fops;
7402 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7403 					       flow_idx);
7404 	enum mlx5_flow_drv_type ftype;
7405 
7406 	if (!flow) {
7407 		return rte_flow_error_set(error, ENOENT,
7408 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7409 			  NULL,
7410 			  "invalid flow handle");
7411 	}
7412 	ftype = flow->drv_type;
7413 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7414 	fops = flow_get_drv_ops(ftype);
7415 
7416 	return fops->query(dev, flow, actions, data, error);
7417 }
7418 
7419 /**
7420  * Query a flow.
7421  *
7422  * @see rte_flow_query()
7423  * @see rte_flow_ops
7424  */
7425 int
7426 mlx5_flow_query(struct rte_eth_dev *dev,
7427 		struct rte_flow *flow,
7428 		const struct rte_flow_action *actions,
7429 		void *data,
7430 		struct rte_flow_error *error)
7431 {
7432 	int ret;
7433 
7434 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7435 			     error);
7436 	if (ret < 0)
7437 		return ret;
7438 	return 0;
7439 }
7440 
7441 /**
7442  * Get rte_flow callbacks.
7443  *
7444  * @param dev
7445  *   Pointer to Ethernet device structure.
7446  * @param ops
7447  *   Pointer to operation-specific structure.
7448  *
7449  * @return 0
7450  */
7451 int
7452 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7453 		  const struct rte_flow_ops **ops)
7454 {
7455 	*ops = &mlx5_flow_ops;
7456 	return 0;
7457 }
7458 
7459 /**
7460  * Validate meter policy actions.
7461  * Dispatcher for action type specific validation.
7462  *
7463  * @param[in] dev
7464  *   Pointer to the Ethernet device structure.
7465  * @param[in] action
7466  *   The meter policy action object to validate.
7467  * @param[in] attr
7468  *   Attributes of flow to determine steering domain.
7469  * @param[out] is_rss
7470  *   Is RSS or not.
7471  * @param[out] domain_bitmap
7472  *   Domain bitmap.
7473  * @param[out] is_def_policy
7474  *   Is default policy or not.
7475  * @param[out] error
7476  *   Perform verbose error reporting if not NULL. Initialized in case of
7477  *   error only.
7478  *
7479  * @return
7480  *   0 on success, otherwise negative errno value.
7481  */
7482 int
7483 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7484 			const struct rte_flow_action *actions[RTE_COLORS],
7485 			struct rte_flow_attr *attr,
7486 			bool *is_rss,
7487 			uint8_t *domain_bitmap,
7488 			uint8_t *policy_mode,
7489 			struct rte_mtr_error *error)
7490 {
7491 	const struct mlx5_flow_driver_ops *fops;
7492 
7493 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7494 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7495 				       domain_bitmap, policy_mode, error);
7496 }
7497 
7498 /**
7499  * Destroy the meter table set.
7500  *
7501  * @param[in] dev
7502  *   Pointer to Ethernet device.
7503  * @param[in] mtr_policy
7504  *   Meter policy struct.
7505  */
7506 void
7507 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7508 		      struct mlx5_flow_meter_policy *mtr_policy)
7509 {
7510 	const struct mlx5_flow_driver_ops *fops;
7511 
7512 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7513 	fops->destroy_mtr_acts(dev, mtr_policy);
7514 }
7515 
7516 /**
7517  * Create policy action, lock free,
7518  * (mutex should be acquired by caller).
7519  * Dispatcher for action type specific call.
7520  *
7521  * @param[in] dev
7522  *   Pointer to the Ethernet device structure.
7523  * @param[in] mtr_policy
7524  *   Meter policy struct.
7525  * @param[in] action
7526  *   Action specification used to create meter actions.
7527  * @param[out] error
7528  *   Perform verbose error reporting if not NULL. Initialized in case of
7529  *   error only.
7530  *
7531  * @return
7532  *   0 on success, otherwise negative errno value.
7533  */
7534 int
7535 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7536 		      struct mlx5_flow_meter_policy *mtr_policy,
7537 		      const struct rte_flow_action *actions[RTE_COLORS],
7538 		      struct rte_mtr_error *error)
7539 {
7540 	const struct mlx5_flow_driver_ops *fops;
7541 
7542 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7543 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7544 }
7545 
7546 /**
7547  * Create policy rules, lock free,
7548  * (mutex should be acquired by caller).
7549  * Dispatcher for action type specific call.
7550  *
7551  * @param[in] dev
7552  *   Pointer to the Ethernet device structure.
7553  * @param[in] mtr_policy
7554  *   Meter policy struct.
7555  *
7556  * @return
7557  *   0 on success, -1 otherwise.
7558  */
7559 int
7560 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7561 			     struct mlx5_flow_meter_policy *mtr_policy)
7562 {
7563 	const struct mlx5_flow_driver_ops *fops;
7564 
7565 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7566 	return fops->create_policy_rules(dev, mtr_policy);
7567 }
7568 
7569 /**
7570  * Destroy policy rules, lock free,
7571  * (mutex should be acquired by caller).
7572  * Dispatcher for action type specific call.
7573  *
7574  * @param[in] dev
7575  *   Pointer to the Ethernet device structure.
7576  * @param[in] mtr_policy
7577  *   Meter policy struct.
7578  */
7579 void
7580 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7581 			     struct mlx5_flow_meter_policy *mtr_policy)
7582 {
7583 	const struct mlx5_flow_driver_ops *fops;
7584 
7585 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7586 	fops->destroy_policy_rules(dev, mtr_policy);
7587 }
7588 
7589 /**
7590  * Destroy the default policy table set.
7591  *
7592  * @param[in] dev
7593  *   Pointer to Ethernet device.
7594  */
7595 void
7596 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7597 {
7598 	const struct mlx5_flow_driver_ops *fops;
7599 
7600 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7601 	fops->destroy_def_policy(dev);
7602 }
7603 
7604 /**
7605  * Destroy the default policy table set.
7606  *
7607  * @param[in] dev
7608  *   Pointer to Ethernet device.
7609  *
7610  * @return
7611  *   0 on success, -1 otherwise.
7612  */
7613 int
7614 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7615 {
7616 	const struct mlx5_flow_driver_ops *fops;
7617 
7618 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7619 	return fops->create_def_policy(dev);
7620 }
7621 
7622 /**
7623  * Create the needed meter and suffix tables.
7624  *
7625  * @param[in] dev
7626  *   Pointer to Ethernet device.
7627  *
7628  * @return
7629  *   0 on success, -1 otherwise.
7630  */
7631 int
7632 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7633 			struct mlx5_flow_meter_info *fm,
7634 			uint32_t mtr_idx,
7635 			uint8_t domain_bitmap)
7636 {
7637 	const struct mlx5_flow_driver_ops *fops;
7638 
7639 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7640 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7641 }
7642 
7643 /**
7644  * Destroy the meter table set.
7645  *
7646  * @param[in] dev
7647  *   Pointer to Ethernet device.
7648  * @param[in] tbl
7649  *   Pointer to the meter table set.
7650  */
7651 void
7652 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7653 			   struct mlx5_flow_meter_info *fm)
7654 {
7655 	const struct mlx5_flow_driver_ops *fops;
7656 
7657 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7658 	fops->destroy_mtr_tbls(dev, fm);
7659 }
7660 
7661 /**
7662  * Destroy the global meter drop table.
7663  *
7664  * @param[in] dev
7665  *   Pointer to Ethernet device.
7666  */
7667 void
7668 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7669 {
7670 	const struct mlx5_flow_driver_ops *fops;
7671 
7672 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7673 	fops->destroy_mtr_drop_tbls(dev);
7674 }
7675 
7676 /**
7677  * Destroy the sub policy table with RX queue.
7678  *
7679  * @param[in] dev
7680  *   Pointer to Ethernet device.
7681  * @param[in] mtr_policy
7682  *   Pointer to meter policy table.
7683  */
7684 void
7685 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7686 		struct mlx5_flow_meter_policy *mtr_policy)
7687 {
7688 	const struct mlx5_flow_driver_ops *fops;
7689 
7690 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7691 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7692 }
7693 
7694 /**
7695  * Allocate the needed aso flow meter id.
7696  *
7697  * @param[in] dev
7698  *   Pointer to Ethernet device.
7699  *
7700  * @return
7701  *   Index to aso flow meter on success, NULL otherwise.
7702  */
7703 uint32_t
7704 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7705 {
7706 	const struct mlx5_flow_driver_ops *fops;
7707 
7708 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7709 	return fops->create_meter(dev);
7710 }
7711 
7712 /**
7713  * Free the aso flow meter id.
7714  *
7715  * @param[in] dev
7716  *   Pointer to Ethernet device.
7717  * @param[in] mtr_idx
7718  *  Index to aso flow meter to be free.
7719  *
7720  * @return
7721  *   0 on success.
7722  */
7723 void
7724 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7725 {
7726 	const struct mlx5_flow_driver_ops *fops;
7727 
7728 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7729 	fops->free_meter(dev, mtr_idx);
7730 }
7731 
7732 /**
7733  * Allocate a counter.
7734  *
7735  * @param[in] dev
7736  *   Pointer to Ethernet device structure.
7737  *
7738  * @return
7739  *   Index to allocated counter  on success, 0 otherwise.
7740  */
7741 uint32_t
7742 mlx5_counter_alloc(struct rte_eth_dev *dev)
7743 {
7744 	const struct mlx5_flow_driver_ops *fops;
7745 	struct rte_flow_attr attr = { .transfer = 0 };
7746 
7747 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7748 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7749 		return fops->counter_alloc(dev);
7750 	}
7751 	DRV_LOG(ERR,
7752 		"port %u counter allocate is not supported.",
7753 		 dev->data->port_id);
7754 	return 0;
7755 }
7756 
7757 /**
7758  * Free a counter.
7759  *
7760  * @param[in] dev
7761  *   Pointer to Ethernet device structure.
7762  * @param[in] cnt
7763  *   Index to counter to be free.
7764  */
7765 void
7766 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7767 {
7768 	const struct mlx5_flow_driver_ops *fops;
7769 	struct rte_flow_attr attr = { .transfer = 0 };
7770 
7771 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7772 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7773 		fops->counter_free(dev, cnt);
7774 		return;
7775 	}
7776 	DRV_LOG(ERR,
7777 		"port %u counter free is not supported.",
7778 		 dev->data->port_id);
7779 }
7780 
7781 /**
7782  * Query counter statistics.
7783  *
7784  * @param[in] dev
7785  *   Pointer to Ethernet device structure.
7786  * @param[in] cnt
7787  *   Index to counter to query.
7788  * @param[in] clear
7789  *   Set to clear counter statistics.
7790  * @param[out] pkts
7791  *   The counter hits packets number to save.
7792  * @param[out] bytes
7793  *   The counter hits bytes number to save.
7794  *
7795  * @return
7796  *   0 on success, a negative errno value otherwise.
7797  */
7798 int
7799 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7800 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7801 {
7802 	const struct mlx5_flow_driver_ops *fops;
7803 	struct rte_flow_attr attr = { .transfer = 0 };
7804 
7805 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7806 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7807 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7808 	}
7809 	DRV_LOG(ERR,
7810 		"port %u counter query is not supported.",
7811 		 dev->data->port_id);
7812 	return -ENOTSUP;
7813 }
7814 
7815 /**
7816  * Allocate a new memory for the counter values wrapped by all the needed
7817  * management.
7818  *
7819  * @param[in] sh
7820  *   Pointer to mlx5_dev_ctx_shared object.
7821  *
7822  * @return
7823  *   0 on success, a negative errno value otherwise.
7824  */
7825 static int
7826 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7827 {
7828 	struct mlx5_counter_stats_mem_mng *mem_mng;
7829 	volatile struct flow_counter_stats *raw_data;
7830 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7831 	int size = (sizeof(struct flow_counter_stats) *
7832 			MLX5_COUNTERS_PER_POOL +
7833 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7834 			sizeof(struct mlx5_counter_stats_mem_mng);
7835 	size_t pgsize = rte_mem_page_size();
7836 	uint8_t *mem;
7837 	int ret;
7838 	int i;
7839 
7840 	if (pgsize == (size_t)-1) {
7841 		DRV_LOG(ERR, "Failed to get mem page size");
7842 		rte_errno = ENOMEM;
7843 		return -ENOMEM;
7844 	}
7845 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7846 	if (!mem) {
7847 		rte_errno = ENOMEM;
7848 		return -ENOMEM;
7849 	}
7850 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7851 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7852 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7853 					  sh->cdev->pdn, mem, size,
7854 					  &mem_mng->wm);
7855 	if (ret) {
7856 		rte_errno = errno;
7857 		mlx5_free(mem);
7858 		return -rte_errno;
7859 	}
7860 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7861 	raw_data = (volatile struct flow_counter_stats *)mem;
7862 	for (i = 0; i < raws_n; ++i) {
7863 		mem_mng->raws[i].mem_mng = mem_mng;
7864 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7865 	}
7866 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7867 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7868 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7869 				 next);
7870 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7871 	sh->cmng.mem_mng = mem_mng;
7872 	return 0;
7873 }
7874 
7875 /**
7876  * Set the statistic memory to the new counter pool.
7877  *
7878  * @param[in] sh
7879  *   Pointer to mlx5_dev_ctx_shared object.
7880  * @param[in] pool
7881  *   Pointer to the pool to set the statistic memory.
7882  *
7883  * @return
7884  *   0 on success, a negative errno value otherwise.
7885  */
7886 static int
7887 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7888 			       struct mlx5_flow_counter_pool *pool)
7889 {
7890 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7891 	/* Resize statistic memory once used out. */
7892 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7893 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7894 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7895 		return -1;
7896 	}
7897 	rte_spinlock_lock(&pool->sl);
7898 	pool->raw = cmng->mem_mng->raws + pool->index %
7899 		    MLX5_CNT_CONTAINER_RESIZE;
7900 	rte_spinlock_unlock(&pool->sl);
7901 	pool->raw_hw = NULL;
7902 	return 0;
7903 }
7904 
7905 #define MLX5_POOL_QUERY_FREQ_US 1000000
7906 
7907 /**
7908  * Set the periodic procedure for triggering asynchronous batch queries for all
7909  * the counter pools.
7910  *
7911  * @param[in] sh
7912  *   Pointer to mlx5_dev_ctx_shared object.
7913  */
7914 void
7915 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7916 {
7917 	uint32_t pools_n, us;
7918 
7919 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7920 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7921 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7922 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7923 		sh->cmng.query_thread_on = 0;
7924 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7925 	} else {
7926 		sh->cmng.query_thread_on = 1;
7927 	}
7928 }
7929 
7930 /**
7931  * The periodic procedure for triggering asynchronous batch queries for all the
7932  * counter pools. This function is probably called by the host thread.
7933  *
7934  * @param[in] arg
7935  *   The parameter for the alarm process.
7936  */
7937 void
7938 mlx5_flow_query_alarm(void *arg)
7939 {
7940 	struct mlx5_dev_ctx_shared *sh = arg;
7941 	int ret;
7942 	uint16_t pool_index = sh->cmng.pool_index;
7943 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7944 	struct mlx5_flow_counter_pool *pool;
7945 	uint16_t n_valid;
7946 
7947 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7948 		goto set_alarm;
7949 	rte_spinlock_lock(&cmng->pool_update_sl);
7950 	pool = cmng->pools[pool_index];
7951 	n_valid = cmng->n_valid;
7952 	rte_spinlock_unlock(&cmng->pool_update_sl);
7953 	/* Set the statistic memory to the new created pool. */
7954 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7955 		goto set_alarm;
7956 	if (pool->raw_hw)
7957 		/* There is a pool query in progress. */
7958 		goto set_alarm;
7959 	pool->raw_hw =
7960 		LIST_FIRST(&sh->cmng.free_stat_raws);
7961 	if (!pool->raw_hw)
7962 		/* No free counter statistics raw memory. */
7963 		goto set_alarm;
7964 	/*
7965 	 * Identify the counters released between query trigger and query
7966 	 * handle more efficiently. The counter released in this gap period
7967 	 * should wait for a new round of query as the new arrived packets
7968 	 * will not be taken into account.
7969 	 */
7970 	pool->query_gen++;
7971 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7972 					       MLX5_COUNTERS_PER_POOL,
7973 					       NULL, NULL,
7974 					       pool->raw_hw->mem_mng->wm.lkey,
7975 					       (void *)(uintptr_t)
7976 					       pool->raw_hw->data,
7977 					       sh->devx_comp,
7978 					       (uint64_t)(uintptr_t)pool);
7979 	if (ret) {
7980 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7981 			" %d", pool->min_dcs->id);
7982 		pool->raw_hw = NULL;
7983 		goto set_alarm;
7984 	}
7985 	LIST_REMOVE(pool->raw_hw, next);
7986 	sh->cmng.pending_queries++;
7987 	pool_index++;
7988 	if (pool_index >= n_valid)
7989 		pool_index = 0;
7990 set_alarm:
7991 	sh->cmng.pool_index = pool_index;
7992 	mlx5_set_query_alarm(sh);
7993 }
7994 
7995 /**
7996  * Check and callback event for new aged flow in the counter pool
7997  *
7998  * @param[in] sh
7999  *   Pointer to mlx5_dev_ctx_shared object.
8000  * @param[in] pool
8001  *   Pointer to Current counter pool.
8002  */
8003 static void
8004 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
8005 		   struct mlx5_flow_counter_pool *pool)
8006 {
8007 	struct mlx5_priv *priv;
8008 	struct mlx5_flow_counter *cnt;
8009 	struct mlx5_age_info *age_info;
8010 	struct mlx5_age_param *age_param;
8011 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
8012 	struct mlx5_counter_stats_raw *prev = pool->raw;
8013 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
8014 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
8015 	uint16_t expected = AGE_CANDIDATE;
8016 	uint32_t i;
8017 
8018 	pool->time_of_last_age_check = curr_time;
8019 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8020 		cnt = MLX5_POOL_GET_CNT(pool, i);
8021 		age_param = MLX5_CNT_TO_AGE(cnt);
8022 		if (__atomic_load_n(&age_param->state,
8023 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
8024 			continue;
8025 		if (cur->data[i].hits != prev->data[i].hits) {
8026 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
8027 					 __ATOMIC_RELAXED);
8028 			continue;
8029 		}
8030 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8031 				       time_delta,
8032 				       __ATOMIC_RELAXED) <= age_param->timeout)
8033 			continue;
8034 		/**
8035 		 * Hold the lock first, or if between the
8036 		 * state AGE_TMOUT and tailq operation the
8037 		 * release happened, the release procedure
8038 		 * may delete a non-existent tailq node.
8039 		 */
8040 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8041 		age_info = GET_PORT_AGE_INFO(priv);
8042 		rte_spinlock_lock(&age_info->aged_sl);
8043 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8044 						AGE_TMOUT, false,
8045 						__ATOMIC_RELAXED,
8046 						__ATOMIC_RELAXED)) {
8047 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8048 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8049 		}
8050 		rte_spinlock_unlock(&age_info->aged_sl);
8051 	}
8052 	mlx5_age_event_prepare(sh);
8053 }
8054 
8055 /**
8056  * Handler for the HW respond about ready values from an asynchronous batch
8057  * query. This function is probably called by the host thread.
8058  *
8059  * @param[in] sh
8060  *   The pointer to the shared device context.
8061  * @param[in] async_id
8062  *   The Devx async ID.
8063  * @param[in] status
8064  *   The status of the completion.
8065  */
8066 void
8067 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8068 				  uint64_t async_id, int status)
8069 {
8070 	struct mlx5_flow_counter_pool *pool =
8071 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8072 	struct mlx5_counter_stats_raw *raw_to_free;
8073 	uint8_t query_gen = pool->query_gen ^ 1;
8074 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8075 	enum mlx5_counter_type cnt_type =
8076 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8077 				MLX5_COUNTER_TYPE_ORIGIN;
8078 
8079 	if (unlikely(status)) {
8080 		raw_to_free = pool->raw_hw;
8081 	} else {
8082 		raw_to_free = pool->raw;
8083 		if (pool->is_aged)
8084 			mlx5_flow_aging_check(sh, pool);
8085 		rte_spinlock_lock(&pool->sl);
8086 		pool->raw = pool->raw_hw;
8087 		rte_spinlock_unlock(&pool->sl);
8088 		/* Be sure the new raw counters data is updated in memory. */
8089 		rte_io_wmb();
8090 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8091 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8092 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8093 				     &pool->counters[query_gen], next);
8094 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8095 		}
8096 	}
8097 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8098 	pool->raw_hw = NULL;
8099 	sh->cmng.pending_queries--;
8100 }
8101 
8102 static int
8103 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8104 		    const struct flow_grp_info *grp_info,
8105 		    struct rte_flow_error *error)
8106 {
8107 	if (grp_info->transfer && grp_info->external &&
8108 	    grp_info->fdb_def_rule) {
8109 		if (group == UINT32_MAX)
8110 			return rte_flow_error_set
8111 						(error, EINVAL,
8112 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8113 						 NULL,
8114 						 "group index not supported");
8115 		*table = group + 1;
8116 	} else {
8117 		*table = group;
8118 	}
8119 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8120 	return 0;
8121 }
8122 
8123 /**
8124  * Translate the rte_flow group index to HW table value.
8125  *
8126  * If tunnel offload is disabled, all group ids converted to flow table
8127  * id using the standard method.
8128  * If tunnel offload is enabled, group id can be converted using the
8129  * standard or tunnel conversion method. Group conversion method
8130  * selection depends on flags in `grp_info` parameter:
8131  * - Internal (grp_info.external == 0) groups conversion uses the
8132  *   standard method.
8133  * - Group ids in JUMP action converted with the tunnel conversion.
8134  * - Group id in rule attribute conversion depends on a rule type and
8135  *   group id value:
8136  *   ** non zero group attributes converted with the tunnel method
8137  *   ** zero group attribute in non-tunnel rule is converted using the
8138  *      standard method - there's only one root table
8139  *   ** zero group attribute in steer tunnel rule is converted with the
8140  *      standard method - single root table
8141  *   ** zero group attribute in match tunnel rule is a special OvS
8142  *      case: that value is used for portability reasons. That group
8143  *      id is converted with the tunnel conversion method.
8144  *
8145  * @param[in] dev
8146  *   Port device
8147  * @param[in] tunnel
8148  *   PMD tunnel offload object
8149  * @param[in] group
8150  *   rte_flow group index value.
8151  * @param[out] table
8152  *   HW table value.
8153  * @param[in] grp_info
8154  *   flags used for conversion
8155  * @param[out] error
8156  *   Pointer to error structure.
8157  *
8158  * @return
8159  *   0 on success, a negative errno value otherwise and rte_errno is set.
8160  */
8161 int
8162 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8163 			 const struct mlx5_flow_tunnel *tunnel,
8164 			 uint32_t group, uint32_t *table,
8165 			 const struct flow_grp_info *grp_info,
8166 			 struct rte_flow_error *error)
8167 {
8168 	int ret;
8169 	bool standard_translation;
8170 
8171 	if (!grp_info->skip_scale && grp_info->external &&
8172 	    group < MLX5_MAX_TABLES_EXTERNAL)
8173 		group *= MLX5_FLOW_TABLE_FACTOR;
8174 	if (is_tunnel_offload_active(dev)) {
8175 		standard_translation = !grp_info->external ||
8176 					grp_info->std_tbl_fix;
8177 	} else {
8178 		standard_translation = true;
8179 	}
8180 	DRV_LOG(DEBUG,
8181 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8182 		dev->data->port_id, group, grp_info->transfer,
8183 		grp_info->external, grp_info->fdb_def_rule,
8184 		standard_translation ? "STANDARD" : "TUNNEL");
8185 	if (standard_translation)
8186 		ret = flow_group_to_table(dev->data->port_id, group, table,
8187 					  grp_info, error);
8188 	else
8189 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8190 						      table, error);
8191 
8192 	return ret;
8193 }
8194 
8195 /**
8196  * Discover availability of metadata reg_c's.
8197  *
8198  * Iteratively use test flows to check availability.
8199  *
8200  * @param[in] dev
8201  *   Pointer to the Ethernet device structure.
8202  *
8203  * @return
8204  *   0 on success, a negative errno value otherwise and rte_errno is set.
8205  */
8206 int
8207 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8208 {
8209 	struct mlx5_priv *priv = dev->data->dev_private;
8210 	enum modify_reg idx;
8211 	int n = 0;
8212 
8213 	/* reg_c[0] and reg_c[1] are reserved. */
8214 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8215 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8216 	/* Discover availability of other reg_c's. */
8217 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8218 		struct rte_flow_attr attr = {
8219 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8220 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8221 			.ingress = 1,
8222 		};
8223 		struct rte_flow_item items[] = {
8224 			[0] = {
8225 				.type = RTE_FLOW_ITEM_TYPE_END,
8226 			},
8227 		};
8228 		struct rte_flow_action actions[] = {
8229 			[0] = {
8230 				.type = (enum rte_flow_action_type)
8231 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8232 				.conf = &(struct mlx5_flow_action_copy_mreg){
8233 					.src = REG_C_1,
8234 					.dst = idx,
8235 				},
8236 			},
8237 			[1] = {
8238 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8239 				.conf = &(struct rte_flow_action_jump){
8240 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8241 				},
8242 			},
8243 			[2] = {
8244 				.type = RTE_FLOW_ACTION_TYPE_END,
8245 			},
8246 		};
8247 		uint32_t flow_idx;
8248 		struct rte_flow *flow;
8249 		struct rte_flow_error error;
8250 
8251 		if (!priv->config.dv_flow_en)
8252 			break;
8253 		/* Create internal flow, validation skips copy action. */
8254 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8255 					items, actions, false, &error);
8256 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8257 				      flow_idx);
8258 		if (!flow)
8259 			continue;
8260 		priv->sh->flow_mreg_c[n++] = idx;
8261 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8262 	}
8263 	for (; n < MLX5_MREG_C_NUM; ++n)
8264 		priv->sh->flow_mreg_c[n] = REG_NON;
8265 	priv->sh->metadata_regc_check_flag = 1;
8266 	return 0;
8267 }
8268 
8269 int
8270 save_dump_file(const uint8_t *data, uint32_t size,
8271 	uint32_t type, uint64_t id, void *arg, FILE *file)
8272 {
8273 	char line[BUF_SIZE];
8274 	uint32_t out = 0;
8275 	uint32_t k;
8276 	uint32_t actions_num;
8277 	struct rte_flow_query_count *count;
8278 
8279 	memset(line, 0, BUF_SIZE);
8280 	switch (type) {
8281 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8282 		actions_num = *(uint32_t *)(arg);
8283 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8284 				type, id, actions_num);
8285 		break;
8286 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8287 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8288 				type, id);
8289 		break;
8290 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8291 		count = (struct rte_flow_query_count *)arg;
8292 		fprintf(file,
8293 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8294 			type, id, count->hits, count->bytes);
8295 		return 0;
8296 	default:
8297 		return -1;
8298 	}
8299 
8300 	for (k = 0; k < size; k++) {
8301 		/* Make sure we do not overrun the line buffer length. */
8302 		if (out >= BUF_SIZE - 4) {
8303 			line[out] = '\0';
8304 			break;
8305 		}
8306 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8307 				(data[k]) & 0xff);
8308 	}
8309 	fprintf(file, "%s\n", line);
8310 	return 0;
8311 }
8312 
8313 int
8314 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8315 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8316 {
8317 	struct rte_flow_action action[2];
8318 	enum mlx5_flow_drv_type ftype;
8319 	const struct mlx5_flow_driver_ops *fops;
8320 
8321 	if (!flow) {
8322 		return rte_flow_error_set(error, ENOENT,
8323 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8324 				NULL,
8325 				"invalid flow handle");
8326 	}
8327 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8328 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8329 	if (flow->counter) {
8330 		memset(count, 0, sizeof(struct rte_flow_query_count));
8331 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8332 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8333 						ftype < MLX5_FLOW_TYPE_MAX);
8334 		fops = flow_get_drv_ops(ftype);
8335 		return fops->query(dev, flow, action, count, error);
8336 	}
8337 	return -1;
8338 }
8339 
8340 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8341 /**
8342  * Dump flow ipool data to file
8343  *
8344  * @param[in] dev
8345  *   The pointer to Ethernet device.
8346  * @param[in] file
8347  *   A pointer to a file for output.
8348  * @param[out] error
8349  *   Perform verbose error reporting if not NULL. PMDs initialize this
8350  *   structure in case of error only.
8351  * @return
8352  *   0 on success, a negative value otherwise.
8353  */
8354 int
8355 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8356 	struct rte_flow *flow, FILE *file,
8357 	struct rte_flow_error *error)
8358 {
8359 	struct mlx5_priv *priv = dev->data->dev_private;
8360 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8361 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8362 	uint32_t handle_idx;
8363 	struct mlx5_flow_handle *dh;
8364 	struct rte_flow_query_count count;
8365 	uint32_t actions_num;
8366 	const uint8_t *data;
8367 	size_t size;
8368 	uint64_t id;
8369 	uint32_t type;
8370 	void *action = NULL;
8371 
8372 	if (!flow) {
8373 		return rte_flow_error_set(error, ENOENT,
8374 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8375 				NULL,
8376 				"invalid flow handle");
8377 	}
8378 	handle_idx = flow->dev_handles;
8379 	while (handle_idx) {
8380 		dh = mlx5_ipool_get(priv->sh->ipool
8381 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8382 		if (!dh)
8383 			continue;
8384 		handle_idx = dh->next.next;
8385 
8386 		/* query counter */
8387 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8388 		flow_dv_query_count_ptr(dev, flow->counter,
8389 						&action, error);
8390 		if (action) {
8391 			id = (uint64_t)(uintptr_t)action;
8392 			if (!mlx5_flow_query_counter(dev, flow, &count, error))
8393 				save_dump_file(NULL, 0, type,
8394 						id, (void *)&count, file);
8395 		}
8396 		/* Get modify_hdr and encap_decap buf from ipools. */
8397 		encap_decap = NULL;
8398 		modify_hdr = dh->dvh.modify_hdr;
8399 
8400 		if (dh->dvh.rix_encap_decap) {
8401 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8402 						[MLX5_IPOOL_DECAP_ENCAP],
8403 						dh->dvh.rix_encap_decap);
8404 		}
8405 		if (modify_hdr) {
8406 			data = (const uint8_t *)modify_hdr->actions;
8407 			size = (size_t)(modify_hdr->actions_num) * 8;
8408 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8409 			actions_num = modify_hdr->actions_num;
8410 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8411 			save_dump_file(data, size, type, id,
8412 						(void *)(&actions_num), file);
8413 		}
8414 		if (encap_decap) {
8415 			data = encap_decap->buf;
8416 			size = encap_decap->size;
8417 			id = (uint64_t)(uintptr_t)encap_decap->action;
8418 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8419 			save_dump_file(data, size, type,
8420 						id, NULL, file);
8421 		}
8422 	}
8423 	return 0;
8424 }
8425 
8426 /**
8427  * Dump all flow's encap_decap/modify_hdr/counter data to file
8428  *
8429  * @param[in] dev
8430  *   The pointer to Ethernet device.
8431  * @param[in] file
8432  *   A pointer to a file for output.
8433  * @param[out] error
8434  *   Perform verbose error reporting if not NULL. PMDs initialize this
8435  *   structure in case of error only.
8436  * @return
8437  *   0 on success, a negative value otherwise.
8438  */
8439 static int
8440 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8441 	FILE *file, struct rte_flow_error *error)
8442 {
8443 	struct mlx5_priv *priv = dev->data->dev_private;
8444 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8445 	struct mlx5_hlist *h;
8446 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8447 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8448 	struct rte_flow_query_count count;
8449 	uint32_t actions_num;
8450 	const uint8_t *data;
8451 	size_t size;
8452 	uint64_t id;
8453 	uint32_t type;
8454 	uint32_t i;
8455 	uint32_t j;
8456 	struct mlx5_list_inconst *l_inconst;
8457 	struct mlx5_list_entry *e;
8458 	int lcore_index;
8459 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8460 	uint32_t max;
8461 	void *action;
8462 
8463 	/* encap_decap hlist is lcore_share, get global core cache. */
8464 	i = MLX5_LIST_GLOBAL;
8465 	h = sh->encaps_decaps;
8466 	if (h) {
8467 		for (j = 0; j <= h->mask; j++) {
8468 			l_inconst = &h->buckets[j].l;
8469 			if (!l_inconst || !l_inconst->cache[i])
8470 				continue;
8471 
8472 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8473 			while (e) {
8474 				encap_decap =
8475 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8476 				data = encap_decap->buf;
8477 				size = encap_decap->size;
8478 				id = (uint64_t)(uintptr_t)encap_decap->action;
8479 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8480 				save_dump_file(data, size, type,
8481 					id, NULL, file);
8482 				e = LIST_NEXT(e, next);
8483 			}
8484 		}
8485 	}
8486 
8487 	/* get modify_hdr */
8488 	h = sh->modify_cmds;
8489 	if (h) {
8490 		lcore_index = rte_lcore_index(rte_lcore_id());
8491 		if (unlikely(lcore_index == -1)) {
8492 			lcore_index = MLX5_LIST_NLCORE;
8493 			rte_spinlock_lock(&h->l_const.lcore_lock);
8494 		}
8495 		i = lcore_index;
8496 
8497 		for (j = 0; j <= h->mask; j++) {
8498 			l_inconst = &h->buckets[j].l;
8499 			if (!l_inconst || !l_inconst->cache[i])
8500 				continue;
8501 
8502 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8503 			while (e) {
8504 				modify_hdr =
8505 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8506 				data = (const uint8_t *)modify_hdr->actions;
8507 				size = (size_t)(modify_hdr->actions_num) * 8;
8508 				actions_num = modify_hdr->actions_num;
8509 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8510 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8511 				save_dump_file(data, size, type, id,
8512 						(void *)(&actions_num), file);
8513 				e = LIST_NEXT(e, next);
8514 			}
8515 		}
8516 
8517 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8518 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8519 	}
8520 
8521 	/* get counter */
8522 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8523 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8524 	for (j = 1; j <= max; j++) {
8525 		action = NULL;
8526 		flow_dv_query_count_ptr(dev, j, &action, error);
8527 		if (action) {
8528 			if (!flow_dv_query_count(dev, j, &count, error)) {
8529 				type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8530 				id = (uint64_t)(uintptr_t)action;
8531 				save_dump_file(NULL, 0, type,
8532 						id, (void *)&count, file);
8533 			}
8534 		}
8535 	}
8536 	return 0;
8537 }
8538 #endif
8539 
8540 /**
8541  * Dump flow raw hw data to file
8542  *
8543  * @param[in] dev
8544  *    The pointer to Ethernet device.
8545  * @param[in] file
8546  *   A pointer to a file for output.
8547  * @param[out] error
8548  *   Perform verbose error reporting if not NULL. PMDs initialize this
8549  *   structure in case of error only.
8550  * @return
8551  *   0 on success, a negative value otherwise.
8552  */
8553 int
8554 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8555 		   FILE *file,
8556 		   struct rte_flow_error *error __rte_unused)
8557 {
8558 	struct mlx5_priv *priv = dev->data->dev_private;
8559 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8560 	uint32_t handle_idx;
8561 	int ret;
8562 	struct mlx5_flow_handle *dh;
8563 	struct rte_flow *flow;
8564 
8565 	if (!priv->config.dv_flow_en) {
8566 		if (fputs("device dv flow disabled\n", file) <= 0)
8567 			return -errno;
8568 		return -ENOTSUP;
8569 	}
8570 
8571 	/* dump all */
8572 	if (!flow_idx) {
8573 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8574 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8575 			return -EINVAL;
8576 #endif
8577 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8578 					sh->rx_domain,
8579 					sh->tx_domain, file);
8580 	}
8581 	/* dump one */
8582 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8583 			(uintptr_t)(void *)flow_idx);
8584 	if (!flow)
8585 		return -EINVAL;
8586 
8587 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8588 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8589 #endif
8590 	handle_idx = flow->dev_handles;
8591 	while (handle_idx) {
8592 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8593 				handle_idx);
8594 		if (!dh)
8595 			return -ENOENT;
8596 		if (dh->drv_flow) {
8597 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8598 					file);
8599 			if (ret)
8600 				return -ENOENT;
8601 		}
8602 		handle_idx = dh->next.next;
8603 	}
8604 	return 0;
8605 }
8606 
8607 /**
8608  * Get aged-out flows.
8609  *
8610  * @param[in] dev
8611  *   Pointer to the Ethernet device structure.
8612  * @param[in] context
8613  *   The address of an array of pointers to the aged-out flows contexts.
8614  * @param[in] nb_countexts
8615  *   The length of context array pointers.
8616  * @param[out] error
8617  *   Perform verbose error reporting if not NULL. Initialized in case of
8618  *   error only.
8619  *
8620  * @return
8621  *   how many contexts get in success, otherwise negative errno value.
8622  *   if nb_contexts is 0, return the amount of all aged contexts.
8623  *   if nb_contexts is not 0 , return the amount of aged flows reported
8624  *   in the context array.
8625  */
8626 int
8627 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8628 			uint32_t nb_contexts, struct rte_flow_error *error)
8629 {
8630 	const struct mlx5_flow_driver_ops *fops;
8631 	struct rte_flow_attr attr = { .transfer = 0 };
8632 
8633 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8634 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8635 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8636 						    error);
8637 	}
8638 	DRV_LOG(ERR,
8639 		"port %u get aged flows is not supported.",
8640 		 dev->data->port_id);
8641 	return -ENOTSUP;
8642 }
8643 
8644 /* Wrapper for driver action_validate op callback */
8645 static int
8646 flow_drv_action_validate(struct rte_eth_dev *dev,
8647 			 const struct rte_flow_indir_action_conf *conf,
8648 			 const struct rte_flow_action *action,
8649 			 const struct mlx5_flow_driver_ops *fops,
8650 			 struct rte_flow_error *error)
8651 {
8652 	static const char err_msg[] = "indirect action validation unsupported";
8653 
8654 	if (!fops->action_validate) {
8655 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8656 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8657 				   NULL, err_msg);
8658 		return -rte_errno;
8659 	}
8660 	return fops->action_validate(dev, conf, action, error);
8661 }
8662 
8663 /**
8664  * Destroys the shared action by handle.
8665  *
8666  * @param dev
8667  *   Pointer to Ethernet device structure.
8668  * @param[in] handle
8669  *   Handle for the indirect action object to be destroyed.
8670  * @param[out] error
8671  *   Perform verbose error reporting if not NULL. PMDs initialize this
8672  *   structure in case of error only.
8673  *
8674  * @return
8675  *   0 on success, a negative errno value otherwise and rte_errno is set.
8676  *
8677  * @note: wrapper for driver action_create op callback.
8678  */
8679 static int
8680 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8681 			   struct rte_flow_action_handle *handle,
8682 			   struct rte_flow_error *error)
8683 {
8684 	static const char err_msg[] = "indirect action destruction unsupported";
8685 	struct rte_flow_attr attr = { .transfer = 0 };
8686 	const struct mlx5_flow_driver_ops *fops =
8687 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8688 
8689 	if (!fops->action_destroy) {
8690 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8691 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8692 				   NULL, err_msg);
8693 		return -rte_errno;
8694 	}
8695 	return fops->action_destroy(dev, handle, error);
8696 }
8697 
8698 /* Wrapper for driver action_destroy op callback */
8699 static int
8700 flow_drv_action_update(struct rte_eth_dev *dev,
8701 		       struct rte_flow_action_handle *handle,
8702 		       const void *update,
8703 		       const struct mlx5_flow_driver_ops *fops,
8704 		       struct rte_flow_error *error)
8705 {
8706 	static const char err_msg[] = "indirect action update unsupported";
8707 
8708 	if (!fops->action_update) {
8709 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8710 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8711 				   NULL, err_msg);
8712 		return -rte_errno;
8713 	}
8714 	return fops->action_update(dev, handle, update, error);
8715 }
8716 
8717 /* Wrapper for driver action_destroy op callback */
8718 static int
8719 flow_drv_action_query(struct rte_eth_dev *dev,
8720 		      const struct rte_flow_action_handle *handle,
8721 		      void *data,
8722 		      const struct mlx5_flow_driver_ops *fops,
8723 		      struct rte_flow_error *error)
8724 {
8725 	static const char err_msg[] = "indirect action query unsupported";
8726 
8727 	if (!fops->action_query) {
8728 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8729 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8730 				   NULL, err_msg);
8731 		return -rte_errno;
8732 	}
8733 	return fops->action_query(dev, handle, data, error);
8734 }
8735 
8736 /**
8737  * Create indirect action for reuse in multiple flow rules.
8738  *
8739  * @param dev
8740  *   Pointer to Ethernet device structure.
8741  * @param conf
8742  *   Pointer to indirect action object configuration.
8743  * @param[in] action
8744  *   Action configuration for indirect action object creation.
8745  * @param[out] error
8746  *   Perform verbose error reporting if not NULL. PMDs initialize this
8747  *   structure in case of error only.
8748  * @return
8749  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8750  */
8751 static struct rte_flow_action_handle *
8752 mlx5_action_handle_create(struct rte_eth_dev *dev,
8753 			  const struct rte_flow_indir_action_conf *conf,
8754 			  const struct rte_flow_action *action,
8755 			  struct rte_flow_error *error)
8756 {
8757 	static const char err_msg[] = "indirect action creation unsupported";
8758 	struct rte_flow_attr attr = { .transfer = 0 };
8759 	const struct mlx5_flow_driver_ops *fops =
8760 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8761 
8762 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8763 		return NULL;
8764 	if (!fops->action_create) {
8765 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8766 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8767 				   NULL, err_msg);
8768 		return NULL;
8769 	}
8770 	return fops->action_create(dev, conf, action, error);
8771 }
8772 
8773 /**
8774  * Updates inplace the indirect action configuration pointed by *handle*
8775  * with the configuration provided as *update* argument.
8776  * The update of the indirect action configuration effects all flow rules
8777  * reusing the action via handle.
8778  *
8779  * @param dev
8780  *   Pointer to Ethernet device structure.
8781  * @param[in] handle
8782  *   Handle for the indirect action to be updated.
8783  * @param[in] update
8784  *   Action specification used to modify the action pointed by handle.
8785  *   *update* could be of same type with the action pointed by the *handle*
8786  *   handle argument, or some other structures like a wrapper, depending on
8787  *   the indirect action type.
8788  * @param[out] error
8789  *   Perform verbose error reporting if not NULL. PMDs initialize this
8790  *   structure in case of error only.
8791  *
8792  * @return
8793  *   0 on success, a negative errno value otherwise and rte_errno is set.
8794  */
8795 static int
8796 mlx5_action_handle_update(struct rte_eth_dev *dev,
8797 		struct rte_flow_action_handle *handle,
8798 		const void *update,
8799 		struct rte_flow_error *error)
8800 {
8801 	struct rte_flow_attr attr = { .transfer = 0 };
8802 	const struct mlx5_flow_driver_ops *fops =
8803 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8804 	int ret;
8805 
8806 	ret = flow_drv_action_validate(dev, NULL,
8807 			(const struct rte_flow_action *)update, fops, error);
8808 	if (ret)
8809 		return ret;
8810 	return flow_drv_action_update(dev, handle, update, fops,
8811 				      error);
8812 }
8813 
8814 /**
8815  * Query the indirect action by handle.
8816  *
8817  * This function allows retrieving action-specific data such as counters.
8818  * Data is gathered by special action which may be present/referenced in
8819  * more than one flow rule definition.
8820  *
8821  * see @RTE_FLOW_ACTION_TYPE_COUNT
8822  *
8823  * @param dev
8824  *   Pointer to Ethernet device structure.
8825  * @param[in] handle
8826  *   Handle for the indirect action to query.
8827  * @param[in, out] data
8828  *   Pointer to storage for the associated query data type.
8829  * @param[out] error
8830  *   Perform verbose error reporting if not NULL. PMDs initialize this
8831  *   structure in case of error only.
8832  *
8833  * @return
8834  *   0 on success, a negative errno value otherwise and rte_errno is set.
8835  */
8836 static int
8837 mlx5_action_handle_query(struct rte_eth_dev *dev,
8838 			 const struct rte_flow_action_handle *handle,
8839 			 void *data,
8840 			 struct rte_flow_error *error)
8841 {
8842 	struct rte_flow_attr attr = { .transfer = 0 };
8843 	const struct mlx5_flow_driver_ops *fops =
8844 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8845 
8846 	return flow_drv_action_query(dev, handle, data, fops, error);
8847 }
8848 
8849 /**
8850  * Destroy all indirect actions (shared RSS).
8851  *
8852  * @param dev
8853  *   Pointer to Ethernet device.
8854  *
8855  * @return
8856  *   0 on success, a negative errno value otherwise and rte_errno is set.
8857  */
8858 int
8859 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8860 {
8861 	struct rte_flow_error error;
8862 	struct mlx5_priv *priv = dev->data->dev_private;
8863 	struct mlx5_shared_action_rss *shared_rss;
8864 	int ret = 0;
8865 	uint32_t idx;
8866 
8867 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8868 		      priv->rss_shared_actions, idx, shared_rss, next) {
8869 		ret |= mlx5_action_handle_destroy(dev,
8870 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8871 	}
8872 	return ret;
8873 }
8874 
8875 /**
8876  * Validate existing indirect actions against current device configuration
8877  * and attach them to device resources.
8878  *
8879  * @param dev
8880  *   Pointer to Ethernet device.
8881  *
8882  * @return
8883  *   0 on success, a negative errno value otherwise and rte_errno is set.
8884  */
8885 int
8886 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8887 {
8888 	struct mlx5_priv *priv = dev->data->dev_private;
8889 	struct mlx5_indexed_pool *ipool =
8890 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8891 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8892 	int ret = 0;
8893 	uint32_t idx;
8894 
8895 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8896 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8897 		const char *message;
8898 		uint32_t queue_idx;
8899 
8900 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8901 					       ind_tbl->queues_n,
8902 					       &message, &queue_idx);
8903 		if (ret != 0) {
8904 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8905 				dev->data->port_id, ind_tbl->queues[queue_idx],
8906 				message);
8907 			break;
8908 		}
8909 	}
8910 	if (ret != 0)
8911 		return ret;
8912 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8913 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8914 
8915 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8916 		if (ret != 0) {
8917 			DRV_LOG(ERR, "Port %u could not attach "
8918 				"indirection table obj %p",
8919 				dev->data->port_id, (void *)ind_tbl);
8920 			goto error;
8921 		}
8922 	}
8923 	return 0;
8924 error:
8925 	shared_rss_last = shared_rss;
8926 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8927 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8928 
8929 		if (shared_rss == shared_rss_last)
8930 			break;
8931 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8932 			DRV_LOG(CRIT, "Port %u could not detach "
8933 				"indirection table obj %p on rollback",
8934 				dev->data->port_id, (void *)ind_tbl);
8935 	}
8936 	return ret;
8937 }
8938 
8939 /**
8940  * Detach indirect actions of the device from its resources.
8941  *
8942  * @param dev
8943  *   Pointer to Ethernet device.
8944  *
8945  * @return
8946  *   0 on success, a negative errno value otherwise and rte_errno is set.
8947  */
8948 int
8949 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8950 {
8951 	struct mlx5_priv *priv = dev->data->dev_private;
8952 	struct mlx5_indexed_pool *ipool =
8953 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8954 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8955 	int ret = 0;
8956 	uint32_t idx;
8957 
8958 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8959 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8960 
8961 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8962 		if (ret != 0) {
8963 			DRV_LOG(ERR, "Port %u could not detach "
8964 				"indirection table obj %p",
8965 				dev->data->port_id, (void *)ind_tbl);
8966 			goto error;
8967 		}
8968 	}
8969 	return 0;
8970 error:
8971 	shared_rss_last = shared_rss;
8972 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8973 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8974 
8975 		if (shared_rss == shared_rss_last)
8976 			break;
8977 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
8978 			DRV_LOG(CRIT, "Port %u could not attach "
8979 				"indirection table obj %p on rollback",
8980 				dev->data->port_id, (void *)ind_tbl);
8981 	}
8982 	return ret;
8983 }
8984 
8985 #ifndef HAVE_MLX5DV_DR
8986 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8987 #else
8988 #define MLX5_DOMAIN_SYNC_FLOW \
8989 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8990 #endif
8991 
8992 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8993 {
8994 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8995 	const struct mlx5_flow_driver_ops *fops;
8996 	int ret;
8997 	struct rte_flow_attr attr = { .transfer = 0 };
8998 
8999 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9000 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
9001 	if (ret > 0)
9002 		ret = -ret;
9003 	return ret;
9004 }
9005 
9006 const struct mlx5_flow_tunnel *
9007 mlx5_get_tof(const struct rte_flow_item *item,
9008 	     const struct rte_flow_action *action,
9009 	     enum mlx5_tof_rule_type *rule_type)
9010 {
9011 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9012 		if (item->type == (typeof(item->type))
9013 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
9014 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
9015 			return flow_items_to_tunnel(item);
9016 		}
9017 	}
9018 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9019 		if (action->type == (typeof(action->type))
9020 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9021 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9022 			return flow_actions_to_tunnel(action);
9023 		}
9024 	}
9025 	return NULL;
9026 }
9027 
9028 /**
9029  * tunnel offload functionality is defined for DV environment only
9030  */
9031 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9032 __extension__
9033 union tunnel_offload_mark {
9034 	uint32_t val;
9035 	struct {
9036 		uint32_t app_reserve:8;
9037 		uint32_t table_id:15;
9038 		uint32_t transfer:1;
9039 		uint32_t _unused_:8;
9040 	};
9041 };
9042 
9043 static bool
9044 mlx5_access_tunnel_offload_db
9045 	(struct rte_eth_dev *dev,
9046 	 bool (*match)(struct rte_eth_dev *,
9047 		       struct mlx5_flow_tunnel *, const void *),
9048 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9049 	 void (*miss)(struct rte_eth_dev *, void *),
9050 	 void *ctx, bool lock_op);
9051 
9052 static int
9053 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9054 			     struct rte_flow *flow,
9055 			     const struct rte_flow_attr *attr,
9056 			     const struct rte_flow_action *app_actions,
9057 			     uint32_t flow_idx,
9058 			     const struct mlx5_flow_tunnel *tunnel,
9059 			     struct tunnel_default_miss_ctx *ctx,
9060 			     struct rte_flow_error *error)
9061 {
9062 	struct mlx5_priv *priv = dev->data->dev_private;
9063 	struct mlx5_flow *dev_flow;
9064 	struct rte_flow_attr miss_attr = *attr;
9065 	const struct rte_flow_item miss_items[2] = {
9066 		{
9067 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9068 			.spec = NULL,
9069 			.last = NULL,
9070 			.mask = NULL
9071 		},
9072 		{
9073 			.type = RTE_FLOW_ITEM_TYPE_END,
9074 			.spec = NULL,
9075 			.last = NULL,
9076 			.mask = NULL
9077 		}
9078 	};
9079 	union tunnel_offload_mark mark_id;
9080 	struct rte_flow_action_mark miss_mark;
9081 	struct rte_flow_action miss_actions[3] = {
9082 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9083 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9084 	};
9085 	const struct rte_flow_action_jump *jump_data;
9086 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9087 	struct flow_grp_info grp_info = {
9088 		.external = 1,
9089 		.transfer = attr->transfer,
9090 		.fdb_def_rule = !!priv->fdb_def_rule,
9091 		.std_tbl_fix = 0,
9092 	};
9093 	int ret;
9094 
9095 	if (!attr->transfer) {
9096 		uint32_t q_size;
9097 
9098 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9099 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9100 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9101 					 0, SOCKET_ID_ANY);
9102 		if (!ctx->queue)
9103 			return rte_flow_error_set
9104 				(error, ENOMEM,
9105 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9106 				NULL, "invalid default miss RSS");
9107 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9108 		ctx->action_rss.level = 0,
9109 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9110 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9111 		ctx->action_rss.queue_num = priv->reta_idx_n,
9112 		ctx->action_rss.key = priv->rss_conf.rss_key,
9113 		ctx->action_rss.queue = ctx->queue;
9114 		if (!priv->reta_idx_n || !priv->rxqs_n)
9115 			return rte_flow_error_set
9116 				(error, EINVAL,
9117 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9118 				NULL, "invalid port configuration");
9119 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9120 			ctx->action_rss.types = 0;
9121 		for (i = 0; i != priv->reta_idx_n; ++i)
9122 			ctx->queue[i] = (*priv->reta_idx)[i];
9123 	} else {
9124 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9125 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9126 	}
9127 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9128 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9129 	jump_data = app_actions->conf;
9130 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9131 	miss_attr.group = jump_data->group;
9132 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9133 				       &flow_table, &grp_info, error);
9134 	if (ret)
9135 		return rte_flow_error_set(error, EINVAL,
9136 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9137 					  NULL, "invalid tunnel id");
9138 	mark_id.app_reserve = 0;
9139 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9140 	mark_id.transfer = !!attr->transfer;
9141 	mark_id._unused_ = 0;
9142 	miss_mark.id = mark_id.val;
9143 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9144 				    miss_items, miss_actions, flow_idx, error);
9145 	if (!dev_flow)
9146 		return -rte_errno;
9147 	dev_flow->flow = flow;
9148 	dev_flow->external = true;
9149 	dev_flow->tunnel = tunnel;
9150 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9151 	/* Subflow object was created, we must include one in the list. */
9152 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9153 		      dev_flow->handle, next);
9154 	DRV_LOG(DEBUG,
9155 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9156 		dev->data->port_id, tunnel->app_tunnel.type,
9157 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9158 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9159 				  miss_actions, error);
9160 	if (!ret)
9161 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9162 						  error);
9163 
9164 	return ret;
9165 }
9166 
9167 static const struct mlx5_flow_tbl_data_entry  *
9168 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9169 {
9170 	struct mlx5_priv *priv = dev->data->dev_private;
9171 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9172 	struct mlx5_list_entry *he;
9173 	union tunnel_offload_mark mbits = { .val = mark };
9174 	union mlx5_flow_tbl_key table_key = {
9175 		{
9176 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9177 			.id = 0,
9178 			.reserved = 0,
9179 			.dummy = 0,
9180 			.is_fdb = !!mbits.transfer,
9181 			.is_egress = 0,
9182 		}
9183 	};
9184 	struct mlx5_flow_cb_ctx ctx = {
9185 		.data = &table_key.v64,
9186 	};
9187 
9188 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9189 	return he ?
9190 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9191 }
9192 
9193 static void
9194 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9195 				   struct mlx5_list_entry *entry)
9196 {
9197 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9198 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9199 
9200 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9201 			tunnel_flow_tbl_to_id(tte->flow_table));
9202 	mlx5_free(tte);
9203 }
9204 
9205 static int
9206 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9207 				  struct mlx5_list_entry *entry, void *cb_ctx)
9208 {
9209 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9210 	union tunnel_tbl_key tbl = {
9211 		.val = *(uint64_t *)(ctx->data),
9212 	};
9213 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9214 
9215 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9216 }
9217 
9218 static struct mlx5_list_entry *
9219 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9220 {
9221 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9222 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9223 	struct tunnel_tbl_entry *tte;
9224 	union tunnel_tbl_key tbl = {
9225 		.val = *(uint64_t *)(ctx->data),
9226 	};
9227 
9228 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9229 			  sizeof(*tte), 0,
9230 			  SOCKET_ID_ANY);
9231 	if (!tte)
9232 		goto err;
9233 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9234 			  &tte->flow_table);
9235 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9236 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9237 			tte->flow_table);
9238 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9239 				tte->flow_table);
9240 		goto err;
9241 	} else if (!tte->flow_table) {
9242 		goto err;
9243 	}
9244 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9245 	tte->tunnel_id = tbl.tunnel_id;
9246 	tte->group = tbl.group;
9247 	return &tte->hash;
9248 err:
9249 	if (tte)
9250 		mlx5_free(tte);
9251 	return NULL;
9252 }
9253 
9254 static struct mlx5_list_entry *
9255 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9256 				  struct mlx5_list_entry *oentry,
9257 				  void *cb_ctx __rte_unused)
9258 {
9259 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9260 						   0, SOCKET_ID_ANY);
9261 
9262 	if (!tte)
9263 		return NULL;
9264 	memcpy(tte, oentry, sizeof(*tte));
9265 	return &tte->hash;
9266 }
9267 
9268 static void
9269 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9270 				       struct mlx5_list_entry *entry)
9271 {
9272 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9273 
9274 	mlx5_free(tte);
9275 }
9276 
9277 static uint32_t
9278 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9279 				const struct mlx5_flow_tunnel *tunnel,
9280 				uint32_t group, uint32_t *table,
9281 				struct rte_flow_error *error)
9282 {
9283 	struct mlx5_list_entry *he;
9284 	struct tunnel_tbl_entry *tte;
9285 	union tunnel_tbl_key key = {
9286 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9287 		.group = group
9288 	};
9289 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9290 	struct mlx5_hlist *group_hash;
9291 	struct mlx5_flow_cb_ctx ctx = {
9292 		.data = &key.val,
9293 	};
9294 
9295 	group_hash = tunnel ? tunnel->groups : thub->groups;
9296 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9297 	if (!he)
9298 		return rte_flow_error_set(error, EINVAL,
9299 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9300 					  NULL,
9301 					  "tunnel group index not supported");
9302 	tte = container_of(he, typeof(*tte), hash);
9303 	*table = tte->flow_table;
9304 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9305 		dev->data->port_id, key.tunnel_id, group, *table);
9306 	return 0;
9307 }
9308 
9309 static void
9310 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9311 		      struct mlx5_flow_tunnel *tunnel)
9312 {
9313 	struct mlx5_priv *priv = dev->data->dev_private;
9314 	struct mlx5_indexed_pool *ipool;
9315 
9316 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9317 		dev->data->port_id, tunnel->tunnel_id);
9318 	LIST_REMOVE(tunnel, chain);
9319 	mlx5_hlist_destroy(tunnel->groups);
9320 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9321 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9322 }
9323 
9324 static bool
9325 mlx5_access_tunnel_offload_db
9326 	(struct rte_eth_dev *dev,
9327 	 bool (*match)(struct rte_eth_dev *,
9328 		       struct mlx5_flow_tunnel *, const void *),
9329 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9330 	 void (*miss)(struct rte_eth_dev *, void *),
9331 	 void *ctx, bool lock_op)
9332 {
9333 	bool verdict = false;
9334 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9335 	struct mlx5_flow_tunnel *tunnel;
9336 
9337 	rte_spinlock_lock(&thub->sl);
9338 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9339 		verdict = match(dev, tunnel, (const void *)ctx);
9340 		if (verdict)
9341 			break;
9342 	}
9343 	if (!lock_op)
9344 		rte_spinlock_unlock(&thub->sl);
9345 	if (verdict && hit)
9346 		hit(dev, tunnel, ctx);
9347 	if (!verdict && miss)
9348 		miss(dev, ctx);
9349 	if (lock_op)
9350 		rte_spinlock_unlock(&thub->sl);
9351 
9352 	return verdict;
9353 }
9354 
9355 struct tunnel_db_find_tunnel_id_ctx {
9356 	uint32_t tunnel_id;
9357 	struct mlx5_flow_tunnel *tunnel;
9358 };
9359 
9360 static bool
9361 find_tunnel_id_match(struct rte_eth_dev *dev,
9362 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9363 {
9364 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9365 
9366 	RTE_SET_USED(dev);
9367 	return tunnel->tunnel_id == ctx->tunnel_id;
9368 }
9369 
9370 static void
9371 find_tunnel_id_hit(struct rte_eth_dev *dev,
9372 		   struct mlx5_flow_tunnel *tunnel, void *x)
9373 {
9374 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9375 	RTE_SET_USED(dev);
9376 	ctx->tunnel = tunnel;
9377 }
9378 
9379 static struct mlx5_flow_tunnel *
9380 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9381 {
9382 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9383 		.tunnel_id = id,
9384 	};
9385 
9386 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9387 				      find_tunnel_id_hit, NULL, &ctx, true);
9388 
9389 	return ctx.tunnel;
9390 }
9391 
9392 static struct mlx5_flow_tunnel *
9393 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9394 			  const struct rte_flow_tunnel *app_tunnel)
9395 {
9396 	struct mlx5_priv *priv = dev->data->dev_private;
9397 	struct mlx5_indexed_pool *ipool;
9398 	struct mlx5_flow_tunnel *tunnel;
9399 	uint32_t id;
9400 
9401 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9402 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9403 	if (!tunnel)
9404 		return NULL;
9405 	if (id >= MLX5_MAX_TUNNELS) {
9406 		mlx5_ipool_free(ipool, id);
9407 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9408 		return NULL;
9409 	}
9410 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9411 					   priv->sh,
9412 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9413 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9414 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9415 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9416 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9417 	if (!tunnel->groups) {
9418 		mlx5_ipool_free(ipool, id);
9419 		return NULL;
9420 	}
9421 	/* initiate new PMD tunnel */
9422 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9423 	tunnel->tunnel_id = id;
9424 	tunnel->action.type = (typeof(tunnel->action.type))
9425 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9426 	tunnel->action.conf = tunnel;
9427 	tunnel->item.type = (typeof(tunnel->item.type))
9428 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9429 	tunnel->item.spec = tunnel;
9430 	tunnel->item.last = NULL;
9431 	tunnel->item.mask = NULL;
9432 
9433 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9434 		dev->data->port_id, tunnel->tunnel_id);
9435 
9436 	return tunnel;
9437 }
9438 
9439 struct tunnel_db_get_tunnel_ctx {
9440 	const struct rte_flow_tunnel *app_tunnel;
9441 	struct mlx5_flow_tunnel *tunnel;
9442 };
9443 
9444 static bool get_tunnel_match(struct rte_eth_dev *dev,
9445 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9446 {
9447 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9448 
9449 	RTE_SET_USED(dev);
9450 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9451 		       sizeof(*ctx->app_tunnel));
9452 }
9453 
9454 static void get_tunnel_hit(struct rte_eth_dev *dev,
9455 			   struct mlx5_flow_tunnel *tunnel, void *x)
9456 {
9457 	/* called under tunnel spinlock protection */
9458 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9459 
9460 	RTE_SET_USED(dev);
9461 	tunnel->refctn++;
9462 	ctx->tunnel = tunnel;
9463 }
9464 
9465 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9466 {
9467 	/* called under tunnel spinlock protection */
9468 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9469 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9470 
9471 	rte_spinlock_unlock(&thub->sl);
9472 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9473 	rte_spinlock_lock(&thub->sl);
9474 	if (ctx->tunnel) {
9475 		ctx->tunnel->refctn = 1;
9476 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9477 	}
9478 }
9479 
9480 
9481 static int
9482 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9483 		     const struct rte_flow_tunnel *app_tunnel,
9484 		     struct mlx5_flow_tunnel **tunnel)
9485 {
9486 	struct tunnel_db_get_tunnel_ctx ctx = {
9487 		.app_tunnel = app_tunnel,
9488 	};
9489 
9490 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9491 				      get_tunnel_miss, &ctx, true);
9492 	*tunnel = ctx.tunnel;
9493 	return ctx.tunnel ? 0 : -ENOMEM;
9494 }
9495 
9496 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9497 {
9498 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9499 
9500 	if (!thub)
9501 		return;
9502 	if (!LIST_EMPTY(&thub->tunnels))
9503 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9504 	mlx5_hlist_destroy(thub->groups);
9505 	mlx5_free(thub);
9506 }
9507 
9508 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9509 {
9510 	int err;
9511 	struct mlx5_flow_tunnel_hub *thub;
9512 
9513 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9514 			   0, SOCKET_ID_ANY);
9515 	if (!thub)
9516 		return -ENOMEM;
9517 	LIST_INIT(&thub->tunnels);
9518 	rte_spinlock_init(&thub->sl);
9519 	thub->groups = mlx5_hlist_create("flow groups", 64,
9520 					 false, true, sh,
9521 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9522 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9523 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9524 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9525 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9526 	if (!thub->groups) {
9527 		err = -rte_errno;
9528 		goto err;
9529 	}
9530 	sh->tunnel_hub = thub;
9531 
9532 	return 0;
9533 
9534 err:
9535 	if (thub->groups)
9536 		mlx5_hlist_destroy(thub->groups);
9537 	if (thub)
9538 		mlx5_free(thub);
9539 	return err;
9540 }
9541 
9542 static inline int
9543 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9544 			  struct rte_flow_tunnel *tunnel,
9545 			  struct rte_flow_error *error)
9546 {
9547 	struct mlx5_priv *priv = dev->data->dev_private;
9548 
9549 	if (!priv->config.dv_flow_en)
9550 		return rte_flow_error_set(error, ENOTSUP,
9551 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9552 					  "flow DV interface is off");
9553 	if (!is_tunnel_offload_active(dev))
9554 		return rte_flow_error_set(error, ENOTSUP,
9555 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9556 					  "tunnel offload was not activated");
9557 	if (!tunnel)
9558 		return rte_flow_error_set(error, EINVAL,
9559 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9560 					  "no application tunnel");
9561 	switch (tunnel->type) {
9562 	default:
9563 		return rte_flow_error_set(error, EINVAL,
9564 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9565 					  "unsupported tunnel type");
9566 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9567 	case RTE_FLOW_ITEM_TYPE_GRE:
9568 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9569 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9570 		break;
9571 	}
9572 	return 0;
9573 }
9574 
9575 static int
9576 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9577 		    struct rte_flow_tunnel *app_tunnel,
9578 		    struct rte_flow_action **actions,
9579 		    uint32_t *num_of_actions,
9580 		    struct rte_flow_error *error)
9581 {
9582 	struct mlx5_flow_tunnel *tunnel;
9583 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9584 
9585 	if (ret)
9586 		return ret;
9587 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9588 	if (ret < 0) {
9589 		return rte_flow_error_set(error, ret,
9590 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9591 					  "failed to initialize pmd tunnel");
9592 	}
9593 	*actions = &tunnel->action;
9594 	*num_of_actions = 1;
9595 	return 0;
9596 }
9597 
9598 static int
9599 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9600 		       struct rte_flow_tunnel *app_tunnel,
9601 		       struct rte_flow_item **items,
9602 		       uint32_t *num_of_items,
9603 		       struct rte_flow_error *error)
9604 {
9605 	struct mlx5_flow_tunnel *tunnel;
9606 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9607 
9608 	if (ret)
9609 		return ret;
9610 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9611 	if (ret < 0) {
9612 		return rte_flow_error_set(error, ret,
9613 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9614 					  "failed to initialize pmd tunnel");
9615 	}
9616 	*items = &tunnel->item;
9617 	*num_of_items = 1;
9618 	return 0;
9619 }
9620 
9621 struct tunnel_db_element_release_ctx {
9622 	struct rte_flow_item *items;
9623 	struct rte_flow_action *actions;
9624 	uint32_t num_elements;
9625 	struct rte_flow_error *error;
9626 	int ret;
9627 };
9628 
9629 static bool
9630 tunnel_element_release_match(struct rte_eth_dev *dev,
9631 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9632 {
9633 	const struct tunnel_db_element_release_ctx *ctx = x;
9634 
9635 	RTE_SET_USED(dev);
9636 	if (ctx->num_elements != 1)
9637 		return false;
9638 	else if (ctx->items)
9639 		return ctx->items == &tunnel->item;
9640 	else if (ctx->actions)
9641 		return ctx->actions == &tunnel->action;
9642 
9643 	return false;
9644 }
9645 
9646 static void
9647 tunnel_element_release_hit(struct rte_eth_dev *dev,
9648 			   struct mlx5_flow_tunnel *tunnel, void *x)
9649 {
9650 	struct tunnel_db_element_release_ctx *ctx = x;
9651 	ctx->ret = 0;
9652 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9653 		mlx5_flow_tunnel_free(dev, tunnel);
9654 }
9655 
9656 static void
9657 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9658 {
9659 	struct tunnel_db_element_release_ctx *ctx = x;
9660 	RTE_SET_USED(dev);
9661 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9662 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9663 				      "invalid argument");
9664 }
9665 
9666 static int
9667 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9668 		       struct rte_flow_item *pmd_items,
9669 		       uint32_t num_items, struct rte_flow_error *err)
9670 {
9671 	struct tunnel_db_element_release_ctx ctx = {
9672 		.items = pmd_items,
9673 		.actions = NULL,
9674 		.num_elements = num_items,
9675 		.error = err,
9676 	};
9677 
9678 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9679 				      tunnel_element_release_hit,
9680 				      tunnel_element_release_miss, &ctx, false);
9681 
9682 	return ctx.ret;
9683 }
9684 
9685 static int
9686 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9687 			 struct rte_flow_action *pmd_actions,
9688 			 uint32_t num_actions, struct rte_flow_error *err)
9689 {
9690 	struct tunnel_db_element_release_ctx ctx = {
9691 		.items = NULL,
9692 		.actions = pmd_actions,
9693 		.num_elements = num_actions,
9694 		.error = err,
9695 	};
9696 
9697 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9698 				      tunnel_element_release_hit,
9699 				      tunnel_element_release_miss, &ctx, false);
9700 
9701 	return ctx.ret;
9702 }
9703 
9704 static int
9705 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9706 				  struct rte_mbuf *m,
9707 				  struct rte_flow_restore_info *info,
9708 				  struct rte_flow_error *err)
9709 {
9710 	uint64_t ol_flags = m->ol_flags;
9711 	const struct mlx5_flow_tbl_data_entry *tble;
9712 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9713 
9714 	if (!is_tunnel_offload_active(dev)) {
9715 		info->flags = 0;
9716 		return 0;
9717 	}
9718 
9719 	if ((ol_flags & mask) != mask)
9720 		goto err;
9721 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9722 	if (!tble) {
9723 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9724 			dev->data->port_id, m->hash.fdir.hi);
9725 		goto err;
9726 	}
9727 	MLX5_ASSERT(tble->tunnel);
9728 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9729 	info->group_id = tble->group_id;
9730 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9731 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9732 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9733 
9734 	return 0;
9735 
9736 err:
9737 	return rte_flow_error_set(err, EINVAL,
9738 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9739 				  "failed to get restore info");
9740 }
9741 
9742 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9743 static int
9744 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9745 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9746 			   __rte_unused struct rte_flow_action **actions,
9747 			   __rte_unused uint32_t *num_of_actions,
9748 			   __rte_unused struct rte_flow_error *error)
9749 {
9750 	return -ENOTSUP;
9751 }
9752 
9753 static int
9754 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9755 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9756 		       __rte_unused struct rte_flow_item **items,
9757 		       __rte_unused uint32_t *num_of_items,
9758 		       __rte_unused struct rte_flow_error *error)
9759 {
9760 	return -ENOTSUP;
9761 }
9762 
9763 static int
9764 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9765 			      __rte_unused struct rte_flow_item *pmd_items,
9766 			      __rte_unused uint32_t num_items,
9767 			      __rte_unused struct rte_flow_error *err)
9768 {
9769 	return -ENOTSUP;
9770 }
9771 
9772 static int
9773 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9774 				__rte_unused struct rte_flow_action *pmd_action,
9775 				__rte_unused uint32_t num_actions,
9776 				__rte_unused struct rte_flow_error *err)
9777 {
9778 	return -ENOTSUP;
9779 }
9780 
9781 static int
9782 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9783 				  __rte_unused struct rte_mbuf *m,
9784 				  __rte_unused struct rte_flow_restore_info *i,
9785 				  __rte_unused struct rte_flow_error *err)
9786 {
9787 	return -ENOTSUP;
9788 }
9789 
9790 static int
9791 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9792 			     __rte_unused struct rte_flow *flow,
9793 			     __rte_unused const struct rte_flow_attr *attr,
9794 			     __rte_unused const struct rte_flow_action *actions,
9795 			     __rte_unused uint32_t flow_idx,
9796 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9797 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9798 			     __rte_unused struct rte_flow_error *error)
9799 {
9800 	return -ENOTSUP;
9801 }
9802 
9803 static struct mlx5_flow_tunnel *
9804 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9805 		    __rte_unused uint32_t id)
9806 {
9807 	return NULL;
9808 }
9809 
9810 static void
9811 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9812 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9813 {
9814 }
9815 
9816 static uint32_t
9817 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9818 				__rte_unused const struct mlx5_flow_tunnel *t,
9819 				__rte_unused uint32_t group,
9820 				__rte_unused uint32_t *table,
9821 				struct rte_flow_error *error)
9822 {
9823 	return rte_flow_error_set(error, ENOTSUP,
9824 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9825 				  "tunnel offload requires DV support");
9826 }
9827 
9828 void
9829 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9830 			__rte_unused  uint16_t port_id)
9831 {
9832 }
9833 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9834 
9835 /* Flex flow item API */
9836 static struct rte_flow_item_flex_handle *
9837 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9838 			   const struct rte_flow_item_flex_conf *conf,
9839 			   struct rte_flow_error *error)
9840 {
9841 	static const char err_msg[] = "flex item creation unsupported";
9842 	struct rte_flow_attr attr = { .transfer = 0 };
9843 	const struct mlx5_flow_driver_ops *fops =
9844 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9845 
9846 	if (!fops->item_create) {
9847 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9848 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9849 				   NULL, err_msg);
9850 		return NULL;
9851 	}
9852 	return fops->item_create(dev, conf, error);
9853 }
9854 
9855 static int
9856 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9857 			    const struct rte_flow_item_flex_handle *handle,
9858 			    struct rte_flow_error *error)
9859 {
9860 	static const char err_msg[] = "flex item release unsupported";
9861 	struct rte_flow_attr attr = { .transfer = 0 };
9862 	const struct mlx5_flow_driver_ops *fops =
9863 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9864 
9865 	if (!fops->item_release) {
9866 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9867 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9868 				   NULL, err_msg);
9869 		return -rte_errno;
9870 	}
9871 	return fops->item_release(dev, handle, error);
9872 }
9873 
9874 static void
9875 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9876 {
9877 	int ret;
9878 	struct rte_flow_error error;
9879 
9880 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9881 		char *item_name;
9882 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9883 				    sizeof(item_name),
9884 				    (void *)(uintptr_t)item->type, &error);
9885 		if (ret > 0)
9886 			printf("%s ", item_name);
9887 		else
9888 			printf("%d\n", (int)item->type);
9889 	}
9890 	printf("END\n");
9891 }
9892 
9893 static int
9894 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9895 {
9896 	const struct rte_flow_item_udp *spec = udp_item->spec;
9897 	const struct rte_flow_item_udp *mask = udp_item->mask;
9898 	uint16_t udp_dport = 0;
9899 
9900 	if (spec != NULL) {
9901 		if (!mask)
9902 			mask = &rte_flow_item_udp_mask;
9903 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9904 				mask->hdr.dst_port);
9905 	}
9906 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9907 }
9908 
9909 static const struct mlx5_flow_expand_node *
9910 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9911 		unsigned int item_idx,
9912 		const struct mlx5_flow_expand_node graph[],
9913 		const struct mlx5_flow_expand_node *node)
9914 {
9915 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9916 
9917 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9918 			node != NULL &&
9919 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9920 		/*
9921 		 * The expansion node is VXLAN and it is also the last
9922 		 * expandable item in the pattern, so need to continue
9923 		 * expansion of the inner tunnel.
9924 		 */
9925 		MLX5_ASSERT(item_idx > 0);
9926 		prev_item = pattern + item_idx - 1;
9927 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9928 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9929 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9930 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9931 	}
9932 	return node;
9933 }
9934 
9935 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9936 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9937 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9938 };
9939 
9940 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9941 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9942 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9943 	{ 9, 10, 11 }, { 12, 13, 14 },
9944 };
9945 
9946 /**
9947  * Discover the number of available flow priorities.
9948  *
9949  * @param dev
9950  *   Ethernet device.
9951  *
9952  * @return
9953  *   On success, number of available flow priorities.
9954  *   On failure, a negative errno-style code and rte_errno is set.
9955  */
9956 int
9957 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9958 {
9959 	static const uint16_t vprio[] = {8, 16};
9960 	const struct mlx5_priv *priv = dev->data->dev_private;
9961 	const struct mlx5_flow_driver_ops *fops;
9962 	enum mlx5_flow_drv_type type;
9963 	int ret;
9964 
9965 	type = mlx5_flow_os_get_type();
9966 	if (type == MLX5_FLOW_TYPE_MAX) {
9967 		type = MLX5_FLOW_TYPE_VERBS;
9968 		if (priv->sh->devx && priv->config.dv_flow_en)
9969 			type = MLX5_FLOW_TYPE_DV;
9970 	}
9971 	fops = flow_get_drv_ops(type);
9972 	if (fops->discover_priorities == NULL) {
9973 		DRV_LOG(ERR, "Priority discovery not supported");
9974 		rte_errno = ENOTSUP;
9975 		return -rte_errno;
9976 	}
9977 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
9978 	if (ret < 0)
9979 		return ret;
9980 	switch (ret) {
9981 	case 8:
9982 		ret = RTE_DIM(priority_map_3);
9983 		break;
9984 	case 16:
9985 		ret = RTE_DIM(priority_map_5);
9986 		break;
9987 	default:
9988 		rte_errno = ENOTSUP;
9989 		DRV_LOG(ERR,
9990 			"port %u maximum priority: %d expected 8/16",
9991 			dev->data->port_id, ret);
9992 		return -rte_errno;
9993 	}
9994 	DRV_LOG(INFO, "port %u supported flow priorities:"
9995 		" 0-%d for ingress or egress root table,"
9996 		" 0-%d for non-root table or transfer root table.",
9997 		dev->data->port_id, ret - 2,
9998 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
9999 	return ret;
10000 }
10001 
10002 /**
10003  * Adjust flow priority based on the highest layer and the request priority.
10004  *
10005  * @param[in] dev
10006  *   Pointer to the Ethernet device structure.
10007  * @param[in] priority
10008  *   The rule base priority.
10009  * @param[in] subpriority
10010  *   The priority based on the items.
10011  *
10012  * @return
10013  *   The new priority.
10014  */
10015 uint32_t
10016 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10017 			  uint32_t subpriority)
10018 {
10019 	uint32_t res = 0;
10020 	struct mlx5_priv *priv = dev->data->dev_private;
10021 
10022 	switch (priv->sh->flow_max_priority) {
10023 	case RTE_DIM(priority_map_3):
10024 		res = priority_map_3[priority][subpriority];
10025 		break;
10026 	case RTE_DIM(priority_map_5):
10027 		res = priority_map_5[priority][subpriority];
10028 		break;
10029 	}
10030 	return  res;
10031 }
10032