xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 8f393c4ffdc1ff9b46702708781723ca0f17f5ac)
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 ETH_RSS_* definitions).
102 	 */
103 	uint8_t optional;
104 	/**< optional expand field. Default 0 to expand, 1 not go deeper. */
105 };
106 
107 /** Object returned by mlx5_flow_expand_rss(). */
108 struct mlx5_flow_expand_rss {
109 	uint32_t entries;
110 	/**< Number of entries @p patterns and @p priorities. */
111 	struct {
112 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
113 		uint32_t priority; /**< Priority offset for each expansion. */
114 	} entry[];
115 };
116 
117 static void
118 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
119 
120 static bool
121 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
122 {
123 	switch (item->type) {
124 	case RTE_FLOW_ITEM_TYPE_ETH:
125 	case RTE_FLOW_ITEM_TYPE_VLAN:
126 	case RTE_FLOW_ITEM_TYPE_IPV4:
127 	case RTE_FLOW_ITEM_TYPE_IPV6:
128 	case RTE_FLOW_ITEM_TYPE_UDP:
129 	case RTE_FLOW_ITEM_TYPE_TCP:
130 	case RTE_FLOW_ITEM_TYPE_VXLAN:
131 	case RTE_FLOW_ITEM_TYPE_NVGRE:
132 	case RTE_FLOW_ITEM_TYPE_GRE:
133 	case RTE_FLOW_ITEM_TYPE_GENEVE:
134 	case RTE_FLOW_ITEM_TYPE_MPLS:
135 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
136 	case RTE_FLOW_ITEM_TYPE_GRE_KEY:
137 	case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
138 		return true;
139 	default:
140 		break;
141 	}
142 	return false;
143 }
144 
145 static enum rte_flow_item_type
146 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
147 {
148 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
149 	uint16_t ether_type = 0;
150 	uint16_t ether_type_m;
151 	uint8_t ip_next_proto = 0;
152 	uint8_t ip_next_proto_m;
153 
154 	if (item == NULL || item->spec == NULL)
155 		return ret;
156 	switch (item->type) {
157 	case RTE_FLOW_ITEM_TYPE_ETH:
158 		if (item->mask)
159 			ether_type_m = ((const struct rte_flow_item_eth *)
160 						(item->mask))->type;
161 		else
162 			ether_type_m = rte_flow_item_eth_mask.type;
163 		if (ether_type_m != RTE_BE16(0xFFFF))
164 			break;
165 		ether_type = ((const struct rte_flow_item_eth *)
166 				(item->spec))->type;
167 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
168 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
169 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
170 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
171 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
172 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
173 		else
174 			ret = RTE_FLOW_ITEM_TYPE_END;
175 		break;
176 	case RTE_FLOW_ITEM_TYPE_VLAN:
177 		if (item->mask)
178 			ether_type_m = ((const struct rte_flow_item_vlan *)
179 						(item->mask))->inner_type;
180 		else
181 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
182 		if (ether_type_m != RTE_BE16(0xFFFF))
183 			break;
184 		ether_type = ((const struct rte_flow_item_vlan *)
185 				(item->spec))->inner_type;
186 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
187 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
188 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
189 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
190 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
191 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
192 		else
193 			ret = RTE_FLOW_ITEM_TYPE_END;
194 		break;
195 	case RTE_FLOW_ITEM_TYPE_IPV4:
196 		if (item->mask)
197 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
198 					(item->mask))->hdr.next_proto_id;
199 		else
200 			ip_next_proto_m =
201 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
202 		if (ip_next_proto_m != 0xFF)
203 			break;
204 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
205 				(item->spec))->hdr.next_proto_id;
206 		if (ip_next_proto == IPPROTO_UDP)
207 			ret = RTE_FLOW_ITEM_TYPE_UDP;
208 		else if (ip_next_proto == IPPROTO_TCP)
209 			ret = RTE_FLOW_ITEM_TYPE_TCP;
210 		else if (ip_next_proto == IPPROTO_IP)
211 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
212 		else if (ip_next_proto == IPPROTO_IPV6)
213 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
214 		else
215 			ret = RTE_FLOW_ITEM_TYPE_END;
216 		break;
217 	case RTE_FLOW_ITEM_TYPE_IPV6:
218 		if (item->mask)
219 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
220 						(item->mask))->hdr.proto;
221 		else
222 			ip_next_proto_m =
223 				rte_flow_item_ipv6_mask.hdr.proto;
224 		if (ip_next_proto_m != 0xFF)
225 			break;
226 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
227 				(item->spec))->hdr.proto;
228 		if (ip_next_proto == IPPROTO_UDP)
229 			ret = RTE_FLOW_ITEM_TYPE_UDP;
230 		else if (ip_next_proto == IPPROTO_TCP)
231 			ret = RTE_FLOW_ITEM_TYPE_TCP;
232 		else if (ip_next_proto == IPPROTO_IP)
233 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
234 		else if (ip_next_proto == IPPROTO_IPV6)
235 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
236 		else
237 			ret = RTE_FLOW_ITEM_TYPE_END;
238 		break;
239 	default:
240 		ret = RTE_FLOW_ITEM_TYPE_VOID;
241 		break;
242 	}
243 	return ret;
244 }
245 
246 #define MLX5_RSS_EXP_ELT_N 16
247 
248 /**
249  * Expand RSS flows into several possible flows according to the RSS hash
250  * fields requested and the driver capabilities.
251  *
252  * @param[out] buf
253  *   Buffer to store the result expansion.
254  * @param[in] size
255  *   Buffer size in bytes. If 0, @p buf can be NULL.
256  * @param[in] pattern
257  *   User flow pattern.
258  * @param[in] types
259  *   RSS types to expand (see ETH_RSS_* definitions).
260  * @param[in] graph
261  *   Input graph to expand @p pattern according to @p types.
262  * @param[in] graph_root_index
263  *   Index of root node in @p graph, typically 0.
264  *
265  * @return
266  *   A positive value representing the size of @p buf in bytes regardless of
267  *   @p size on success, a negative errno value otherwise and rte_errno is
268  *   set, the following errors are defined:
269  *
270  *   -E2BIG: graph-depth @p graph is too deep.
271  *   -EINVAL: @p size has not enough space for expanded pattern.
272  */
273 static int
274 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
275 		     const struct rte_flow_item *pattern, uint64_t types,
276 		     const struct mlx5_flow_expand_node graph[],
277 		     int graph_root_index)
278 {
279 	const struct rte_flow_item *item;
280 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
281 	const int *next_node;
282 	const int *stack[MLX5_RSS_EXP_ELT_N];
283 	int stack_pos = 0;
284 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
285 	unsigned int i;
286 	size_t lsize;
287 	size_t user_pattern_size = 0;
288 	void *addr = NULL;
289 	const struct mlx5_flow_expand_node *next = NULL;
290 	struct rte_flow_item missed_item;
291 	int missed = 0;
292 	int elt = 0;
293 	const struct rte_flow_item *last_item = NULL;
294 
295 	memset(&missed_item, 0, sizeof(missed_item));
296 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
297 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
298 	if (lsize > size)
299 		return -EINVAL;
300 	buf->entry[0].priority = 0;
301 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
302 	buf->entries = 0;
303 	addr = buf->entry[0].pattern;
304 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
305 		if (!mlx5_flow_is_rss_expandable_item(item)) {
306 			user_pattern_size += sizeof(*item);
307 			continue;
308 		}
309 		last_item = item;
310 		for (i = 0; node->next && node->next[i]; ++i) {
311 			next = &graph[node->next[i]];
312 			if (next->type == item->type)
313 				break;
314 		}
315 		if (next)
316 			node = next;
317 		user_pattern_size += sizeof(*item);
318 	}
319 	user_pattern_size += sizeof(*item); /* Handle END item. */
320 	lsize += user_pattern_size;
321 	if (lsize > size)
322 		return -EINVAL;
323 	/* Copy the user pattern in the first entry of the buffer. */
324 	rte_memcpy(addr, pattern, user_pattern_size);
325 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
326 	buf->entries = 1;
327 	/* Start expanding. */
328 	memset(flow_items, 0, sizeof(flow_items));
329 	user_pattern_size -= sizeof(*item);
330 	/*
331 	 * Check if the last valid item has spec set, need complete pattern,
332 	 * and the pattern can be used for expansion.
333 	 */
334 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
335 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
336 		/* Item type END indicates expansion is not required. */
337 		return lsize;
338 	}
339 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
340 		next = NULL;
341 		missed = 1;
342 		for (i = 0; node->next && node->next[i]; ++i) {
343 			next = &graph[node->next[i]];
344 			if (next->type == missed_item.type) {
345 				flow_items[0].type = missed_item.type;
346 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
347 				break;
348 			}
349 			next = NULL;
350 		}
351 	}
352 	if (next && missed) {
353 		elt = 2; /* missed item + item end. */
354 		node = next;
355 		lsize += elt * sizeof(*item) + user_pattern_size;
356 		if (lsize > size)
357 			return -EINVAL;
358 		if (node->rss_types & types) {
359 			buf->entry[buf->entries].priority = 1;
360 			buf->entry[buf->entries].pattern = addr;
361 			buf->entries++;
362 			rte_memcpy(addr, buf->entry[0].pattern,
363 				   user_pattern_size);
364 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
365 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
366 			addr = (void *)(((uintptr_t)addr) +
367 					elt * sizeof(*item));
368 		}
369 	}
370 	memset(flow_items, 0, sizeof(flow_items));
371 	next_node = node->next;
372 	stack[stack_pos] = next_node;
373 	node = next_node ? &graph[*next_node] : NULL;
374 	while (node) {
375 		flow_items[stack_pos].type = node->type;
376 		if (node->rss_types & types) {
377 			size_t n;
378 			/*
379 			 * compute the number of items to copy from the
380 			 * expansion and copy it.
381 			 * When the stack_pos is 0, there are 1 element in it,
382 			 * plus the addition END item.
383 			 */
384 			elt = stack_pos + 2;
385 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
386 			lsize += elt * sizeof(*item) + user_pattern_size;
387 			if (lsize > size)
388 				return -EINVAL;
389 			n = elt * sizeof(*item);
390 			buf->entry[buf->entries].priority =
391 				stack_pos + 1 + missed;
392 			buf->entry[buf->entries].pattern = addr;
393 			buf->entries++;
394 			rte_memcpy(addr, buf->entry[0].pattern,
395 				   user_pattern_size);
396 			addr = (void *)(((uintptr_t)addr) +
397 					user_pattern_size);
398 			rte_memcpy(addr, &missed_item,
399 				   missed * sizeof(*item));
400 			addr = (void *)(((uintptr_t)addr) +
401 				missed * sizeof(*item));
402 			rte_memcpy(addr, flow_items, n);
403 			addr = (void *)(((uintptr_t)addr) + n);
404 		}
405 		/* Go deeper. */
406 		if (!node->optional && node->next) {
407 			next_node = node->next;
408 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
409 				rte_errno = E2BIG;
410 				return -rte_errno;
411 			}
412 			stack[stack_pos] = next_node;
413 		} else if (*(next_node + 1)) {
414 			/* Follow up with the next possibility. */
415 			++next_node;
416 		} else {
417 			/* Move to the next path. */
418 			if (stack_pos)
419 				next_node = stack[--stack_pos];
420 			next_node++;
421 			stack[stack_pos] = next_node;
422 		}
423 		node = *next_node ? &graph[*next_node] : NULL;
424 	};
425 	return lsize;
426 }
427 
428 enum mlx5_expansion {
429 	MLX5_EXPANSION_ROOT,
430 	MLX5_EXPANSION_ROOT_OUTER,
431 	MLX5_EXPANSION_ROOT_ETH_VLAN,
432 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
433 	MLX5_EXPANSION_OUTER_ETH,
434 	MLX5_EXPANSION_OUTER_ETH_VLAN,
435 	MLX5_EXPANSION_OUTER_VLAN,
436 	MLX5_EXPANSION_OUTER_IPV4,
437 	MLX5_EXPANSION_OUTER_IPV4_UDP,
438 	MLX5_EXPANSION_OUTER_IPV4_TCP,
439 	MLX5_EXPANSION_OUTER_IPV6,
440 	MLX5_EXPANSION_OUTER_IPV6_UDP,
441 	MLX5_EXPANSION_OUTER_IPV6_TCP,
442 	MLX5_EXPANSION_VXLAN,
443 	MLX5_EXPANSION_VXLAN_GPE,
444 	MLX5_EXPANSION_GRE,
445 	MLX5_EXPANSION_NVGRE,
446 	MLX5_EXPANSION_GRE_KEY,
447 	MLX5_EXPANSION_MPLS,
448 	MLX5_EXPANSION_ETH,
449 	MLX5_EXPANSION_ETH_VLAN,
450 	MLX5_EXPANSION_VLAN,
451 	MLX5_EXPANSION_IPV4,
452 	MLX5_EXPANSION_IPV4_UDP,
453 	MLX5_EXPANSION_IPV4_TCP,
454 	MLX5_EXPANSION_IPV6,
455 	MLX5_EXPANSION_IPV6_UDP,
456 	MLX5_EXPANSION_IPV6_TCP,
457 	MLX5_EXPANSION_IPV6_FRAG_EXT,
458 };
459 
460 /** Supported expansion of items. */
461 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
462 	[MLX5_EXPANSION_ROOT] = {
463 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
464 						  MLX5_EXPANSION_IPV4,
465 						  MLX5_EXPANSION_IPV6),
466 		.type = RTE_FLOW_ITEM_TYPE_END,
467 	},
468 	[MLX5_EXPANSION_ROOT_OUTER] = {
469 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
470 						  MLX5_EXPANSION_OUTER_IPV4,
471 						  MLX5_EXPANSION_OUTER_IPV6),
472 		.type = RTE_FLOW_ITEM_TYPE_END,
473 	},
474 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
475 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
476 		.type = RTE_FLOW_ITEM_TYPE_END,
477 	},
478 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
479 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
480 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
481 		.type = RTE_FLOW_ITEM_TYPE_END,
482 	},
483 	[MLX5_EXPANSION_OUTER_ETH] = {
484 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
485 						  MLX5_EXPANSION_OUTER_IPV6),
486 		.type = RTE_FLOW_ITEM_TYPE_ETH,
487 		.rss_types = 0,
488 	},
489 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
490 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
491 		.type = RTE_FLOW_ITEM_TYPE_ETH,
492 		.rss_types = 0,
493 	},
494 	[MLX5_EXPANSION_OUTER_VLAN] = {
495 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
496 						  MLX5_EXPANSION_OUTER_IPV6),
497 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
498 	},
499 	[MLX5_EXPANSION_OUTER_IPV4] = {
500 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
501 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
502 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
503 			 MLX5_EXPANSION_GRE,
504 			 MLX5_EXPANSION_NVGRE,
505 			 MLX5_EXPANSION_IPV4,
506 			 MLX5_EXPANSION_IPV6),
507 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
508 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
509 			ETH_RSS_NONFRAG_IPV4_OTHER,
510 	},
511 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
512 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
513 						  MLX5_EXPANSION_VXLAN_GPE,
514 						  MLX5_EXPANSION_MPLS),
515 		.type = RTE_FLOW_ITEM_TYPE_UDP,
516 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
517 	},
518 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
519 		.type = RTE_FLOW_ITEM_TYPE_TCP,
520 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
521 	},
522 	[MLX5_EXPANSION_OUTER_IPV6] = {
523 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
524 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
525 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
526 			 MLX5_EXPANSION_IPV4,
527 			 MLX5_EXPANSION_IPV6,
528 			 MLX5_EXPANSION_GRE,
529 			 MLX5_EXPANSION_NVGRE),
530 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
531 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
532 			ETH_RSS_NONFRAG_IPV6_OTHER,
533 	},
534 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
535 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
536 						  MLX5_EXPANSION_VXLAN_GPE,
537 						  MLX5_EXPANSION_MPLS),
538 		.type = RTE_FLOW_ITEM_TYPE_UDP,
539 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
540 	},
541 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
542 		.type = RTE_FLOW_ITEM_TYPE_TCP,
543 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
544 	},
545 	[MLX5_EXPANSION_VXLAN] = {
546 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
547 						  MLX5_EXPANSION_IPV4,
548 						  MLX5_EXPANSION_IPV6),
549 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
550 	},
551 	[MLX5_EXPANSION_VXLAN_GPE] = {
552 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
553 						  MLX5_EXPANSION_IPV4,
554 						  MLX5_EXPANSION_IPV6),
555 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
556 	},
557 	[MLX5_EXPANSION_GRE] = {
558 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
559 						  MLX5_EXPANSION_IPV6,
560 						  MLX5_EXPANSION_GRE_KEY,
561 						  MLX5_EXPANSION_MPLS),
562 		.type = RTE_FLOW_ITEM_TYPE_GRE,
563 	},
564 	[MLX5_EXPANSION_GRE_KEY] = {
565 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
566 						  MLX5_EXPANSION_IPV6,
567 						  MLX5_EXPANSION_MPLS),
568 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
569 		.optional = 1,
570 	},
571 	[MLX5_EXPANSION_NVGRE] = {
572 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
573 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
574 	},
575 	[MLX5_EXPANSION_MPLS] = {
576 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
577 						  MLX5_EXPANSION_IPV6,
578 						  MLX5_EXPANSION_ETH),
579 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
580 	},
581 	[MLX5_EXPANSION_ETH] = {
582 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
583 						  MLX5_EXPANSION_IPV6),
584 		.type = RTE_FLOW_ITEM_TYPE_ETH,
585 	},
586 	[MLX5_EXPANSION_ETH_VLAN] = {
587 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
588 		.type = RTE_FLOW_ITEM_TYPE_ETH,
589 	},
590 	[MLX5_EXPANSION_VLAN] = {
591 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
592 						  MLX5_EXPANSION_IPV6),
593 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
594 	},
595 	[MLX5_EXPANSION_IPV4] = {
596 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
597 						  MLX5_EXPANSION_IPV4_TCP),
598 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
599 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
600 			ETH_RSS_NONFRAG_IPV4_OTHER,
601 	},
602 	[MLX5_EXPANSION_IPV4_UDP] = {
603 		.type = RTE_FLOW_ITEM_TYPE_UDP,
604 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
605 	},
606 	[MLX5_EXPANSION_IPV4_TCP] = {
607 		.type = RTE_FLOW_ITEM_TYPE_TCP,
608 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
609 	},
610 	[MLX5_EXPANSION_IPV6] = {
611 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
612 						  MLX5_EXPANSION_IPV6_TCP,
613 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
614 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
615 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
616 			ETH_RSS_NONFRAG_IPV6_OTHER,
617 	},
618 	[MLX5_EXPANSION_IPV6_UDP] = {
619 		.type = RTE_FLOW_ITEM_TYPE_UDP,
620 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
621 	},
622 	[MLX5_EXPANSION_IPV6_TCP] = {
623 		.type = RTE_FLOW_ITEM_TYPE_TCP,
624 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
625 	},
626 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
627 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
628 	},
629 };
630 
631 static struct rte_flow_action_handle *
632 mlx5_action_handle_create(struct rte_eth_dev *dev,
633 			  const struct rte_flow_indir_action_conf *conf,
634 			  const struct rte_flow_action *action,
635 			  struct rte_flow_error *error);
636 static int mlx5_action_handle_destroy
637 				(struct rte_eth_dev *dev,
638 				 struct rte_flow_action_handle *handle,
639 				 struct rte_flow_error *error);
640 static int mlx5_action_handle_update
641 				(struct rte_eth_dev *dev,
642 				 struct rte_flow_action_handle *handle,
643 				 const void *update,
644 				 struct rte_flow_error *error);
645 static int mlx5_action_handle_query
646 				(struct rte_eth_dev *dev,
647 				 const struct rte_flow_action_handle *handle,
648 				 void *data,
649 				 struct rte_flow_error *error);
650 static int
651 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
652 		    struct rte_flow_tunnel *app_tunnel,
653 		    struct rte_flow_action **actions,
654 		    uint32_t *num_of_actions,
655 		    struct rte_flow_error *error);
656 static int
657 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
658 		       struct rte_flow_tunnel *app_tunnel,
659 		       struct rte_flow_item **items,
660 		       uint32_t *num_of_items,
661 		       struct rte_flow_error *error);
662 static int
663 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
664 			      struct rte_flow_item *pmd_items,
665 			      uint32_t num_items, struct rte_flow_error *err);
666 static int
667 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
668 				struct rte_flow_action *pmd_actions,
669 				uint32_t num_actions,
670 				struct rte_flow_error *err);
671 static int
672 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
673 				  struct rte_mbuf *m,
674 				  struct rte_flow_restore_info *info,
675 				  struct rte_flow_error *err);
676 
677 static const struct rte_flow_ops mlx5_flow_ops = {
678 	.validate = mlx5_flow_validate,
679 	.create = mlx5_flow_create,
680 	.destroy = mlx5_flow_destroy,
681 	.flush = mlx5_flow_flush,
682 	.isolate = mlx5_flow_isolate,
683 	.query = mlx5_flow_query,
684 	.dev_dump = mlx5_flow_dev_dump,
685 	.get_aged_flows = mlx5_flow_get_aged_flows,
686 	.action_handle_create = mlx5_action_handle_create,
687 	.action_handle_destroy = mlx5_action_handle_destroy,
688 	.action_handle_update = mlx5_action_handle_update,
689 	.action_handle_query = mlx5_action_handle_query,
690 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
691 	.tunnel_match = mlx5_flow_tunnel_match,
692 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
693 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
694 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
695 };
696 
697 /* Tunnel information. */
698 struct mlx5_flow_tunnel_info {
699 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
700 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
701 };
702 
703 static struct mlx5_flow_tunnel_info tunnels_info[] = {
704 	{
705 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
706 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
707 	},
708 	{
709 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
710 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
711 	},
712 	{
713 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
714 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
715 	},
716 	{
717 		.tunnel = MLX5_FLOW_LAYER_GRE,
718 		.ptype = RTE_PTYPE_TUNNEL_GRE,
719 	},
720 	{
721 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
722 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
723 	},
724 	{
725 		.tunnel = MLX5_FLOW_LAYER_MPLS,
726 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
727 	},
728 	{
729 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
730 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
731 	},
732 	{
733 		.tunnel = MLX5_FLOW_LAYER_IPIP,
734 		.ptype = RTE_PTYPE_TUNNEL_IP,
735 	},
736 	{
737 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
738 		.ptype = RTE_PTYPE_TUNNEL_IP,
739 	},
740 	{
741 		.tunnel = MLX5_FLOW_LAYER_GTP,
742 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
743 	},
744 };
745 
746 
747 
748 /**
749  * Translate tag ID to register.
750  *
751  * @param[in] dev
752  *   Pointer to the Ethernet device structure.
753  * @param[in] feature
754  *   The feature that request the register.
755  * @param[in] id
756  *   The request register ID.
757  * @param[out] error
758  *   Error description in case of any.
759  *
760  * @return
761  *   The request register on success, a negative errno
762  *   value otherwise and rte_errno is set.
763  */
764 int
765 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
766 		     enum mlx5_feature_name feature,
767 		     uint32_t id,
768 		     struct rte_flow_error *error)
769 {
770 	struct mlx5_priv *priv = dev->data->dev_private;
771 	struct mlx5_dev_config *config = &priv->config;
772 	enum modify_reg start_reg;
773 	bool skip_mtr_reg = false;
774 
775 	switch (feature) {
776 	case MLX5_HAIRPIN_RX:
777 		return REG_B;
778 	case MLX5_HAIRPIN_TX:
779 		return REG_A;
780 	case MLX5_METADATA_RX:
781 		switch (config->dv_xmeta_en) {
782 		case MLX5_XMETA_MODE_LEGACY:
783 			return REG_B;
784 		case MLX5_XMETA_MODE_META16:
785 			return REG_C_0;
786 		case MLX5_XMETA_MODE_META32:
787 			return REG_C_1;
788 		}
789 		break;
790 	case MLX5_METADATA_TX:
791 		return REG_A;
792 	case MLX5_METADATA_FDB:
793 		switch (config->dv_xmeta_en) {
794 		case MLX5_XMETA_MODE_LEGACY:
795 			return REG_NON;
796 		case MLX5_XMETA_MODE_META16:
797 			return REG_C_0;
798 		case MLX5_XMETA_MODE_META32:
799 			return REG_C_1;
800 		}
801 		break;
802 	case MLX5_FLOW_MARK:
803 		switch (config->dv_xmeta_en) {
804 		case MLX5_XMETA_MODE_LEGACY:
805 			return REG_NON;
806 		case MLX5_XMETA_MODE_META16:
807 			return REG_C_1;
808 		case MLX5_XMETA_MODE_META32:
809 			return REG_C_0;
810 		}
811 		break;
812 	case MLX5_MTR_ID:
813 		/*
814 		 * If meter color and meter id share one register, flow match
815 		 * should use the meter color register for match.
816 		 */
817 		if (priv->mtr_reg_share)
818 			return priv->mtr_color_reg;
819 		else
820 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
821 			       REG_C_3;
822 	case MLX5_MTR_COLOR:
823 	case MLX5_ASO_FLOW_HIT:
824 	case MLX5_ASO_CONNTRACK:
825 		/* All features use the same REG_C. */
826 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
827 		return priv->mtr_color_reg;
828 	case MLX5_COPY_MARK:
829 		/*
830 		 * Metadata COPY_MARK register using is in meter suffix sub
831 		 * flow while with meter. It's safe to share the same register.
832 		 */
833 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
834 	case MLX5_APP_TAG:
835 		/*
836 		 * If meter is enable, it will engage the register for color
837 		 * match and flow match. If meter color match is not using the
838 		 * REG_C_2, need to skip the REG_C_x be used by meter color
839 		 * match.
840 		 * If meter is disable, free to use all available registers.
841 		 */
842 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
843 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
844 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
845 		if (id > (uint32_t)(REG_C_7 - start_reg))
846 			return rte_flow_error_set(error, EINVAL,
847 						  RTE_FLOW_ERROR_TYPE_ITEM,
848 						  NULL, "invalid tag id");
849 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
850 			return rte_flow_error_set(error, ENOTSUP,
851 						  RTE_FLOW_ERROR_TYPE_ITEM,
852 						  NULL, "unsupported tag id");
853 		/*
854 		 * This case means meter is using the REG_C_x great than 2.
855 		 * Take care not to conflict with meter color REG_C_x.
856 		 * If the available index REG_C_y >= REG_C_x, skip the
857 		 * color register.
858 		 */
859 		if (skip_mtr_reg && config->flow_mreg_c
860 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
861 			if (id >= (uint32_t)(REG_C_7 - start_reg))
862 				return rte_flow_error_set(error, EINVAL,
863 						       RTE_FLOW_ERROR_TYPE_ITEM,
864 							NULL, "invalid tag id");
865 			if (config->flow_mreg_c
866 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
867 				return config->flow_mreg_c
868 					       [id + 1 + start_reg - REG_C_0];
869 			return rte_flow_error_set(error, ENOTSUP,
870 						  RTE_FLOW_ERROR_TYPE_ITEM,
871 						  NULL, "unsupported tag id");
872 		}
873 		return config->flow_mreg_c[id + start_reg - REG_C_0];
874 	}
875 	MLX5_ASSERT(false);
876 	return rte_flow_error_set(error, EINVAL,
877 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
878 				  NULL, "invalid feature name");
879 }
880 
881 /**
882  * Check extensive flow metadata register support.
883  *
884  * @param dev
885  *   Pointer to rte_eth_dev structure.
886  *
887  * @return
888  *   True if device supports extensive flow metadata register, otherwise false.
889  */
890 bool
891 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
892 {
893 	struct mlx5_priv *priv = dev->data->dev_private;
894 	struct mlx5_dev_config *config = &priv->config;
895 
896 	/*
897 	 * Having available reg_c can be regarded inclusively as supporting
898 	 * extensive flow metadata register, which could mean,
899 	 * - metadata register copy action by modify header.
900 	 * - 16 modify header actions is supported.
901 	 * - reg_c's are preserved across different domain (FDB and NIC) on
902 	 *   packet loopback by flow lookup miss.
903 	 */
904 	return config->flow_mreg_c[2] != REG_NON;
905 }
906 
907 /**
908  * Get the lowest priority.
909  *
910  * @param[in] dev
911  *   Pointer to the Ethernet device structure.
912  * @param[in] attributes
913  *   Pointer to device flow rule attributes.
914  *
915  * @return
916  *   The value of lowest priority of flow.
917  */
918 uint32_t
919 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
920 			  const struct rte_flow_attr *attr)
921 {
922 	struct mlx5_priv *priv = dev->data->dev_private;
923 
924 	if (!attr->group && !attr->transfer)
925 		return priv->config.flow_prio - 2;
926 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
927 }
928 
929 /**
930  * Calculate matcher priority of the flow.
931  *
932  * @param[in] dev
933  *   Pointer to the Ethernet device structure.
934  * @param[in] attr
935  *   Pointer to device flow rule attributes.
936  * @param[in] subpriority
937  *   The priority based on the items.
938  * @return
939  *   The matcher priority of the flow.
940  */
941 uint16_t
942 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
943 			  const struct rte_flow_attr *attr,
944 			  uint32_t subpriority)
945 {
946 	uint16_t priority = (uint16_t)attr->priority;
947 	struct mlx5_priv *priv = dev->data->dev_private;
948 
949 	if (!attr->group && !attr->transfer) {
950 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
951 			priority = priv->config.flow_prio - 1;
952 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
953 	}
954 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
955 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
956 	return priority * 3 + subpriority;
957 }
958 
959 /**
960  * Verify the @p item specifications (spec, last, mask) are compatible with the
961  * NIC capabilities.
962  *
963  * @param[in] item
964  *   Item specification.
965  * @param[in] mask
966  *   @p item->mask or flow default bit-masks.
967  * @param[in] nic_mask
968  *   Bit-masks covering supported fields by the NIC to compare with user mask.
969  * @param[in] size
970  *   Bit-masks size in bytes.
971  * @param[in] range_accepted
972  *   True if range of values is accepted for specific fields, false otherwise.
973  * @param[out] error
974  *   Pointer to error structure.
975  *
976  * @return
977  *   0 on success, a negative errno value otherwise and rte_errno is set.
978  */
979 int
980 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
981 			  const uint8_t *mask,
982 			  const uint8_t *nic_mask,
983 			  unsigned int size,
984 			  bool range_accepted,
985 			  struct rte_flow_error *error)
986 {
987 	unsigned int i;
988 
989 	MLX5_ASSERT(nic_mask);
990 	for (i = 0; i < size; ++i)
991 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
992 			return rte_flow_error_set(error, ENOTSUP,
993 						  RTE_FLOW_ERROR_TYPE_ITEM,
994 						  item,
995 						  "mask enables non supported"
996 						  " bits");
997 	if (!item->spec && (item->mask || item->last))
998 		return rte_flow_error_set(error, EINVAL,
999 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1000 					  "mask/last without a spec is not"
1001 					  " supported");
1002 	if (item->spec && item->last && !range_accepted) {
1003 		uint8_t spec[size];
1004 		uint8_t last[size];
1005 		unsigned int i;
1006 		int ret;
1007 
1008 		for (i = 0; i < size; ++i) {
1009 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1010 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1011 		}
1012 		ret = memcmp(spec, last, size);
1013 		if (ret != 0)
1014 			return rte_flow_error_set(error, EINVAL,
1015 						  RTE_FLOW_ERROR_TYPE_ITEM,
1016 						  item,
1017 						  "range is not valid");
1018 	}
1019 	return 0;
1020 }
1021 
1022 /**
1023  * Adjust the hash fields according to the @p flow information.
1024  *
1025  * @param[in] dev_flow.
1026  *   Pointer to the mlx5_flow.
1027  * @param[in] tunnel
1028  *   1 when the hash field is for a tunnel item.
1029  * @param[in] layer_types
1030  *   ETH_RSS_* types.
1031  * @param[in] hash_fields
1032  *   Item hash fields.
1033  *
1034  * @return
1035  *   The hash fields that should be used.
1036  */
1037 uint64_t
1038 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1039 			    int tunnel __rte_unused, uint64_t layer_types,
1040 			    uint64_t hash_fields)
1041 {
1042 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1043 	int rss_request_inner = rss_desc->level >= 2;
1044 
1045 	/* Check RSS hash level for tunnel. */
1046 	if (tunnel && rss_request_inner)
1047 		hash_fields |= IBV_RX_HASH_INNER;
1048 	else if (tunnel || rss_request_inner)
1049 		return 0;
1050 #endif
1051 	/* Check if requested layer matches RSS hash fields. */
1052 	if (!(rss_desc->types & layer_types))
1053 		return 0;
1054 	return hash_fields;
1055 }
1056 
1057 /**
1058  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1059  * if several tunnel rules are used on this queue, the tunnel ptype will be
1060  * cleared.
1061  *
1062  * @param rxq_ctrl
1063  *   Rx queue to update.
1064  */
1065 static void
1066 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1067 {
1068 	unsigned int i;
1069 	uint32_t tunnel_ptype = 0;
1070 
1071 	/* Look up for the ptype to use. */
1072 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1073 		if (!rxq_ctrl->flow_tunnels_n[i])
1074 			continue;
1075 		if (!tunnel_ptype) {
1076 			tunnel_ptype = tunnels_info[i].ptype;
1077 		} else {
1078 			tunnel_ptype = 0;
1079 			break;
1080 		}
1081 	}
1082 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1083 }
1084 
1085 /**
1086  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1087  * flow.
1088  *
1089  * @param[in] dev
1090  *   Pointer to the Ethernet device structure.
1091  * @param[in] dev_handle
1092  *   Pointer to device flow handle structure.
1093  */
1094 void
1095 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1096 		       struct mlx5_flow_handle *dev_handle)
1097 {
1098 	struct mlx5_priv *priv = dev->data->dev_private;
1099 	const int mark = dev_handle->mark;
1100 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1101 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1102 	unsigned int i;
1103 
1104 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1105 		struct mlx5_hrxq *hrxq;
1106 
1107 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1108 			      dev_handle->rix_hrxq);
1109 		if (hrxq)
1110 			ind_tbl = hrxq->ind_table;
1111 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1112 		struct mlx5_shared_action_rss *shared_rss;
1113 
1114 		shared_rss = mlx5_ipool_get
1115 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1116 			 dev_handle->rix_srss);
1117 		if (shared_rss)
1118 			ind_tbl = shared_rss->ind_tbl;
1119 	}
1120 	if (!ind_tbl)
1121 		return;
1122 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1123 		int idx = ind_tbl->queues[i];
1124 		struct mlx5_rxq_ctrl *rxq_ctrl =
1125 			container_of((*priv->rxqs)[idx],
1126 				     struct mlx5_rxq_ctrl, rxq);
1127 
1128 		/*
1129 		 * To support metadata register copy on Tx loopback,
1130 		 * this must be always enabled (metadata may arive
1131 		 * from other port - not from local flows only.
1132 		 */
1133 		if (priv->config.dv_flow_en &&
1134 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1135 		    mlx5_flow_ext_mreg_supported(dev)) {
1136 			rxq_ctrl->rxq.mark = 1;
1137 			rxq_ctrl->flow_mark_n = 1;
1138 		} else if (mark) {
1139 			rxq_ctrl->rxq.mark = 1;
1140 			rxq_ctrl->flow_mark_n++;
1141 		}
1142 		if (tunnel) {
1143 			unsigned int j;
1144 
1145 			/* Increase the counter matching the flow. */
1146 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1147 				if ((tunnels_info[j].tunnel &
1148 				     dev_handle->layers) ==
1149 				    tunnels_info[j].tunnel) {
1150 					rxq_ctrl->flow_tunnels_n[j]++;
1151 					break;
1152 				}
1153 			}
1154 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1155 		}
1156 	}
1157 }
1158 
1159 /**
1160  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1161  *
1162  * @param[in] dev
1163  *   Pointer to the Ethernet device structure.
1164  * @param[in] flow
1165  *   Pointer to flow structure.
1166  */
1167 static void
1168 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1169 {
1170 	struct mlx5_priv *priv = dev->data->dev_private;
1171 	uint32_t handle_idx;
1172 	struct mlx5_flow_handle *dev_handle;
1173 
1174 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1175 		       handle_idx, dev_handle, next)
1176 		flow_drv_rxq_flags_set(dev, dev_handle);
1177 }
1178 
1179 /**
1180  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1181  * device flow if no other flow uses it with the same kind of request.
1182  *
1183  * @param dev
1184  *   Pointer to Ethernet device.
1185  * @param[in] dev_handle
1186  *   Pointer to the device flow handle structure.
1187  */
1188 static void
1189 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1190 			struct mlx5_flow_handle *dev_handle)
1191 {
1192 	struct mlx5_priv *priv = dev->data->dev_private;
1193 	const int mark = dev_handle->mark;
1194 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1195 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1196 	unsigned int i;
1197 
1198 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1199 		struct mlx5_hrxq *hrxq;
1200 
1201 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1202 			      dev_handle->rix_hrxq);
1203 		if (hrxq)
1204 			ind_tbl = hrxq->ind_table;
1205 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1206 		struct mlx5_shared_action_rss *shared_rss;
1207 
1208 		shared_rss = mlx5_ipool_get
1209 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1210 			 dev_handle->rix_srss);
1211 		if (shared_rss)
1212 			ind_tbl = shared_rss->ind_tbl;
1213 	}
1214 	if (!ind_tbl)
1215 		return;
1216 	MLX5_ASSERT(dev->data->dev_started);
1217 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1218 		int idx = ind_tbl->queues[i];
1219 		struct mlx5_rxq_ctrl *rxq_ctrl =
1220 			container_of((*priv->rxqs)[idx],
1221 				     struct mlx5_rxq_ctrl, rxq);
1222 
1223 		if (priv->config.dv_flow_en &&
1224 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1225 		    mlx5_flow_ext_mreg_supported(dev)) {
1226 			rxq_ctrl->rxq.mark = 1;
1227 			rxq_ctrl->flow_mark_n = 1;
1228 		} else if (mark) {
1229 			rxq_ctrl->flow_mark_n--;
1230 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1231 		}
1232 		if (tunnel) {
1233 			unsigned int j;
1234 
1235 			/* Decrease the counter matching the flow. */
1236 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1237 				if ((tunnels_info[j].tunnel &
1238 				     dev_handle->layers) ==
1239 				    tunnels_info[j].tunnel) {
1240 					rxq_ctrl->flow_tunnels_n[j]--;
1241 					break;
1242 				}
1243 			}
1244 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1245 		}
1246 	}
1247 }
1248 
1249 /**
1250  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1251  * @p flow if no other flow uses it with the same kind of request.
1252  *
1253  * @param dev
1254  *   Pointer to Ethernet device.
1255  * @param[in] flow
1256  *   Pointer to the flow.
1257  */
1258 static void
1259 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1260 {
1261 	struct mlx5_priv *priv = dev->data->dev_private;
1262 	uint32_t handle_idx;
1263 	struct mlx5_flow_handle *dev_handle;
1264 
1265 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1266 		       handle_idx, dev_handle, next)
1267 		flow_drv_rxq_flags_trim(dev, dev_handle);
1268 }
1269 
1270 /**
1271  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1272  *
1273  * @param dev
1274  *   Pointer to Ethernet device.
1275  */
1276 static void
1277 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1278 {
1279 	struct mlx5_priv *priv = dev->data->dev_private;
1280 	unsigned int i;
1281 
1282 	for (i = 0; i != priv->rxqs_n; ++i) {
1283 		struct mlx5_rxq_ctrl *rxq_ctrl;
1284 		unsigned int j;
1285 
1286 		if (!(*priv->rxqs)[i])
1287 			continue;
1288 		rxq_ctrl = container_of((*priv->rxqs)[i],
1289 					struct mlx5_rxq_ctrl, rxq);
1290 		rxq_ctrl->flow_mark_n = 0;
1291 		rxq_ctrl->rxq.mark = 0;
1292 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1293 			rxq_ctrl->flow_tunnels_n[j] = 0;
1294 		rxq_ctrl->rxq.tunnel = 0;
1295 	}
1296 }
1297 
1298 /**
1299  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1300  *
1301  * @param[in] dev
1302  *   Pointer to the Ethernet device structure.
1303  */
1304 void
1305 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1306 {
1307 	struct mlx5_priv *priv = dev->data->dev_private;
1308 	struct mlx5_rxq_data *data;
1309 	unsigned int i;
1310 
1311 	for (i = 0; i != priv->rxqs_n; ++i) {
1312 		if (!(*priv->rxqs)[i])
1313 			continue;
1314 		data = (*priv->rxqs)[i];
1315 		if (!rte_flow_dynf_metadata_avail()) {
1316 			data->dynf_meta = 0;
1317 			data->flow_meta_mask = 0;
1318 			data->flow_meta_offset = -1;
1319 			data->flow_meta_port_mask = 0;
1320 		} else {
1321 			data->dynf_meta = 1;
1322 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1323 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1324 			data->flow_meta_port_mask = (uint32_t)~0;
1325 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1326 				data->flow_meta_port_mask >>= 16;
1327 		}
1328 	}
1329 }
1330 
1331 /*
1332  * return a pointer to the desired action in the list of actions.
1333  *
1334  * @param[in] actions
1335  *   The list of actions to search the action in.
1336  * @param[in] action
1337  *   The action to find.
1338  *
1339  * @return
1340  *   Pointer to the action in the list, if found. NULL otherwise.
1341  */
1342 const struct rte_flow_action *
1343 mlx5_flow_find_action(const struct rte_flow_action *actions,
1344 		      enum rte_flow_action_type action)
1345 {
1346 	if (actions == NULL)
1347 		return NULL;
1348 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1349 		if (actions->type == action)
1350 			return actions;
1351 	return NULL;
1352 }
1353 
1354 /*
1355  * Validate the flag action.
1356  *
1357  * @param[in] action_flags
1358  *   Bit-fields that holds the actions detected until now.
1359  * @param[in] attr
1360  *   Attributes of flow that includes this action.
1361  * @param[out] error
1362  *   Pointer to error structure.
1363  *
1364  * @return
1365  *   0 on success, a negative errno value otherwise and rte_errno is set.
1366  */
1367 int
1368 mlx5_flow_validate_action_flag(uint64_t action_flags,
1369 			       const struct rte_flow_attr *attr,
1370 			       struct rte_flow_error *error)
1371 {
1372 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1373 		return rte_flow_error_set(error, EINVAL,
1374 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1375 					  "can't mark and flag in same flow");
1376 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1377 		return rte_flow_error_set(error, EINVAL,
1378 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1379 					  "can't have 2 flag"
1380 					  " actions in same flow");
1381 	if (attr->egress)
1382 		return rte_flow_error_set(error, ENOTSUP,
1383 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1384 					  "flag action not supported for "
1385 					  "egress");
1386 	return 0;
1387 }
1388 
1389 /*
1390  * Validate the mark action.
1391  *
1392  * @param[in] action
1393  *   Pointer to the queue action.
1394  * @param[in] action_flags
1395  *   Bit-fields that holds the actions detected until now.
1396  * @param[in] attr
1397  *   Attributes of flow that includes this action.
1398  * @param[out] error
1399  *   Pointer to error structure.
1400  *
1401  * @return
1402  *   0 on success, a negative errno value otherwise and rte_errno is set.
1403  */
1404 int
1405 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1406 			       uint64_t action_flags,
1407 			       const struct rte_flow_attr *attr,
1408 			       struct rte_flow_error *error)
1409 {
1410 	const struct rte_flow_action_mark *mark = action->conf;
1411 
1412 	if (!mark)
1413 		return rte_flow_error_set(error, EINVAL,
1414 					  RTE_FLOW_ERROR_TYPE_ACTION,
1415 					  action,
1416 					  "configuration cannot be null");
1417 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1418 		return rte_flow_error_set(error, EINVAL,
1419 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1420 					  &mark->id,
1421 					  "mark id must in 0 <= id < "
1422 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1423 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1424 		return rte_flow_error_set(error, EINVAL,
1425 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1426 					  "can't flag and mark in same flow");
1427 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1428 		return rte_flow_error_set(error, EINVAL,
1429 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1430 					  "can't have 2 mark actions in same"
1431 					  " flow");
1432 	if (attr->egress)
1433 		return rte_flow_error_set(error, ENOTSUP,
1434 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1435 					  "mark action not supported for "
1436 					  "egress");
1437 	return 0;
1438 }
1439 
1440 /*
1441  * Validate the drop action.
1442  *
1443  * @param[in] action_flags
1444  *   Bit-fields that holds the actions detected until now.
1445  * @param[in] attr
1446  *   Attributes of flow that includes this action.
1447  * @param[out] error
1448  *   Pointer to error structure.
1449  *
1450  * @return
1451  *   0 on success, a negative errno value otherwise and rte_errno is set.
1452  */
1453 int
1454 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1455 			       const struct rte_flow_attr *attr,
1456 			       struct rte_flow_error *error)
1457 {
1458 	if (attr->egress)
1459 		return rte_flow_error_set(error, ENOTSUP,
1460 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1461 					  "drop action not supported for "
1462 					  "egress");
1463 	return 0;
1464 }
1465 
1466 /*
1467  * Validate the queue action.
1468  *
1469  * @param[in] action
1470  *   Pointer to the queue action.
1471  * @param[in] action_flags
1472  *   Bit-fields that holds the actions detected until now.
1473  * @param[in] dev
1474  *   Pointer to the Ethernet device structure.
1475  * @param[in] attr
1476  *   Attributes of flow that includes this action.
1477  * @param[out] error
1478  *   Pointer to error structure.
1479  *
1480  * @return
1481  *   0 on success, a negative errno value otherwise and rte_errno is set.
1482  */
1483 int
1484 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1485 				uint64_t action_flags,
1486 				struct rte_eth_dev *dev,
1487 				const struct rte_flow_attr *attr,
1488 				struct rte_flow_error *error)
1489 {
1490 	struct mlx5_priv *priv = dev->data->dev_private;
1491 	const struct rte_flow_action_queue *queue = action->conf;
1492 
1493 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1494 		return rte_flow_error_set(error, EINVAL,
1495 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1496 					  "can't have 2 fate actions in"
1497 					  " same flow");
1498 	if (!priv->rxqs_n)
1499 		return rte_flow_error_set(error, EINVAL,
1500 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1501 					  NULL, "No Rx queues configured");
1502 	if (queue->index >= priv->rxqs_n)
1503 		return rte_flow_error_set(error, EINVAL,
1504 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1505 					  &queue->index,
1506 					  "queue index out of range");
1507 	if (!(*priv->rxqs)[queue->index])
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1510 					  &queue->index,
1511 					  "queue is not configured");
1512 	if (attr->egress)
1513 		return rte_flow_error_set(error, ENOTSUP,
1514 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1515 					  "queue action not supported for "
1516 					  "egress");
1517 	return 0;
1518 }
1519 
1520 /*
1521  * Validate the rss action.
1522  *
1523  * @param[in] dev
1524  *   Pointer to the Ethernet device structure.
1525  * @param[in] action
1526  *   Pointer to the queue action.
1527  * @param[out] error
1528  *   Pointer to error structure.
1529  *
1530  * @return
1531  *   0 on success, a negative errno value otherwise and rte_errno is set.
1532  */
1533 int
1534 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1535 			 const struct rte_flow_action *action,
1536 			 struct rte_flow_error *error)
1537 {
1538 	struct mlx5_priv *priv = dev->data->dev_private;
1539 	const struct rte_flow_action_rss *rss = action->conf;
1540 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1541 	unsigned int i;
1542 
1543 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1544 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1545 		return rte_flow_error_set(error, ENOTSUP,
1546 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1547 					  &rss->func,
1548 					  "RSS hash function not supported");
1549 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1550 	if (rss->level > 2)
1551 #else
1552 	if (rss->level > 1)
1553 #endif
1554 		return rte_flow_error_set(error, ENOTSUP,
1555 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1556 					  &rss->level,
1557 					  "tunnel RSS is not supported");
1558 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1559 	if (rss->key_len == 0 && rss->key != NULL)
1560 		return rte_flow_error_set(error, ENOTSUP,
1561 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1562 					  &rss->key_len,
1563 					  "RSS hash key length 0");
1564 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1565 		return rte_flow_error_set(error, ENOTSUP,
1566 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1567 					  &rss->key_len,
1568 					  "RSS hash key too small");
1569 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1570 		return rte_flow_error_set(error, ENOTSUP,
1571 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1572 					  &rss->key_len,
1573 					  "RSS hash key too large");
1574 	if (rss->queue_num > priv->config.ind_table_max_size)
1575 		return rte_flow_error_set(error, ENOTSUP,
1576 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1577 					  &rss->queue_num,
1578 					  "number of queues too large");
1579 	if (rss->types & MLX5_RSS_HF_MASK)
1580 		return rte_flow_error_set(error, ENOTSUP,
1581 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1582 					  &rss->types,
1583 					  "some RSS protocols are not"
1584 					  " supported");
1585 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1586 	    !(rss->types & ETH_RSS_IP))
1587 		return rte_flow_error_set(error, EINVAL,
1588 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1589 					  "L3 partial RSS requested but L3 RSS"
1590 					  " type not specified");
1591 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1592 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1595 					  "L4 partial RSS requested but L4 RSS"
1596 					  " type not specified");
1597 	if (!priv->rxqs_n)
1598 		return rte_flow_error_set(error, EINVAL,
1599 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1600 					  NULL, "No Rx queues configured");
1601 	if (!rss->queue_num)
1602 		return rte_flow_error_set(error, EINVAL,
1603 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1604 					  NULL, "No queues configured");
1605 	for (i = 0; i != rss->queue_num; ++i) {
1606 		struct mlx5_rxq_ctrl *rxq_ctrl;
1607 
1608 		if (rss->queue[i] >= priv->rxqs_n)
1609 			return rte_flow_error_set
1610 				(error, EINVAL,
1611 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1612 				 &rss->queue[i], "queue index out of range");
1613 		if (!(*priv->rxqs)[rss->queue[i]])
1614 			return rte_flow_error_set
1615 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1616 				 &rss->queue[i], "queue is not configured");
1617 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1618 					struct mlx5_rxq_ctrl, rxq);
1619 		if (i == 0)
1620 			rxq_type = rxq_ctrl->type;
1621 		if (rxq_type != rxq_ctrl->type)
1622 			return rte_flow_error_set
1623 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1624 				 &rss->queue[i],
1625 				 "combining hairpin and regular RSS queues is not supported");
1626 	}
1627 	return 0;
1628 }
1629 
1630 /*
1631  * Validate the rss action.
1632  *
1633  * @param[in] action
1634  *   Pointer to the queue action.
1635  * @param[in] action_flags
1636  *   Bit-fields that holds the actions detected until now.
1637  * @param[in] dev
1638  *   Pointer to the Ethernet device structure.
1639  * @param[in] attr
1640  *   Attributes of flow that includes this action.
1641  * @param[in] item_flags
1642  *   Items that were detected.
1643  * @param[out] error
1644  *   Pointer to error structure.
1645  *
1646  * @return
1647  *   0 on success, a negative errno value otherwise and rte_errno is set.
1648  */
1649 int
1650 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1651 			      uint64_t action_flags,
1652 			      struct rte_eth_dev *dev,
1653 			      const struct rte_flow_attr *attr,
1654 			      uint64_t item_flags,
1655 			      struct rte_flow_error *error)
1656 {
1657 	const struct rte_flow_action_rss *rss = action->conf;
1658 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1659 	int ret;
1660 
1661 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1662 		return rte_flow_error_set(error, EINVAL,
1663 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1664 					  "can't have 2 fate actions"
1665 					  " in same flow");
1666 	ret = mlx5_validate_action_rss(dev, action, error);
1667 	if (ret)
1668 		return ret;
1669 	if (attr->egress)
1670 		return rte_flow_error_set(error, ENOTSUP,
1671 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1672 					  "rss action not supported for "
1673 					  "egress");
1674 	if (rss->level > 1 && !tunnel)
1675 		return rte_flow_error_set(error, EINVAL,
1676 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1677 					  "inner RSS is not supported for "
1678 					  "non-tunnel flows");
1679 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1680 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1683 					  "RSS on eCPRI is not supported now");
1684 	}
1685 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1686 	    !(item_flags &
1687 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1688 	    rss->level > 1)
1689 		return rte_flow_error_set(error, EINVAL,
1690 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1691 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1692 	return 0;
1693 }
1694 
1695 /*
1696  * Validate the default miss action.
1697  *
1698  * @param[in] action_flags
1699  *   Bit-fields that holds the actions detected until now.
1700  * @param[out] error
1701  *   Pointer to error structure.
1702  *
1703  * @return
1704  *   0 on success, a negative errno value otherwise and rte_errno is set.
1705  */
1706 int
1707 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1708 				const struct rte_flow_attr *attr,
1709 				struct rte_flow_error *error)
1710 {
1711 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1712 		return rte_flow_error_set(error, EINVAL,
1713 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1714 					  "can't have 2 fate actions in"
1715 					  " same flow");
1716 	if (attr->egress)
1717 		return rte_flow_error_set(error, ENOTSUP,
1718 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1719 					  "default miss action not supported "
1720 					  "for egress");
1721 	if (attr->group)
1722 		return rte_flow_error_set(error, ENOTSUP,
1723 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1724 					  "only group 0 is supported");
1725 	if (attr->transfer)
1726 		return rte_flow_error_set(error, ENOTSUP,
1727 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1728 					  NULL, "transfer is not supported");
1729 	return 0;
1730 }
1731 
1732 /*
1733  * Validate the count action.
1734  *
1735  * @param[in] dev
1736  *   Pointer to the Ethernet device structure.
1737  * @param[in] attr
1738  *   Attributes of flow that includes this action.
1739  * @param[out] error
1740  *   Pointer to error structure.
1741  *
1742  * @return
1743  *   0 on success, a negative errno value otherwise and rte_errno is set.
1744  */
1745 int
1746 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1747 				const struct rte_flow_attr *attr,
1748 				struct rte_flow_error *error)
1749 {
1750 	if (attr->egress)
1751 		return rte_flow_error_set(error, ENOTSUP,
1752 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1753 					  "count action not supported for "
1754 					  "egress");
1755 	return 0;
1756 }
1757 
1758 /*
1759  * Validate the ASO CT action.
1760  *
1761  * @param[in] dev
1762  *   Pointer to the Ethernet device structure.
1763  * @param[in] conntrack
1764  *   Pointer to the CT action profile.
1765  * @param[out] error
1766  *   Pointer to error structure.
1767  *
1768  * @return
1769  *   0 on success, a negative errno value otherwise and rte_errno is set.
1770  */
1771 int
1772 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1773 			const struct rte_flow_action_conntrack *conntrack,
1774 			struct rte_flow_error *error)
1775 {
1776 	RTE_SET_USED(dev);
1777 
1778 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1779 		return rte_flow_error_set(error, EINVAL,
1780 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1781 					  "Invalid CT state");
1782 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1783 		return rte_flow_error_set(error, EINVAL,
1784 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1785 					  "Invalid last TCP packet flag");
1786 	return 0;
1787 }
1788 
1789 /**
1790  * Verify the @p attributes will be correctly understood by the NIC and store
1791  * them in the @p flow if everything is correct.
1792  *
1793  * @param[in] dev
1794  *   Pointer to the Ethernet device structure.
1795  * @param[in] attributes
1796  *   Pointer to flow attributes
1797  * @param[out] error
1798  *   Pointer to error structure.
1799  *
1800  * @return
1801  *   0 on success, a negative errno value otherwise and rte_errno is set.
1802  */
1803 int
1804 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1805 			      const struct rte_flow_attr *attributes,
1806 			      struct rte_flow_error *error)
1807 {
1808 	struct mlx5_priv *priv = dev->data->dev_private;
1809 	uint32_t priority_max = priv->config.flow_prio - 1;
1810 
1811 	if (attributes->group)
1812 		return rte_flow_error_set(error, ENOTSUP,
1813 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1814 					  NULL, "groups is not supported");
1815 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1816 	    attributes->priority >= priority_max)
1817 		return rte_flow_error_set(error, ENOTSUP,
1818 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1819 					  NULL, "priority out of range");
1820 	if (attributes->egress)
1821 		return rte_flow_error_set(error, ENOTSUP,
1822 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1823 					  "egress is not supported");
1824 	if (attributes->transfer && !priv->config.dv_esw_en)
1825 		return rte_flow_error_set(error, ENOTSUP,
1826 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1827 					  NULL, "transfer is not supported");
1828 	if (!attributes->ingress)
1829 		return rte_flow_error_set(error, EINVAL,
1830 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1831 					  NULL,
1832 					  "ingress attribute is mandatory");
1833 	return 0;
1834 }
1835 
1836 /**
1837  * Validate ICMP6 item.
1838  *
1839  * @param[in] item
1840  *   Item specification.
1841  * @param[in] item_flags
1842  *   Bit-fields that holds the items detected until now.
1843  * @param[in] ext_vlan_sup
1844  *   Whether extended VLAN features are supported or not.
1845  * @param[out] error
1846  *   Pointer to error structure.
1847  *
1848  * @return
1849  *   0 on success, a negative errno value otherwise and rte_errno is set.
1850  */
1851 int
1852 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1853 			       uint64_t item_flags,
1854 			       uint8_t target_protocol,
1855 			       struct rte_flow_error *error)
1856 {
1857 	const struct rte_flow_item_icmp6 *mask = item->mask;
1858 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1859 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1860 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1861 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1862 				      MLX5_FLOW_LAYER_OUTER_L4;
1863 	int ret;
1864 
1865 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1866 		return rte_flow_error_set(error, EINVAL,
1867 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1868 					  "protocol filtering not compatible"
1869 					  " with ICMP6 layer");
1870 	if (!(item_flags & l3m))
1871 		return rte_flow_error_set(error, EINVAL,
1872 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1873 					  "IPv6 is mandatory to filter on"
1874 					  " ICMP6");
1875 	if (item_flags & l4m)
1876 		return rte_flow_error_set(error, EINVAL,
1877 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1878 					  "multiple L4 layers not supported");
1879 	if (!mask)
1880 		mask = &rte_flow_item_icmp6_mask;
1881 	ret = mlx5_flow_item_acceptable
1882 		(item, (const uint8_t *)mask,
1883 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1884 		 sizeof(struct rte_flow_item_icmp6),
1885 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1886 	if (ret < 0)
1887 		return ret;
1888 	return 0;
1889 }
1890 
1891 /**
1892  * Validate ICMP item.
1893  *
1894  * @param[in] item
1895  *   Item specification.
1896  * @param[in] item_flags
1897  *   Bit-fields that holds the items detected until now.
1898  * @param[out] error
1899  *   Pointer to error structure.
1900  *
1901  * @return
1902  *   0 on success, a negative errno value otherwise and rte_errno is set.
1903  */
1904 int
1905 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1906 			     uint64_t item_flags,
1907 			     uint8_t target_protocol,
1908 			     struct rte_flow_error *error)
1909 {
1910 	const struct rte_flow_item_icmp *mask = item->mask;
1911 	const struct rte_flow_item_icmp nic_mask = {
1912 		.hdr.icmp_type = 0xff,
1913 		.hdr.icmp_code = 0xff,
1914 		.hdr.icmp_ident = RTE_BE16(0xffff),
1915 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1916 	};
1917 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1918 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1919 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1920 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1921 				      MLX5_FLOW_LAYER_OUTER_L4;
1922 	int ret;
1923 
1924 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1925 		return rte_flow_error_set(error, EINVAL,
1926 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 					  "protocol filtering not compatible"
1928 					  " with ICMP layer");
1929 	if (!(item_flags & l3m))
1930 		return rte_flow_error_set(error, EINVAL,
1931 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1932 					  "IPv4 is mandatory to filter"
1933 					  " on ICMP");
1934 	if (item_flags & l4m)
1935 		return rte_flow_error_set(error, EINVAL,
1936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1937 					  "multiple L4 layers not supported");
1938 	if (!mask)
1939 		mask = &nic_mask;
1940 	ret = mlx5_flow_item_acceptable
1941 		(item, (const uint8_t *)mask,
1942 		 (const uint8_t *)&nic_mask,
1943 		 sizeof(struct rte_flow_item_icmp),
1944 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1945 	if (ret < 0)
1946 		return ret;
1947 	return 0;
1948 }
1949 
1950 /**
1951  * Validate Ethernet item.
1952  *
1953  * @param[in] item
1954  *   Item specification.
1955  * @param[in] item_flags
1956  *   Bit-fields that holds the items detected until now.
1957  * @param[out] error
1958  *   Pointer to error structure.
1959  *
1960  * @return
1961  *   0 on success, a negative errno value otherwise and rte_errno is set.
1962  */
1963 int
1964 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1965 			    uint64_t item_flags, bool ext_vlan_sup,
1966 			    struct rte_flow_error *error)
1967 {
1968 	const struct rte_flow_item_eth *mask = item->mask;
1969 	const struct rte_flow_item_eth nic_mask = {
1970 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1971 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1972 		.type = RTE_BE16(0xffff),
1973 		.has_vlan = ext_vlan_sup ? 1 : 0,
1974 	};
1975 	int ret;
1976 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1977 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1978 				       MLX5_FLOW_LAYER_OUTER_L2;
1979 
1980 	if (item_flags & ethm)
1981 		return rte_flow_error_set(error, ENOTSUP,
1982 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1983 					  "multiple L2 layers not supported");
1984 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1985 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1988 					  "L2 layer should not follow "
1989 					  "L3 layers");
1990 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1991 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1992 		return rte_flow_error_set(error, EINVAL,
1993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 					  "L2 layer should not follow VLAN");
1995 	if (!mask)
1996 		mask = &rte_flow_item_eth_mask;
1997 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1998 					(const uint8_t *)&nic_mask,
1999 					sizeof(struct rte_flow_item_eth),
2000 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2001 	return ret;
2002 }
2003 
2004 /**
2005  * Validate VLAN item.
2006  *
2007  * @param[in] item
2008  *   Item specification.
2009  * @param[in] item_flags
2010  *   Bit-fields that holds the items detected until now.
2011  * @param[in] dev
2012  *   Ethernet device flow is being created on.
2013  * @param[out] error
2014  *   Pointer to error structure.
2015  *
2016  * @return
2017  *   0 on success, a negative errno value otherwise and rte_errno is set.
2018  */
2019 int
2020 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2021 			     uint64_t item_flags,
2022 			     struct rte_eth_dev *dev,
2023 			     struct rte_flow_error *error)
2024 {
2025 	const struct rte_flow_item_vlan *spec = item->spec;
2026 	const struct rte_flow_item_vlan *mask = item->mask;
2027 	const struct rte_flow_item_vlan nic_mask = {
2028 		.tci = RTE_BE16(UINT16_MAX),
2029 		.inner_type = RTE_BE16(UINT16_MAX),
2030 	};
2031 	uint16_t vlan_tag = 0;
2032 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 	int ret;
2034 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2035 					MLX5_FLOW_LAYER_INNER_L4) :
2036 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2037 					MLX5_FLOW_LAYER_OUTER_L4);
2038 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2039 					MLX5_FLOW_LAYER_OUTER_VLAN;
2040 
2041 	if (item_flags & vlanm)
2042 		return rte_flow_error_set(error, EINVAL,
2043 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2044 					  "multiple VLAN layers not supported");
2045 	else if ((item_flags & l34m) != 0)
2046 		return rte_flow_error_set(error, EINVAL,
2047 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2048 					  "VLAN cannot follow L3/L4 layer");
2049 	if (!mask)
2050 		mask = &rte_flow_item_vlan_mask;
2051 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2052 					(const uint8_t *)&nic_mask,
2053 					sizeof(struct rte_flow_item_vlan),
2054 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2055 	if (ret)
2056 		return ret;
2057 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2058 		struct mlx5_priv *priv = dev->data->dev_private;
2059 
2060 		if (priv->vmwa_context) {
2061 			/*
2062 			 * Non-NULL context means we have a virtual machine
2063 			 * and SR-IOV enabled, we have to create VLAN interface
2064 			 * to make hypervisor to setup E-Switch vport
2065 			 * context correctly. We avoid creating the multiple
2066 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2067 			 */
2068 			return rte_flow_error_set(error, EINVAL,
2069 						  RTE_FLOW_ERROR_TYPE_ITEM,
2070 						  item,
2071 						  "VLAN tag mask is not"
2072 						  " supported in virtual"
2073 						  " environment");
2074 		}
2075 	}
2076 	if (spec) {
2077 		vlan_tag = spec->tci;
2078 		vlan_tag &= mask->tci;
2079 	}
2080 	/*
2081 	 * From verbs perspective an empty VLAN is equivalent
2082 	 * to a packet without VLAN layer.
2083 	 */
2084 	if (!vlan_tag)
2085 		return rte_flow_error_set(error, EINVAL,
2086 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2087 					  item->spec,
2088 					  "VLAN cannot be empty");
2089 	return 0;
2090 }
2091 
2092 /**
2093  * Validate IPV4 item.
2094  *
2095  * @param[in] item
2096  *   Item specification.
2097  * @param[in] item_flags
2098  *   Bit-fields that holds the items detected until now.
2099  * @param[in] last_item
2100  *   Previous validated item in the pattern items.
2101  * @param[in] ether_type
2102  *   Type in the ethernet layer header (including dot1q).
2103  * @param[in] acc_mask
2104  *   Acceptable mask, if NULL default internal default mask
2105  *   will be used to check whether item fields are supported.
2106  * @param[in] range_accepted
2107  *   True if range of values is accepted for specific fields, false otherwise.
2108  * @param[out] error
2109  *   Pointer to error structure.
2110  *
2111  * @return
2112  *   0 on success, a negative errno value otherwise and rte_errno is set.
2113  */
2114 int
2115 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2116 			     uint64_t item_flags,
2117 			     uint64_t last_item,
2118 			     uint16_t ether_type,
2119 			     const struct rte_flow_item_ipv4 *acc_mask,
2120 			     bool range_accepted,
2121 			     struct rte_flow_error *error)
2122 {
2123 	const struct rte_flow_item_ipv4 *mask = item->mask;
2124 	const struct rte_flow_item_ipv4 *spec = item->spec;
2125 	const struct rte_flow_item_ipv4 nic_mask = {
2126 		.hdr = {
2127 			.src_addr = RTE_BE32(0xffffffff),
2128 			.dst_addr = RTE_BE32(0xffffffff),
2129 			.type_of_service = 0xff,
2130 			.next_proto_id = 0xff,
2131 		},
2132 	};
2133 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2134 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2135 				      MLX5_FLOW_LAYER_OUTER_L3;
2136 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2137 				      MLX5_FLOW_LAYER_OUTER_L4;
2138 	int ret;
2139 	uint8_t next_proto = 0xFF;
2140 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2141 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2142 				  MLX5_FLOW_LAYER_INNER_VLAN);
2143 
2144 	if ((last_item & l2_vlan) && ether_type &&
2145 	    ether_type != RTE_ETHER_TYPE_IPV4)
2146 		return rte_flow_error_set(error, EINVAL,
2147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2148 					  "IPv4 cannot follow L2/VLAN layer "
2149 					  "which ether type is not IPv4");
2150 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2151 		if (mask && spec)
2152 			next_proto = mask->hdr.next_proto_id &
2153 				     spec->hdr.next_proto_id;
2154 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2155 			return rte_flow_error_set(error, EINVAL,
2156 						  RTE_FLOW_ERROR_TYPE_ITEM,
2157 						  item,
2158 						  "multiple tunnel "
2159 						  "not supported");
2160 	}
2161 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2162 		return rte_flow_error_set(error, EINVAL,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "wrong tunnel type - IPv6 specified "
2165 					  "but IPv4 item provided");
2166 	if (item_flags & l3m)
2167 		return rte_flow_error_set(error, ENOTSUP,
2168 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2169 					  "multiple L3 layers not supported");
2170 	else if (item_flags & l4m)
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "L3 cannot follow an L4 layer.");
2174 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2175 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2176 		return rte_flow_error_set(error, EINVAL,
2177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2178 					  "L3 cannot follow an NVGRE layer.");
2179 	if (!mask)
2180 		mask = &rte_flow_item_ipv4_mask;
2181 	else if (mask->hdr.next_proto_id != 0 &&
2182 		 mask->hdr.next_proto_id != 0xff)
2183 		return rte_flow_error_set(error, EINVAL,
2184 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2185 					  "partial mask is not supported"
2186 					  " for protocol");
2187 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2188 					acc_mask ? (const uint8_t *)acc_mask
2189 						 : (const uint8_t *)&nic_mask,
2190 					sizeof(struct rte_flow_item_ipv4),
2191 					range_accepted, error);
2192 	if (ret < 0)
2193 		return ret;
2194 	return 0;
2195 }
2196 
2197 /**
2198  * Validate IPV6 item.
2199  *
2200  * @param[in] item
2201  *   Item specification.
2202  * @param[in] item_flags
2203  *   Bit-fields that holds the items detected until now.
2204  * @param[in] last_item
2205  *   Previous validated item in the pattern items.
2206  * @param[in] ether_type
2207  *   Type in the ethernet layer header (including dot1q).
2208  * @param[in] acc_mask
2209  *   Acceptable mask, if NULL default internal default mask
2210  *   will be used to check whether item fields are supported.
2211  * @param[out] error
2212  *   Pointer to error structure.
2213  *
2214  * @return
2215  *   0 on success, a negative errno value otherwise and rte_errno is set.
2216  */
2217 int
2218 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2219 			     uint64_t item_flags,
2220 			     uint64_t last_item,
2221 			     uint16_t ether_type,
2222 			     const struct rte_flow_item_ipv6 *acc_mask,
2223 			     struct rte_flow_error *error)
2224 {
2225 	const struct rte_flow_item_ipv6 *mask = item->mask;
2226 	const struct rte_flow_item_ipv6 *spec = item->spec;
2227 	const struct rte_flow_item_ipv6 nic_mask = {
2228 		.hdr = {
2229 			.src_addr =
2230 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2231 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2232 			.dst_addr =
2233 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2234 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2235 			.vtc_flow = RTE_BE32(0xffffffff),
2236 			.proto = 0xff,
2237 		},
2238 	};
2239 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2240 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2241 				      MLX5_FLOW_LAYER_OUTER_L3;
2242 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2243 				      MLX5_FLOW_LAYER_OUTER_L4;
2244 	int ret;
2245 	uint8_t next_proto = 0xFF;
2246 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2247 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2248 				  MLX5_FLOW_LAYER_INNER_VLAN);
2249 
2250 	if ((last_item & l2_vlan) && ether_type &&
2251 	    ether_type != RTE_ETHER_TYPE_IPV6)
2252 		return rte_flow_error_set(error, EINVAL,
2253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 					  "IPv6 cannot follow L2/VLAN layer "
2255 					  "which ether type is not IPv6");
2256 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2257 		next_proto = spec->hdr.proto;
2258 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2259 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2260 			return rte_flow_error_set(error, EINVAL,
2261 						  RTE_FLOW_ERROR_TYPE_ITEM,
2262 						  item,
2263 						  "multiple tunnel "
2264 						  "not supported");
2265 	}
2266 	if (next_proto == IPPROTO_HOPOPTS  ||
2267 	    next_proto == IPPROTO_ROUTING  ||
2268 	    next_proto == IPPROTO_FRAGMENT ||
2269 	    next_proto == IPPROTO_ESP	   ||
2270 	    next_proto == IPPROTO_AH	   ||
2271 	    next_proto == IPPROTO_DSTOPTS)
2272 		return rte_flow_error_set(error, EINVAL,
2273 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2274 					  "IPv6 proto (next header) should "
2275 					  "not be set as extension header");
2276 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2277 		return rte_flow_error_set(error, EINVAL,
2278 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2279 					  "wrong tunnel type - IPv4 specified "
2280 					  "but IPv6 item provided");
2281 	if (item_flags & l3m)
2282 		return rte_flow_error_set(error, ENOTSUP,
2283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2284 					  "multiple L3 layers not supported");
2285 	else if (item_flags & l4m)
2286 		return rte_flow_error_set(error, EINVAL,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "L3 cannot follow an L4 layer.");
2289 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2290 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2291 		return rte_flow_error_set(error, EINVAL,
2292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 					  "L3 cannot follow an NVGRE layer.");
2294 	if (!mask)
2295 		mask = &rte_flow_item_ipv6_mask;
2296 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2297 					acc_mask ? (const uint8_t *)acc_mask
2298 						 : (const uint8_t *)&nic_mask,
2299 					sizeof(struct rte_flow_item_ipv6),
2300 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2301 	if (ret < 0)
2302 		return ret;
2303 	return 0;
2304 }
2305 
2306 /**
2307  * Validate UDP item.
2308  *
2309  * @param[in] item
2310  *   Item specification.
2311  * @param[in] item_flags
2312  *   Bit-fields that holds the items detected until now.
2313  * @param[in] target_protocol
2314  *   The next protocol in the previous item.
2315  * @param[in] flow_mask
2316  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2317  * @param[out] error
2318  *   Pointer to error structure.
2319  *
2320  * @return
2321  *   0 on success, a negative errno value otherwise and rte_errno is set.
2322  */
2323 int
2324 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2325 			    uint64_t item_flags,
2326 			    uint8_t target_protocol,
2327 			    struct rte_flow_error *error)
2328 {
2329 	const struct rte_flow_item_udp *mask = item->mask;
2330 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2331 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2332 				      MLX5_FLOW_LAYER_OUTER_L3;
2333 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2334 				      MLX5_FLOW_LAYER_OUTER_L4;
2335 	int ret;
2336 
2337 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2338 		return rte_flow_error_set(error, EINVAL,
2339 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2340 					  "protocol filtering not compatible"
2341 					  " with UDP layer");
2342 	if (!(item_flags & l3m))
2343 		return rte_flow_error_set(error, EINVAL,
2344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2345 					  "L3 is mandatory to filter on L4");
2346 	if (item_flags & l4m)
2347 		return rte_flow_error_set(error, EINVAL,
2348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2349 					  "multiple L4 layers not supported");
2350 	if (!mask)
2351 		mask = &rte_flow_item_udp_mask;
2352 	ret = mlx5_flow_item_acceptable
2353 		(item, (const uint8_t *)mask,
2354 		 (const uint8_t *)&rte_flow_item_udp_mask,
2355 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2356 		 error);
2357 	if (ret < 0)
2358 		return ret;
2359 	return 0;
2360 }
2361 
2362 /**
2363  * Validate TCP item.
2364  *
2365  * @param[in] item
2366  *   Item specification.
2367  * @param[in] item_flags
2368  *   Bit-fields that holds the items detected until now.
2369  * @param[in] target_protocol
2370  *   The next protocol in the previous item.
2371  * @param[out] error
2372  *   Pointer to error structure.
2373  *
2374  * @return
2375  *   0 on success, a negative errno value otherwise and rte_errno is set.
2376  */
2377 int
2378 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2379 			    uint64_t item_flags,
2380 			    uint8_t target_protocol,
2381 			    const struct rte_flow_item_tcp *flow_mask,
2382 			    struct rte_flow_error *error)
2383 {
2384 	const struct rte_flow_item_tcp *mask = item->mask;
2385 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2386 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2387 				      MLX5_FLOW_LAYER_OUTER_L3;
2388 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2389 				      MLX5_FLOW_LAYER_OUTER_L4;
2390 	int ret;
2391 
2392 	MLX5_ASSERT(flow_mask);
2393 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2394 		return rte_flow_error_set(error, EINVAL,
2395 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2396 					  "protocol filtering not compatible"
2397 					  " with TCP layer");
2398 	if (!(item_flags & l3m))
2399 		return rte_flow_error_set(error, EINVAL,
2400 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2401 					  "L3 is mandatory to filter on L4");
2402 	if (item_flags & l4m)
2403 		return rte_flow_error_set(error, EINVAL,
2404 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2405 					  "multiple L4 layers not supported");
2406 	if (!mask)
2407 		mask = &rte_flow_item_tcp_mask;
2408 	ret = mlx5_flow_item_acceptable
2409 		(item, (const uint8_t *)mask,
2410 		 (const uint8_t *)flow_mask,
2411 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2412 		 error);
2413 	if (ret < 0)
2414 		return ret;
2415 	return 0;
2416 }
2417 
2418 /**
2419  * Validate VXLAN item.
2420  *
2421  * @param[in] dev
2422  *   Pointer to the Ethernet device structure.
2423  * @param[in] item
2424  *   Item specification.
2425  * @param[in] item_flags
2426  *   Bit-fields that holds the items detected until now.
2427  * @param[in] attr
2428  *   Flow rule attributes.
2429  * @param[out] error
2430  *   Pointer to error structure.
2431  *
2432  * @return
2433  *   0 on success, a negative errno value otherwise and rte_errno is set.
2434  */
2435 int
2436 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2437 			      const struct rte_flow_item *item,
2438 			      uint64_t item_flags,
2439 			      const struct rte_flow_attr *attr,
2440 			      struct rte_flow_error *error)
2441 {
2442 	const struct rte_flow_item_vxlan *spec = item->spec;
2443 	const struct rte_flow_item_vxlan *mask = item->mask;
2444 	int ret;
2445 	struct mlx5_priv *priv = dev->data->dev_private;
2446 	union vni {
2447 		uint32_t vlan_id;
2448 		uint8_t vni[4];
2449 	} id = { .vlan_id = 0, };
2450 	const struct rte_flow_item_vxlan nic_mask = {
2451 		.vni = "\xff\xff\xff",
2452 		.rsvd1 = 0xff,
2453 	};
2454 	const struct rte_flow_item_vxlan *valid_mask;
2455 
2456 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2457 		return rte_flow_error_set(error, ENOTSUP,
2458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2459 					  "multiple tunnel layers not"
2460 					  " supported");
2461 	valid_mask = &rte_flow_item_vxlan_mask;
2462 	/*
2463 	 * Verify only UDPv4 is present as defined in
2464 	 * https://tools.ietf.org/html/rfc7348
2465 	 */
2466 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2467 		return rte_flow_error_set(error, EINVAL,
2468 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2469 					  "no outer UDP layer found");
2470 	if (!mask)
2471 		mask = &rte_flow_item_vxlan_mask;
2472 	/* FDB domain & NIC domain non-zero group */
2473 	if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2474 		valid_mask = &nic_mask;
2475 	/* Group zero in NIC domain */
2476 	if (!attr->group && !attr->transfer && priv->sh->tunnel_header_0_1)
2477 		valid_mask = &nic_mask;
2478 	ret = mlx5_flow_item_acceptable
2479 		(item, (const uint8_t *)mask,
2480 		 (const uint8_t *)valid_mask,
2481 		 sizeof(struct rte_flow_item_vxlan),
2482 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2483 	if (ret < 0)
2484 		return ret;
2485 	if (spec) {
2486 		memcpy(&id.vni[1], spec->vni, 3);
2487 		memcpy(&id.vni[1], mask->vni, 3);
2488 	}
2489 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2490 		return rte_flow_error_set(error, ENOTSUP,
2491 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2492 					  "VXLAN tunnel must be fully defined");
2493 	return 0;
2494 }
2495 
2496 /**
2497  * Validate VXLAN_GPE item.
2498  *
2499  * @param[in] item
2500  *   Item specification.
2501  * @param[in] item_flags
2502  *   Bit-fields that holds the items detected until now.
2503  * @param[in] priv
2504  *   Pointer to the private data structure.
2505  * @param[in] target_protocol
2506  *   The next protocol in the previous item.
2507  * @param[out] error
2508  *   Pointer to error structure.
2509  *
2510  * @return
2511  *   0 on success, a negative errno value otherwise and rte_errno is set.
2512  */
2513 int
2514 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2515 				  uint64_t item_flags,
2516 				  struct rte_eth_dev *dev,
2517 				  struct rte_flow_error *error)
2518 {
2519 	struct mlx5_priv *priv = dev->data->dev_private;
2520 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2521 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2522 	int ret;
2523 	union vni {
2524 		uint32_t vlan_id;
2525 		uint8_t vni[4];
2526 	} id = { .vlan_id = 0, };
2527 
2528 	if (!priv->config.l3_vxlan_en)
2529 		return rte_flow_error_set(error, ENOTSUP,
2530 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2531 					  "L3 VXLAN is not enabled by device"
2532 					  " parameter and/or not configured in"
2533 					  " firmware");
2534 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2535 		return rte_flow_error_set(error, ENOTSUP,
2536 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2537 					  "multiple tunnel layers not"
2538 					  " supported");
2539 	/*
2540 	 * Verify only UDPv4 is present as defined in
2541 	 * https://tools.ietf.org/html/rfc7348
2542 	 */
2543 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2544 		return rte_flow_error_set(error, EINVAL,
2545 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2546 					  "no outer UDP layer found");
2547 	if (!mask)
2548 		mask = &rte_flow_item_vxlan_gpe_mask;
2549 	ret = mlx5_flow_item_acceptable
2550 		(item, (const uint8_t *)mask,
2551 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2552 		 sizeof(struct rte_flow_item_vxlan_gpe),
2553 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2554 	if (ret < 0)
2555 		return ret;
2556 	if (spec) {
2557 		if (spec->protocol)
2558 			return rte_flow_error_set(error, ENOTSUP,
2559 						  RTE_FLOW_ERROR_TYPE_ITEM,
2560 						  item,
2561 						  "VxLAN-GPE protocol"
2562 						  " not supported");
2563 		memcpy(&id.vni[1], spec->vni, 3);
2564 		memcpy(&id.vni[1], mask->vni, 3);
2565 	}
2566 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2567 		return rte_flow_error_set(error, ENOTSUP,
2568 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2569 					  "VXLAN-GPE tunnel must be fully"
2570 					  " defined");
2571 	return 0;
2572 }
2573 /**
2574  * Validate GRE Key item.
2575  *
2576  * @param[in] item
2577  *   Item specification.
2578  * @param[in] item_flags
2579  *   Bit flags to mark detected items.
2580  * @param[in] gre_item
2581  *   Pointer to gre_item
2582  * @param[out] error
2583  *   Pointer to error structure.
2584  *
2585  * @return
2586  *   0 on success, a negative errno value otherwise and rte_errno is set.
2587  */
2588 int
2589 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2590 				uint64_t item_flags,
2591 				const struct rte_flow_item *gre_item,
2592 				struct rte_flow_error *error)
2593 {
2594 	const rte_be32_t *mask = item->mask;
2595 	int ret = 0;
2596 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2597 	const struct rte_flow_item_gre *gre_spec;
2598 	const struct rte_flow_item_gre *gre_mask;
2599 
2600 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2601 		return rte_flow_error_set(error, ENOTSUP,
2602 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2603 					  "Multiple GRE key not support");
2604 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2605 		return rte_flow_error_set(error, ENOTSUP,
2606 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2607 					  "No preceding GRE header");
2608 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2609 		return rte_flow_error_set(error, ENOTSUP,
2610 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2611 					  "GRE key following a wrong item");
2612 	gre_mask = gre_item->mask;
2613 	if (!gre_mask)
2614 		gre_mask = &rte_flow_item_gre_mask;
2615 	gre_spec = gre_item->spec;
2616 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2617 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2618 		return rte_flow_error_set(error, EINVAL,
2619 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2620 					  "Key bit must be on");
2621 
2622 	if (!mask)
2623 		mask = &gre_key_default_mask;
2624 	ret = mlx5_flow_item_acceptable
2625 		(item, (const uint8_t *)mask,
2626 		 (const uint8_t *)&gre_key_default_mask,
2627 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2628 	return ret;
2629 }
2630 
2631 /**
2632  * Validate GRE item.
2633  *
2634  * @param[in] item
2635  *   Item specification.
2636  * @param[in] item_flags
2637  *   Bit flags to mark detected items.
2638  * @param[in] target_protocol
2639  *   The next protocol in the previous item.
2640  * @param[out] error
2641  *   Pointer to error structure.
2642  *
2643  * @return
2644  *   0 on success, a negative errno value otherwise and rte_errno is set.
2645  */
2646 int
2647 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2648 			    uint64_t item_flags,
2649 			    uint8_t target_protocol,
2650 			    struct rte_flow_error *error)
2651 {
2652 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2653 	const struct rte_flow_item_gre *mask = item->mask;
2654 	int ret;
2655 	const struct rte_flow_item_gre nic_mask = {
2656 		.c_rsvd0_ver = RTE_BE16(0xB000),
2657 		.protocol = RTE_BE16(UINT16_MAX),
2658 	};
2659 
2660 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2661 		return rte_flow_error_set(error, EINVAL,
2662 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2663 					  "protocol filtering not compatible"
2664 					  " with this GRE layer");
2665 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2666 		return rte_flow_error_set(error, ENOTSUP,
2667 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2668 					  "multiple tunnel layers not"
2669 					  " supported");
2670 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2671 		return rte_flow_error_set(error, ENOTSUP,
2672 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2673 					  "L3 Layer is missing");
2674 	if (!mask)
2675 		mask = &rte_flow_item_gre_mask;
2676 	ret = mlx5_flow_item_acceptable
2677 		(item, (const uint8_t *)mask,
2678 		 (const uint8_t *)&nic_mask,
2679 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2680 		 error);
2681 	if (ret < 0)
2682 		return ret;
2683 #ifndef HAVE_MLX5DV_DR
2684 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2685 	if (spec && (spec->protocol & mask->protocol))
2686 		return rte_flow_error_set(error, ENOTSUP,
2687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2688 					  "without MPLS support the"
2689 					  " specification cannot be used for"
2690 					  " filtering");
2691 #endif
2692 #endif
2693 	return 0;
2694 }
2695 
2696 /**
2697  * Validate Geneve item.
2698  *
2699  * @param[in] item
2700  *   Item specification.
2701  * @param[in] itemFlags
2702  *   Bit-fields that holds the items detected until now.
2703  * @param[in] enPriv
2704  *   Pointer to the private data structure.
2705  * @param[out] error
2706  *   Pointer to error structure.
2707  *
2708  * @return
2709  *   0 on success, a negative errno value otherwise and rte_errno is set.
2710  */
2711 
2712 int
2713 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2714 			       uint64_t item_flags,
2715 			       struct rte_eth_dev *dev,
2716 			       struct rte_flow_error *error)
2717 {
2718 	struct mlx5_priv *priv = dev->data->dev_private;
2719 	const struct rte_flow_item_geneve *spec = item->spec;
2720 	const struct rte_flow_item_geneve *mask = item->mask;
2721 	int ret;
2722 	uint16_t gbhdr;
2723 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2724 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2725 	const struct rte_flow_item_geneve nic_mask = {
2726 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2727 		.vni = "\xff\xff\xff",
2728 		.protocol = RTE_BE16(UINT16_MAX),
2729 	};
2730 
2731 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2732 		return rte_flow_error_set(error, ENOTSUP,
2733 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2734 					  "L3 Geneve is not enabled by device"
2735 					  " parameter and/or not configured in"
2736 					  " firmware");
2737 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2738 		return rte_flow_error_set(error, ENOTSUP,
2739 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2740 					  "multiple tunnel layers not"
2741 					  " supported");
2742 	/*
2743 	 * Verify only UDPv4 is present as defined in
2744 	 * https://tools.ietf.org/html/rfc7348
2745 	 */
2746 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2747 		return rte_flow_error_set(error, EINVAL,
2748 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2749 					  "no outer UDP layer found");
2750 	if (!mask)
2751 		mask = &rte_flow_item_geneve_mask;
2752 	ret = mlx5_flow_item_acceptable
2753 				  (item, (const uint8_t *)mask,
2754 				   (const uint8_t *)&nic_mask,
2755 				   sizeof(struct rte_flow_item_geneve),
2756 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2757 	if (ret)
2758 		return ret;
2759 	if (spec) {
2760 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2761 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2762 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2763 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2764 			return rte_flow_error_set(error, ENOTSUP,
2765 						  RTE_FLOW_ERROR_TYPE_ITEM,
2766 						  item,
2767 						  "Geneve protocol unsupported"
2768 						  " fields are being used");
2769 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2770 			return rte_flow_error_set
2771 					(error, ENOTSUP,
2772 					 RTE_FLOW_ERROR_TYPE_ITEM,
2773 					 item,
2774 					 "Unsupported Geneve options length");
2775 	}
2776 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2777 		return rte_flow_error_set
2778 				    (error, ENOTSUP,
2779 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2780 				     "Geneve tunnel must be fully defined");
2781 	return 0;
2782 }
2783 
2784 /**
2785  * Validate Geneve TLV option item.
2786  *
2787  * @param[in] item
2788  *   Item specification.
2789  * @param[in] last_item
2790  *   Previous validated item in the pattern items.
2791  * @param[in] geneve_item
2792  *   Previous GENEVE item specification.
2793  * @param[in] dev
2794  *   Pointer to the rte_eth_dev structure.
2795  * @param[out] error
2796  *   Pointer to error structure.
2797  *
2798  * @return
2799  *   0 on success, a negative errno value otherwise and rte_errno is set.
2800  */
2801 int
2802 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2803 				   uint64_t last_item,
2804 				   const struct rte_flow_item *geneve_item,
2805 				   struct rte_eth_dev *dev,
2806 				   struct rte_flow_error *error)
2807 {
2808 	struct mlx5_priv *priv = dev->data->dev_private;
2809 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2810 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2811 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2812 	uint8_t data_max_supported =
2813 			hca_attr->max_geneve_tlv_option_data_len * 4;
2814 	struct mlx5_dev_config *config = &priv->config;
2815 	const struct rte_flow_item_geneve *geneve_spec;
2816 	const struct rte_flow_item_geneve *geneve_mask;
2817 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2818 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2819 	unsigned int i;
2820 	unsigned int data_len;
2821 	uint8_t tlv_option_len;
2822 	uint16_t optlen_m, optlen_v;
2823 	const struct rte_flow_item_geneve_opt full_mask = {
2824 		.option_class = RTE_BE16(0xffff),
2825 		.option_type = 0xff,
2826 		.option_len = 0x1f,
2827 	};
2828 
2829 	if (!mask)
2830 		mask = &rte_flow_item_geneve_opt_mask;
2831 	if (!spec)
2832 		return rte_flow_error_set
2833 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2834 			"Geneve TLV opt class/type/length must be specified");
2835 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2836 		return rte_flow_error_set
2837 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2838 			"Geneve TLV opt length exceeeds the limit (31)");
2839 	/* Check if class type and length masks are full. */
2840 	if (full_mask.option_class != mask->option_class ||
2841 	    full_mask.option_type != mask->option_type ||
2842 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2843 		return rte_flow_error_set
2844 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2845 			"Geneve TLV opt class/type/length masks must be full");
2846 	/* Check if length is supported */
2847 	if ((uint32_t)spec->option_len >
2848 			config->hca_attr.max_geneve_tlv_option_data_len)
2849 		return rte_flow_error_set
2850 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2851 			"Geneve TLV opt length not supported");
2852 	if (config->hca_attr.max_geneve_tlv_options > 1)
2853 		DRV_LOG(DEBUG,
2854 			"max_geneve_tlv_options supports more than 1 option");
2855 	/* Check GENEVE item preceding. */
2856 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2857 		return rte_flow_error_set
2858 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 			"Geneve opt item must be preceded with Geneve item");
2860 	geneve_spec = geneve_item->spec;
2861 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2862 					  &rte_flow_item_geneve_mask;
2863 	/* Check if GENEVE TLV option size doesn't exceed option length */
2864 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2865 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2866 		tlv_option_len = spec->option_len & mask->option_len;
2867 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2868 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2869 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2870 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2871 		if ((optlen_v & optlen_m) <= tlv_option_len)
2872 			return rte_flow_error_set
2873 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 				 "GENEVE TLV option length exceeds optlen");
2875 	}
2876 	/* Check if length is 0 or data is 0. */
2877 	if (spec->data == NULL || spec->option_len == 0)
2878 		return rte_flow_error_set
2879 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2880 			"Geneve TLV opt with zero data/length not supported");
2881 	/* Check not all data & mask are 0. */
2882 	data_len = spec->option_len * 4;
2883 	if (mask->data == NULL) {
2884 		for (i = 0; i < data_len; i++)
2885 			if (spec->data[i])
2886 				break;
2887 		if (i == data_len)
2888 			return rte_flow_error_set(error, ENOTSUP,
2889 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2890 				"Can't match on Geneve option data 0");
2891 	} else {
2892 		for (i = 0; i < data_len; i++)
2893 			if (spec->data[i] & mask->data[i])
2894 				break;
2895 		if (i == data_len)
2896 			return rte_flow_error_set(error, ENOTSUP,
2897 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2898 				"Can't match on Geneve option data and mask 0");
2899 		/* Check data mask supported. */
2900 		for (i = data_max_supported; i < data_len ; i++)
2901 			if (mask->data[i])
2902 				return rte_flow_error_set(error, ENOTSUP,
2903 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2904 					"Data mask is of unsupported size");
2905 	}
2906 	/* Check GENEVE option is supported in NIC. */
2907 	if (!config->hca_attr.geneve_tlv_opt)
2908 		return rte_flow_error_set
2909 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2910 			"Geneve TLV opt not supported");
2911 	/* Check if we already have geneve option with different type/class. */
2912 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2913 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2914 	if (geneve_opt_resource != NULL)
2915 		if (geneve_opt_resource->option_class != spec->option_class ||
2916 		    geneve_opt_resource->option_type != spec->option_type ||
2917 		    geneve_opt_resource->length != spec->option_len) {
2918 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2919 			return rte_flow_error_set(error, ENOTSUP,
2920 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2921 				"Only one Geneve TLV option supported");
2922 		}
2923 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2924 	return 0;
2925 }
2926 
2927 /**
2928  * Validate MPLS item.
2929  *
2930  * @param[in] dev
2931  *   Pointer to the rte_eth_dev structure.
2932  * @param[in] item
2933  *   Item specification.
2934  * @param[in] item_flags
2935  *   Bit-fields that holds the items detected until now.
2936  * @param[in] prev_layer
2937  *   The protocol layer indicated in previous item.
2938  * @param[out] error
2939  *   Pointer to error structure.
2940  *
2941  * @return
2942  *   0 on success, a negative errno value otherwise and rte_errno is set.
2943  */
2944 int
2945 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2946 			     const struct rte_flow_item *item __rte_unused,
2947 			     uint64_t item_flags __rte_unused,
2948 			     uint64_t prev_layer __rte_unused,
2949 			     struct rte_flow_error *error)
2950 {
2951 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2952 	const struct rte_flow_item_mpls *mask = item->mask;
2953 	struct mlx5_priv *priv = dev->data->dev_private;
2954 	int ret;
2955 
2956 	if (!priv->config.mpls_en)
2957 		return rte_flow_error_set(error, ENOTSUP,
2958 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2959 					  "MPLS not supported or"
2960 					  " disabled in firmware"
2961 					  " configuration.");
2962 	/* MPLS over UDP, GRE is allowed */
2963 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2964 			    MLX5_FLOW_LAYER_GRE |
2965 			    MLX5_FLOW_LAYER_GRE_KEY)))
2966 		return rte_flow_error_set(error, EINVAL,
2967 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2968 					  "protocol filtering not compatible"
2969 					  " with MPLS layer");
2970 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2971 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2972 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2973 		return rte_flow_error_set(error, ENOTSUP,
2974 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2975 					  "multiple tunnel layers not"
2976 					  " supported");
2977 	if (!mask)
2978 		mask = &rte_flow_item_mpls_mask;
2979 	ret = mlx5_flow_item_acceptable
2980 		(item, (const uint8_t *)mask,
2981 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2982 		 sizeof(struct rte_flow_item_mpls),
2983 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2984 	if (ret < 0)
2985 		return ret;
2986 	return 0;
2987 #else
2988 	return rte_flow_error_set(error, ENOTSUP,
2989 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2990 				  "MPLS is not supported by Verbs, please"
2991 				  " update.");
2992 #endif
2993 }
2994 
2995 /**
2996  * Validate NVGRE item.
2997  *
2998  * @param[in] item
2999  *   Item specification.
3000  * @param[in] item_flags
3001  *   Bit flags to mark detected items.
3002  * @param[in] target_protocol
3003  *   The next protocol in the previous item.
3004  * @param[out] error
3005  *   Pointer to error structure.
3006  *
3007  * @return
3008  *   0 on success, a negative errno value otherwise and rte_errno is set.
3009  */
3010 int
3011 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3012 			      uint64_t item_flags,
3013 			      uint8_t target_protocol,
3014 			      struct rte_flow_error *error)
3015 {
3016 	const struct rte_flow_item_nvgre *mask = item->mask;
3017 	int ret;
3018 
3019 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3020 		return rte_flow_error_set(error, EINVAL,
3021 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3022 					  "protocol filtering not compatible"
3023 					  " with this GRE layer");
3024 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3025 		return rte_flow_error_set(error, ENOTSUP,
3026 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3027 					  "multiple tunnel layers not"
3028 					  " supported");
3029 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3030 		return rte_flow_error_set(error, ENOTSUP,
3031 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3032 					  "L3 Layer is missing");
3033 	if (!mask)
3034 		mask = &rte_flow_item_nvgre_mask;
3035 	ret = mlx5_flow_item_acceptable
3036 		(item, (const uint8_t *)mask,
3037 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3038 		 sizeof(struct rte_flow_item_nvgre),
3039 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3040 	if (ret < 0)
3041 		return ret;
3042 	return 0;
3043 }
3044 
3045 /**
3046  * Validate eCPRI item.
3047  *
3048  * @param[in] item
3049  *   Item specification.
3050  * @param[in] item_flags
3051  *   Bit-fields that holds the items detected until now.
3052  * @param[in] last_item
3053  *   Previous validated item in the pattern items.
3054  * @param[in] ether_type
3055  *   Type in the ethernet layer header (including dot1q).
3056  * @param[in] acc_mask
3057  *   Acceptable mask, if NULL default internal default mask
3058  *   will be used to check whether item fields are supported.
3059  * @param[out] error
3060  *   Pointer to error structure.
3061  *
3062  * @return
3063  *   0 on success, a negative errno value otherwise and rte_errno is set.
3064  */
3065 int
3066 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3067 			      uint64_t item_flags,
3068 			      uint64_t last_item,
3069 			      uint16_t ether_type,
3070 			      const struct rte_flow_item_ecpri *acc_mask,
3071 			      struct rte_flow_error *error)
3072 {
3073 	const struct rte_flow_item_ecpri *mask = item->mask;
3074 	const struct rte_flow_item_ecpri nic_mask = {
3075 		.hdr = {
3076 			.common = {
3077 				.u32 =
3078 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3079 					.type = 0xFF,
3080 					}).u32),
3081 			},
3082 			.dummy[0] = 0xFFFFFFFF,
3083 		},
3084 	};
3085 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3086 					MLX5_FLOW_LAYER_OUTER_VLAN);
3087 	struct rte_flow_item_ecpri mask_lo;
3088 
3089 	if (!(last_item & outer_l2_vlan) &&
3090 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3091 		return rte_flow_error_set(error, EINVAL,
3092 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3093 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3094 	if ((last_item & outer_l2_vlan) && ether_type &&
3095 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3096 		return rte_flow_error_set(error, EINVAL,
3097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3098 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3099 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3100 		return rte_flow_error_set(error, EINVAL,
3101 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3102 					  "eCPRI with tunnel is not supported right now");
3103 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3104 		return rte_flow_error_set(error, ENOTSUP,
3105 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3106 					  "multiple L3 layers not supported");
3107 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3108 		return rte_flow_error_set(error, EINVAL,
3109 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3110 					  "eCPRI cannot coexist with a TCP layer");
3111 	/* In specification, eCPRI could be over UDP layer. */
3112 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3113 		return rte_flow_error_set(error, EINVAL,
3114 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3115 					  "eCPRI over UDP layer is not yet supported right now");
3116 	/* Mask for type field in common header could be zero. */
3117 	if (!mask)
3118 		mask = &rte_flow_item_ecpri_mask;
3119 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3120 	/* Input mask is in big-endian format. */
3121 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3122 		return rte_flow_error_set(error, EINVAL,
3123 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3124 					  "partial mask is not supported for protocol");
3125 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3126 		return rte_flow_error_set(error, EINVAL,
3127 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3128 					  "message header mask must be after a type mask");
3129 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3130 					 acc_mask ? (const uint8_t *)acc_mask
3131 						  : (const uint8_t *)&nic_mask,
3132 					 sizeof(struct rte_flow_item_ecpri),
3133 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3134 }
3135 
3136 static int
3137 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3138 		   const struct rte_flow_attr *attr __rte_unused,
3139 		   const struct rte_flow_item items[] __rte_unused,
3140 		   const struct rte_flow_action actions[] __rte_unused,
3141 		   bool external __rte_unused,
3142 		   int hairpin __rte_unused,
3143 		   struct rte_flow_error *error)
3144 {
3145 	return rte_flow_error_set(error, ENOTSUP,
3146 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3147 }
3148 
3149 static struct mlx5_flow *
3150 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3151 		  const struct rte_flow_attr *attr __rte_unused,
3152 		  const struct rte_flow_item items[] __rte_unused,
3153 		  const struct rte_flow_action actions[] __rte_unused,
3154 		  struct rte_flow_error *error)
3155 {
3156 	rte_flow_error_set(error, ENOTSUP,
3157 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3158 	return NULL;
3159 }
3160 
3161 static int
3162 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3163 		    struct mlx5_flow *dev_flow __rte_unused,
3164 		    const struct rte_flow_attr *attr __rte_unused,
3165 		    const struct rte_flow_item items[] __rte_unused,
3166 		    const struct rte_flow_action actions[] __rte_unused,
3167 		    struct rte_flow_error *error)
3168 {
3169 	return rte_flow_error_set(error, ENOTSUP,
3170 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3171 }
3172 
3173 static int
3174 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3175 		struct rte_flow *flow __rte_unused,
3176 		struct rte_flow_error *error)
3177 {
3178 	return rte_flow_error_set(error, ENOTSUP,
3179 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3180 }
3181 
3182 static void
3183 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3184 		 struct rte_flow *flow __rte_unused)
3185 {
3186 }
3187 
3188 static void
3189 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3190 		  struct rte_flow *flow __rte_unused)
3191 {
3192 }
3193 
3194 static int
3195 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3196 		struct rte_flow *flow __rte_unused,
3197 		const struct rte_flow_action *actions __rte_unused,
3198 		void *data __rte_unused,
3199 		struct rte_flow_error *error)
3200 {
3201 	return rte_flow_error_set(error, ENOTSUP,
3202 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3203 }
3204 
3205 static int
3206 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3207 		      uint32_t domains __rte_unused,
3208 		      uint32_t flags __rte_unused)
3209 {
3210 	return 0;
3211 }
3212 
3213 /* Void driver to protect from null pointer reference. */
3214 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3215 	.validate = flow_null_validate,
3216 	.prepare = flow_null_prepare,
3217 	.translate = flow_null_translate,
3218 	.apply = flow_null_apply,
3219 	.remove = flow_null_remove,
3220 	.destroy = flow_null_destroy,
3221 	.query = flow_null_query,
3222 	.sync_domain = flow_null_sync_domain,
3223 };
3224 
3225 /**
3226  * Select flow driver type according to flow attributes and device
3227  * configuration.
3228  *
3229  * @param[in] dev
3230  *   Pointer to the dev structure.
3231  * @param[in] attr
3232  *   Pointer to the flow attributes.
3233  *
3234  * @return
3235  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3236  */
3237 static enum mlx5_flow_drv_type
3238 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3239 {
3240 	struct mlx5_priv *priv = dev->data->dev_private;
3241 	/* The OS can determine first a specific flow type (DV, VERBS) */
3242 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3243 
3244 	if (type != MLX5_FLOW_TYPE_MAX)
3245 		return type;
3246 	/* If no OS specific type - continue with DV/VERBS selection */
3247 	if (attr->transfer && priv->config.dv_esw_en)
3248 		type = MLX5_FLOW_TYPE_DV;
3249 	if (!attr->transfer)
3250 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3251 						 MLX5_FLOW_TYPE_VERBS;
3252 	return type;
3253 }
3254 
3255 #define flow_get_drv_ops(type) flow_drv_ops[type]
3256 
3257 /**
3258  * Flow driver validation API. This abstracts calling driver specific functions.
3259  * The type of flow driver is determined according to flow attributes.
3260  *
3261  * @param[in] dev
3262  *   Pointer to the dev structure.
3263  * @param[in] attr
3264  *   Pointer to the flow attributes.
3265  * @param[in] items
3266  *   Pointer to the list of items.
3267  * @param[in] actions
3268  *   Pointer to the list of actions.
3269  * @param[in] external
3270  *   This flow rule is created by request external to PMD.
3271  * @param[in] hairpin
3272  *   Number of hairpin TX actions, 0 means classic flow.
3273  * @param[out] error
3274  *   Pointer to the error structure.
3275  *
3276  * @return
3277  *   0 on success, a negative errno value otherwise and rte_errno is set.
3278  */
3279 static inline int
3280 flow_drv_validate(struct rte_eth_dev *dev,
3281 		  const struct rte_flow_attr *attr,
3282 		  const struct rte_flow_item items[],
3283 		  const struct rte_flow_action actions[],
3284 		  bool external, int hairpin, struct rte_flow_error *error)
3285 {
3286 	const struct mlx5_flow_driver_ops *fops;
3287 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3288 
3289 	fops = flow_get_drv_ops(type);
3290 	return fops->validate(dev, attr, items, actions, external,
3291 			      hairpin, error);
3292 }
3293 
3294 /**
3295  * Flow driver preparation API. This abstracts calling driver specific
3296  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3297  * calculates the size of memory required for device flow, allocates the memory,
3298  * initializes the device flow and returns the pointer.
3299  *
3300  * @note
3301  *   This function initializes device flow structure such as dv or verbs in
3302  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3303  *   rest. For example, adding returning device flow to flow->dev_flow list and
3304  *   setting backward reference to the flow should be done out of this function.
3305  *   layers field is not filled either.
3306  *
3307  * @param[in] dev
3308  *   Pointer to the dev structure.
3309  * @param[in] attr
3310  *   Pointer to the flow attributes.
3311  * @param[in] items
3312  *   Pointer to the list of items.
3313  * @param[in] actions
3314  *   Pointer to the list of actions.
3315  * @param[in] flow_idx
3316  *   This memory pool index to the flow.
3317  * @param[out] error
3318  *   Pointer to the error structure.
3319  *
3320  * @return
3321  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3322  */
3323 static inline struct mlx5_flow *
3324 flow_drv_prepare(struct rte_eth_dev *dev,
3325 		 const struct rte_flow *flow,
3326 		 const struct rte_flow_attr *attr,
3327 		 const struct rte_flow_item items[],
3328 		 const struct rte_flow_action actions[],
3329 		 uint32_t flow_idx,
3330 		 struct rte_flow_error *error)
3331 {
3332 	const struct mlx5_flow_driver_ops *fops;
3333 	enum mlx5_flow_drv_type type = flow->drv_type;
3334 	struct mlx5_flow *mlx5_flow = NULL;
3335 
3336 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3337 	fops = flow_get_drv_ops(type);
3338 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3339 	if (mlx5_flow)
3340 		mlx5_flow->flow_idx = flow_idx;
3341 	return mlx5_flow;
3342 }
3343 
3344 /**
3345  * Flow driver translation API. This abstracts calling driver specific
3346  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3347  * translates a generic flow into a driver flow. flow_drv_prepare() must
3348  * precede.
3349  *
3350  * @note
3351  *   dev_flow->layers could be filled as a result of parsing during translation
3352  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3353  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3354  *   flow->actions could be overwritten even though all the expanded dev_flows
3355  *   have the same actions.
3356  *
3357  * @param[in] dev
3358  *   Pointer to the rte dev structure.
3359  * @param[in, out] dev_flow
3360  *   Pointer to the mlx5 flow.
3361  * @param[in] attr
3362  *   Pointer to the flow attributes.
3363  * @param[in] items
3364  *   Pointer to the list of items.
3365  * @param[in] actions
3366  *   Pointer to the list of actions.
3367  * @param[out] error
3368  *   Pointer to the error structure.
3369  *
3370  * @return
3371  *   0 on success, a negative errno value otherwise and rte_errno is set.
3372  */
3373 static inline int
3374 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3375 		   const struct rte_flow_attr *attr,
3376 		   const struct rte_flow_item items[],
3377 		   const struct rte_flow_action actions[],
3378 		   struct rte_flow_error *error)
3379 {
3380 	const struct mlx5_flow_driver_ops *fops;
3381 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3382 
3383 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3384 	fops = flow_get_drv_ops(type);
3385 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3386 }
3387 
3388 /**
3389  * Flow driver apply API. This abstracts calling driver specific functions.
3390  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3391  * translated driver flows on to device. flow_drv_translate() must precede.
3392  *
3393  * @param[in] dev
3394  *   Pointer to Ethernet device structure.
3395  * @param[in, out] flow
3396  *   Pointer to flow structure.
3397  * @param[out] error
3398  *   Pointer to error structure.
3399  *
3400  * @return
3401  *   0 on success, a negative errno value otherwise and rte_errno is set.
3402  */
3403 static inline int
3404 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3405 	       struct rte_flow_error *error)
3406 {
3407 	const struct mlx5_flow_driver_ops *fops;
3408 	enum mlx5_flow_drv_type type = flow->drv_type;
3409 
3410 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3411 	fops = flow_get_drv_ops(type);
3412 	return fops->apply(dev, flow, error);
3413 }
3414 
3415 /**
3416  * Flow driver destroy API. This abstracts calling driver specific functions.
3417  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3418  * on device and releases resources of the flow.
3419  *
3420  * @param[in] dev
3421  *   Pointer to Ethernet device.
3422  * @param[in, out] flow
3423  *   Pointer to flow structure.
3424  */
3425 static inline void
3426 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3427 {
3428 	const struct mlx5_flow_driver_ops *fops;
3429 	enum mlx5_flow_drv_type type = flow->drv_type;
3430 
3431 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3432 	fops = flow_get_drv_ops(type);
3433 	fops->destroy(dev, flow);
3434 }
3435 
3436 /**
3437  * Flow driver find RSS policy tbl API. This abstracts calling driver
3438  * specific functions. Parent flow (rte_flow) should have driver
3439  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3440  *
3441  * @param[in] dev
3442  *   Pointer to Ethernet device.
3443  * @param[in, out] flow
3444  *   Pointer to flow structure.
3445  * @param[in] policy
3446  *   Pointer to meter policy table.
3447  * @param[in] rss_desc
3448  *   Pointer to rss_desc
3449  */
3450 static struct mlx5_flow_meter_sub_policy *
3451 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3452 		struct rte_flow *flow,
3453 		struct mlx5_flow_meter_policy *policy,
3454 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3455 {
3456 	const struct mlx5_flow_driver_ops *fops;
3457 	enum mlx5_flow_drv_type type = flow->drv_type;
3458 
3459 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3460 	fops = flow_get_drv_ops(type);
3461 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3462 }
3463 
3464 /**
3465  * Flow driver color tag rule API. This abstracts calling driver
3466  * specific functions. Parent flow (rte_flow) should have driver
3467  * type (drv_type). It will create the color tag rules in hierarchy meter.
3468  *
3469  * @param[in] dev
3470  *   Pointer to Ethernet device.
3471  * @param[in, out] flow
3472  *   Pointer to flow structure.
3473  * @param[in] fm
3474  *   Pointer to flow meter structure.
3475  * @param[in] src_port
3476  *   The src port this extra rule should use.
3477  * @param[in] item
3478  *   The src port id match item.
3479  * @param[out] error
3480  *   Pointer to error structure.
3481  */
3482 static int
3483 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3484 		struct rte_flow *flow,
3485 		struct mlx5_flow_meter_info *fm,
3486 		int32_t src_port,
3487 		const struct rte_flow_item *item,
3488 		struct rte_flow_error *error)
3489 {
3490 	const struct mlx5_flow_driver_ops *fops;
3491 	enum mlx5_flow_drv_type type = flow->drv_type;
3492 
3493 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3494 	fops = flow_get_drv_ops(type);
3495 	return fops->meter_hierarchy_rule_create(dev, fm,
3496 						src_port, item, error);
3497 }
3498 
3499 /**
3500  * Get RSS action from the action list.
3501  *
3502  * @param[in] dev
3503  *   Pointer to Ethernet device.
3504  * @param[in] actions
3505  *   Pointer to the list of actions.
3506  * @param[in] flow
3507  *   Parent flow structure pointer.
3508  *
3509  * @return
3510  *   Pointer to the RSS action if exist, else return NULL.
3511  */
3512 static const struct rte_flow_action_rss*
3513 flow_get_rss_action(struct rte_eth_dev *dev,
3514 		    const struct rte_flow_action actions[])
3515 {
3516 	struct mlx5_priv *priv = dev->data->dev_private;
3517 	const struct rte_flow_action_rss *rss = NULL;
3518 
3519 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3520 		switch (actions->type) {
3521 		case RTE_FLOW_ACTION_TYPE_RSS:
3522 			rss = actions->conf;
3523 			break;
3524 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3525 		{
3526 			const struct rte_flow_action_sample *sample =
3527 								actions->conf;
3528 			const struct rte_flow_action *act = sample->actions;
3529 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3530 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3531 					rss = act->conf;
3532 			break;
3533 		}
3534 		case RTE_FLOW_ACTION_TYPE_METER:
3535 		{
3536 			uint32_t mtr_idx;
3537 			struct mlx5_flow_meter_info *fm;
3538 			struct mlx5_flow_meter_policy *policy;
3539 			const struct rte_flow_action_meter *mtr = actions->conf;
3540 
3541 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3542 			if (fm && !fm->def_policy) {
3543 				policy = mlx5_flow_meter_policy_find(dev,
3544 						fm->policy_id, NULL);
3545 				MLX5_ASSERT(policy);
3546 				if (policy->is_hierarchy) {
3547 					policy =
3548 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3549 									policy);
3550 					if (!policy)
3551 						return NULL;
3552 				}
3553 				if (policy->is_rss)
3554 					rss =
3555 				policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3556 			}
3557 			break;
3558 		}
3559 		default:
3560 			break;
3561 		}
3562 	}
3563 	return rss;
3564 }
3565 
3566 /**
3567  * Get ASO age action by index.
3568  *
3569  * @param[in] dev
3570  *   Pointer to the Ethernet device structure.
3571  * @param[in] age_idx
3572  *   Index to the ASO age action.
3573  *
3574  * @return
3575  *   The specified ASO age action.
3576  */
3577 struct mlx5_aso_age_action*
3578 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3579 {
3580 	uint16_t pool_idx = age_idx & UINT16_MAX;
3581 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3582 	struct mlx5_priv *priv = dev->data->dev_private;
3583 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3584 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3585 
3586 	return &pool->actions[offset - 1];
3587 }
3588 
3589 /* maps indirect action to translated direct in some actions array */
3590 struct mlx5_translated_action_handle {
3591 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3592 	int index; /**< Index in related array of rte_flow_action. */
3593 };
3594 
3595 /**
3596  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3597  * direct action if translation possible.
3598  * This functionality used to run same execution path for both direct and
3599  * indirect actions on flow create. All necessary preparations for indirect
3600  * action handling should be performed on *handle* actions list returned
3601  * from this call.
3602  *
3603  * @param[in] dev
3604  *   Pointer to Ethernet device.
3605  * @param[in] actions
3606  *   List of actions to translate.
3607  * @param[out] handle
3608  *   List to store translated indirect action object handles.
3609  * @param[in, out] indir_n
3610  *   Size of *handle* array. On return should be updated with number of
3611  *   indirect actions retrieved from the *actions* list.
3612  * @param[out] translated_actions
3613  *   List of actions where all indirect actions were translated to direct
3614  *   if possible. NULL if no translation took place.
3615  * @param[out] error
3616  *   Pointer to the error structure.
3617  *
3618  * @return
3619  *   0 on success, a negative errno value otherwise and rte_errno is set.
3620  */
3621 static int
3622 flow_action_handles_translate(struct rte_eth_dev *dev,
3623 			      const struct rte_flow_action actions[],
3624 			      struct mlx5_translated_action_handle *handle,
3625 			      int *indir_n,
3626 			      struct rte_flow_action **translated_actions,
3627 			      struct rte_flow_error *error)
3628 {
3629 	struct mlx5_priv *priv = dev->data->dev_private;
3630 	struct rte_flow_action *translated = NULL;
3631 	size_t actions_size;
3632 	int n;
3633 	int copied_n = 0;
3634 	struct mlx5_translated_action_handle *handle_end = NULL;
3635 
3636 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3637 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3638 			continue;
3639 		if (copied_n == *indir_n) {
3640 			return rte_flow_error_set
3641 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3642 				 NULL, "too many shared actions");
3643 		}
3644 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3645 			   sizeof(actions[n].conf));
3646 		handle[copied_n].index = n;
3647 		copied_n++;
3648 	}
3649 	n++;
3650 	*indir_n = copied_n;
3651 	if (!copied_n)
3652 		return 0;
3653 	actions_size = sizeof(struct rte_flow_action) * n;
3654 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3655 	if (!translated) {
3656 		rte_errno = ENOMEM;
3657 		return -ENOMEM;
3658 	}
3659 	memcpy(translated, actions, actions_size);
3660 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3661 		struct mlx5_shared_action_rss *shared_rss;
3662 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3663 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3664 		uint32_t idx = act_idx &
3665 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3666 
3667 		switch (type) {
3668 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3669 			shared_rss = mlx5_ipool_get
3670 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3671 			translated[handle->index].type =
3672 				RTE_FLOW_ACTION_TYPE_RSS;
3673 			translated[handle->index].conf =
3674 				&shared_rss->origin;
3675 			break;
3676 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3677 			translated[handle->index].type =
3678 						(enum rte_flow_action_type)
3679 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3680 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3681 			break;
3682 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3683 			if (priv->sh->flow_hit_aso_en) {
3684 				translated[handle->index].type =
3685 					(enum rte_flow_action_type)
3686 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3687 				translated[handle->index].conf =
3688 							 (void *)(uintptr_t)idx;
3689 				break;
3690 			}
3691 			/* Fall-through */
3692 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3693 			if (priv->sh->ct_aso_en) {
3694 				translated[handle->index].type =
3695 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3696 				translated[handle->index].conf =
3697 							 (void *)(uintptr_t)idx;
3698 				break;
3699 			}
3700 			/* Fall-through */
3701 		default:
3702 			mlx5_free(translated);
3703 			return rte_flow_error_set
3704 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3705 				 NULL, "invalid indirect action type");
3706 		}
3707 	}
3708 	*translated_actions = translated;
3709 	return 0;
3710 }
3711 
3712 /**
3713  * Get Shared RSS action from the action list.
3714  *
3715  * @param[in] dev
3716  *   Pointer to Ethernet device.
3717  * @param[in] shared
3718  *   Pointer to the list of actions.
3719  * @param[in] shared_n
3720  *   Actions list length.
3721  *
3722  * @return
3723  *   The MLX5 RSS action ID if exists, otherwise return 0.
3724  */
3725 static uint32_t
3726 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3727 			   struct mlx5_translated_action_handle *handle,
3728 			   int shared_n)
3729 {
3730 	struct mlx5_translated_action_handle *handle_end;
3731 	struct mlx5_priv *priv = dev->data->dev_private;
3732 	struct mlx5_shared_action_rss *shared_rss;
3733 
3734 
3735 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3736 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3737 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3738 		uint32_t idx = act_idx &
3739 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3740 		switch (type) {
3741 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3742 			shared_rss = mlx5_ipool_get
3743 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3744 									   idx);
3745 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3746 					   __ATOMIC_RELAXED);
3747 			return idx;
3748 		default:
3749 			break;
3750 		}
3751 	}
3752 	return 0;
3753 }
3754 
3755 static unsigned int
3756 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3757 {
3758 	const struct rte_flow_item *item;
3759 	unsigned int has_vlan = 0;
3760 
3761 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3762 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3763 			has_vlan = 1;
3764 			break;
3765 		}
3766 	}
3767 	if (has_vlan)
3768 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3769 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3770 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3771 			       MLX5_EXPANSION_ROOT_OUTER;
3772 }
3773 
3774 /**
3775  *  Get layer flags from the prefix flow.
3776  *
3777  *  Some flows may be split to several subflows, the prefix subflow gets the
3778  *  match items and the suffix sub flow gets the actions.
3779  *  Some actions need the user defined match item flags to get the detail for
3780  *  the action.
3781  *  This function helps the suffix flow to get the item layer flags from prefix
3782  *  subflow.
3783  *
3784  * @param[in] dev_flow
3785  *   Pointer the created preifx subflow.
3786  *
3787  * @return
3788  *   The layers get from prefix subflow.
3789  */
3790 static inline uint64_t
3791 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3792 {
3793 	uint64_t layers = 0;
3794 
3795 	/*
3796 	 * Layers bits could be localization, but usually the compiler will
3797 	 * help to do the optimization work for source code.
3798 	 * If no decap actions, use the layers directly.
3799 	 */
3800 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3801 		return dev_flow->handle->layers;
3802 	/* Convert L3 layers with decap action. */
3803 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3804 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3805 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3806 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3807 	/* Convert L4 layers with decap action.  */
3808 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3809 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3810 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3811 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3812 	return layers;
3813 }
3814 
3815 /**
3816  * Get metadata split action information.
3817  *
3818  * @param[in] actions
3819  *   Pointer to the list of actions.
3820  * @param[out] qrss
3821  *   Pointer to the return pointer.
3822  * @param[out] qrss_type
3823  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3824  *   if no QUEUE/RSS is found.
3825  * @param[out] encap_idx
3826  *   Pointer to the index of the encap action if exists, otherwise the last
3827  *   action index.
3828  *
3829  * @return
3830  *   Total number of actions.
3831  */
3832 static int
3833 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3834 				       const struct rte_flow_action **qrss,
3835 				       int *encap_idx)
3836 {
3837 	const struct rte_flow_action_raw_encap *raw_encap;
3838 	int actions_n = 0;
3839 	int raw_decap_idx = -1;
3840 
3841 	*encap_idx = -1;
3842 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3843 		switch (actions->type) {
3844 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3845 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3846 			*encap_idx = actions_n;
3847 			break;
3848 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3849 			raw_decap_idx = actions_n;
3850 			break;
3851 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3852 			raw_encap = actions->conf;
3853 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3854 				*encap_idx = raw_decap_idx != -1 ?
3855 						      raw_decap_idx : actions_n;
3856 			break;
3857 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3858 		case RTE_FLOW_ACTION_TYPE_RSS:
3859 			*qrss = actions;
3860 			break;
3861 		default:
3862 			break;
3863 		}
3864 		actions_n++;
3865 	}
3866 	if (*encap_idx == -1)
3867 		*encap_idx = actions_n;
3868 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3869 	return actions_n + 1;
3870 }
3871 
3872 /**
3873  * Check if the action will change packet.
3874  *
3875  * @param dev
3876  *   Pointer to Ethernet device.
3877  * @param[in] type
3878  *   action type.
3879  *
3880  * @return
3881  *   true if action will change packet, false otherwise.
3882  */
3883 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3884 					  enum rte_flow_action_type type)
3885 {
3886 	struct mlx5_priv *priv = dev->data->dev_private;
3887 
3888 	switch (type) {
3889 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3890 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3891 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3892 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3893 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3894 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3895 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3896 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3897 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3898 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
3899 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3900 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3901 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3902 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3903 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3904 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3905 	case RTE_FLOW_ACTION_TYPE_SET_META:
3906 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
3907 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3908 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3909 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3910 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3911 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3912 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3913 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3914 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3915 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3916 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3917 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3918 		return true;
3919 	case RTE_FLOW_ACTION_TYPE_FLAG:
3920 	case RTE_FLOW_ACTION_TYPE_MARK:
3921 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3922 			return true;
3923 		else
3924 			return false;
3925 	default:
3926 		return false;
3927 	}
3928 }
3929 
3930 /**
3931  * Check meter action from the action list.
3932  *
3933  * @param dev
3934  *   Pointer to Ethernet device.
3935  * @param[in] actions
3936  *   Pointer to the list of actions.
3937  * @param[out] has_mtr
3938  *   Pointer to the meter exist flag.
3939  * @param[out] has_modify
3940  *   Pointer to the flag showing there's packet change action.
3941  * @param[out] meter_id
3942  *   Pointer to the meter id.
3943  *
3944  * @return
3945  *   Total number of actions.
3946  */
3947 static int
3948 flow_check_meter_action(struct rte_eth_dev *dev,
3949 			const struct rte_flow_action actions[],
3950 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3951 {
3952 	const struct rte_flow_action_meter *mtr = NULL;
3953 	int actions_n = 0;
3954 
3955 	MLX5_ASSERT(has_mtr);
3956 	*has_mtr = false;
3957 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3958 		switch (actions->type) {
3959 		case RTE_FLOW_ACTION_TYPE_METER:
3960 			mtr = actions->conf;
3961 			*meter_id = mtr->mtr_id;
3962 			*has_mtr = true;
3963 			break;
3964 		default:
3965 			break;
3966 		}
3967 		if (!*has_mtr)
3968 			*has_modify |= flow_check_modify_action_type(dev,
3969 								actions->type);
3970 		actions_n++;
3971 	}
3972 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3973 	return actions_n + 1;
3974 }
3975 
3976 /**
3977  * Check if the flow should be split due to hairpin.
3978  * The reason for the split is that in current HW we can't
3979  * support encap and push-vlan on Rx, so if a flow contains
3980  * these actions we move it to Tx.
3981  *
3982  * @param dev
3983  *   Pointer to Ethernet device.
3984  * @param[in] attr
3985  *   Flow rule attributes.
3986  * @param[in] actions
3987  *   Associated actions (list terminated by the END action).
3988  *
3989  * @return
3990  *   > 0 the number of actions and the flow should be split,
3991  *   0 when no split required.
3992  */
3993 static int
3994 flow_check_hairpin_split(struct rte_eth_dev *dev,
3995 			 const struct rte_flow_attr *attr,
3996 			 const struct rte_flow_action actions[])
3997 {
3998 	int queue_action = 0;
3999 	int action_n = 0;
4000 	int split = 0;
4001 	const struct rte_flow_action_queue *queue;
4002 	const struct rte_flow_action_rss *rss;
4003 	const struct rte_flow_action_raw_encap *raw_encap;
4004 	const struct rte_eth_hairpin_conf *conf;
4005 
4006 	if (!attr->ingress)
4007 		return 0;
4008 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4009 		switch (actions->type) {
4010 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4011 			queue = actions->conf;
4012 			if (queue == NULL)
4013 				return 0;
4014 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4015 			if (conf == NULL || conf->tx_explicit != 0)
4016 				return 0;
4017 			queue_action = 1;
4018 			action_n++;
4019 			break;
4020 		case RTE_FLOW_ACTION_TYPE_RSS:
4021 			rss = actions->conf;
4022 			if (rss == NULL || rss->queue_num == 0)
4023 				return 0;
4024 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4025 			if (conf == NULL || conf->tx_explicit != 0)
4026 				return 0;
4027 			queue_action = 1;
4028 			action_n++;
4029 			break;
4030 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4031 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4032 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4033 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4034 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4035 			split++;
4036 			action_n++;
4037 			break;
4038 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4039 			raw_encap = actions->conf;
4040 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4041 				split++;
4042 			action_n++;
4043 			break;
4044 		default:
4045 			action_n++;
4046 			break;
4047 		}
4048 	}
4049 	if (split && queue_action)
4050 		return action_n;
4051 	return 0;
4052 }
4053 
4054 /* Declare flow create/destroy prototype in advance. */
4055 static uint32_t
4056 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4057 		 const struct rte_flow_attr *attr,
4058 		 const struct rte_flow_item items[],
4059 		 const struct rte_flow_action actions[],
4060 		 bool external, struct rte_flow_error *error);
4061 
4062 static void
4063 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4064 		  uint32_t flow_idx);
4065 
4066 int
4067 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4068 		      struct mlx5_list_entry *entry, void *cb_ctx)
4069 {
4070 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4071 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4072 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4073 
4074 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4075 }
4076 
4077 struct mlx5_list_entry *
4078 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4079 {
4080 	struct rte_eth_dev *dev = tool_ctx;
4081 	struct mlx5_priv *priv = dev->data->dev_private;
4082 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4083 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4084 	struct rte_flow_error *error = ctx->error;
4085 	uint32_t idx = 0;
4086 	int ret;
4087 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4088 	struct rte_flow_attr attr = {
4089 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4090 		.ingress = 1,
4091 	};
4092 	struct mlx5_rte_flow_item_tag tag_spec = {
4093 		.data = mark_id,
4094 	};
4095 	struct rte_flow_item items[] = {
4096 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4097 	};
4098 	struct rte_flow_action_mark ftag = {
4099 		.id = mark_id,
4100 	};
4101 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4102 		.dst = REG_B,
4103 		.src = REG_NON,
4104 	};
4105 	struct rte_flow_action_jump jump = {
4106 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4107 	};
4108 	struct rte_flow_action actions[] = {
4109 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4110 	};
4111 
4112 	/* Fill the register fileds in the flow. */
4113 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4114 	if (ret < 0)
4115 		return NULL;
4116 	tag_spec.id = ret;
4117 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4118 	if (ret < 0)
4119 		return NULL;
4120 	cp_mreg.src = ret;
4121 	/* Provide the full width of FLAG specific value. */
4122 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4123 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4124 	/* Build a new flow. */
4125 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4126 		items[0] = (struct rte_flow_item){
4127 			.type = (enum rte_flow_item_type)
4128 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4129 			.spec = &tag_spec,
4130 		};
4131 		items[1] = (struct rte_flow_item){
4132 			.type = RTE_FLOW_ITEM_TYPE_END,
4133 		};
4134 		actions[0] = (struct rte_flow_action){
4135 			.type = (enum rte_flow_action_type)
4136 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4137 			.conf = &ftag,
4138 		};
4139 		actions[1] = (struct rte_flow_action){
4140 			.type = (enum rte_flow_action_type)
4141 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4142 			.conf = &cp_mreg,
4143 		};
4144 		actions[2] = (struct rte_flow_action){
4145 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4146 			.conf = &jump,
4147 		};
4148 		actions[3] = (struct rte_flow_action){
4149 			.type = RTE_FLOW_ACTION_TYPE_END,
4150 		};
4151 	} else {
4152 		/* Default rule, wildcard match. */
4153 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4154 		items[0] = (struct rte_flow_item){
4155 			.type = RTE_FLOW_ITEM_TYPE_END,
4156 		};
4157 		actions[0] = (struct rte_flow_action){
4158 			.type = (enum rte_flow_action_type)
4159 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4160 			.conf = &cp_mreg,
4161 		};
4162 		actions[1] = (struct rte_flow_action){
4163 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4164 			.conf = &jump,
4165 		};
4166 		actions[2] = (struct rte_flow_action){
4167 			.type = RTE_FLOW_ACTION_TYPE_END,
4168 		};
4169 	}
4170 	/* Build a new entry. */
4171 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4172 	if (!mcp_res) {
4173 		rte_errno = ENOMEM;
4174 		return NULL;
4175 	}
4176 	mcp_res->idx = idx;
4177 	mcp_res->mark_id = mark_id;
4178 	/*
4179 	 * The copy Flows are not included in any list. There
4180 	 * ones are referenced from other Flows and can not
4181 	 * be applied, removed, deleted in ardbitrary order
4182 	 * by list traversing.
4183 	 */
4184 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4185 					&attr, items, actions, false, error);
4186 	if (!mcp_res->rix_flow) {
4187 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4188 		return NULL;
4189 	}
4190 	return &mcp_res->hlist_ent;
4191 }
4192 
4193 struct mlx5_list_entry *
4194 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4195 		      void *cb_ctx __rte_unused)
4196 {
4197 	struct rte_eth_dev *dev = tool_ctx;
4198 	struct mlx5_priv *priv = dev->data->dev_private;
4199 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4200 	uint32_t idx = 0;
4201 
4202 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4203 	if (!mcp_res) {
4204 		rte_errno = ENOMEM;
4205 		return NULL;
4206 	}
4207 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4208 	mcp_res->idx = idx;
4209 	return &mcp_res->hlist_ent;
4210 }
4211 
4212 void
4213 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4214 {
4215 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4216 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4217 	struct rte_eth_dev *dev = tool_ctx;
4218 	struct mlx5_priv *priv = dev->data->dev_private;
4219 
4220 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4221 }
4222 
4223 /**
4224  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4225  *
4226  * As mark_id is unique, if there's already a registered flow for the mark_id,
4227  * return by increasing the reference counter of the resource. Otherwise, create
4228  * the resource (mcp_res) and flow.
4229  *
4230  * Flow looks like,
4231  *   - If ingress port is ANY and reg_c[1] is mark_id,
4232  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4233  *
4234  * For default flow (zero mark_id), flow is like,
4235  *   - If ingress port is ANY,
4236  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4237  *
4238  * @param dev
4239  *   Pointer to Ethernet device.
4240  * @param mark_id
4241  *   ID of MARK action, zero means default flow for META.
4242  * @param[out] error
4243  *   Perform verbose error reporting if not NULL.
4244  *
4245  * @return
4246  *   Associated resource on success, NULL otherwise and rte_errno is set.
4247  */
4248 static struct mlx5_flow_mreg_copy_resource *
4249 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4250 			  struct rte_flow_error *error)
4251 {
4252 	struct mlx5_priv *priv = dev->data->dev_private;
4253 	struct mlx5_list_entry *entry;
4254 	struct mlx5_flow_cb_ctx ctx = {
4255 		.dev = dev,
4256 		.error = error,
4257 		.data = &mark_id,
4258 	};
4259 
4260 	/* Check if already registered. */
4261 	MLX5_ASSERT(priv->mreg_cp_tbl);
4262 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4263 	if (!entry)
4264 		return NULL;
4265 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4266 			    hlist_ent);
4267 }
4268 
4269 void
4270 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4271 {
4272 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4273 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4274 	struct rte_eth_dev *dev = tool_ctx;
4275 	struct mlx5_priv *priv = dev->data->dev_private;
4276 
4277 	MLX5_ASSERT(mcp_res->rix_flow);
4278 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4279 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4280 }
4281 
4282 /**
4283  * Release flow in RX_CP_TBL.
4284  *
4285  * @param dev
4286  *   Pointer to Ethernet device.
4287  * @flow
4288  *   Parent flow for wich copying is provided.
4289  */
4290 static void
4291 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4292 			  struct rte_flow *flow)
4293 {
4294 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4295 	struct mlx5_priv *priv = dev->data->dev_private;
4296 
4297 	if (!flow->rix_mreg_copy)
4298 		return;
4299 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4300 				 flow->rix_mreg_copy);
4301 	if (!mcp_res || !priv->mreg_cp_tbl)
4302 		return;
4303 	MLX5_ASSERT(mcp_res->rix_flow);
4304 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4305 	flow->rix_mreg_copy = 0;
4306 }
4307 
4308 /**
4309  * Remove the default copy action from RX_CP_TBL.
4310  *
4311  * This functions is called in the mlx5_dev_start(). No thread safe
4312  * is guaranteed.
4313  *
4314  * @param dev
4315  *   Pointer to Ethernet device.
4316  */
4317 static void
4318 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4319 {
4320 	struct mlx5_list_entry *entry;
4321 	struct mlx5_priv *priv = dev->data->dev_private;
4322 	struct mlx5_flow_cb_ctx ctx;
4323 	uint32_t mark_id;
4324 
4325 	/* Check if default flow is registered. */
4326 	if (!priv->mreg_cp_tbl)
4327 		return;
4328 	mark_id = MLX5_DEFAULT_COPY_ID;
4329 	ctx.data = &mark_id;
4330 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4331 	if (!entry)
4332 		return;
4333 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4334 }
4335 
4336 /**
4337  * Add the default copy action in in RX_CP_TBL.
4338  *
4339  * This functions is called in the mlx5_dev_start(). No thread safe
4340  * is guaranteed.
4341  *
4342  * @param dev
4343  *   Pointer to Ethernet device.
4344  * @param[out] error
4345  *   Perform verbose error reporting if not NULL.
4346  *
4347  * @return
4348  *   0 for success, negative value otherwise and rte_errno is set.
4349  */
4350 static int
4351 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4352 				  struct rte_flow_error *error)
4353 {
4354 	struct mlx5_priv *priv = dev->data->dev_private;
4355 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4356 	struct mlx5_flow_cb_ctx ctx;
4357 	uint32_t mark_id;
4358 
4359 	/* Check whether extensive metadata feature is engaged. */
4360 	if (!priv->config.dv_flow_en ||
4361 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4362 	    !mlx5_flow_ext_mreg_supported(dev) ||
4363 	    !priv->sh->dv_regc0_mask)
4364 		return 0;
4365 	/*
4366 	 * Add default mreg copy flow may be called multiple time, but
4367 	 * only be called once in stop. Avoid register it twice.
4368 	 */
4369 	mark_id = MLX5_DEFAULT_COPY_ID;
4370 	ctx.data = &mark_id;
4371 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4372 		return 0;
4373 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4374 	if (!mcp_res)
4375 		return -rte_errno;
4376 	return 0;
4377 }
4378 
4379 /**
4380  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4381  *
4382  * All the flow having Q/RSS action should be split by
4383  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4384  * performs the following,
4385  *   - CQE->flow_tag := reg_c[1] (MARK)
4386  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4387  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4388  * but there should be a flow per each MARK ID set by MARK action.
4389  *
4390  * For the aforementioned reason, if there's a MARK action in flow's action
4391  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4392  * the MARK ID to CQE's flow_tag like,
4393  *   - If reg_c[1] is mark_id,
4394  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4395  *
4396  * For SET_META action which stores value in reg_c[0], as the destination is
4397  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4398  * MARK ID means the default flow. The default flow looks like,
4399  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4400  *
4401  * @param dev
4402  *   Pointer to Ethernet device.
4403  * @param flow
4404  *   Pointer to flow structure.
4405  * @param[in] actions
4406  *   Pointer to the list of actions.
4407  * @param[out] error
4408  *   Perform verbose error reporting if not NULL.
4409  *
4410  * @return
4411  *   0 on success, negative value otherwise and rte_errno is set.
4412  */
4413 static int
4414 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4415 			    struct rte_flow *flow,
4416 			    const struct rte_flow_action *actions,
4417 			    struct rte_flow_error *error)
4418 {
4419 	struct mlx5_priv *priv = dev->data->dev_private;
4420 	struct mlx5_dev_config *config = &priv->config;
4421 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4422 	const struct rte_flow_action_mark *mark;
4423 
4424 	/* Check whether extensive metadata feature is engaged. */
4425 	if (!config->dv_flow_en ||
4426 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4427 	    !mlx5_flow_ext_mreg_supported(dev) ||
4428 	    !priv->sh->dv_regc0_mask)
4429 		return 0;
4430 	/* Find MARK action. */
4431 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4432 		switch (actions->type) {
4433 		case RTE_FLOW_ACTION_TYPE_FLAG:
4434 			mcp_res = flow_mreg_add_copy_action
4435 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4436 			if (!mcp_res)
4437 				return -rte_errno;
4438 			flow->rix_mreg_copy = mcp_res->idx;
4439 			return 0;
4440 		case RTE_FLOW_ACTION_TYPE_MARK:
4441 			mark = (const struct rte_flow_action_mark *)
4442 				actions->conf;
4443 			mcp_res =
4444 				flow_mreg_add_copy_action(dev, mark->id, error);
4445 			if (!mcp_res)
4446 				return -rte_errno;
4447 			flow->rix_mreg_copy = mcp_res->idx;
4448 			return 0;
4449 		default:
4450 			break;
4451 		}
4452 	}
4453 	return 0;
4454 }
4455 
4456 #define MLX5_MAX_SPLIT_ACTIONS 24
4457 #define MLX5_MAX_SPLIT_ITEMS 24
4458 
4459 /**
4460  * Split the hairpin flow.
4461  * Since HW can't support encap and push-vlan on Rx, we move these
4462  * actions to Tx.
4463  * If the count action is after the encap then we also
4464  * move the count action. in this case the count will also measure
4465  * the outer bytes.
4466  *
4467  * @param dev
4468  *   Pointer to Ethernet device.
4469  * @param[in] actions
4470  *   Associated actions (list terminated by the END action).
4471  * @param[out] actions_rx
4472  *   Rx flow actions.
4473  * @param[out] actions_tx
4474  *   Tx flow actions..
4475  * @param[out] pattern_tx
4476  *   The pattern items for the Tx flow.
4477  * @param[out] flow_id
4478  *   The flow ID connected to this flow.
4479  *
4480  * @return
4481  *   0 on success.
4482  */
4483 static int
4484 flow_hairpin_split(struct rte_eth_dev *dev,
4485 		   const struct rte_flow_action actions[],
4486 		   struct rte_flow_action actions_rx[],
4487 		   struct rte_flow_action actions_tx[],
4488 		   struct rte_flow_item pattern_tx[],
4489 		   uint32_t flow_id)
4490 {
4491 	const struct rte_flow_action_raw_encap *raw_encap;
4492 	const struct rte_flow_action_raw_decap *raw_decap;
4493 	struct mlx5_rte_flow_action_set_tag *set_tag;
4494 	struct rte_flow_action *tag_action;
4495 	struct mlx5_rte_flow_item_tag *tag_item;
4496 	struct rte_flow_item *item;
4497 	char *addr;
4498 	int encap = 0;
4499 
4500 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4501 		switch (actions->type) {
4502 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4503 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4504 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4505 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4506 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4507 			rte_memcpy(actions_tx, actions,
4508 			       sizeof(struct rte_flow_action));
4509 			actions_tx++;
4510 			break;
4511 		case RTE_FLOW_ACTION_TYPE_COUNT:
4512 			if (encap) {
4513 				rte_memcpy(actions_tx, actions,
4514 					   sizeof(struct rte_flow_action));
4515 				actions_tx++;
4516 			} else {
4517 				rte_memcpy(actions_rx, actions,
4518 					   sizeof(struct rte_flow_action));
4519 				actions_rx++;
4520 			}
4521 			break;
4522 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4523 			raw_encap = actions->conf;
4524 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4525 				memcpy(actions_tx, actions,
4526 				       sizeof(struct rte_flow_action));
4527 				actions_tx++;
4528 				encap = 1;
4529 			} else {
4530 				rte_memcpy(actions_rx, actions,
4531 					   sizeof(struct rte_flow_action));
4532 				actions_rx++;
4533 			}
4534 			break;
4535 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4536 			raw_decap = actions->conf;
4537 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4538 				memcpy(actions_tx, actions,
4539 				       sizeof(struct rte_flow_action));
4540 				actions_tx++;
4541 			} else {
4542 				rte_memcpy(actions_rx, actions,
4543 					   sizeof(struct rte_flow_action));
4544 				actions_rx++;
4545 			}
4546 			break;
4547 		default:
4548 			rte_memcpy(actions_rx, actions,
4549 				   sizeof(struct rte_flow_action));
4550 			actions_rx++;
4551 			break;
4552 		}
4553 	}
4554 	/* Add set meta action and end action for the Rx flow. */
4555 	tag_action = actions_rx;
4556 	tag_action->type = (enum rte_flow_action_type)
4557 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4558 	actions_rx++;
4559 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4560 	actions_rx++;
4561 	set_tag = (void *)actions_rx;
4562 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4563 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4564 		.data = flow_id,
4565 	};
4566 	MLX5_ASSERT(set_tag->id > REG_NON);
4567 	tag_action->conf = set_tag;
4568 	/* Create Tx item list. */
4569 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4570 	addr = (void *)&pattern_tx[2];
4571 	item = pattern_tx;
4572 	item->type = (enum rte_flow_item_type)
4573 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4574 	tag_item = (void *)addr;
4575 	tag_item->data = flow_id;
4576 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4577 	MLX5_ASSERT(set_tag->id > REG_NON);
4578 	item->spec = tag_item;
4579 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4580 	tag_item = (void *)addr;
4581 	tag_item->data = UINT32_MAX;
4582 	tag_item->id = UINT16_MAX;
4583 	item->mask = tag_item;
4584 	item->last = NULL;
4585 	item++;
4586 	item->type = RTE_FLOW_ITEM_TYPE_END;
4587 	return 0;
4588 }
4589 
4590 /**
4591  * The last stage of splitting chain, just creates the subflow
4592  * without any modification.
4593  *
4594  * @param[in] dev
4595  *   Pointer to Ethernet device.
4596  * @param[in] flow
4597  *   Parent flow structure pointer.
4598  * @param[in, out] sub_flow
4599  *   Pointer to return the created subflow, may be NULL.
4600  * @param[in] attr
4601  *   Flow rule attributes.
4602  * @param[in] items
4603  *   Pattern specification (list terminated by the END pattern item).
4604  * @param[in] actions
4605  *   Associated actions (list terminated by the END action).
4606  * @param[in] flow_split_info
4607  *   Pointer to flow split info structure.
4608  * @param[out] error
4609  *   Perform verbose error reporting if not NULL.
4610  * @return
4611  *   0 on success, negative value otherwise
4612  */
4613 static int
4614 flow_create_split_inner(struct rte_eth_dev *dev,
4615 			struct rte_flow *flow,
4616 			struct mlx5_flow **sub_flow,
4617 			const struct rte_flow_attr *attr,
4618 			const struct rte_flow_item items[],
4619 			const struct rte_flow_action actions[],
4620 			struct mlx5_flow_split_info *flow_split_info,
4621 			struct rte_flow_error *error)
4622 {
4623 	struct mlx5_flow *dev_flow;
4624 
4625 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4626 				    flow_split_info->flow_idx, error);
4627 	if (!dev_flow)
4628 		return -rte_errno;
4629 	dev_flow->flow = flow;
4630 	dev_flow->external = flow_split_info->external;
4631 	dev_flow->skip_scale = flow_split_info->skip_scale;
4632 	/* Subflow object was created, we must include one in the list. */
4633 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4634 		      dev_flow->handle, next);
4635 	/*
4636 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4637 	 * flow may need some user defined item layer flags, and pass the
4638 	 * Metadate rxq mark flag to suffix flow as well.
4639 	 */
4640 	if (flow_split_info->prefix_layers)
4641 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4642 	if (flow_split_info->prefix_mark)
4643 		dev_flow->handle->mark = 1;
4644 	if (sub_flow)
4645 		*sub_flow = dev_flow;
4646 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4647 	dev_flow->dv.table_id = flow_split_info->table_id;
4648 #endif
4649 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4650 }
4651 
4652 /**
4653  * Get the sub policy of a meter.
4654  *
4655  * @param[in] dev
4656  *   Pointer to Ethernet device.
4657  * @param[in] flow
4658  *   Parent flow structure pointer.
4659  * @param wks
4660  *   Pointer to thread flow work space.
4661  * @param[in] attr
4662  *   Flow rule attributes.
4663  * @param[in] items
4664  *   Pattern specification (list terminated by the END pattern item).
4665  * @param[out] error
4666  *   Perform verbose error reporting if not NULL.
4667  *
4668  * @return
4669  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4670  */
4671 static struct mlx5_flow_meter_sub_policy *
4672 get_meter_sub_policy(struct rte_eth_dev *dev,
4673 		     struct rte_flow *flow,
4674 		     struct mlx5_flow_workspace *wks,
4675 		     const struct rte_flow_attr *attr,
4676 		     const struct rte_flow_item items[],
4677 		     struct rte_flow_error *error)
4678 {
4679 	struct mlx5_flow_meter_policy *policy;
4680 	struct mlx5_flow_meter_policy *final_policy;
4681 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4682 
4683 	policy = wks->policy;
4684 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4685 	if (final_policy->is_rss || final_policy->is_queue) {
4686 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4687 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4688 		uint32_t i;
4689 
4690 		/**
4691 		 * This is a tmp dev_flow,
4692 		 * no need to register any matcher for it in translate.
4693 		 */
4694 		wks->skip_matcher_reg = 1;
4695 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4696 			struct mlx5_flow dev_flow = {0};
4697 			struct mlx5_flow_handle dev_handle = { {0} };
4698 
4699 			if (final_policy->is_rss) {
4700 				const void *rss_act =
4701 					final_policy->act_cnt[i].rss->conf;
4702 				struct rte_flow_action rss_actions[2] = {
4703 					[0] = {
4704 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4705 					.conf = rss_act
4706 					},
4707 					[1] = {
4708 					.type = RTE_FLOW_ACTION_TYPE_END,
4709 					.conf = NULL
4710 					}
4711 				};
4712 
4713 				dev_flow.handle = &dev_handle;
4714 				dev_flow.ingress = attr->ingress;
4715 				dev_flow.flow = flow;
4716 				dev_flow.external = 0;
4717 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4718 				dev_flow.dv.transfer = attr->transfer;
4719 #endif
4720 				/**
4721 				 * Translate RSS action to get rss hash fields.
4722 				 */
4723 				if (flow_drv_translate(dev, &dev_flow, attr,
4724 						items, rss_actions, error))
4725 					goto exit;
4726 				rss_desc_v[i] = wks->rss_desc;
4727 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4728 				rss_desc_v[i].hash_fields =
4729 						dev_flow.hash_fields;
4730 				rss_desc_v[i].queue_num =
4731 						rss_desc_v[i].hash_fields ?
4732 						rss_desc_v[i].queue_num : 1;
4733 				rss_desc_v[i].tunnel =
4734 					!!(dev_flow.handle->layers &
4735 					MLX5_FLOW_LAYER_TUNNEL);
4736 			} else {
4737 				/* This is queue action. */
4738 				rss_desc_v[i] = wks->rss_desc;
4739 				rss_desc_v[i].key_len = 0;
4740 				rss_desc_v[i].hash_fields = 0;
4741 				rss_desc_v[i].queue =
4742 					&final_policy->act_cnt[i].queue;
4743 				rss_desc_v[i].queue_num = 1;
4744 			}
4745 			rss_desc[i] = &rss_desc_v[i];
4746 		}
4747 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4748 						flow, policy, rss_desc);
4749 	} else {
4750 		enum mlx5_meter_domain mtr_domain =
4751 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4752 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4753 					MLX5_MTR_DOMAIN_INGRESS;
4754 		sub_policy = policy->sub_policys[mtr_domain][0];
4755 	}
4756 	if (!sub_policy) {
4757 		rte_flow_error_set(error, EINVAL,
4758 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4759 			"Failed to get meter sub-policy.");
4760 		goto exit;
4761 	}
4762 exit:
4763 	return sub_policy;
4764 }
4765 
4766 /**
4767  * Split the meter flow.
4768  *
4769  * As meter flow will split to three sub flow, other than meter
4770  * action, the other actions make sense to only meter accepts
4771  * the packet. If it need to be dropped, no other additional
4772  * actions should be take.
4773  *
4774  * One kind of special action which decapsulates the L3 tunnel
4775  * header will be in the prefix sub flow, as not to take the
4776  * L3 tunnel header into account.
4777  *
4778  * @param[in] dev
4779  *   Pointer to Ethernet device.
4780  * @param[in] flow
4781  *   Parent flow structure pointer.
4782  * @param wks
4783  *   Pointer to thread flow work space.
4784  * @param[in] attr
4785  *   Flow rule attributes.
4786  * @param[in] items
4787  *   Pattern specification (list terminated by the END pattern item).
4788  * @param[out] sfx_items
4789  *   Suffix flow match items (list terminated by the END pattern item).
4790  * @param[in] actions
4791  *   Associated actions (list terminated by the END action).
4792  * @param[out] actions_sfx
4793  *   Suffix flow actions.
4794  * @param[out] actions_pre
4795  *   Prefix flow actions.
4796  * @param[out] mtr_flow_id
4797  *   Pointer to meter flow id.
4798  * @param[out] error
4799  *   Perform verbose error reporting if not NULL.
4800  *
4801  * @return
4802  *   0 on success, a negative errno value otherwise and rte_errno is set.
4803  */
4804 static int
4805 flow_meter_split_prep(struct rte_eth_dev *dev,
4806 		      struct rte_flow *flow,
4807 		      struct mlx5_flow_workspace *wks,
4808 		      const struct rte_flow_attr *attr,
4809 		      const struct rte_flow_item items[],
4810 		      struct rte_flow_item sfx_items[],
4811 		      const struct rte_flow_action actions[],
4812 		      struct rte_flow_action actions_sfx[],
4813 		      struct rte_flow_action actions_pre[],
4814 		      uint32_t *mtr_flow_id,
4815 		      struct rte_flow_error *error)
4816 {
4817 	struct mlx5_priv *priv = dev->data->dev_private;
4818 	struct mlx5_flow_meter_info *fm = wks->fm;
4819 	struct rte_flow_action *tag_action = NULL;
4820 	struct rte_flow_item *tag_item;
4821 	struct mlx5_rte_flow_action_set_tag *set_tag;
4822 	const struct rte_flow_action_raw_encap *raw_encap;
4823 	const struct rte_flow_action_raw_decap *raw_decap;
4824 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4825 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4826 	uint32_t tag_id = 0;
4827 	struct rte_flow_item *vlan_item_dst = NULL;
4828 	const struct rte_flow_item *vlan_item_src = NULL;
4829 	struct rte_flow_action *hw_mtr_action;
4830 	struct rte_flow_action *action_pre_head = NULL;
4831 	int32_t flow_src_port = priv->representor_id;
4832 	bool mtr_first;
4833 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4834 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4835 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4836 	uint32_t flow_id = 0;
4837 	uint32_t flow_id_reversed = 0;
4838 	uint8_t flow_id_bits = 0;
4839 	int shift;
4840 
4841 	/* Prepare the suffix subflow items. */
4842 	tag_item = sfx_items++;
4843 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4844 		struct mlx5_priv *port_priv;
4845 		const struct rte_flow_item_port_id *pid_v;
4846 		int item_type = items->type;
4847 
4848 		switch (item_type) {
4849 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4850 			pid_v = items->spec;
4851 			MLX5_ASSERT(pid_v);
4852 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4853 			if (!port_priv)
4854 				return rte_flow_error_set(error,
4855 						rte_errno,
4856 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4857 						pid_v,
4858 						"Failed to get port info.");
4859 			flow_src_port = port_priv->representor_id;
4860 			if (!fm->def_policy && wks->policy->is_hierarchy &&
4861 			    flow_src_port != priv->representor_id) {
4862 				if (flow_drv_mtr_hierarchy_rule_create(dev,
4863 								flow, fm,
4864 								flow_src_port,
4865 								items,
4866 								error))
4867 					return -rte_errno;
4868 			}
4869 			memcpy(sfx_items, items, sizeof(*sfx_items));
4870 			sfx_items++;
4871 			break;
4872 		case RTE_FLOW_ITEM_TYPE_VLAN:
4873 			/* Determine if copy vlan item below. */
4874 			vlan_item_src = items;
4875 			vlan_item_dst = sfx_items++;
4876 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4877 			break;
4878 		default:
4879 			break;
4880 		}
4881 	}
4882 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4883 	sfx_items++;
4884 	mtr_first = priv->sh->meter_aso_en &&
4885 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4886 	/* For ASO meter, meter must be before tag in TX direction. */
4887 	if (mtr_first) {
4888 		action_pre_head = actions_pre++;
4889 		/* Leave space for tag action. */
4890 		tag_action = actions_pre++;
4891 	}
4892 	/* Prepare the actions for prefix and suffix flow. */
4893 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4894 		struct rte_flow_action *action_cur = NULL;
4895 
4896 		switch (actions->type) {
4897 		case RTE_FLOW_ACTION_TYPE_METER:
4898 			if (mtr_first) {
4899 				action_cur = action_pre_head;
4900 			} else {
4901 				/* Leave space for tag action. */
4902 				tag_action = actions_pre++;
4903 				action_cur = actions_pre++;
4904 			}
4905 			break;
4906 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4907 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4908 			action_cur = actions_pre++;
4909 			break;
4910 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4911 			raw_encap = actions->conf;
4912 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4913 				action_cur = actions_pre++;
4914 			break;
4915 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4916 			raw_decap = actions->conf;
4917 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4918 				action_cur = actions_pre++;
4919 			break;
4920 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4921 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4922 			if (vlan_item_dst && vlan_item_src) {
4923 				memcpy(vlan_item_dst, vlan_item_src,
4924 					sizeof(*vlan_item_dst));
4925 				/*
4926 				 * Convert to internal match item, it is used
4927 				 * for vlan push and set vid.
4928 				 */
4929 				vlan_item_dst->type = (enum rte_flow_item_type)
4930 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4931 			}
4932 			break;
4933 		default:
4934 			break;
4935 		}
4936 		if (!action_cur)
4937 			action_cur = (fm->def_policy) ?
4938 					actions_sfx++ : actions_pre++;
4939 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4940 	}
4941 	/* Add end action to the actions. */
4942 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4943 	if (priv->sh->meter_aso_en) {
4944 		/**
4945 		 * For ASO meter, need to add an extra jump action explicitly,
4946 		 * to jump from meter to policer table.
4947 		 */
4948 		struct mlx5_flow_meter_sub_policy *sub_policy;
4949 		struct mlx5_flow_tbl_data_entry *tbl_data;
4950 
4951 		if (!fm->def_policy) {
4952 			sub_policy = get_meter_sub_policy(dev, flow, wks,
4953 							  attr, items, error);
4954 			if (!sub_policy)
4955 				return -rte_errno;
4956 		} else {
4957 			enum mlx5_meter_domain mtr_domain =
4958 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4959 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4960 					MLX5_MTR_DOMAIN_INGRESS;
4961 
4962 			sub_policy =
4963 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4964 		}
4965 		tbl_data = container_of(sub_policy->tbl_rsc,
4966 					struct mlx5_flow_tbl_data_entry, tbl);
4967 		hw_mtr_action = actions_pre++;
4968 		hw_mtr_action->type = (enum rte_flow_action_type)
4969 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4970 		hw_mtr_action->conf = tbl_data->jump.action;
4971 	}
4972 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4973 	actions_pre++;
4974 	if (!tag_action)
4975 		return rte_flow_error_set(error, ENOMEM,
4976 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4977 					"No tag action space.");
4978 	if (!mtr_flow_id) {
4979 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4980 		goto exit;
4981 	}
4982 	/* Only default-policy Meter creates mtr flow id. */
4983 	if (fm->def_policy) {
4984 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4985 		if (!tag_id)
4986 			return rte_flow_error_set(error, ENOMEM,
4987 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4988 					"Failed to allocate meter flow id.");
4989 		flow_id = tag_id - 1;
4990 		flow_id_bits = (!flow_id) ? 1 :
4991 				(MLX5_REG_BITS - __builtin_clz(flow_id));
4992 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4993 		    mtr_reg_bits) {
4994 			mlx5_ipool_free(fm->flow_ipool, tag_id);
4995 			return rte_flow_error_set(error, EINVAL,
4996 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4997 					"Meter flow id exceeds max limit.");
4998 		}
4999 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5000 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5001 	}
5002 	/* Build tag actions and items for meter_id/meter flow_id. */
5003 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5004 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5005 	tag_item_mask = tag_item_spec + 1;
5006 	/* Both flow_id and meter_id share the same register. */
5007 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5008 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5009 							    0, error),
5010 		.offset = mtr_id_offset,
5011 		.length = mtr_reg_bits,
5012 		.data = flow->meter,
5013 	};
5014 	/*
5015 	 * The color Reg bits used by flow_id are growing from
5016 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5017 	 */
5018 	for (shift = 0; shift < flow_id_bits; shift++)
5019 		flow_id_reversed = (flow_id_reversed << 1) |
5020 				((flow_id >> shift) & 0x1);
5021 	set_tag->data |=
5022 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5023 	tag_item_spec->id = set_tag->id;
5024 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5025 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5026 	tag_action->type = (enum rte_flow_action_type)
5027 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5028 	tag_action->conf = set_tag;
5029 	tag_item->type = (enum rte_flow_item_type)
5030 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5031 	tag_item->spec = tag_item_spec;
5032 	tag_item->last = NULL;
5033 	tag_item->mask = tag_item_mask;
5034 exit:
5035 	if (mtr_flow_id)
5036 		*mtr_flow_id = tag_id;
5037 	return 0;
5038 }
5039 
5040 /**
5041  * Split action list having QUEUE/RSS for metadata register copy.
5042  *
5043  * Once Q/RSS action is detected in user's action list, the flow action
5044  * should be split in order to copy metadata registers, which will happen in
5045  * RX_CP_TBL like,
5046  *   - CQE->flow_tag := reg_c[1] (MARK)
5047  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5048  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5049  * This is because the last action of each flow must be a terminal action
5050  * (QUEUE, RSS or DROP).
5051  *
5052  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5053  * stored and kept in the mlx5_flow structure per each sub_flow.
5054  *
5055  * The Q/RSS action is replaced with,
5056  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5057  * And the following JUMP action is added at the end,
5058  *   - JUMP, to RX_CP_TBL.
5059  *
5060  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5061  * flow_create_split_metadata() routine. The flow will look like,
5062  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5063  *
5064  * @param dev
5065  *   Pointer to Ethernet device.
5066  * @param[out] split_actions
5067  *   Pointer to store split actions to jump to CP_TBL.
5068  * @param[in] actions
5069  *   Pointer to the list of original flow actions.
5070  * @param[in] qrss
5071  *   Pointer to the Q/RSS action.
5072  * @param[in] actions_n
5073  *   Number of original actions.
5074  * @param[out] error
5075  *   Perform verbose error reporting if not NULL.
5076  *
5077  * @return
5078  *   non-zero unique flow_id on success, otherwise 0 and
5079  *   error/rte_error are set.
5080  */
5081 static uint32_t
5082 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5083 			  struct rte_flow_action *split_actions,
5084 			  const struct rte_flow_action *actions,
5085 			  const struct rte_flow_action *qrss,
5086 			  int actions_n, struct rte_flow_error *error)
5087 {
5088 	struct mlx5_priv *priv = dev->data->dev_private;
5089 	struct mlx5_rte_flow_action_set_tag *set_tag;
5090 	struct rte_flow_action_jump *jump;
5091 	const int qrss_idx = qrss - actions;
5092 	uint32_t flow_id = 0;
5093 	int ret = 0;
5094 
5095 	/*
5096 	 * Given actions will be split
5097 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5098 	 * - Add jump to mreg CP_TBL.
5099 	 * As a result, there will be one more action.
5100 	 */
5101 	++actions_n;
5102 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5103 	set_tag = (void *)(split_actions + actions_n);
5104 	/*
5105 	 * If tag action is not set to void(it means we are not the meter
5106 	 * suffix flow), add the tag action. Since meter suffix flow already
5107 	 * has the tag added.
5108 	 */
5109 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5110 		/*
5111 		 * Allocate the new subflow ID. This one is unique within
5112 		 * device and not shared with representors. Otherwise,
5113 		 * we would have to resolve multi-thread access synch
5114 		 * issue. Each flow on the shared device is appended
5115 		 * with source vport identifier, so the resulting
5116 		 * flows will be unique in the shared (by master and
5117 		 * representors) domain even if they have coinciding
5118 		 * IDs.
5119 		 */
5120 		mlx5_ipool_malloc(priv->sh->ipool
5121 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5122 		if (!flow_id)
5123 			return rte_flow_error_set(error, ENOMEM,
5124 						  RTE_FLOW_ERROR_TYPE_ACTION,
5125 						  NULL, "can't allocate id "
5126 						  "for split Q/RSS subflow");
5127 		/* Internal SET_TAG action to set flow ID. */
5128 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5129 			.data = flow_id,
5130 		};
5131 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5132 		if (ret < 0)
5133 			return ret;
5134 		set_tag->id = ret;
5135 		/* Construct new actions array. */
5136 		/* Replace QUEUE/RSS action. */
5137 		split_actions[qrss_idx] = (struct rte_flow_action){
5138 			.type = (enum rte_flow_action_type)
5139 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5140 			.conf = set_tag,
5141 		};
5142 	}
5143 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5144 	jump = (void *)(set_tag + 1);
5145 	*jump = (struct rte_flow_action_jump){
5146 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5147 	};
5148 	split_actions[actions_n - 2] = (struct rte_flow_action){
5149 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5150 		.conf = jump,
5151 	};
5152 	split_actions[actions_n - 1] = (struct rte_flow_action){
5153 		.type = RTE_FLOW_ACTION_TYPE_END,
5154 	};
5155 	return flow_id;
5156 }
5157 
5158 /**
5159  * Extend the given action list for Tx metadata copy.
5160  *
5161  * Copy the given action list to the ext_actions and add flow metadata register
5162  * copy action in order to copy reg_a set by WQE to reg_c[0].
5163  *
5164  * @param[out] ext_actions
5165  *   Pointer to the extended action list.
5166  * @param[in] actions
5167  *   Pointer to the list of actions.
5168  * @param[in] actions_n
5169  *   Number of actions in the list.
5170  * @param[out] error
5171  *   Perform verbose error reporting if not NULL.
5172  * @param[in] encap_idx
5173  *   The encap action inndex.
5174  *
5175  * @return
5176  *   0 on success, negative value otherwise
5177  */
5178 static int
5179 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5180 		       struct rte_flow_action *ext_actions,
5181 		       const struct rte_flow_action *actions,
5182 		       int actions_n, struct rte_flow_error *error,
5183 		       int encap_idx)
5184 {
5185 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5186 		(struct mlx5_flow_action_copy_mreg *)
5187 			(ext_actions + actions_n + 1);
5188 	int ret;
5189 
5190 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5191 	if (ret < 0)
5192 		return ret;
5193 	cp_mreg->dst = ret;
5194 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5195 	if (ret < 0)
5196 		return ret;
5197 	cp_mreg->src = ret;
5198 	if (encap_idx != 0)
5199 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5200 	if (encap_idx == actions_n - 1) {
5201 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5202 			.type = (enum rte_flow_action_type)
5203 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5204 			.conf = cp_mreg,
5205 		};
5206 		ext_actions[actions_n] = (struct rte_flow_action){
5207 			.type = RTE_FLOW_ACTION_TYPE_END,
5208 		};
5209 	} else {
5210 		ext_actions[encap_idx] = (struct rte_flow_action){
5211 			.type = (enum rte_flow_action_type)
5212 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5213 			.conf = cp_mreg,
5214 		};
5215 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5216 				sizeof(*ext_actions) * (actions_n - encap_idx));
5217 	}
5218 	return 0;
5219 }
5220 
5221 /**
5222  * Check the match action from the action list.
5223  *
5224  * @param[in] actions
5225  *   Pointer to the list of actions.
5226  * @param[in] attr
5227  *   Flow rule attributes.
5228  * @param[in] action
5229  *   The action to be check if exist.
5230  * @param[out] match_action_pos
5231  *   Pointer to the position of the matched action if exists, otherwise is -1.
5232  * @param[out] qrss_action_pos
5233  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5234  * @param[out] modify_after_mirror
5235  *   Pointer to the flag of modify action after FDB mirroring.
5236  *
5237  * @return
5238  *   > 0 the total number of actions.
5239  *   0 if not found match action in action list.
5240  */
5241 static int
5242 flow_check_match_action(const struct rte_flow_action actions[],
5243 			const struct rte_flow_attr *attr,
5244 			enum rte_flow_action_type action,
5245 			int *match_action_pos, int *qrss_action_pos,
5246 			int *modify_after_mirror)
5247 {
5248 	const struct rte_flow_action_sample *sample;
5249 	int actions_n = 0;
5250 	uint32_t ratio = 0;
5251 	int sub_type = 0;
5252 	int flag = 0;
5253 	int fdb_mirror = 0;
5254 
5255 	*match_action_pos = -1;
5256 	*qrss_action_pos = -1;
5257 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5258 		if (actions->type == action) {
5259 			flag = 1;
5260 			*match_action_pos = actions_n;
5261 		}
5262 		switch (actions->type) {
5263 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5264 		case RTE_FLOW_ACTION_TYPE_RSS:
5265 			*qrss_action_pos = actions_n;
5266 			break;
5267 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5268 			sample = actions->conf;
5269 			ratio = sample->ratio;
5270 			sub_type = ((const struct rte_flow_action *)
5271 					(sample->actions))->type;
5272 			if (ratio == 1 && attr->transfer)
5273 				fdb_mirror = 1;
5274 			break;
5275 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5276 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5277 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5278 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5279 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5280 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5281 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5282 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5283 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5284 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5285 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5286 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5287 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5288 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5289 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5290 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5291 		case RTE_FLOW_ACTION_TYPE_FLAG:
5292 		case RTE_FLOW_ACTION_TYPE_MARK:
5293 		case RTE_FLOW_ACTION_TYPE_SET_META:
5294 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5295 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5296 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5297 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5298 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5299 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5300 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5301 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5302 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5303 		case RTE_FLOW_ACTION_TYPE_METER:
5304 			if (fdb_mirror)
5305 				*modify_after_mirror = 1;
5306 			break;
5307 		default:
5308 			break;
5309 		}
5310 		actions_n++;
5311 	}
5312 	if (flag && fdb_mirror && !*modify_after_mirror) {
5313 		/* FDB mirroring uses the destination array to implement
5314 		 * instead of FLOW_SAMPLER object.
5315 		 */
5316 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5317 			flag = 0;
5318 	}
5319 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5320 	return flag ? actions_n + 1 : 0;
5321 }
5322 
5323 #define SAMPLE_SUFFIX_ITEM 2
5324 
5325 /**
5326  * Split the sample flow.
5327  *
5328  * As sample flow will split to two sub flow, sample flow with
5329  * sample action, the other actions will move to new suffix flow.
5330  *
5331  * Also add unique tag id with tag action in the sample flow,
5332  * the same tag id will be as match in the suffix flow.
5333  *
5334  * @param dev
5335  *   Pointer to Ethernet device.
5336  * @param[in] add_tag
5337  *   Add extra tag action flag.
5338  * @param[out] sfx_items
5339  *   Suffix flow match items (list terminated by the END pattern item).
5340  * @param[in] actions
5341  *   Associated actions (list terminated by the END action).
5342  * @param[out] actions_sfx
5343  *   Suffix flow actions.
5344  * @param[out] actions_pre
5345  *   Prefix flow actions.
5346  * @param[in] actions_n
5347  *  The total number of actions.
5348  * @param[in] sample_action_pos
5349  *   The sample action position.
5350  * @param[in] qrss_action_pos
5351  *   The Queue/RSS action position.
5352  * @param[in] jump_table
5353  *   Add extra jump action flag.
5354  * @param[out] error
5355  *   Perform verbose error reporting if not NULL.
5356  *
5357  * @return
5358  *   0 on success, or unique flow_id, a negative errno value
5359  *   otherwise and rte_errno is set.
5360  */
5361 static int
5362 flow_sample_split_prep(struct rte_eth_dev *dev,
5363 		       int add_tag,
5364 		       struct rte_flow_item sfx_items[],
5365 		       const struct rte_flow_action actions[],
5366 		       struct rte_flow_action actions_sfx[],
5367 		       struct rte_flow_action actions_pre[],
5368 		       int actions_n,
5369 		       int sample_action_pos,
5370 		       int qrss_action_pos,
5371 		       int jump_table,
5372 		       struct rte_flow_error *error)
5373 {
5374 	struct mlx5_priv *priv = dev->data->dev_private;
5375 	struct mlx5_rte_flow_action_set_tag *set_tag;
5376 	struct mlx5_rte_flow_item_tag *tag_spec;
5377 	struct mlx5_rte_flow_item_tag *tag_mask;
5378 	struct rte_flow_action_jump *jump_action;
5379 	uint32_t tag_id = 0;
5380 	int index;
5381 	int append_index = 0;
5382 	int ret;
5383 
5384 	if (sample_action_pos < 0)
5385 		return rte_flow_error_set(error, EINVAL,
5386 					  RTE_FLOW_ERROR_TYPE_ACTION,
5387 					  NULL, "invalid position of sample "
5388 					  "action in list");
5389 	/* Prepare the actions for prefix and suffix flow. */
5390 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5391 		index = qrss_action_pos;
5392 		/* Put the preceding the Queue/RSS action into prefix flow. */
5393 		if (index != 0)
5394 			memcpy(actions_pre, actions,
5395 			       sizeof(struct rte_flow_action) * index);
5396 		/* Put others preceding the sample action into prefix flow. */
5397 		if (sample_action_pos > index + 1)
5398 			memcpy(actions_pre + index, actions + index + 1,
5399 			       sizeof(struct rte_flow_action) *
5400 			       (sample_action_pos - index - 1));
5401 		index = sample_action_pos - 1;
5402 		/* Put Queue/RSS action into Suffix flow. */
5403 		memcpy(actions_sfx, actions + qrss_action_pos,
5404 		       sizeof(struct rte_flow_action));
5405 		actions_sfx++;
5406 	} else {
5407 		index = sample_action_pos;
5408 		if (index != 0)
5409 			memcpy(actions_pre, actions,
5410 			       sizeof(struct rte_flow_action) * index);
5411 	}
5412 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5413 	 * For CX6DX and above, metadata registers Cx preserve their value,
5414 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5415 	 */
5416 	if (add_tag) {
5417 		/* Prepare the prefix tag action. */
5418 		append_index++;
5419 		set_tag = (void *)(actions_pre + actions_n + append_index);
5420 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5421 		if (ret < 0)
5422 			return ret;
5423 		mlx5_ipool_malloc(priv->sh->ipool
5424 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5425 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5426 			.id = ret,
5427 			.data = tag_id,
5428 		};
5429 		/* Prepare the suffix subflow items. */
5430 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5431 		tag_spec->data = tag_id;
5432 		tag_spec->id = set_tag->id;
5433 		tag_mask = tag_spec + 1;
5434 		tag_mask->data = UINT32_MAX;
5435 		sfx_items[0] = (struct rte_flow_item){
5436 			.type = (enum rte_flow_item_type)
5437 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5438 			.spec = tag_spec,
5439 			.last = NULL,
5440 			.mask = tag_mask,
5441 		};
5442 		sfx_items[1] = (struct rte_flow_item){
5443 			.type = (enum rte_flow_item_type)
5444 				RTE_FLOW_ITEM_TYPE_END,
5445 		};
5446 		/* Prepare the tag action in prefix subflow. */
5447 		actions_pre[index++] =
5448 			(struct rte_flow_action){
5449 			.type = (enum rte_flow_action_type)
5450 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5451 			.conf = set_tag,
5452 		};
5453 	}
5454 	memcpy(actions_pre + index, actions + sample_action_pos,
5455 	       sizeof(struct rte_flow_action));
5456 	index += 1;
5457 	/* For the modify action after the sample action in E-Switch mirroring,
5458 	 * Add the extra jump action in prefix subflow and jump into the next
5459 	 * table, then do the modify action in the new table.
5460 	 */
5461 	if (jump_table) {
5462 		/* Prepare the prefix jump action. */
5463 		append_index++;
5464 		jump_action = (void *)(actions_pre + actions_n + append_index);
5465 		jump_action->group = jump_table;
5466 		actions_pre[index++] =
5467 			(struct rte_flow_action){
5468 			.type = (enum rte_flow_action_type)
5469 				RTE_FLOW_ACTION_TYPE_JUMP,
5470 			.conf = jump_action,
5471 		};
5472 	}
5473 	actions_pre[index] = (struct rte_flow_action){
5474 		.type = (enum rte_flow_action_type)
5475 			RTE_FLOW_ACTION_TYPE_END,
5476 	};
5477 	/* Put the actions after sample into Suffix flow. */
5478 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5479 	       sizeof(struct rte_flow_action) *
5480 	       (actions_n - sample_action_pos - 1));
5481 	return tag_id;
5482 }
5483 
5484 /**
5485  * The splitting for metadata feature.
5486  *
5487  * - Q/RSS action on NIC Rx should be split in order to pass by
5488  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5489  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5490  *
5491  * - All the actions on NIC Tx should have a mreg copy action to
5492  *   copy reg_a from WQE to reg_c[0].
5493  *
5494  * @param dev
5495  *   Pointer to Ethernet device.
5496  * @param[in] flow
5497  *   Parent flow structure pointer.
5498  * @param[in] attr
5499  *   Flow rule attributes.
5500  * @param[in] items
5501  *   Pattern specification (list terminated by the END pattern item).
5502  * @param[in] actions
5503  *   Associated actions (list terminated by the END action).
5504  * @param[in] flow_split_info
5505  *   Pointer to flow split info structure.
5506  * @param[out] error
5507  *   Perform verbose error reporting if not NULL.
5508  * @return
5509  *   0 on success, negative value otherwise
5510  */
5511 static int
5512 flow_create_split_metadata(struct rte_eth_dev *dev,
5513 			   struct rte_flow *flow,
5514 			   const struct rte_flow_attr *attr,
5515 			   const struct rte_flow_item items[],
5516 			   const struct rte_flow_action actions[],
5517 			   struct mlx5_flow_split_info *flow_split_info,
5518 			   struct rte_flow_error *error)
5519 {
5520 	struct mlx5_priv *priv = dev->data->dev_private;
5521 	struct mlx5_dev_config *config = &priv->config;
5522 	const struct rte_flow_action *qrss = NULL;
5523 	struct rte_flow_action *ext_actions = NULL;
5524 	struct mlx5_flow *dev_flow = NULL;
5525 	uint32_t qrss_id = 0;
5526 	int mtr_sfx = 0;
5527 	size_t act_size;
5528 	int actions_n;
5529 	int encap_idx;
5530 	int ret;
5531 
5532 	/* Check whether extensive metadata feature is engaged. */
5533 	if (!config->dv_flow_en ||
5534 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5535 	    !mlx5_flow_ext_mreg_supported(dev))
5536 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5537 					       actions, flow_split_info, error);
5538 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5539 							   &encap_idx);
5540 	if (qrss) {
5541 		/* Exclude hairpin flows from splitting. */
5542 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5543 			const struct rte_flow_action_queue *queue;
5544 
5545 			queue = qrss->conf;
5546 			if (mlx5_rxq_get_type(dev, queue->index) ==
5547 			    MLX5_RXQ_TYPE_HAIRPIN)
5548 				qrss = NULL;
5549 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5550 			const struct rte_flow_action_rss *rss;
5551 
5552 			rss = qrss->conf;
5553 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5554 			    MLX5_RXQ_TYPE_HAIRPIN)
5555 				qrss = NULL;
5556 		}
5557 	}
5558 	if (qrss) {
5559 		/* Check if it is in meter suffix table. */
5560 		mtr_sfx = attr->group == (attr->transfer ?
5561 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5562 			  MLX5_FLOW_TABLE_LEVEL_METER);
5563 		/*
5564 		 * Q/RSS action on NIC Rx should be split in order to pass by
5565 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5566 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5567 		 */
5568 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5569 			   sizeof(struct rte_flow_action_set_tag) +
5570 			   sizeof(struct rte_flow_action_jump);
5571 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5572 					  SOCKET_ID_ANY);
5573 		if (!ext_actions)
5574 			return rte_flow_error_set(error, ENOMEM,
5575 						  RTE_FLOW_ERROR_TYPE_ACTION,
5576 						  NULL, "no memory to split "
5577 						  "metadata flow");
5578 		/*
5579 		 * If we are the suffix flow of meter, tag already exist.
5580 		 * Set the tag action to void.
5581 		 */
5582 		if (mtr_sfx)
5583 			ext_actions[qrss - actions].type =
5584 						RTE_FLOW_ACTION_TYPE_VOID;
5585 		else
5586 			ext_actions[qrss - actions].type =
5587 						(enum rte_flow_action_type)
5588 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5589 		/*
5590 		 * Create the new actions list with removed Q/RSS action
5591 		 * and appended set tag and jump to register copy table
5592 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5593 		 * in advance, because it is needed for set tag action.
5594 		 */
5595 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5596 						    qrss, actions_n, error);
5597 		if (!mtr_sfx && !qrss_id) {
5598 			ret = -rte_errno;
5599 			goto exit;
5600 		}
5601 	} else if (attr->egress && !attr->transfer) {
5602 		/*
5603 		 * All the actions on NIC Tx should have a metadata register
5604 		 * copy action to copy reg_a from WQE to reg_c[meta]
5605 		 */
5606 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5607 			   sizeof(struct mlx5_flow_action_copy_mreg);
5608 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5609 					  SOCKET_ID_ANY);
5610 		if (!ext_actions)
5611 			return rte_flow_error_set(error, ENOMEM,
5612 						  RTE_FLOW_ERROR_TYPE_ACTION,
5613 						  NULL, "no memory to split "
5614 						  "metadata flow");
5615 		/* Create the action list appended with copy register. */
5616 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5617 					     actions_n, error, encap_idx);
5618 		if (ret < 0)
5619 			goto exit;
5620 	}
5621 	/* Add the unmodified original or prefix subflow. */
5622 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5623 				      items, ext_actions ? ext_actions :
5624 				      actions, flow_split_info, error);
5625 	if (ret < 0)
5626 		goto exit;
5627 	MLX5_ASSERT(dev_flow);
5628 	if (qrss) {
5629 		const struct rte_flow_attr q_attr = {
5630 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5631 			.ingress = 1,
5632 		};
5633 		/* Internal PMD action to set register. */
5634 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5635 			.data = qrss_id,
5636 			.id = REG_NON,
5637 		};
5638 		struct rte_flow_item q_items[] = {
5639 			{
5640 				.type = (enum rte_flow_item_type)
5641 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5642 				.spec = &q_tag_spec,
5643 				.last = NULL,
5644 				.mask = NULL,
5645 			},
5646 			{
5647 				.type = RTE_FLOW_ITEM_TYPE_END,
5648 			},
5649 		};
5650 		struct rte_flow_action q_actions[] = {
5651 			{
5652 				.type = qrss->type,
5653 				.conf = qrss->conf,
5654 			},
5655 			{
5656 				.type = RTE_FLOW_ACTION_TYPE_END,
5657 			},
5658 		};
5659 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5660 
5661 		/*
5662 		 * Configure the tag item only if there is no meter subflow.
5663 		 * Since tag is already marked in the meter suffix subflow
5664 		 * we can just use the meter suffix items as is.
5665 		 */
5666 		if (qrss_id) {
5667 			/* Not meter subflow. */
5668 			MLX5_ASSERT(!mtr_sfx);
5669 			/*
5670 			 * Put unique id in prefix flow due to it is destroyed
5671 			 * after suffix flow and id will be freed after there
5672 			 * is no actual flows with this id and identifier
5673 			 * reallocation becomes possible (for example, for
5674 			 * other flows in other threads).
5675 			 */
5676 			dev_flow->handle->split_flow_id = qrss_id;
5677 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5678 						   error);
5679 			if (ret < 0)
5680 				goto exit;
5681 			q_tag_spec.id = ret;
5682 		}
5683 		dev_flow = NULL;
5684 		/* Add suffix subflow to execute Q/RSS. */
5685 		flow_split_info->prefix_layers = layers;
5686 		flow_split_info->prefix_mark = 0;
5687 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5688 					      &q_attr, mtr_sfx ? items :
5689 					      q_items, q_actions,
5690 					      flow_split_info, error);
5691 		if (ret < 0)
5692 			goto exit;
5693 		/* qrss ID should be freed if failed. */
5694 		qrss_id = 0;
5695 		MLX5_ASSERT(dev_flow);
5696 	}
5697 
5698 exit:
5699 	/*
5700 	 * We do not destroy the partially created sub_flows in case of error.
5701 	 * These ones are included into parent flow list and will be destroyed
5702 	 * by flow_drv_destroy.
5703 	 */
5704 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5705 			qrss_id);
5706 	mlx5_free(ext_actions);
5707 	return ret;
5708 }
5709 
5710 /**
5711  * Create meter internal drop flow with the original pattern.
5712  *
5713  * @param dev
5714  *   Pointer to Ethernet device.
5715  * @param[in] flow
5716  *   Parent flow structure pointer.
5717  * @param[in] attr
5718  *   Flow rule attributes.
5719  * @param[in] items
5720  *   Pattern specification (list terminated by the END pattern item).
5721  * @param[in] flow_split_info
5722  *   Pointer to flow split info structure.
5723  * @param[in] fm
5724  *   Pointer to flow meter structure.
5725  * @param[out] error
5726  *   Perform verbose error reporting if not NULL.
5727  * @return
5728  *   0 on success, negative value otherwise
5729  */
5730 static uint32_t
5731 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5732 			struct rte_flow *flow,
5733 			const struct rte_flow_attr *attr,
5734 			const struct rte_flow_item items[],
5735 			struct mlx5_flow_split_info *flow_split_info,
5736 			struct mlx5_flow_meter_info *fm,
5737 			struct rte_flow_error *error)
5738 {
5739 	struct mlx5_flow *dev_flow = NULL;
5740 	struct rte_flow_attr drop_attr = *attr;
5741 	struct rte_flow_action drop_actions[3];
5742 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5743 
5744 	MLX5_ASSERT(fm->drop_cnt);
5745 	drop_actions[0].type =
5746 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5747 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5748 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5749 	drop_actions[1].conf = NULL;
5750 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5751 	drop_actions[2].conf = NULL;
5752 	drop_split_info.external = false;
5753 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5754 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5755 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5756 	return flow_create_split_inner(dev, flow, &dev_flow,
5757 				&drop_attr, items, drop_actions,
5758 				&drop_split_info, error);
5759 }
5760 
5761 /**
5762  * The splitting for meter feature.
5763  *
5764  * - The meter flow will be split to two flows as prefix and
5765  *   suffix flow. The packets make sense only it pass the prefix
5766  *   meter action.
5767  *
5768  * - Reg_C_5 is used for the packet to match betweend prefix and
5769  *   suffix flow.
5770  *
5771  * @param dev
5772  *   Pointer to Ethernet device.
5773  * @param[in] flow
5774  *   Parent flow structure pointer.
5775  * @param[in] attr
5776  *   Flow rule attributes.
5777  * @param[in] items
5778  *   Pattern specification (list terminated by the END pattern item).
5779  * @param[in] actions
5780  *   Associated actions (list terminated by the END action).
5781  * @param[in] flow_split_info
5782  *   Pointer to flow split info structure.
5783  * @param[out] error
5784  *   Perform verbose error reporting if not NULL.
5785  * @return
5786  *   0 on success, negative value otherwise
5787  */
5788 static int
5789 flow_create_split_meter(struct rte_eth_dev *dev,
5790 			struct rte_flow *flow,
5791 			const struct rte_flow_attr *attr,
5792 			const struct rte_flow_item items[],
5793 			const struct rte_flow_action actions[],
5794 			struct mlx5_flow_split_info *flow_split_info,
5795 			struct rte_flow_error *error)
5796 {
5797 	struct mlx5_priv *priv = dev->data->dev_private;
5798 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5799 	struct rte_flow_action *sfx_actions = NULL;
5800 	struct rte_flow_action *pre_actions = NULL;
5801 	struct rte_flow_item *sfx_items = NULL;
5802 	struct mlx5_flow *dev_flow = NULL;
5803 	struct rte_flow_attr sfx_attr = *attr;
5804 	struct mlx5_flow_meter_info *fm = NULL;
5805 	uint8_t skip_scale_restore;
5806 	bool has_mtr = false;
5807 	bool has_modify = false;
5808 	bool set_mtr_reg = true;
5809 	bool is_mtr_hierarchy = false;
5810 	uint32_t meter_id = 0;
5811 	uint32_t mtr_idx = 0;
5812 	uint32_t mtr_flow_id = 0;
5813 	size_t act_size;
5814 	size_t item_size;
5815 	int actions_n = 0;
5816 	int ret = 0;
5817 
5818 	if (priv->mtr_en)
5819 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5820 						    &has_modify, &meter_id);
5821 	if (has_mtr) {
5822 		if (flow->meter) {
5823 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5824 			if (!fm)
5825 				return rte_flow_error_set(error, EINVAL,
5826 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5827 						NULL, "Meter not found.");
5828 		} else {
5829 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5830 			if (!fm)
5831 				return rte_flow_error_set(error, EINVAL,
5832 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5833 						NULL, "Meter not found.");
5834 			ret = mlx5_flow_meter_attach(priv, fm,
5835 						     &sfx_attr, error);
5836 			if (ret)
5837 				return -rte_errno;
5838 			flow->meter = mtr_idx;
5839 		}
5840 		MLX5_ASSERT(wks);
5841 		wks->fm = fm;
5842 		if (!fm->def_policy) {
5843 			wks->policy = mlx5_flow_meter_policy_find(dev,
5844 								  fm->policy_id,
5845 								  NULL);
5846 			MLX5_ASSERT(wks->policy);
5847 			if (wks->policy->is_hierarchy) {
5848 				wks->final_policy =
5849 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
5850 								wks->policy);
5851 				if (!wks->final_policy)
5852 					return rte_flow_error_set(error,
5853 					EINVAL,
5854 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5855 				"Failed to find terminal policy of hierarchy.");
5856 				is_mtr_hierarchy = true;
5857 			}
5858 		}
5859 		/*
5860 		 * If it isn't default-policy Meter, and
5861 		 * 1. There's no action in flow to change
5862 		 *    packet (modify/encap/decap etc.), OR
5863 		 * 2. No drop count needed for this meter.
5864 		 * 3. It's not meter hierarchy.
5865 		 * Then no need to use regC to save meter id anymore.
5866 		 */
5867 		if (!fm->def_policy && !is_mtr_hierarchy &&
5868 		    (!has_modify || !fm->drop_cnt))
5869 			set_mtr_reg = false;
5870 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5871 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5872 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5873 		/* Suffix items: tag, vlan, port id, end. */
5874 #define METER_SUFFIX_ITEM 4
5875 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5876 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5877 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5878 					  0, SOCKET_ID_ANY);
5879 		if (!sfx_actions)
5880 			return rte_flow_error_set(error, ENOMEM,
5881 						  RTE_FLOW_ERROR_TYPE_ACTION,
5882 						  NULL, "no memory to split "
5883 						  "meter flow");
5884 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5885 			     act_size);
5886 		/* There's no suffix flow for meter of non-default policy. */
5887 		if (!fm->def_policy)
5888 			pre_actions = sfx_actions + 1;
5889 		else
5890 			pre_actions = sfx_actions + actions_n;
5891 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
5892 					    items, sfx_items, actions,
5893 					    sfx_actions, pre_actions,
5894 					    (set_mtr_reg ? &mtr_flow_id : NULL),
5895 					    error);
5896 		if (ret) {
5897 			ret = -rte_errno;
5898 			goto exit;
5899 		}
5900 		/* Add the prefix subflow. */
5901 		flow_split_info->prefix_mark = 0;
5902 		skip_scale_restore = flow_split_info->skip_scale;
5903 		flow_split_info->skip_scale |=
5904 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5905 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5906 					      attr, items, pre_actions,
5907 					      flow_split_info, error);
5908 		flow_split_info->skip_scale = skip_scale_restore;
5909 		if (ret) {
5910 			if (mtr_flow_id)
5911 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5912 			ret = -rte_errno;
5913 			goto exit;
5914 		}
5915 		if (mtr_flow_id) {
5916 			dev_flow->handle->split_flow_id = mtr_flow_id;
5917 			dev_flow->handle->is_meter_flow_id = 1;
5918 		}
5919 		if (!fm->def_policy) {
5920 			if (!set_mtr_reg && fm->drop_cnt)
5921 				ret =
5922 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5923 							&sfx_attr, items,
5924 							flow_split_info,
5925 							fm, error);
5926 			goto exit;
5927 		}
5928 		/* Setting the sfx group atrr. */
5929 		sfx_attr.group = sfx_attr.transfer ?
5930 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5931 				 MLX5_FLOW_TABLE_LEVEL_METER;
5932 		flow_split_info->prefix_layers =
5933 				flow_get_prefix_layer_flags(dev_flow);
5934 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5935 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5936 	}
5937 	/* Add the prefix subflow. */
5938 	ret = flow_create_split_metadata(dev, flow,
5939 					 &sfx_attr, sfx_items ?
5940 					 sfx_items : items,
5941 					 sfx_actions ? sfx_actions : actions,
5942 					 flow_split_info, error);
5943 exit:
5944 	if (sfx_actions)
5945 		mlx5_free(sfx_actions);
5946 	return ret;
5947 }
5948 
5949 /**
5950  * The splitting for sample feature.
5951  *
5952  * Once Sample action is detected in the action list, the flow actions should
5953  * be split into prefix sub flow and suffix sub flow.
5954  *
5955  * The original items remain in the prefix sub flow, all actions preceding the
5956  * sample action and the sample action itself will be copied to the prefix
5957  * sub flow, the actions following the sample action will be copied to the
5958  * suffix sub flow, Queue action always be located in the suffix sub flow.
5959  *
5960  * In order to make the packet from prefix sub flow matches with suffix sub
5961  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5962  * flow uses tag item with the unique flow id.
5963  *
5964  * @param dev
5965  *   Pointer to Ethernet device.
5966  * @param[in] flow
5967  *   Parent flow structure pointer.
5968  * @param[in] attr
5969  *   Flow rule attributes.
5970  * @param[in] items
5971  *   Pattern specification (list terminated by the END pattern item).
5972  * @param[in] actions
5973  *   Associated actions (list terminated by the END action).
5974  * @param[in] flow_split_info
5975  *   Pointer to flow split info structure.
5976  * @param[out] error
5977  *   Perform verbose error reporting if not NULL.
5978  * @return
5979  *   0 on success, negative value otherwise
5980  */
5981 static int
5982 flow_create_split_sample(struct rte_eth_dev *dev,
5983 			 struct rte_flow *flow,
5984 			 const struct rte_flow_attr *attr,
5985 			 const struct rte_flow_item items[],
5986 			 const struct rte_flow_action actions[],
5987 			 struct mlx5_flow_split_info *flow_split_info,
5988 			 struct rte_flow_error *error)
5989 {
5990 	struct mlx5_priv *priv = dev->data->dev_private;
5991 	struct rte_flow_action *sfx_actions = NULL;
5992 	struct rte_flow_action *pre_actions = NULL;
5993 	struct rte_flow_item *sfx_items = NULL;
5994 	struct mlx5_flow *dev_flow = NULL;
5995 	struct rte_flow_attr sfx_attr = *attr;
5996 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5997 	struct mlx5_flow_dv_sample_resource *sample_res;
5998 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5999 	struct mlx5_flow_tbl_resource *sfx_tbl;
6000 #endif
6001 	size_t act_size;
6002 	size_t item_size;
6003 	uint32_t fdb_tx = 0;
6004 	int32_t tag_id = 0;
6005 	int actions_n = 0;
6006 	int sample_action_pos;
6007 	int qrss_action_pos;
6008 	int add_tag = 0;
6009 	int modify_after_mirror = 0;
6010 	uint16_t jump_table = 0;
6011 	const uint32_t next_ft_step = 1;
6012 	int ret = 0;
6013 
6014 	if (priv->sampler_en)
6015 		actions_n = flow_check_match_action(actions, attr,
6016 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6017 					&sample_action_pos, &qrss_action_pos,
6018 					&modify_after_mirror);
6019 	if (actions_n) {
6020 		/* The prefix actions must includes sample, tag, end. */
6021 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6022 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6023 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6024 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6025 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6026 					  item_size), 0, SOCKET_ID_ANY);
6027 		if (!sfx_actions)
6028 			return rte_flow_error_set(error, ENOMEM,
6029 						  RTE_FLOW_ERROR_TYPE_ACTION,
6030 						  NULL, "no memory to split "
6031 						  "sample flow");
6032 		/* The representor_id is UINT16_MAX for uplink. */
6033 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6034 		/*
6035 		 * When reg_c_preserve is set, metadata registers Cx preserve
6036 		 * their value even through packet duplication.
6037 		 */
6038 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
6039 		if (add_tag)
6040 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6041 					+ act_size);
6042 		if (modify_after_mirror)
6043 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6044 				     next_ft_step;
6045 		pre_actions = sfx_actions + actions_n;
6046 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6047 						actions, sfx_actions,
6048 						pre_actions, actions_n,
6049 						sample_action_pos,
6050 						qrss_action_pos, jump_table,
6051 						error);
6052 		if (tag_id < 0 || (add_tag && !tag_id)) {
6053 			ret = -rte_errno;
6054 			goto exit;
6055 		}
6056 		if (modify_after_mirror)
6057 			flow_split_info->skip_scale =
6058 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6059 		/* Add the prefix subflow. */
6060 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6061 					      items, pre_actions,
6062 					      flow_split_info, error);
6063 		if (ret) {
6064 			ret = -rte_errno;
6065 			goto exit;
6066 		}
6067 		dev_flow->handle->split_flow_id = tag_id;
6068 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6069 		if (!modify_after_mirror) {
6070 			/* Set the sfx group attr. */
6071 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6072 						dev_flow->dv.sample_res;
6073 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6074 						sample_res->normal_path_tbl;
6075 			sfx_tbl_data = container_of(sfx_tbl,
6076 						struct mlx5_flow_tbl_data_entry,
6077 						tbl);
6078 			sfx_attr.group = sfx_attr.transfer ?
6079 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6080 		} else {
6081 			MLX5_ASSERT(attr->transfer);
6082 			sfx_attr.group = jump_table;
6083 		}
6084 		flow_split_info->prefix_layers =
6085 				flow_get_prefix_layer_flags(dev_flow);
6086 		flow_split_info->prefix_mark = dev_flow->handle->mark;
6087 		/* Suffix group level already be scaled with factor, set
6088 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6089 		 * again in translation.
6090 		 */
6091 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6092 #endif
6093 	}
6094 	/* Add the suffix subflow. */
6095 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6096 				      sfx_items ? sfx_items : items,
6097 				      sfx_actions ? sfx_actions : actions,
6098 				      flow_split_info, error);
6099 exit:
6100 	if (sfx_actions)
6101 		mlx5_free(sfx_actions);
6102 	return ret;
6103 }
6104 
6105 /**
6106  * Split the flow to subflow set. The splitters might be linked
6107  * in the chain, like this:
6108  * flow_create_split_outer() calls:
6109  *   flow_create_split_meter() calls:
6110  *     flow_create_split_metadata(meter_subflow_0) calls:
6111  *       flow_create_split_inner(metadata_subflow_0)
6112  *       flow_create_split_inner(metadata_subflow_1)
6113  *       flow_create_split_inner(metadata_subflow_2)
6114  *     flow_create_split_metadata(meter_subflow_1) calls:
6115  *       flow_create_split_inner(metadata_subflow_0)
6116  *       flow_create_split_inner(metadata_subflow_1)
6117  *       flow_create_split_inner(metadata_subflow_2)
6118  *
6119  * This provide flexible way to add new levels of flow splitting.
6120  * The all of successfully created subflows are included to the
6121  * parent flow dev_flow list.
6122  *
6123  * @param dev
6124  *   Pointer to Ethernet device.
6125  * @param[in] flow
6126  *   Parent flow structure pointer.
6127  * @param[in] attr
6128  *   Flow rule attributes.
6129  * @param[in] items
6130  *   Pattern specification (list terminated by the END pattern item).
6131  * @param[in] actions
6132  *   Associated actions (list terminated by the END action).
6133  * @param[in] flow_split_info
6134  *   Pointer to flow split info structure.
6135  * @param[out] error
6136  *   Perform verbose error reporting if not NULL.
6137  * @return
6138  *   0 on success, negative value otherwise
6139  */
6140 static int
6141 flow_create_split_outer(struct rte_eth_dev *dev,
6142 			struct rte_flow *flow,
6143 			const struct rte_flow_attr *attr,
6144 			const struct rte_flow_item items[],
6145 			const struct rte_flow_action actions[],
6146 			struct mlx5_flow_split_info *flow_split_info,
6147 			struct rte_flow_error *error)
6148 {
6149 	int ret;
6150 
6151 	ret = flow_create_split_sample(dev, flow, attr, items,
6152 				       actions, flow_split_info, error);
6153 	MLX5_ASSERT(ret <= 0);
6154 	return ret;
6155 }
6156 
6157 static inline struct mlx5_flow_tunnel *
6158 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6159 {
6160 	struct mlx5_flow_tunnel *tunnel;
6161 
6162 #pragma GCC diagnostic push
6163 #pragma GCC diagnostic ignored "-Wcast-qual"
6164 	tunnel = (typeof(tunnel))flow->tunnel;
6165 #pragma GCC diagnostic pop
6166 
6167 	return tunnel;
6168 }
6169 
6170 /**
6171  * Adjust flow RSS workspace if needed.
6172  *
6173  * @param wks
6174  *   Pointer to thread flow work space.
6175  * @param rss_desc
6176  *   Pointer to RSS descriptor.
6177  * @param[in] nrssq_num
6178  *   New RSS queue number.
6179  *
6180  * @return
6181  *   0 on success, -1 otherwise and rte_errno is set.
6182  */
6183 static int
6184 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6185 			  struct mlx5_flow_rss_desc *rss_desc,
6186 			  uint32_t nrssq_num)
6187 {
6188 	if (likely(nrssq_num <= wks->rssq_num))
6189 		return 0;
6190 	rss_desc->queue = realloc(rss_desc->queue,
6191 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6192 	if (!rss_desc->queue) {
6193 		rte_errno = ENOMEM;
6194 		return -1;
6195 	}
6196 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6197 	return 0;
6198 }
6199 
6200 /**
6201  * Create a flow and add it to @p list.
6202  *
6203  * @param dev
6204  *   Pointer to Ethernet device.
6205  * @param list
6206  *   Pointer to a TAILQ flow list. If this parameter NULL,
6207  *   no list insertion occurred, flow is just created,
6208  *   this is caller's responsibility to track the
6209  *   created flow.
6210  * @param[in] attr
6211  *   Flow rule attributes.
6212  * @param[in] items
6213  *   Pattern specification (list terminated by the END pattern item).
6214  * @param[in] actions
6215  *   Associated actions (list terminated by the END action).
6216  * @param[in] external
6217  *   This flow rule is created by request external to PMD.
6218  * @param[out] error
6219  *   Perform verbose error reporting if not NULL.
6220  *
6221  * @return
6222  *   A flow index on success, 0 otherwise and rte_errno is set.
6223  */
6224 static uint32_t
6225 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6226 		 const struct rte_flow_attr *attr,
6227 		 const struct rte_flow_item items[],
6228 		 const struct rte_flow_action original_actions[],
6229 		 bool external, struct rte_flow_error *error)
6230 {
6231 	struct mlx5_priv *priv = dev->data->dev_private;
6232 	struct rte_flow *flow = NULL;
6233 	struct mlx5_flow *dev_flow;
6234 	const struct rte_flow_action_rss *rss = NULL;
6235 	struct mlx5_translated_action_handle
6236 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6237 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6238 	union {
6239 		struct mlx5_flow_expand_rss buf;
6240 		uint8_t buffer[2048];
6241 	} expand_buffer;
6242 	union {
6243 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6244 		uint8_t buffer[2048];
6245 	} actions_rx;
6246 	union {
6247 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6248 		uint8_t buffer[2048];
6249 	} actions_hairpin_tx;
6250 	union {
6251 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6252 		uint8_t buffer[2048];
6253 	} items_tx;
6254 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6255 	struct mlx5_flow_rss_desc *rss_desc;
6256 	const struct rte_flow_action *p_actions_rx;
6257 	uint32_t i;
6258 	uint32_t idx = 0;
6259 	int hairpin_flow;
6260 	struct rte_flow_attr attr_tx = { .priority = 0 };
6261 	const struct rte_flow_action *actions;
6262 	struct rte_flow_action *translated_actions = NULL;
6263 	struct mlx5_flow_tunnel *tunnel;
6264 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6265 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6266 	struct mlx5_flow_split_info flow_split_info = {
6267 		.external = !!external,
6268 		.skip_scale = 0,
6269 		.flow_idx = 0,
6270 		.prefix_mark = 0,
6271 		.prefix_layers = 0,
6272 		.table_id = 0
6273 	};
6274 	int ret;
6275 
6276 	MLX5_ASSERT(wks);
6277 	rss_desc = &wks->rss_desc;
6278 	ret = flow_action_handles_translate(dev, original_actions,
6279 					    indir_actions,
6280 					    &indir_actions_n,
6281 					    &translated_actions, error);
6282 	if (ret < 0) {
6283 		MLX5_ASSERT(translated_actions == NULL);
6284 		return 0;
6285 	}
6286 	actions = translated_actions ? translated_actions : original_actions;
6287 	p_actions_rx = actions;
6288 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6289 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6290 				external, hairpin_flow, error);
6291 	if (ret < 0)
6292 		goto error_before_hairpin_split;
6293 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6294 	if (!flow) {
6295 		rte_errno = ENOMEM;
6296 		goto error_before_hairpin_split;
6297 	}
6298 	if (hairpin_flow > 0) {
6299 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6300 			rte_errno = EINVAL;
6301 			goto error_before_hairpin_split;
6302 		}
6303 		flow_hairpin_split(dev, actions, actions_rx.actions,
6304 				   actions_hairpin_tx.actions, items_tx.items,
6305 				   idx);
6306 		p_actions_rx = actions_rx.actions;
6307 	}
6308 	flow_split_info.flow_idx = idx;
6309 	flow->drv_type = flow_get_drv_type(dev, attr);
6310 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6311 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6312 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6313 	/* RSS Action only works on NIC RX domain */
6314 	if (attr->ingress && !attr->transfer)
6315 		rss = flow_get_rss_action(dev, p_actions_rx);
6316 	if (rss) {
6317 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6318 			return 0;
6319 		/*
6320 		 * The following information is required by
6321 		 * mlx5_flow_hashfields_adjust() in advance.
6322 		 */
6323 		rss_desc->level = rss->level;
6324 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6325 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6326 	}
6327 	flow->dev_handles = 0;
6328 	if (rss && rss->types) {
6329 		unsigned int graph_root;
6330 
6331 		graph_root = find_graph_root(items, rss->level);
6332 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6333 					   items, rss->types,
6334 					   mlx5_support_expansion, graph_root);
6335 		MLX5_ASSERT(ret > 0 &&
6336 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6337 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6338 			for (i = 0; i < buf->entries; ++i)
6339 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6340 		}
6341 	} else {
6342 		buf->entries = 1;
6343 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6344 	}
6345 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6346 						      indir_actions_n);
6347 	for (i = 0; i < buf->entries; ++i) {
6348 		/* Initialize flow split data. */
6349 		flow_split_info.prefix_layers = 0;
6350 		flow_split_info.prefix_mark = 0;
6351 		flow_split_info.skip_scale = 0;
6352 		/*
6353 		 * The splitter may create multiple dev_flows,
6354 		 * depending on configuration. In the simplest
6355 		 * case it just creates unmodified original flow.
6356 		 */
6357 		ret = flow_create_split_outer(dev, flow, attr,
6358 					      buf->entry[i].pattern,
6359 					      p_actions_rx, &flow_split_info,
6360 					      error);
6361 		if (ret < 0)
6362 			goto error;
6363 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6364 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6365 							   p_actions_rx,
6366 							   idx,
6367 							   wks->flows[0].tunnel,
6368 							   &default_miss_ctx,
6369 							   error);
6370 			if (ret < 0) {
6371 				mlx5_free(default_miss_ctx.queue);
6372 				goto error;
6373 			}
6374 		}
6375 	}
6376 	/* Create the tx flow. */
6377 	if (hairpin_flow) {
6378 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6379 		attr_tx.ingress = 0;
6380 		attr_tx.egress = 1;
6381 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6382 					 actions_hairpin_tx.actions,
6383 					 idx, error);
6384 		if (!dev_flow)
6385 			goto error;
6386 		dev_flow->flow = flow;
6387 		dev_flow->external = 0;
6388 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6389 			      dev_flow->handle, next);
6390 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6391 					 items_tx.items,
6392 					 actions_hairpin_tx.actions, error);
6393 		if (ret < 0)
6394 			goto error;
6395 	}
6396 	/*
6397 	 * Update the metadata register copy table. If extensive
6398 	 * metadata feature is enabled and registers are supported
6399 	 * we might create the extra rte_flow for each unique
6400 	 * MARK/FLAG action ID.
6401 	 *
6402 	 * The table is updated for ingress Flows only, because
6403 	 * the egress Flows belong to the different device and
6404 	 * copy table should be updated in peer NIC Rx domain.
6405 	 */
6406 	if (attr->ingress &&
6407 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6408 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6409 		if (ret)
6410 			goto error;
6411 	}
6412 	/*
6413 	 * If the flow is external (from application) OR device is started,
6414 	 * OR mreg discover, then apply immediately.
6415 	 */
6416 	if (external || dev->data->dev_started ||
6417 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6418 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6419 		ret = flow_drv_apply(dev, flow, error);
6420 		if (ret < 0)
6421 			goto error;
6422 	}
6423 	flow->type = type;
6424 	flow_rxq_flags_set(dev, flow);
6425 	rte_free(translated_actions);
6426 	tunnel = flow_tunnel_from_rule(wks->flows);
6427 	if (tunnel) {
6428 		flow->tunnel = 1;
6429 		flow->tunnel_id = tunnel->tunnel_id;
6430 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6431 		mlx5_free(default_miss_ctx.queue);
6432 	}
6433 	mlx5_flow_pop_thread_workspace();
6434 	return idx;
6435 error:
6436 	MLX5_ASSERT(flow);
6437 	ret = rte_errno; /* Save rte_errno before cleanup. */
6438 	flow_mreg_del_copy_action(dev, flow);
6439 	flow_drv_destroy(dev, flow);
6440 	if (rss_desc->shared_rss)
6441 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6442 			mlx5_ipool_get
6443 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6444 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6445 	mlx5_ipool_free(priv->flows[type], idx);
6446 	rte_errno = ret; /* Restore rte_errno. */
6447 	ret = rte_errno;
6448 	rte_errno = ret;
6449 	mlx5_flow_pop_thread_workspace();
6450 error_before_hairpin_split:
6451 	rte_free(translated_actions);
6452 	return 0;
6453 }
6454 
6455 /**
6456  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6457  * incoming packets to table 1.
6458  *
6459  * Other flow rules, requested for group n, will be created in
6460  * e-switch table n+1.
6461  * Jump action to e-switch group n will be created to group n+1.
6462  *
6463  * Used when working in switchdev mode, to utilise advantages of table 1
6464  * and above.
6465  *
6466  * @param dev
6467  *   Pointer to Ethernet device.
6468  *
6469  * @return
6470  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6471  */
6472 struct rte_flow *
6473 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6474 {
6475 	const struct rte_flow_attr attr = {
6476 		.group = 0,
6477 		.priority = 0,
6478 		.ingress = 1,
6479 		.egress = 0,
6480 		.transfer = 1,
6481 	};
6482 	const struct rte_flow_item pattern = {
6483 		.type = RTE_FLOW_ITEM_TYPE_END,
6484 	};
6485 	struct rte_flow_action_jump jump = {
6486 		.group = 1,
6487 	};
6488 	const struct rte_flow_action actions[] = {
6489 		{
6490 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6491 			.conf = &jump,
6492 		},
6493 		{
6494 			.type = RTE_FLOW_ACTION_TYPE_END,
6495 		},
6496 	};
6497 	struct rte_flow_error error;
6498 
6499 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6500 						   &attr, &pattern,
6501 						   actions, false, &error);
6502 }
6503 
6504 /**
6505  * Validate a flow supported by the NIC.
6506  *
6507  * @see rte_flow_validate()
6508  * @see rte_flow_ops
6509  */
6510 int
6511 mlx5_flow_validate(struct rte_eth_dev *dev,
6512 		   const struct rte_flow_attr *attr,
6513 		   const struct rte_flow_item items[],
6514 		   const struct rte_flow_action original_actions[],
6515 		   struct rte_flow_error *error)
6516 {
6517 	int hairpin_flow;
6518 	struct mlx5_translated_action_handle
6519 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6520 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6521 	const struct rte_flow_action *actions;
6522 	struct rte_flow_action *translated_actions = NULL;
6523 	int ret = flow_action_handles_translate(dev, original_actions,
6524 						indir_actions,
6525 						&indir_actions_n,
6526 						&translated_actions, error);
6527 
6528 	if (ret)
6529 		return ret;
6530 	actions = translated_actions ? translated_actions : original_actions;
6531 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6532 	ret = flow_drv_validate(dev, attr, items, actions,
6533 				true, hairpin_flow, error);
6534 	rte_free(translated_actions);
6535 	return ret;
6536 }
6537 
6538 /**
6539  * Create a flow.
6540  *
6541  * @see rte_flow_create()
6542  * @see rte_flow_ops
6543  */
6544 struct rte_flow *
6545 mlx5_flow_create(struct rte_eth_dev *dev,
6546 		 const struct rte_flow_attr *attr,
6547 		 const struct rte_flow_item items[],
6548 		 const struct rte_flow_action actions[],
6549 		 struct rte_flow_error *error)
6550 {
6551 	/*
6552 	 * If the device is not started yet, it is not allowed to created a
6553 	 * flow from application. PMD default flows and traffic control flows
6554 	 * are not affected.
6555 	 */
6556 	if (unlikely(!dev->data->dev_started)) {
6557 		DRV_LOG(DEBUG, "port %u is not started when "
6558 			"inserting a flow", dev->data->port_id);
6559 		rte_flow_error_set(error, ENODEV,
6560 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6561 				   NULL,
6562 				   "port not started");
6563 		return NULL;
6564 	}
6565 
6566 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6567 						   attr, items, actions,
6568 						   true, error);
6569 }
6570 
6571 /**
6572  * Destroy a flow in a list.
6573  *
6574  * @param dev
6575  *   Pointer to Ethernet device.
6576  * @param[in] flow_idx
6577  *   Index of flow to destroy.
6578  */
6579 static void
6580 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6581 		  uint32_t flow_idx)
6582 {
6583 	struct mlx5_priv *priv = dev->data->dev_private;
6584 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6585 
6586 	if (!flow)
6587 		return;
6588 	MLX5_ASSERT(flow->type == type);
6589 	/*
6590 	 * Update RX queue flags only if port is started, otherwise it is
6591 	 * already clean.
6592 	 */
6593 	if (dev->data->dev_started)
6594 		flow_rxq_flags_trim(dev, flow);
6595 	flow_drv_destroy(dev, flow);
6596 	if (flow->tunnel) {
6597 		struct mlx5_flow_tunnel *tunnel;
6598 
6599 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6600 		RTE_VERIFY(tunnel);
6601 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6602 			mlx5_flow_tunnel_free(dev, tunnel);
6603 	}
6604 	flow_mreg_del_copy_action(dev, flow);
6605 	mlx5_ipool_free(priv->flows[type], flow_idx);
6606 }
6607 
6608 /**
6609  * Destroy all flows.
6610  *
6611  * @param dev
6612  *   Pointer to Ethernet device.
6613  * @param type
6614  *   Flow type to be flushed.
6615  * @param active
6616  *   If flushing is called avtively.
6617  */
6618 void
6619 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6620 		     bool active)
6621 {
6622 	struct mlx5_priv *priv = dev->data->dev_private;
6623 	uint32_t num_flushed = 0, fidx = 1;
6624 	struct rte_flow *flow;
6625 
6626 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6627 		flow_list_destroy(dev, type, fidx);
6628 		num_flushed++;
6629 	}
6630 	if (active) {
6631 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6632 			dev->data->port_id, num_flushed);
6633 	}
6634 }
6635 
6636 /**
6637  * Stop all default actions for flows.
6638  *
6639  * @param dev
6640  *   Pointer to Ethernet device.
6641  */
6642 void
6643 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6644 {
6645 	flow_mreg_del_default_copy_action(dev);
6646 	flow_rxq_flags_clear(dev);
6647 }
6648 
6649 /**
6650  * Start all default actions for flows.
6651  *
6652  * @param dev
6653  *   Pointer to Ethernet device.
6654  * @return
6655  *   0 on success, a negative errno value otherwise and rte_errno is set.
6656  */
6657 int
6658 mlx5_flow_start_default(struct rte_eth_dev *dev)
6659 {
6660 	struct rte_flow_error error;
6661 
6662 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6663 	return flow_mreg_add_default_copy_action(dev, &error);
6664 }
6665 
6666 /**
6667  * Release key of thread specific flow workspace data.
6668  */
6669 void
6670 flow_release_workspace(void *data)
6671 {
6672 	struct mlx5_flow_workspace *wks = data;
6673 	struct mlx5_flow_workspace *next;
6674 
6675 	while (wks) {
6676 		next = wks->next;
6677 		free(wks->rss_desc.queue);
6678 		free(wks);
6679 		wks = next;
6680 	}
6681 }
6682 
6683 /**
6684  * Get thread specific current flow workspace.
6685  *
6686  * @return pointer to thread specific flow workspace data, NULL on error.
6687  */
6688 struct mlx5_flow_workspace*
6689 mlx5_flow_get_thread_workspace(void)
6690 {
6691 	struct mlx5_flow_workspace *data;
6692 
6693 	data = mlx5_flow_os_get_specific_workspace();
6694 	MLX5_ASSERT(data && data->inuse);
6695 	if (!data || !data->inuse)
6696 		DRV_LOG(ERR, "flow workspace not initialized.");
6697 	return data;
6698 }
6699 
6700 /**
6701  * Allocate and init new flow workspace.
6702  *
6703  * @return pointer to flow workspace data, NULL on error.
6704  */
6705 static struct mlx5_flow_workspace*
6706 flow_alloc_thread_workspace(void)
6707 {
6708 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6709 
6710 	if (!data) {
6711 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6712 			"memory.");
6713 		return NULL;
6714 	}
6715 	data->rss_desc.queue = calloc(1,
6716 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6717 	if (!data->rss_desc.queue)
6718 		goto err;
6719 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6720 	return data;
6721 err:
6722 	if (data->rss_desc.queue)
6723 		free(data->rss_desc.queue);
6724 	free(data);
6725 	return NULL;
6726 }
6727 
6728 /**
6729  * Get new thread specific flow workspace.
6730  *
6731  * If current workspace inuse, create new one and set as current.
6732  *
6733  * @return pointer to thread specific flow workspace data, NULL on error.
6734  */
6735 static struct mlx5_flow_workspace*
6736 mlx5_flow_push_thread_workspace(void)
6737 {
6738 	struct mlx5_flow_workspace *curr;
6739 	struct mlx5_flow_workspace *data;
6740 
6741 	curr = mlx5_flow_os_get_specific_workspace();
6742 	if (!curr) {
6743 		data = flow_alloc_thread_workspace();
6744 		if (!data)
6745 			return NULL;
6746 	} else if (!curr->inuse) {
6747 		data = curr;
6748 	} else if (curr->next) {
6749 		data = curr->next;
6750 	} else {
6751 		data = flow_alloc_thread_workspace();
6752 		if (!data)
6753 			return NULL;
6754 		curr->next = data;
6755 		data->prev = curr;
6756 	}
6757 	data->inuse = 1;
6758 	data->flow_idx = 0;
6759 	/* Set as current workspace */
6760 	if (mlx5_flow_os_set_specific_workspace(data))
6761 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6762 	return data;
6763 }
6764 
6765 /**
6766  * Close current thread specific flow workspace.
6767  *
6768  * If previous workspace available, set it as current.
6769  *
6770  * @return pointer to thread specific flow workspace data, NULL on error.
6771  */
6772 static void
6773 mlx5_flow_pop_thread_workspace(void)
6774 {
6775 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6776 
6777 	if (!data)
6778 		return;
6779 	if (!data->inuse) {
6780 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6781 		return;
6782 	}
6783 	data->inuse = 0;
6784 	if (!data->prev)
6785 		return;
6786 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6787 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6788 }
6789 
6790 /**
6791  * Verify the flow list is empty
6792  *
6793  * @param dev
6794  *  Pointer to Ethernet device.
6795  *
6796  * @return the number of flows not released.
6797  */
6798 int
6799 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
6800 {
6801 	struct mlx5_priv *priv = dev->data->dev_private;
6802 	struct rte_flow *flow;
6803 	uint32_t idx = 0;
6804 	int ret = 0, i;
6805 
6806 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
6807 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
6808 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
6809 				dev->data->port_id, (void *)flow);
6810 			ret++;
6811 		}
6812 	}
6813 	return ret;
6814 }
6815 
6816 /**
6817  * Enable default hairpin egress flow.
6818  *
6819  * @param dev
6820  *   Pointer to Ethernet device.
6821  * @param queue
6822  *   The queue index.
6823  *
6824  * @return
6825  *   0 on success, a negative errno value otherwise and rte_errno is set.
6826  */
6827 int
6828 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6829 			    uint32_t queue)
6830 {
6831 	const struct rte_flow_attr attr = {
6832 		.egress = 1,
6833 		.priority = 0,
6834 	};
6835 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6836 		.queue = queue,
6837 	};
6838 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6839 		.queue = UINT32_MAX,
6840 	};
6841 	struct rte_flow_item items[] = {
6842 		{
6843 			.type = (enum rte_flow_item_type)
6844 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6845 			.spec = &queue_spec,
6846 			.last = NULL,
6847 			.mask = &queue_mask,
6848 		},
6849 		{
6850 			.type = RTE_FLOW_ITEM_TYPE_END,
6851 		},
6852 	};
6853 	struct rte_flow_action_jump jump = {
6854 		.group = MLX5_HAIRPIN_TX_TABLE,
6855 	};
6856 	struct rte_flow_action actions[2];
6857 	uint32_t flow_idx;
6858 	struct rte_flow_error error;
6859 
6860 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6861 	actions[0].conf = &jump;
6862 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6863 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6864 				    &attr, items, actions, false, &error);
6865 	if (!flow_idx) {
6866 		DRV_LOG(DEBUG,
6867 			"Failed to create ctrl flow: rte_errno(%d),"
6868 			" type(%d), message(%s)",
6869 			rte_errno, error.type,
6870 			error.message ? error.message : " (no stated reason)");
6871 		return -rte_errno;
6872 	}
6873 	return 0;
6874 }
6875 
6876 /**
6877  * Enable a control flow configured from the control plane.
6878  *
6879  * @param dev
6880  *   Pointer to Ethernet device.
6881  * @param eth_spec
6882  *   An Ethernet flow spec to apply.
6883  * @param eth_mask
6884  *   An Ethernet flow mask to apply.
6885  * @param vlan_spec
6886  *   A VLAN flow spec to apply.
6887  * @param vlan_mask
6888  *   A VLAN flow mask to apply.
6889  *
6890  * @return
6891  *   0 on success, a negative errno value otherwise and rte_errno is set.
6892  */
6893 int
6894 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6895 		    struct rte_flow_item_eth *eth_spec,
6896 		    struct rte_flow_item_eth *eth_mask,
6897 		    struct rte_flow_item_vlan *vlan_spec,
6898 		    struct rte_flow_item_vlan *vlan_mask)
6899 {
6900 	struct mlx5_priv *priv = dev->data->dev_private;
6901 	const struct rte_flow_attr attr = {
6902 		.ingress = 1,
6903 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6904 	};
6905 	struct rte_flow_item items[] = {
6906 		{
6907 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6908 			.spec = eth_spec,
6909 			.last = NULL,
6910 			.mask = eth_mask,
6911 		},
6912 		{
6913 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6914 					      RTE_FLOW_ITEM_TYPE_END,
6915 			.spec = vlan_spec,
6916 			.last = NULL,
6917 			.mask = vlan_mask,
6918 		},
6919 		{
6920 			.type = RTE_FLOW_ITEM_TYPE_END,
6921 		},
6922 	};
6923 	uint16_t queue[priv->reta_idx_n];
6924 	struct rte_flow_action_rss action_rss = {
6925 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6926 		.level = 0,
6927 		.types = priv->rss_conf.rss_hf,
6928 		.key_len = priv->rss_conf.rss_key_len,
6929 		.queue_num = priv->reta_idx_n,
6930 		.key = priv->rss_conf.rss_key,
6931 		.queue = queue,
6932 	};
6933 	struct rte_flow_action actions[] = {
6934 		{
6935 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6936 			.conf = &action_rss,
6937 		},
6938 		{
6939 			.type = RTE_FLOW_ACTION_TYPE_END,
6940 		},
6941 	};
6942 	uint32_t flow_idx;
6943 	struct rte_flow_error error;
6944 	unsigned int i;
6945 
6946 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6947 		return 0;
6948 	}
6949 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6950 		action_rss.types = 0;
6951 	for (i = 0; i != priv->reta_idx_n; ++i)
6952 		queue[i] = (*priv->reta_idx)[i];
6953 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6954 				    &attr, items, actions, false, &error);
6955 	if (!flow_idx)
6956 		return -rte_errno;
6957 	return 0;
6958 }
6959 
6960 /**
6961  * Enable a flow control configured from the control plane.
6962  *
6963  * @param dev
6964  *   Pointer to Ethernet device.
6965  * @param eth_spec
6966  *   An Ethernet flow spec to apply.
6967  * @param eth_mask
6968  *   An Ethernet flow mask to apply.
6969  *
6970  * @return
6971  *   0 on success, a negative errno value otherwise and rte_errno is set.
6972  */
6973 int
6974 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6975 	       struct rte_flow_item_eth *eth_spec,
6976 	       struct rte_flow_item_eth *eth_mask)
6977 {
6978 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6979 }
6980 
6981 /**
6982  * Create default miss flow rule matching lacp traffic
6983  *
6984  * @param dev
6985  *   Pointer to Ethernet device.
6986  * @param eth_spec
6987  *   An Ethernet flow spec to apply.
6988  *
6989  * @return
6990  *   0 on success, a negative errno value otherwise and rte_errno is set.
6991  */
6992 int
6993 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6994 {
6995 	/*
6996 	 * The LACP matching is done by only using ether type since using
6997 	 * a multicast dst mac causes kernel to give low priority to this flow.
6998 	 */
6999 	static const struct rte_flow_item_eth lacp_spec = {
7000 		.type = RTE_BE16(0x8809),
7001 	};
7002 	static const struct rte_flow_item_eth lacp_mask = {
7003 		.type = 0xffff,
7004 	};
7005 	const struct rte_flow_attr attr = {
7006 		.ingress = 1,
7007 	};
7008 	struct rte_flow_item items[] = {
7009 		{
7010 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7011 			.spec = &lacp_spec,
7012 			.mask = &lacp_mask,
7013 		},
7014 		{
7015 			.type = RTE_FLOW_ITEM_TYPE_END,
7016 		},
7017 	};
7018 	struct rte_flow_action actions[] = {
7019 		{
7020 			.type = (enum rte_flow_action_type)
7021 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7022 		},
7023 		{
7024 			.type = RTE_FLOW_ACTION_TYPE_END,
7025 		},
7026 	};
7027 	struct rte_flow_error error;
7028 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7029 					&attr, items, actions,
7030 					false, &error);
7031 
7032 	if (!flow_idx)
7033 		return -rte_errno;
7034 	return 0;
7035 }
7036 
7037 /**
7038  * Destroy a flow.
7039  *
7040  * @see rte_flow_destroy()
7041  * @see rte_flow_ops
7042  */
7043 int
7044 mlx5_flow_destroy(struct rte_eth_dev *dev,
7045 		  struct rte_flow *flow,
7046 		  struct rte_flow_error *error __rte_unused)
7047 {
7048 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7049 				(uintptr_t)(void *)flow);
7050 	return 0;
7051 }
7052 
7053 /**
7054  * Destroy all flows.
7055  *
7056  * @see rte_flow_flush()
7057  * @see rte_flow_ops
7058  */
7059 int
7060 mlx5_flow_flush(struct rte_eth_dev *dev,
7061 		struct rte_flow_error *error __rte_unused)
7062 {
7063 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7064 	return 0;
7065 }
7066 
7067 /**
7068  * Isolated mode.
7069  *
7070  * @see rte_flow_isolate()
7071  * @see rte_flow_ops
7072  */
7073 int
7074 mlx5_flow_isolate(struct rte_eth_dev *dev,
7075 		  int enable,
7076 		  struct rte_flow_error *error)
7077 {
7078 	struct mlx5_priv *priv = dev->data->dev_private;
7079 
7080 	if (dev->data->dev_started) {
7081 		rte_flow_error_set(error, EBUSY,
7082 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7083 				   NULL,
7084 				   "port must be stopped first");
7085 		return -rte_errno;
7086 	}
7087 	priv->isolated = !!enable;
7088 	if (enable)
7089 		dev->dev_ops = &mlx5_dev_ops_isolate;
7090 	else
7091 		dev->dev_ops = &mlx5_dev_ops;
7092 
7093 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7094 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7095 
7096 	return 0;
7097 }
7098 
7099 /**
7100  * Query a flow.
7101  *
7102  * @see rte_flow_query()
7103  * @see rte_flow_ops
7104  */
7105 static int
7106 flow_drv_query(struct rte_eth_dev *dev,
7107 	       uint32_t flow_idx,
7108 	       const struct rte_flow_action *actions,
7109 	       void *data,
7110 	       struct rte_flow_error *error)
7111 {
7112 	struct mlx5_priv *priv = dev->data->dev_private;
7113 	const struct mlx5_flow_driver_ops *fops;
7114 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7115 					       flow_idx);
7116 	enum mlx5_flow_drv_type ftype;
7117 
7118 	if (!flow) {
7119 		return rte_flow_error_set(error, ENOENT,
7120 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7121 			  NULL,
7122 			  "invalid flow handle");
7123 	}
7124 	ftype = flow->drv_type;
7125 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7126 	fops = flow_get_drv_ops(ftype);
7127 
7128 	return fops->query(dev, flow, actions, data, error);
7129 }
7130 
7131 /**
7132  * Query a flow.
7133  *
7134  * @see rte_flow_query()
7135  * @see rte_flow_ops
7136  */
7137 int
7138 mlx5_flow_query(struct rte_eth_dev *dev,
7139 		struct rte_flow *flow,
7140 		const struct rte_flow_action *actions,
7141 		void *data,
7142 		struct rte_flow_error *error)
7143 {
7144 	int ret;
7145 
7146 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7147 			     error);
7148 	if (ret < 0)
7149 		return ret;
7150 	return 0;
7151 }
7152 
7153 /**
7154  * Get rte_flow callbacks.
7155  *
7156  * @param dev
7157  *   Pointer to Ethernet device structure.
7158  * @param ops
7159  *   Pointer to operation-specific structure.
7160  *
7161  * @return 0
7162  */
7163 int
7164 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7165 		  const struct rte_flow_ops **ops)
7166 {
7167 	*ops = &mlx5_flow_ops;
7168 	return 0;
7169 }
7170 
7171 /**
7172  * Validate meter policy actions.
7173  * Dispatcher for action type specific validation.
7174  *
7175  * @param[in] dev
7176  *   Pointer to the Ethernet device structure.
7177  * @param[in] action
7178  *   The meter policy action object to validate.
7179  * @param[in] attr
7180  *   Attributes of flow to determine steering domain.
7181  * @param[out] is_rss
7182  *   Is RSS or not.
7183  * @param[out] domain_bitmap
7184  *   Domain bitmap.
7185  * @param[out] is_def_policy
7186  *   Is default policy or not.
7187  * @param[out] error
7188  *   Perform verbose error reporting if not NULL. Initialized in case of
7189  *   error only.
7190  *
7191  * @return
7192  *   0 on success, otherwise negative errno value.
7193  */
7194 int
7195 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7196 			const struct rte_flow_action *actions[RTE_COLORS],
7197 			struct rte_flow_attr *attr,
7198 			bool *is_rss,
7199 			uint8_t *domain_bitmap,
7200 			bool *is_def_policy,
7201 			struct rte_mtr_error *error)
7202 {
7203 	const struct mlx5_flow_driver_ops *fops;
7204 
7205 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7206 	return fops->validate_mtr_acts(dev, actions, attr,
7207 			is_rss, domain_bitmap, is_def_policy, error);
7208 }
7209 
7210 /**
7211  * Destroy the meter table set.
7212  *
7213  * @param[in] dev
7214  *   Pointer to Ethernet device.
7215  * @param[in] mtr_policy
7216  *   Meter policy struct.
7217  */
7218 void
7219 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7220 		      struct mlx5_flow_meter_policy *mtr_policy)
7221 {
7222 	const struct mlx5_flow_driver_ops *fops;
7223 
7224 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7225 	fops->destroy_mtr_acts(dev, mtr_policy);
7226 }
7227 
7228 /**
7229  * Create policy action, lock free,
7230  * (mutex should be acquired by caller).
7231  * Dispatcher for action type specific call.
7232  *
7233  * @param[in] dev
7234  *   Pointer to the Ethernet device structure.
7235  * @param[in] mtr_policy
7236  *   Meter policy struct.
7237  * @param[in] action
7238  *   Action specification used to create meter actions.
7239  * @param[out] error
7240  *   Perform verbose error reporting if not NULL. Initialized in case of
7241  *   error only.
7242  *
7243  * @return
7244  *   0 on success, otherwise negative errno value.
7245  */
7246 int
7247 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7248 		      struct mlx5_flow_meter_policy *mtr_policy,
7249 		      const struct rte_flow_action *actions[RTE_COLORS],
7250 		      struct rte_mtr_error *error)
7251 {
7252 	const struct mlx5_flow_driver_ops *fops;
7253 
7254 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7255 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7256 }
7257 
7258 /**
7259  * Create policy rules, lock free,
7260  * (mutex should be acquired by caller).
7261  * Dispatcher for action type specific call.
7262  *
7263  * @param[in] dev
7264  *   Pointer to the Ethernet device structure.
7265  * @param[in] mtr_policy
7266  *   Meter policy struct.
7267  *
7268  * @return
7269  *   0 on success, -1 otherwise.
7270  */
7271 int
7272 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7273 			     struct mlx5_flow_meter_policy *mtr_policy)
7274 {
7275 	const struct mlx5_flow_driver_ops *fops;
7276 
7277 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7278 	return fops->create_policy_rules(dev, mtr_policy);
7279 }
7280 
7281 /**
7282  * Destroy policy rules, lock free,
7283  * (mutex should be acquired by caller).
7284  * Dispatcher for action type specific call.
7285  *
7286  * @param[in] dev
7287  *   Pointer to the Ethernet device structure.
7288  * @param[in] mtr_policy
7289  *   Meter policy struct.
7290  */
7291 void
7292 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7293 			     struct mlx5_flow_meter_policy *mtr_policy)
7294 {
7295 	const struct mlx5_flow_driver_ops *fops;
7296 
7297 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7298 	fops->destroy_policy_rules(dev, mtr_policy);
7299 }
7300 
7301 /**
7302  * Destroy the default policy table set.
7303  *
7304  * @param[in] dev
7305  *   Pointer to Ethernet device.
7306  */
7307 void
7308 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7309 {
7310 	const struct mlx5_flow_driver_ops *fops;
7311 
7312 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7313 	fops->destroy_def_policy(dev);
7314 }
7315 
7316 /**
7317  * Destroy the default policy table set.
7318  *
7319  * @param[in] dev
7320  *   Pointer to Ethernet device.
7321  *
7322  * @return
7323  *   0 on success, -1 otherwise.
7324  */
7325 int
7326 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7327 {
7328 	const struct mlx5_flow_driver_ops *fops;
7329 
7330 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7331 	return fops->create_def_policy(dev);
7332 }
7333 
7334 /**
7335  * Create the needed meter and suffix tables.
7336  *
7337  * @param[in] dev
7338  *   Pointer to Ethernet device.
7339  *
7340  * @return
7341  *   0 on success, -1 otherwise.
7342  */
7343 int
7344 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7345 			struct mlx5_flow_meter_info *fm,
7346 			uint32_t mtr_idx,
7347 			uint8_t domain_bitmap)
7348 {
7349 	const struct mlx5_flow_driver_ops *fops;
7350 
7351 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7352 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7353 }
7354 
7355 /**
7356  * Destroy the meter table set.
7357  *
7358  * @param[in] dev
7359  *   Pointer to Ethernet device.
7360  * @param[in] tbl
7361  *   Pointer to the meter table set.
7362  */
7363 void
7364 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7365 			   struct mlx5_flow_meter_info *fm)
7366 {
7367 	const struct mlx5_flow_driver_ops *fops;
7368 
7369 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7370 	fops->destroy_mtr_tbls(dev, fm);
7371 }
7372 
7373 /**
7374  * Destroy the global meter drop table.
7375  *
7376  * @param[in] dev
7377  *   Pointer to Ethernet device.
7378  */
7379 void
7380 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7381 {
7382 	const struct mlx5_flow_driver_ops *fops;
7383 
7384 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7385 	fops->destroy_mtr_drop_tbls(dev);
7386 }
7387 
7388 /**
7389  * Destroy the sub policy table with RX queue.
7390  *
7391  * @param[in] dev
7392  *   Pointer to Ethernet device.
7393  * @param[in] mtr_policy
7394  *   Pointer to meter policy table.
7395  */
7396 void
7397 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7398 		struct mlx5_flow_meter_policy *mtr_policy)
7399 {
7400 	const struct mlx5_flow_driver_ops *fops;
7401 
7402 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7403 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7404 }
7405 
7406 /**
7407  * Allocate the needed aso flow meter id.
7408  *
7409  * @param[in] dev
7410  *   Pointer to Ethernet device.
7411  *
7412  * @return
7413  *   Index to aso flow meter on success, NULL otherwise.
7414  */
7415 uint32_t
7416 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7417 {
7418 	const struct mlx5_flow_driver_ops *fops;
7419 
7420 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7421 	return fops->create_meter(dev);
7422 }
7423 
7424 /**
7425  * Free the aso flow meter id.
7426  *
7427  * @param[in] dev
7428  *   Pointer to Ethernet device.
7429  * @param[in] mtr_idx
7430  *  Index to aso flow meter to be free.
7431  *
7432  * @return
7433  *   0 on success.
7434  */
7435 void
7436 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7437 {
7438 	const struct mlx5_flow_driver_ops *fops;
7439 
7440 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7441 	fops->free_meter(dev, mtr_idx);
7442 }
7443 
7444 /**
7445  * Allocate a counter.
7446  *
7447  * @param[in] dev
7448  *   Pointer to Ethernet device structure.
7449  *
7450  * @return
7451  *   Index to allocated counter  on success, 0 otherwise.
7452  */
7453 uint32_t
7454 mlx5_counter_alloc(struct rte_eth_dev *dev)
7455 {
7456 	const struct mlx5_flow_driver_ops *fops;
7457 	struct rte_flow_attr attr = { .transfer = 0 };
7458 
7459 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7460 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7461 		return fops->counter_alloc(dev);
7462 	}
7463 	DRV_LOG(ERR,
7464 		"port %u counter allocate is not supported.",
7465 		 dev->data->port_id);
7466 	return 0;
7467 }
7468 
7469 /**
7470  * Free a counter.
7471  *
7472  * @param[in] dev
7473  *   Pointer to Ethernet device structure.
7474  * @param[in] cnt
7475  *   Index to counter to be free.
7476  */
7477 void
7478 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7479 {
7480 	const struct mlx5_flow_driver_ops *fops;
7481 	struct rte_flow_attr attr = { .transfer = 0 };
7482 
7483 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7484 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7485 		fops->counter_free(dev, cnt);
7486 		return;
7487 	}
7488 	DRV_LOG(ERR,
7489 		"port %u counter free is not supported.",
7490 		 dev->data->port_id);
7491 }
7492 
7493 /**
7494  * Query counter statistics.
7495  *
7496  * @param[in] dev
7497  *   Pointer to Ethernet device structure.
7498  * @param[in] cnt
7499  *   Index to counter to query.
7500  * @param[in] clear
7501  *   Set to clear counter statistics.
7502  * @param[out] pkts
7503  *   The counter hits packets number to save.
7504  * @param[out] bytes
7505  *   The counter hits bytes number to save.
7506  *
7507  * @return
7508  *   0 on success, a negative errno value otherwise.
7509  */
7510 int
7511 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7512 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7513 {
7514 	const struct mlx5_flow_driver_ops *fops;
7515 	struct rte_flow_attr attr = { .transfer = 0 };
7516 
7517 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7518 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7519 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7520 	}
7521 	DRV_LOG(ERR,
7522 		"port %u counter query is not supported.",
7523 		 dev->data->port_id);
7524 	return -ENOTSUP;
7525 }
7526 
7527 /**
7528  * Allocate a new memory for the counter values wrapped by all the needed
7529  * management.
7530  *
7531  * @param[in] sh
7532  *   Pointer to mlx5_dev_ctx_shared object.
7533  *
7534  * @return
7535  *   0 on success, a negative errno value otherwise.
7536  */
7537 static int
7538 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7539 {
7540 	struct mlx5_devx_mkey_attr mkey_attr;
7541 	struct mlx5_counter_stats_mem_mng *mem_mng;
7542 	volatile struct flow_counter_stats *raw_data;
7543 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7544 	int size = (sizeof(struct flow_counter_stats) *
7545 			MLX5_COUNTERS_PER_POOL +
7546 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7547 			sizeof(struct mlx5_counter_stats_mem_mng);
7548 	size_t pgsize = rte_mem_page_size();
7549 	uint8_t *mem;
7550 	int i;
7551 
7552 	if (pgsize == (size_t)-1) {
7553 		DRV_LOG(ERR, "Failed to get mem page size");
7554 		rte_errno = ENOMEM;
7555 		return -ENOMEM;
7556 	}
7557 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7558 	if (!mem) {
7559 		rte_errno = ENOMEM;
7560 		return -ENOMEM;
7561 	}
7562 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7563 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7564 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7565 						 IBV_ACCESS_LOCAL_WRITE);
7566 	if (!mem_mng->umem) {
7567 		rte_errno = errno;
7568 		mlx5_free(mem);
7569 		return -rte_errno;
7570 	}
7571 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7572 	mkey_attr.addr = (uintptr_t)mem;
7573 	mkey_attr.size = size;
7574 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7575 	mkey_attr.pd = sh->pdn;
7576 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7577 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7578 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7579 	if (!mem_mng->dm) {
7580 		mlx5_os_umem_dereg(mem_mng->umem);
7581 		rte_errno = errno;
7582 		mlx5_free(mem);
7583 		return -rte_errno;
7584 	}
7585 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7586 	raw_data = (volatile struct flow_counter_stats *)mem;
7587 	for (i = 0; i < raws_n; ++i) {
7588 		mem_mng->raws[i].mem_mng = mem_mng;
7589 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7590 	}
7591 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7592 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7593 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7594 				 next);
7595 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7596 	sh->cmng.mem_mng = mem_mng;
7597 	return 0;
7598 }
7599 
7600 /**
7601  * Set the statistic memory to the new counter pool.
7602  *
7603  * @param[in] sh
7604  *   Pointer to mlx5_dev_ctx_shared object.
7605  * @param[in] pool
7606  *   Pointer to the pool to set the statistic memory.
7607  *
7608  * @return
7609  *   0 on success, a negative errno value otherwise.
7610  */
7611 static int
7612 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7613 			       struct mlx5_flow_counter_pool *pool)
7614 {
7615 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7616 	/* Resize statistic memory once used out. */
7617 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7618 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7619 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7620 		return -1;
7621 	}
7622 	rte_spinlock_lock(&pool->sl);
7623 	pool->raw = cmng->mem_mng->raws + pool->index %
7624 		    MLX5_CNT_CONTAINER_RESIZE;
7625 	rte_spinlock_unlock(&pool->sl);
7626 	pool->raw_hw = NULL;
7627 	return 0;
7628 }
7629 
7630 #define MLX5_POOL_QUERY_FREQ_US 1000000
7631 
7632 /**
7633  * Set the periodic procedure for triggering asynchronous batch queries for all
7634  * the counter pools.
7635  *
7636  * @param[in] sh
7637  *   Pointer to mlx5_dev_ctx_shared object.
7638  */
7639 void
7640 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7641 {
7642 	uint32_t pools_n, us;
7643 
7644 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7645 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7646 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7647 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7648 		sh->cmng.query_thread_on = 0;
7649 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7650 	} else {
7651 		sh->cmng.query_thread_on = 1;
7652 	}
7653 }
7654 
7655 /**
7656  * The periodic procedure for triggering asynchronous batch queries for all the
7657  * counter pools. This function is probably called by the host thread.
7658  *
7659  * @param[in] arg
7660  *   The parameter for the alarm process.
7661  */
7662 void
7663 mlx5_flow_query_alarm(void *arg)
7664 {
7665 	struct mlx5_dev_ctx_shared *sh = arg;
7666 	int ret;
7667 	uint16_t pool_index = sh->cmng.pool_index;
7668 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7669 	struct mlx5_flow_counter_pool *pool;
7670 	uint16_t n_valid;
7671 
7672 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7673 		goto set_alarm;
7674 	rte_spinlock_lock(&cmng->pool_update_sl);
7675 	pool = cmng->pools[pool_index];
7676 	n_valid = cmng->n_valid;
7677 	rte_spinlock_unlock(&cmng->pool_update_sl);
7678 	/* Set the statistic memory to the new created pool. */
7679 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7680 		goto set_alarm;
7681 	if (pool->raw_hw)
7682 		/* There is a pool query in progress. */
7683 		goto set_alarm;
7684 	pool->raw_hw =
7685 		LIST_FIRST(&sh->cmng.free_stat_raws);
7686 	if (!pool->raw_hw)
7687 		/* No free counter statistics raw memory. */
7688 		goto set_alarm;
7689 	/*
7690 	 * Identify the counters released between query trigger and query
7691 	 * handle more efficiently. The counter released in this gap period
7692 	 * should wait for a new round of query as the new arrived packets
7693 	 * will not be taken into account.
7694 	 */
7695 	pool->query_gen++;
7696 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7697 					       MLX5_COUNTERS_PER_POOL,
7698 					       NULL, NULL,
7699 					       pool->raw_hw->mem_mng->dm->id,
7700 					       (void *)(uintptr_t)
7701 					       pool->raw_hw->data,
7702 					       sh->devx_comp,
7703 					       (uint64_t)(uintptr_t)pool);
7704 	if (ret) {
7705 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7706 			" %d", pool->min_dcs->id);
7707 		pool->raw_hw = NULL;
7708 		goto set_alarm;
7709 	}
7710 	LIST_REMOVE(pool->raw_hw, next);
7711 	sh->cmng.pending_queries++;
7712 	pool_index++;
7713 	if (pool_index >= n_valid)
7714 		pool_index = 0;
7715 set_alarm:
7716 	sh->cmng.pool_index = pool_index;
7717 	mlx5_set_query_alarm(sh);
7718 }
7719 
7720 /**
7721  * Check and callback event for new aged flow in the counter pool
7722  *
7723  * @param[in] sh
7724  *   Pointer to mlx5_dev_ctx_shared object.
7725  * @param[in] pool
7726  *   Pointer to Current counter pool.
7727  */
7728 static void
7729 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7730 		   struct mlx5_flow_counter_pool *pool)
7731 {
7732 	struct mlx5_priv *priv;
7733 	struct mlx5_flow_counter *cnt;
7734 	struct mlx5_age_info *age_info;
7735 	struct mlx5_age_param *age_param;
7736 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7737 	struct mlx5_counter_stats_raw *prev = pool->raw;
7738 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7739 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7740 	uint16_t expected = AGE_CANDIDATE;
7741 	uint32_t i;
7742 
7743 	pool->time_of_last_age_check = curr_time;
7744 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7745 		cnt = MLX5_POOL_GET_CNT(pool, i);
7746 		age_param = MLX5_CNT_TO_AGE(cnt);
7747 		if (__atomic_load_n(&age_param->state,
7748 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7749 			continue;
7750 		if (cur->data[i].hits != prev->data[i].hits) {
7751 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7752 					 __ATOMIC_RELAXED);
7753 			continue;
7754 		}
7755 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7756 				       time_delta,
7757 				       __ATOMIC_RELAXED) <= age_param->timeout)
7758 			continue;
7759 		/**
7760 		 * Hold the lock first, or if between the
7761 		 * state AGE_TMOUT and tailq operation the
7762 		 * release happened, the release procedure
7763 		 * may delete a non-existent tailq node.
7764 		 */
7765 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7766 		age_info = GET_PORT_AGE_INFO(priv);
7767 		rte_spinlock_lock(&age_info->aged_sl);
7768 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7769 						AGE_TMOUT, false,
7770 						__ATOMIC_RELAXED,
7771 						__ATOMIC_RELAXED)) {
7772 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7773 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7774 		}
7775 		rte_spinlock_unlock(&age_info->aged_sl);
7776 	}
7777 	mlx5_age_event_prepare(sh);
7778 }
7779 
7780 /**
7781  * Handler for the HW respond about ready values from an asynchronous batch
7782  * query. This function is probably called by the host thread.
7783  *
7784  * @param[in] sh
7785  *   The pointer to the shared device context.
7786  * @param[in] async_id
7787  *   The Devx async ID.
7788  * @param[in] status
7789  *   The status of the completion.
7790  */
7791 void
7792 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7793 				  uint64_t async_id, int status)
7794 {
7795 	struct mlx5_flow_counter_pool *pool =
7796 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7797 	struct mlx5_counter_stats_raw *raw_to_free;
7798 	uint8_t query_gen = pool->query_gen ^ 1;
7799 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7800 	enum mlx5_counter_type cnt_type =
7801 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7802 				MLX5_COUNTER_TYPE_ORIGIN;
7803 
7804 	if (unlikely(status)) {
7805 		raw_to_free = pool->raw_hw;
7806 	} else {
7807 		raw_to_free = pool->raw;
7808 		if (pool->is_aged)
7809 			mlx5_flow_aging_check(sh, pool);
7810 		rte_spinlock_lock(&pool->sl);
7811 		pool->raw = pool->raw_hw;
7812 		rte_spinlock_unlock(&pool->sl);
7813 		/* Be sure the new raw counters data is updated in memory. */
7814 		rte_io_wmb();
7815 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7816 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7817 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7818 				     &pool->counters[query_gen], next);
7819 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7820 		}
7821 	}
7822 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7823 	pool->raw_hw = NULL;
7824 	sh->cmng.pending_queries--;
7825 }
7826 
7827 static int
7828 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7829 		    const struct flow_grp_info *grp_info,
7830 		    struct rte_flow_error *error)
7831 {
7832 	if (grp_info->transfer && grp_info->external &&
7833 	    grp_info->fdb_def_rule) {
7834 		if (group == UINT32_MAX)
7835 			return rte_flow_error_set
7836 						(error, EINVAL,
7837 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7838 						 NULL,
7839 						 "group index not supported");
7840 		*table = group + 1;
7841 	} else {
7842 		*table = group;
7843 	}
7844 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7845 	return 0;
7846 }
7847 
7848 /**
7849  * Translate the rte_flow group index to HW table value.
7850  *
7851  * If tunnel offload is disabled, all group ids converted to flow table
7852  * id using the standard method.
7853  * If tunnel offload is enabled, group id can be converted using the
7854  * standard or tunnel conversion method. Group conversion method
7855  * selection depends on flags in `grp_info` parameter:
7856  * - Internal (grp_info.external == 0) groups conversion uses the
7857  *   standard method.
7858  * - Group ids in JUMP action converted with the tunnel conversion.
7859  * - Group id in rule attribute conversion depends on a rule type and
7860  *   group id value:
7861  *   ** non zero group attributes converted with the tunnel method
7862  *   ** zero group attribute in non-tunnel rule is converted using the
7863  *      standard method - there's only one root table
7864  *   ** zero group attribute in steer tunnel rule is converted with the
7865  *      standard method - single root table
7866  *   ** zero group attribute in match tunnel rule is a special OvS
7867  *      case: that value is used for portability reasons. That group
7868  *      id is converted with the tunnel conversion method.
7869  *
7870  * @param[in] dev
7871  *   Port device
7872  * @param[in] tunnel
7873  *   PMD tunnel offload object
7874  * @param[in] group
7875  *   rte_flow group index value.
7876  * @param[out] table
7877  *   HW table value.
7878  * @param[in] grp_info
7879  *   flags used for conversion
7880  * @param[out] error
7881  *   Pointer to error structure.
7882  *
7883  * @return
7884  *   0 on success, a negative errno value otherwise and rte_errno is set.
7885  */
7886 int
7887 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7888 			 const struct mlx5_flow_tunnel *tunnel,
7889 			 uint32_t group, uint32_t *table,
7890 			 const struct flow_grp_info *grp_info,
7891 			 struct rte_flow_error *error)
7892 {
7893 	int ret;
7894 	bool standard_translation;
7895 
7896 	if (!grp_info->skip_scale && grp_info->external &&
7897 	    group < MLX5_MAX_TABLES_EXTERNAL)
7898 		group *= MLX5_FLOW_TABLE_FACTOR;
7899 	if (is_tunnel_offload_active(dev)) {
7900 		standard_translation = !grp_info->external ||
7901 					grp_info->std_tbl_fix;
7902 	} else {
7903 		standard_translation = true;
7904 	}
7905 	DRV_LOG(DEBUG,
7906 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7907 		dev->data->port_id, group, grp_info->transfer,
7908 		grp_info->external, grp_info->fdb_def_rule,
7909 		standard_translation ? "STANDARD" : "TUNNEL");
7910 	if (standard_translation)
7911 		ret = flow_group_to_table(dev->data->port_id, group, table,
7912 					  grp_info, error);
7913 	else
7914 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7915 						      table, error);
7916 
7917 	return ret;
7918 }
7919 
7920 /**
7921  * Discover availability of metadata reg_c's.
7922  *
7923  * Iteratively use test flows to check availability.
7924  *
7925  * @param[in] dev
7926  *   Pointer to the Ethernet device structure.
7927  *
7928  * @return
7929  *   0 on success, a negative errno value otherwise and rte_errno is set.
7930  */
7931 int
7932 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7933 {
7934 	struct mlx5_priv *priv = dev->data->dev_private;
7935 	struct mlx5_dev_config *config = &priv->config;
7936 	enum modify_reg idx;
7937 	int n = 0;
7938 
7939 	/* reg_c[0] and reg_c[1] are reserved. */
7940 	config->flow_mreg_c[n++] = REG_C_0;
7941 	config->flow_mreg_c[n++] = REG_C_1;
7942 	/* Discover availability of other reg_c's. */
7943 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7944 		struct rte_flow_attr attr = {
7945 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7946 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7947 			.ingress = 1,
7948 		};
7949 		struct rte_flow_item items[] = {
7950 			[0] = {
7951 				.type = RTE_FLOW_ITEM_TYPE_END,
7952 			},
7953 		};
7954 		struct rte_flow_action actions[] = {
7955 			[0] = {
7956 				.type = (enum rte_flow_action_type)
7957 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7958 				.conf = &(struct mlx5_flow_action_copy_mreg){
7959 					.src = REG_C_1,
7960 					.dst = idx,
7961 				},
7962 			},
7963 			[1] = {
7964 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7965 				.conf = &(struct rte_flow_action_jump){
7966 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7967 				},
7968 			},
7969 			[2] = {
7970 				.type = RTE_FLOW_ACTION_TYPE_END,
7971 			},
7972 		};
7973 		uint32_t flow_idx;
7974 		struct rte_flow *flow;
7975 		struct rte_flow_error error;
7976 
7977 		if (!config->dv_flow_en)
7978 			break;
7979 		/* Create internal flow, validation skips copy action. */
7980 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
7981 					items, actions, false, &error);
7982 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7983 				      flow_idx);
7984 		if (!flow)
7985 			continue;
7986 		config->flow_mreg_c[n++] = idx;
7987 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
7988 	}
7989 	for (; n < MLX5_MREG_C_NUM; ++n)
7990 		config->flow_mreg_c[n] = REG_NON;
7991 	return 0;
7992 }
7993 
7994 int
7995 save_dump_file(const uint8_t *data, uint32_t size,
7996 	uint32_t type, uint32_t id, void *arg, FILE *file)
7997 {
7998 	char line[BUF_SIZE];
7999 	uint32_t out = 0;
8000 	uint32_t k;
8001 	uint32_t actions_num;
8002 	struct rte_flow_query_count *count;
8003 
8004 	memset(line, 0, BUF_SIZE);
8005 	switch (type) {
8006 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8007 		actions_num = *(uint32_t *)(arg);
8008 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
8009 				type, id, actions_num);
8010 		break;
8011 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8012 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
8013 				type, id);
8014 		break;
8015 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8016 		count = (struct rte_flow_query_count *)arg;
8017 		fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
8018 				id, count->hits, count->bytes);
8019 		return 0;
8020 	default:
8021 		return -1;
8022 	}
8023 
8024 	for (k = 0; k < size; k++) {
8025 		/* Make sure we do not overrun the line buffer length. */
8026 		if (out >= BUF_SIZE - 4) {
8027 			line[out] = '\0';
8028 			break;
8029 		}
8030 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8031 				(data[k]) & 0xff);
8032 	}
8033 	fprintf(file, "%s\n", line);
8034 	return 0;
8035 }
8036 
8037 int
8038 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8039 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8040 {
8041 	struct rte_flow_action action[2];
8042 	enum mlx5_flow_drv_type ftype;
8043 	const struct mlx5_flow_driver_ops *fops;
8044 
8045 	if (!flow) {
8046 		return rte_flow_error_set(error, ENOENT,
8047 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8048 				NULL,
8049 				"invalid flow handle");
8050 	}
8051 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8052 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8053 	if (flow->counter) {
8054 		memset(count, 0, sizeof(struct rte_flow_query_count));
8055 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8056 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8057 						ftype < MLX5_FLOW_TYPE_MAX);
8058 		fops = flow_get_drv_ops(ftype);
8059 		return fops->query(dev, flow, action, count, error);
8060 	}
8061 	return -1;
8062 }
8063 
8064 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8065 /**
8066  * Dump flow ipool data to file
8067  *
8068  * @param[in] dev
8069  *   The pointer to Ethernet device.
8070  * @param[in] file
8071  *   A pointer to a file for output.
8072  * @param[out] error
8073  *   Perform verbose error reporting if not NULL. PMDs initialize this
8074  *   structure in case of error only.
8075  * @return
8076  *   0 on success, a negative value otherwise.
8077  */
8078 int
8079 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8080 	struct rte_flow *flow, FILE *file,
8081 	struct rte_flow_error *error)
8082 {
8083 	struct mlx5_priv *priv = dev->data->dev_private;
8084 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8085 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8086 	uint32_t handle_idx;
8087 	struct mlx5_flow_handle *dh;
8088 	struct rte_flow_query_count count;
8089 	uint32_t actions_num;
8090 	const uint8_t *data;
8091 	size_t size;
8092 	uint32_t id;
8093 	uint32_t type;
8094 
8095 	if (!flow) {
8096 		return rte_flow_error_set(error, ENOENT,
8097 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8098 			NULL,
8099 			"invalid flow handle");
8100 	}
8101 	handle_idx = flow->dev_handles;
8102 	while (handle_idx) {
8103 		dh = mlx5_ipool_get(priv->sh->ipool
8104 			[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8105 		if (!dh)
8106 			continue;
8107 		handle_idx = dh->next.next;
8108 		id = (uint32_t)(uintptr_t)dh->drv_flow;
8109 
8110 		/* query counter */
8111 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8112 		if (!mlx5_flow_query_counter(dev, flow, &count, error))
8113 			save_dump_file(NULL, 0, type,
8114 					id, (void *)&count, file);
8115 
8116 		/* Get modify_hdr and encap_decap buf from ipools. */
8117 		encap_decap = NULL;
8118 		modify_hdr = dh->dvh.modify_hdr;
8119 
8120 		if (dh->dvh.rix_encap_decap) {
8121 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8122 						[MLX5_IPOOL_DECAP_ENCAP],
8123 						dh->dvh.rix_encap_decap);
8124 		}
8125 		if (modify_hdr) {
8126 			data = (const uint8_t *)modify_hdr->actions;
8127 			size = (size_t)(modify_hdr->actions_num) * 8;
8128 			actions_num = modify_hdr->actions_num;
8129 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8130 			save_dump_file(data, size, type, id,
8131 					(void *)(&actions_num), file);
8132 		}
8133 		if (encap_decap) {
8134 			data = encap_decap->buf;
8135 			size = encap_decap->size;
8136 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8137 			save_dump_file(data, size, type,
8138 						id, NULL, file);
8139 		}
8140 	}
8141 	return 0;
8142 }
8143 #endif
8144 
8145 /**
8146  * Dump flow raw hw data to file
8147  *
8148  * @param[in] dev
8149  *    The pointer to Ethernet device.
8150  * @param[in] file
8151  *   A pointer to a file for output.
8152  * @param[out] error
8153  *   Perform verbose error reporting if not NULL. PMDs initialize this
8154  *   structure in case of error only.
8155  * @return
8156  *   0 on success, a nagative value otherwise.
8157  */
8158 int
8159 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8160 		   FILE *file,
8161 		   struct rte_flow_error *error __rte_unused)
8162 {
8163 	struct mlx5_priv *priv = dev->data->dev_private;
8164 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8165 	uint32_t handle_idx;
8166 	int ret;
8167 	struct mlx5_flow_handle *dh;
8168 	struct rte_flow *flow;
8169 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8170 	uint32_t idx;
8171 #endif
8172 
8173 	if (!priv->config.dv_flow_en) {
8174 		if (fputs("device dv flow disabled\n", file) <= 0)
8175 			return -errno;
8176 		return -ENOTSUP;
8177 	}
8178 
8179 	/* dump all */
8180 	if (!flow_idx) {
8181 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8182 		MLX5_IPOOL_FOREACH(priv->flows[MLX5_FLOW_TYPE_GEN], idx, flow)
8183 			mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8184 #endif
8185 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8186 					sh->rx_domain,
8187 					sh->tx_domain, file);
8188 	}
8189 	/* dump one */
8190 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8191 			(uintptr_t)(void *)flow_idx);
8192 	if (!flow)
8193 		return -ENOENT;
8194 
8195 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8196 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8197 #endif
8198 	handle_idx = flow->dev_handles;
8199 	while (handle_idx) {
8200 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8201 				handle_idx);
8202 		if (!dh)
8203 			return -ENOENT;
8204 		if (dh->drv_flow) {
8205 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8206 					file);
8207 			if (ret)
8208 				return -ENOENT;
8209 		}
8210 		handle_idx = dh->next.next;
8211 	}
8212 	return 0;
8213 }
8214 
8215 /**
8216  * Get aged-out flows.
8217  *
8218  * @param[in] dev
8219  *   Pointer to the Ethernet device structure.
8220  * @param[in] context
8221  *   The address of an array of pointers to the aged-out flows contexts.
8222  * @param[in] nb_countexts
8223  *   The length of context array pointers.
8224  * @param[out] error
8225  *   Perform verbose error reporting if not NULL. Initialized in case of
8226  *   error only.
8227  *
8228  * @return
8229  *   how many contexts get in success, otherwise negative errno value.
8230  *   if nb_contexts is 0, return the amount of all aged contexts.
8231  *   if nb_contexts is not 0 , return the amount of aged flows reported
8232  *   in the context array.
8233  */
8234 int
8235 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8236 			uint32_t nb_contexts, struct rte_flow_error *error)
8237 {
8238 	const struct mlx5_flow_driver_ops *fops;
8239 	struct rte_flow_attr attr = { .transfer = 0 };
8240 
8241 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8242 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8243 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8244 						    error);
8245 	}
8246 	DRV_LOG(ERR,
8247 		"port %u get aged flows is not supported.",
8248 		 dev->data->port_id);
8249 	return -ENOTSUP;
8250 }
8251 
8252 /* Wrapper for driver action_validate op callback */
8253 static int
8254 flow_drv_action_validate(struct rte_eth_dev *dev,
8255 			 const struct rte_flow_indir_action_conf *conf,
8256 			 const struct rte_flow_action *action,
8257 			 const struct mlx5_flow_driver_ops *fops,
8258 			 struct rte_flow_error *error)
8259 {
8260 	static const char err_msg[] = "indirect action validation unsupported";
8261 
8262 	if (!fops->action_validate) {
8263 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8264 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8265 				   NULL, err_msg);
8266 		return -rte_errno;
8267 	}
8268 	return fops->action_validate(dev, conf, action, error);
8269 }
8270 
8271 /**
8272  * Destroys the shared action by handle.
8273  *
8274  * @param dev
8275  *   Pointer to Ethernet device structure.
8276  * @param[in] handle
8277  *   Handle for the indirect action object to be destroyed.
8278  * @param[out] error
8279  *   Perform verbose error reporting if not NULL. PMDs initialize this
8280  *   structure in case of error only.
8281  *
8282  * @return
8283  *   0 on success, a negative errno value otherwise and rte_errno is set.
8284  *
8285  * @note: wrapper for driver action_create op callback.
8286  */
8287 static int
8288 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8289 			   struct rte_flow_action_handle *handle,
8290 			   struct rte_flow_error *error)
8291 {
8292 	static const char err_msg[] = "indirect action destruction unsupported";
8293 	struct rte_flow_attr attr = { .transfer = 0 };
8294 	const struct mlx5_flow_driver_ops *fops =
8295 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8296 
8297 	if (!fops->action_destroy) {
8298 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8299 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8300 				   NULL, err_msg);
8301 		return -rte_errno;
8302 	}
8303 	return fops->action_destroy(dev, handle, error);
8304 }
8305 
8306 /* Wrapper for driver action_destroy op callback */
8307 static int
8308 flow_drv_action_update(struct rte_eth_dev *dev,
8309 		       struct rte_flow_action_handle *handle,
8310 		       const void *update,
8311 		       const struct mlx5_flow_driver_ops *fops,
8312 		       struct rte_flow_error *error)
8313 {
8314 	static const char err_msg[] = "indirect action update unsupported";
8315 
8316 	if (!fops->action_update) {
8317 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8318 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8319 				   NULL, err_msg);
8320 		return -rte_errno;
8321 	}
8322 	return fops->action_update(dev, handle, update, error);
8323 }
8324 
8325 /* Wrapper for driver action_destroy op callback */
8326 static int
8327 flow_drv_action_query(struct rte_eth_dev *dev,
8328 		      const struct rte_flow_action_handle *handle,
8329 		      void *data,
8330 		      const struct mlx5_flow_driver_ops *fops,
8331 		      struct rte_flow_error *error)
8332 {
8333 	static const char err_msg[] = "indirect action query unsupported";
8334 
8335 	if (!fops->action_query) {
8336 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8337 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8338 				   NULL, err_msg);
8339 		return -rte_errno;
8340 	}
8341 	return fops->action_query(dev, handle, data, error);
8342 }
8343 
8344 /**
8345  * Create indirect action for reuse in multiple flow rules.
8346  *
8347  * @param dev
8348  *   Pointer to Ethernet device structure.
8349  * @param conf
8350  *   Pointer to indirect action object configuration.
8351  * @param[in] action
8352  *   Action configuration for indirect action object creation.
8353  * @param[out] error
8354  *   Perform verbose error reporting if not NULL. PMDs initialize this
8355  *   structure in case of error only.
8356  * @return
8357  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8358  */
8359 static struct rte_flow_action_handle *
8360 mlx5_action_handle_create(struct rte_eth_dev *dev,
8361 			  const struct rte_flow_indir_action_conf *conf,
8362 			  const struct rte_flow_action *action,
8363 			  struct rte_flow_error *error)
8364 {
8365 	static const char err_msg[] = "indirect action creation unsupported";
8366 	struct rte_flow_attr attr = { .transfer = 0 };
8367 	const struct mlx5_flow_driver_ops *fops =
8368 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8369 
8370 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8371 		return NULL;
8372 	if (!fops->action_create) {
8373 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8374 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8375 				   NULL, err_msg);
8376 		return NULL;
8377 	}
8378 	return fops->action_create(dev, conf, action, error);
8379 }
8380 
8381 /**
8382  * Updates inplace the indirect action configuration pointed by *handle*
8383  * with the configuration provided as *update* argument.
8384  * The update of the indirect action configuration effects all flow rules
8385  * reusing the action via handle.
8386  *
8387  * @param dev
8388  *   Pointer to Ethernet device structure.
8389  * @param[in] handle
8390  *   Handle for the indirect action to be updated.
8391  * @param[in] update
8392  *   Action specification used to modify the action pointed by handle.
8393  *   *update* could be of same type with the action pointed by the *handle*
8394  *   handle argument, or some other structures like a wrapper, depending on
8395  *   the indirect action type.
8396  * @param[out] error
8397  *   Perform verbose error reporting if not NULL. PMDs initialize this
8398  *   structure in case of error only.
8399  *
8400  * @return
8401  *   0 on success, a negative errno value otherwise and rte_errno is set.
8402  */
8403 static int
8404 mlx5_action_handle_update(struct rte_eth_dev *dev,
8405 		struct rte_flow_action_handle *handle,
8406 		const void *update,
8407 		struct rte_flow_error *error)
8408 {
8409 	struct rte_flow_attr attr = { .transfer = 0 };
8410 	const struct mlx5_flow_driver_ops *fops =
8411 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8412 	int ret;
8413 
8414 	ret = flow_drv_action_validate(dev, NULL,
8415 			(const struct rte_flow_action *)update, fops, error);
8416 	if (ret)
8417 		return ret;
8418 	return flow_drv_action_update(dev, handle, update, fops,
8419 				      error);
8420 }
8421 
8422 /**
8423  * Query the indirect action by handle.
8424  *
8425  * This function allows retrieving action-specific data such as counters.
8426  * Data is gathered by special action which may be present/referenced in
8427  * more than one flow rule definition.
8428  *
8429  * see @RTE_FLOW_ACTION_TYPE_COUNT
8430  *
8431  * @param dev
8432  *   Pointer to Ethernet device structure.
8433  * @param[in] handle
8434  *   Handle for the indirect action to query.
8435  * @param[in, out] data
8436  *   Pointer to storage for the associated query data type.
8437  * @param[out] error
8438  *   Perform verbose error reporting if not NULL. PMDs initialize this
8439  *   structure in case of error only.
8440  *
8441  * @return
8442  *   0 on success, a negative errno value otherwise and rte_errno is set.
8443  */
8444 static int
8445 mlx5_action_handle_query(struct rte_eth_dev *dev,
8446 			 const struct rte_flow_action_handle *handle,
8447 			 void *data,
8448 			 struct rte_flow_error *error)
8449 {
8450 	struct rte_flow_attr attr = { .transfer = 0 };
8451 	const struct mlx5_flow_driver_ops *fops =
8452 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8453 
8454 	return flow_drv_action_query(dev, handle, data, fops, error);
8455 }
8456 
8457 /**
8458  * Destroy all indirect actions (shared RSS).
8459  *
8460  * @param dev
8461  *   Pointer to Ethernet device.
8462  *
8463  * @return
8464  *   0 on success, a negative errno value otherwise and rte_errno is set.
8465  */
8466 int
8467 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8468 {
8469 	struct rte_flow_error error;
8470 	struct mlx5_priv *priv = dev->data->dev_private;
8471 	struct mlx5_shared_action_rss *shared_rss;
8472 	int ret = 0;
8473 	uint32_t idx;
8474 
8475 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8476 		      priv->rss_shared_actions, idx, shared_rss, next) {
8477 		ret |= mlx5_action_handle_destroy(dev,
8478 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8479 	}
8480 	return ret;
8481 }
8482 
8483 #ifndef HAVE_MLX5DV_DR
8484 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8485 #else
8486 #define MLX5_DOMAIN_SYNC_FLOW \
8487 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8488 #endif
8489 
8490 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8491 {
8492 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8493 	const struct mlx5_flow_driver_ops *fops;
8494 	int ret;
8495 	struct rte_flow_attr attr = { .transfer = 0 };
8496 
8497 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8498 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8499 	if (ret > 0)
8500 		ret = -ret;
8501 	return ret;
8502 }
8503 
8504 const struct mlx5_flow_tunnel *
8505 mlx5_get_tof(const struct rte_flow_item *item,
8506 	     const struct rte_flow_action *action,
8507 	     enum mlx5_tof_rule_type *rule_type)
8508 {
8509 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8510 		if (item->type == (typeof(item->type))
8511 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8512 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8513 			return flow_items_to_tunnel(item);
8514 		}
8515 	}
8516 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8517 		if (action->type == (typeof(action->type))
8518 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8519 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8520 			return flow_actions_to_tunnel(action);
8521 		}
8522 	}
8523 	return NULL;
8524 }
8525 
8526 /**
8527  * tunnel offload functionalilty is defined for DV environment only
8528  */
8529 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8530 __extension__
8531 union tunnel_offload_mark {
8532 	uint32_t val;
8533 	struct {
8534 		uint32_t app_reserve:8;
8535 		uint32_t table_id:15;
8536 		uint32_t transfer:1;
8537 		uint32_t _unused_:8;
8538 	};
8539 };
8540 
8541 static bool
8542 mlx5_access_tunnel_offload_db
8543 	(struct rte_eth_dev *dev,
8544 	 bool (*match)(struct rte_eth_dev *,
8545 		       struct mlx5_flow_tunnel *, const void *),
8546 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8547 	 void (*miss)(struct rte_eth_dev *, void *),
8548 	 void *ctx, bool lock_op);
8549 
8550 static int
8551 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8552 			     struct rte_flow *flow,
8553 			     const struct rte_flow_attr *attr,
8554 			     const struct rte_flow_action *app_actions,
8555 			     uint32_t flow_idx,
8556 			     const struct mlx5_flow_tunnel *tunnel,
8557 			     struct tunnel_default_miss_ctx *ctx,
8558 			     struct rte_flow_error *error)
8559 {
8560 	struct mlx5_priv *priv = dev->data->dev_private;
8561 	struct mlx5_flow *dev_flow;
8562 	struct rte_flow_attr miss_attr = *attr;
8563 	const struct rte_flow_item miss_items[2] = {
8564 		{
8565 			.type = RTE_FLOW_ITEM_TYPE_ETH,
8566 			.spec = NULL,
8567 			.last = NULL,
8568 			.mask = NULL
8569 		},
8570 		{
8571 			.type = RTE_FLOW_ITEM_TYPE_END,
8572 			.spec = NULL,
8573 			.last = NULL,
8574 			.mask = NULL
8575 		}
8576 	};
8577 	union tunnel_offload_mark mark_id;
8578 	struct rte_flow_action_mark miss_mark;
8579 	struct rte_flow_action miss_actions[3] = {
8580 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8581 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
8582 	};
8583 	const struct rte_flow_action_jump *jump_data;
8584 	uint32_t i, flow_table = 0; /* prevent compilation warning */
8585 	struct flow_grp_info grp_info = {
8586 		.external = 1,
8587 		.transfer = attr->transfer,
8588 		.fdb_def_rule = !!priv->fdb_def_rule,
8589 		.std_tbl_fix = 0,
8590 	};
8591 	int ret;
8592 
8593 	if (!attr->transfer) {
8594 		uint32_t q_size;
8595 
8596 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8597 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8598 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8599 					 0, SOCKET_ID_ANY);
8600 		if (!ctx->queue)
8601 			return rte_flow_error_set
8602 				(error, ENOMEM,
8603 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8604 				NULL, "invalid default miss RSS");
8605 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8606 		ctx->action_rss.level = 0,
8607 		ctx->action_rss.types = priv->rss_conf.rss_hf,
8608 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8609 		ctx->action_rss.queue_num = priv->reta_idx_n,
8610 		ctx->action_rss.key = priv->rss_conf.rss_key,
8611 		ctx->action_rss.queue = ctx->queue;
8612 		if (!priv->reta_idx_n || !priv->rxqs_n)
8613 			return rte_flow_error_set
8614 				(error, EINVAL,
8615 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8616 				NULL, "invalid port configuration");
8617 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8618 			ctx->action_rss.types = 0;
8619 		for (i = 0; i != priv->reta_idx_n; ++i)
8620 			ctx->queue[i] = (*priv->reta_idx)[i];
8621 	} else {
8622 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8623 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8624 	}
8625 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8626 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8627 	jump_data = app_actions->conf;
8628 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8629 	miss_attr.group = jump_data->group;
8630 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8631 				       &flow_table, &grp_info, error);
8632 	if (ret)
8633 		return rte_flow_error_set(error, EINVAL,
8634 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8635 					  NULL, "invalid tunnel id");
8636 	mark_id.app_reserve = 0;
8637 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8638 	mark_id.transfer = !!attr->transfer;
8639 	mark_id._unused_ = 0;
8640 	miss_mark.id = mark_id.val;
8641 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8642 				    miss_items, miss_actions, flow_idx, error);
8643 	if (!dev_flow)
8644 		return -rte_errno;
8645 	dev_flow->flow = flow;
8646 	dev_flow->external = true;
8647 	dev_flow->tunnel = tunnel;
8648 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8649 	/* Subflow object was created, we must include one in the list. */
8650 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8651 		      dev_flow->handle, next);
8652 	DRV_LOG(DEBUG,
8653 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8654 		dev->data->port_id, tunnel->app_tunnel.type,
8655 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8656 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8657 				  miss_actions, error);
8658 	if (!ret)
8659 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8660 						  error);
8661 
8662 	return ret;
8663 }
8664 
8665 static const struct mlx5_flow_tbl_data_entry  *
8666 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8667 {
8668 	struct mlx5_priv *priv = dev->data->dev_private;
8669 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8670 	struct mlx5_list_entry *he;
8671 	union tunnel_offload_mark mbits = { .val = mark };
8672 	union mlx5_flow_tbl_key table_key = {
8673 		{
8674 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
8675 			.id = 0,
8676 			.reserved = 0,
8677 			.dummy = 0,
8678 			.is_fdb = !!mbits.transfer,
8679 			.is_egress = 0,
8680 		}
8681 	};
8682 	struct mlx5_flow_cb_ctx ctx = {
8683 		.data = &table_key.v64,
8684 	};
8685 
8686 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
8687 	return he ?
8688 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8689 }
8690 
8691 static void
8692 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
8693 				   struct mlx5_list_entry *entry)
8694 {
8695 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
8696 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8697 
8698 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8699 			tunnel_flow_tbl_to_id(tte->flow_table));
8700 	mlx5_free(tte);
8701 }
8702 
8703 static int
8704 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
8705 				  struct mlx5_list_entry *entry, void *cb_ctx)
8706 {
8707 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8708 	union tunnel_tbl_key tbl = {
8709 		.val = *(uint64_t *)(ctx->data),
8710 	};
8711 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8712 
8713 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8714 }
8715 
8716 static struct mlx5_list_entry *
8717 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
8718 {
8719 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
8720 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
8721 	struct tunnel_tbl_entry *tte;
8722 	union tunnel_tbl_key tbl = {
8723 		.val = *(uint64_t *)(ctx->data),
8724 	};
8725 
8726 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8727 			  sizeof(*tte), 0,
8728 			  SOCKET_ID_ANY);
8729 	if (!tte)
8730 		goto err;
8731 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8732 			  &tte->flow_table);
8733 	if (tte->flow_table >= MLX5_MAX_TABLES) {
8734 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8735 			tte->flow_table);
8736 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8737 				tte->flow_table);
8738 		goto err;
8739 	} else if (!tte->flow_table) {
8740 		goto err;
8741 	}
8742 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8743 	tte->tunnel_id = tbl.tunnel_id;
8744 	tte->group = tbl.group;
8745 	return &tte->hash;
8746 err:
8747 	if (tte)
8748 		mlx5_free(tte);
8749 	return NULL;
8750 }
8751 
8752 static struct mlx5_list_entry *
8753 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
8754 				  struct mlx5_list_entry *oentry,
8755 				  void *cb_ctx __rte_unused)
8756 {
8757 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
8758 						   0, SOCKET_ID_ANY);
8759 
8760 	if (!tte)
8761 		return NULL;
8762 	memcpy(tte, oentry, sizeof(*tte));
8763 	return &tte->hash;
8764 }
8765 
8766 static void
8767 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
8768 				       struct mlx5_list_entry *entry)
8769 {
8770 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8771 
8772 	mlx5_free(tte);
8773 }
8774 
8775 static uint32_t
8776 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8777 				const struct mlx5_flow_tunnel *tunnel,
8778 				uint32_t group, uint32_t *table,
8779 				struct rte_flow_error *error)
8780 {
8781 	struct mlx5_list_entry *he;
8782 	struct tunnel_tbl_entry *tte;
8783 	union tunnel_tbl_key key = {
8784 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8785 		.group = group
8786 	};
8787 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8788 	struct mlx5_hlist *group_hash;
8789 	struct mlx5_flow_cb_ctx ctx = {
8790 		.data = &key.val,
8791 	};
8792 
8793 	group_hash = tunnel ? tunnel->groups : thub->groups;
8794 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
8795 	if (!he)
8796 		return rte_flow_error_set(error, EINVAL,
8797 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8798 					  NULL,
8799 					  "tunnel group index not supported");
8800 	tte = container_of(he, typeof(*tte), hash);
8801 	*table = tte->flow_table;
8802 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8803 		dev->data->port_id, key.tunnel_id, group, *table);
8804 	return 0;
8805 }
8806 
8807 static void
8808 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8809 		      struct mlx5_flow_tunnel *tunnel)
8810 {
8811 	struct mlx5_priv *priv = dev->data->dev_private;
8812 	struct mlx5_indexed_pool *ipool;
8813 
8814 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8815 		dev->data->port_id, tunnel->tunnel_id);
8816 	LIST_REMOVE(tunnel, chain);
8817 	mlx5_hlist_destroy(tunnel->groups);
8818 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8819 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
8820 }
8821 
8822 static bool
8823 mlx5_access_tunnel_offload_db
8824 	(struct rte_eth_dev *dev,
8825 	 bool (*match)(struct rte_eth_dev *,
8826 		       struct mlx5_flow_tunnel *, const void *),
8827 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8828 	 void (*miss)(struct rte_eth_dev *, void *),
8829 	 void *ctx, bool lock_op)
8830 {
8831 	bool verdict = false;
8832 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8833 	struct mlx5_flow_tunnel *tunnel;
8834 
8835 	rte_spinlock_lock(&thub->sl);
8836 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8837 		verdict = match(dev, tunnel, (const void *)ctx);
8838 		if (verdict)
8839 			break;
8840 	}
8841 	if (!lock_op)
8842 		rte_spinlock_unlock(&thub->sl);
8843 	if (verdict && hit)
8844 		hit(dev, tunnel, ctx);
8845 	if (!verdict && miss)
8846 		miss(dev, ctx);
8847 	if (lock_op)
8848 		rte_spinlock_unlock(&thub->sl);
8849 
8850 	return verdict;
8851 }
8852 
8853 struct tunnel_db_find_tunnel_id_ctx {
8854 	uint32_t tunnel_id;
8855 	struct mlx5_flow_tunnel *tunnel;
8856 };
8857 
8858 static bool
8859 find_tunnel_id_match(struct rte_eth_dev *dev,
8860 		     struct mlx5_flow_tunnel *tunnel, const void *x)
8861 {
8862 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8863 
8864 	RTE_SET_USED(dev);
8865 	return tunnel->tunnel_id == ctx->tunnel_id;
8866 }
8867 
8868 static void
8869 find_tunnel_id_hit(struct rte_eth_dev *dev,
8870 		   struct mlx5_flow_tunnel *tunnel, void *x)
8871 {
8872 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8873 	RTE_SET_USED(dev);
8874 	ctx->tunnel = tunnel;
8875 }
8876 
8877 static struct mlx5_flow_tunnel *
8878 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8879 {
8880 	struct tunnel_db_find_tunnel_id_ctx ctx = {
8881 		.tunnel_id = id,
8882 	};
8883 
8884 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8885 				      find_tunnel_id_hit, NULL, &ctx, true);
8886 
8887 	return ctx.tunnel;
8888 }
8889 
8890 static struct mlx5_flow_tunnel *
8891 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8892 			  const struct rte_flow_tunnel *app_tunnel)
8893 {
8894 	struct mlx5_priv *priv = dev->data->dev_private;
8895 	struct mlx5_indexed_pool *ipool;
8896 	struct mlx5_flow_tunnel *tunnel;
8897 	uint32_t id;
8898 
8899 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8900 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
8901 	if (!tunnel)
8902 		return NULL;
8903 	if (id >= MLX5_MAX_TUNNELS) {
8904 		mlx5_ipool_free(ipool, id);
8905 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8906 		return NULL;
8907 	}
8908 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
8909 					   priv->sh,
8910 					   mlx5_flow_tunnel_grp2tbl_create_cb,
8911 					   mlx5_flow_tunnel_grp2tbl_match_cb,
8912 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
8913 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
8914 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
8915 	if (!tunnel->groups) {
8916 		mlx5_ipool_free(ipool, id);
8917 		return NULL;
8918 	}
8919 	/* initiate new PMD tunnel */
8920 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8921 	tunnel->tunnel_id = id;
8922 	tunnel->action.type = (typeof(tunnel->action.type))
8923 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8924 	tunnel->action.conf = tunnel;
8925 	tunnel->item.type = (typeof(tunnel->item.type))
8926 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8927 	tunnel->item.spec = tunnel;
8928 	tunnel->item.last = NULL;
8929 	tunnel->item.mask = NULL;
8930 
8931 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8932 		dev->data->port_id, tunnel->tunnel_id);
8933 
8934 	return tunnel;
8935 }
8936 
8937 struct tunnel_db_get_tunnel_ctx {
8938 	const struct rte_flow_tunnel *app_tunnel;
8939 	struct mlx5_flow_tunnel *tunnel;
8940 };
8941 
8942 static bool get_tunnel_match(struct rte_eth_dev *dev,
8943 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8944 {
8945 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
8946 
8947 	RTE_SET_USED(dev);
8948 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8949 		       sizeof(*ctx->app_tunnel));
8950 }
8951 
8952 static void get_tunnel_hit(struct rte_eth_dev *dev,
8953 			   struct mlx5_flow_tunnel *tunnel, void *x)
8954 {
8955 	/* called under tunnel spinlock protection */
8956 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8957 
8958 	RTE_SET_USED(dev);
8959 	tunnel->refctn++;
8960 	ctx->tunnel = tunnel;
8961 }
8962 
8963 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8964 {
8965 	/* called under tunnel spinlock protection */
8966 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8967 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8968 
8969 	rte_spinlock_unlock(&thub->sl);
8970 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8971 	rte_spinlock_lock(&thub->sl);
8972 	if (ctx->tunnel) {
8973 		ctx->tunnel->refctn = 1;
8974 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8975 	}
8976 }
8977 
8978 
8979 static int
8980 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8981 		     const struct rte_flow_tunnel *app_tunnel,
8982 		     struct mlx5_flow_tunnel **tunnel)
8983 {
8984 	struct tunnel_db_get_tunnel_ctx ctx = {
8985 		.app_tunnel = app_tunnel,
8986 	};
8987 
8988 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8989 				      get_tunnel_miss, &ctx, true);
8990 	*tunnel = ctx.tunnel;
8991 	return ctx.tunnel ? 0 : -ENOMEM;
8992 }
8993 
8994 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8995 {
8996 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8997 
8998 	if (!thub)
8999 		return;
9000 	if (!LIST_EMPTY(&thub->tunnels))
9001 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9002 	mlx5_hlist_destroy(thub->groups);
9003 	mlx5_free(thub);
9004 }
9005 
9006 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9007 {
9008 	int err;
9009 	struct mlx5_flow_tunnel_hub *thub;
9010 
9011 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9012 			   0, SOCKET_ID_ANY);
9013 	if (!thub)
9014 		return -ENOMEM;
9015 	LIST_INIT(&thub->tunnels);
9016 	rte_spinlock_init(&thub->sl);
9017 	thub->groups = mlx5_hlist_create("flow groups", 64,
9018 					 false, true, sh,
9019 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9020 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9021 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9022 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9023 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9024 	if (!thub->groups) {
9025 		err = -rte_errno;
9026 		goto err;
9027 	}
9028 	sh->tunnel_hub = thub;
9029 
9030 	return 0;
9031 
9032 err:
9033 	if (thub->groups)
9034 		mlx5_hlist_destroy(thub->groups);
9035 	if (thub)
9036 		mlx5_free(thub);
9037 	return err;
9038 }
9039 
9040 static inline bool
9041 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9042 			  struct rte_flow_tunnel *tunnel,
9043 			  const char *err_msg)
9044 {
9045 	err_msg = NULL;
9046 	if (!is_tunnel_offload_active(dev)) {
9047 		err_msg = "tunnel offload was not activated";
9048 		goto out;
9049 	} else if (!tunnel) {
9050 		err_msg = "no application tunnel";
9051 		goto out;
9052 	}
9053 
9054 	switch (tunnel->type) {
9055 	default:
9056 		err_msg = "unsupported tunnel type";
9057 		goto out;
9058 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9059 		break;
9060 	}
9061 
9062 out:
9063 	return !err_msg;
9064 }
9065 
9066 static int
9067 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9068 		    struct rte_flow_tunnel *app_tunnel,
9069 		    struct rte_flow_action **actions,
9070 		    uint32_t *num_of_actions,
9071 		    struct rte_flow_error *error)
9072 {
9073 	int ret;
9074 	struct mlx5_flow_tunnel *tunnel;
9075 	const char *err_msg = NULL;
9076 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9077 
9078 	if (!verdict)
9079 		return rte_flow_error_set(error, EINVAL,
9080 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9081 					  err_msg);
9082 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9083 	if (ret < 0) {
9084 		return rte_flow_error_set(error, ret,
9085 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9086 					  "failed to initialize pmd tunnel");
9087 	}
9088 	*actions = &tunnel->action;
9089 	*num_of_actions = 1;
9090 	return 0;
9091 }
9092 
9093 static int
9094 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9095 		       struct rte_flow_tunnel *app_tunnel,
9096 		       struct rte_flow_item **items,
9097 		       uint32_t *num_of_items,
9098 		       struct rte_flow_error *error)
9099 {
9100 	int ret;
9101 	struct mlx5_flow_tunnel *tunnel;
9102 	const char *err_msg = NULL;
9103 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
9104 
9105 	if (!verdict)
9106 		return rte_flow_error_set(error, EINVAL,
9107 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9108 					  err_msg);
9109 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9110 	if (ret < 0) {
9111 		return rte_flow_error_set(error, ret,
9112 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9113 					  "failed to initialize pmd tunnel");
9114 	}
9115 	*items = &tunnel->item;
9116 	*num_of_items = 1;
9117 	return 0;
9118 }
9119 
9120 struct tunnel_db_element_release_ctx {
9121 	struct rte_flow_item *items;
9122 	struct rte_flow_action *actions;
9123 	uint32_t num_elements;
9124 	struct rte_flow_error *error;
9125 	int ret;
9126 };
9127 
9128 static bool
9129 tunnel_element_release_match(struct rte_eth_dev *dev,
9130 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9131 {
9132 	const struct tunnel_db_element_release_ctx *ctx = x;
9133 
9134 	RTE_SET_USED(dev);
9135 	if (ctx->num_elements != 1)
9136 		return false;
9137 	else if (ctx->items)
9138 		return ctx->items == &tunnel->item;
9139 	else if (ctx->actions)
9140 		return ctx->actions == &tunnel->action;
9141 
9142 	return false;
9143 }
9144 
9145 static void
9146 tunnel_element_release_hit(struct rte_eth_dev *dev,
9147 			   struct mlx5_flow_tunnel *tunnel, void *x)
9148 {
9149 	struct tunnel_db_element_release_ctx *ctx = x;
9150 	ctx->ret = 0;
9151 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9152 		mlx5_flow_tunnel_free(dev, tunnel);
9153 }
9154 
9155 static void
9156 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9157 {
9158 	struct tunnel_db_element_release_ctx *ctx = x;
9159 	RTE_SET_USED(dev);
9160 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9161 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9162 				      "invalid argument");
9163 }
9164 
9165 static int
9166 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9167 		       struct rte_flow_item *pmd_items,
9168 		       uint32_t num_items, struct rte_flow_error *err)
9169 {
9170 	struct tunnel_db_element_release_ctx ctx = {
9171 		.items = pmd_items,
9172 		.actions = NULL,
9173 		.num_elements = num_items,
9174 		.error = err,
9175 	};
9176 
9177 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9178 				      tunnel_element_release_hit,
9179 				      tunnel_element_release_miss, &ctx, false);
9180 
9181 	return ctx.ret;
9182 }
9183 
9184 static int
9185 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9186 			 struct rte_flow_action *pmd_actions,
9187 			 uint32_t num_actions, struct rte_flow_error *err)
9188 {
9189 	struct tunnel_db_element_release_ctx ctx = {
9190 		.items = NULL,
9191 		.actions = pmd_actions,
9192 		.num_elements = num_actions,
9193 		.error = err,
9194 	};
9195 
9196 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9197 				      tunnel_element_release_hit,
9198 				      tunnel_element_release_miss, &ctx, false);
9199 
9200 	return ctx.ret;
9201 }
9202 
9203 static int
9204 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9205 				  struct rte_mbuf *m,
9206 				  struct rte_flow_restore_info *info,
9207 				  struct rte_flow_error *err)
9208 {
9209 	uint64_t ol_flags = m->ol_flags;
9210 	const struct mlx5_flow_tbl_data_entry *tble;
9211 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
9212 
9213 	if (!is_tunnel_offload_active(dev)) {
9214 		info->flags = 0;
9215 		return 0;
9216 	}
9217 
9218 	if ((ol_flags & mask) != mask)
9219 		goto err;
9220 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9221 	if (!tble) {
9222 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9223 			dev->data->port_id, m->hash.fdir.hi);
9224 		goto err;
9225 	}
9226 	MLX5_ASSERT(tble->tunnel);
9227 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9228 	info->group_id = tble->group_id;
9229 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9230 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9231 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9232 
9233 	return 0;
9234 
9235 err:
9236 	return rte_flow_error_set(err, EINVAL,
9237 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9238 				  "failed to get restore info");
9239 }
9240 
9241 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9242 static int
9243 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9244 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9245 			   __rte_unused struct rte_flow_action **actions,
9246 			   __rte_unused uint32_t *num_of_actions,
9247 			   __rte_unused struct rte_flow_error *error)
9248 {
9249 	return -ENOTSUP;
9250 }
9251 
9252 static int
9253 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9254 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9255 		       __rte_unused struct rte_flow_item **items,
9256 		       __rte_unused uint32_t *num_of_items,
9257 		       __rte_unused struct rte_flow_error *error)
9258 {
9259 	return -ENOTSUP;
9260 }
9261 
9262 static int
9263 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9264 			      __rte_unused struct rte_flow_item *pmd_items,
9265 			      __rte_unused uint32_t num_items,
9266 			      __rte_unused struct rte_flow_error *err)
9267 {
9268 	return -ENOTSUP;
9269 }
9270 
9271 static int
9272 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9273 				__rte_unused struct rte_flow_action *pmd_action,
9274 				__rte_unused uint32_t num_actions,
9275 				__rte_unused struct rte_flow_error *err)
9276 {
9277 	return -ENOTSUP;
9278 }
9279 
9280 static int
9281 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9282 				  __rte_unused struct rte_mbuf *m,
9283 				  __rte_unused struct rte_flow_restore_info *i,
9284 				  __rte_unused struct rte_flow_error *err)
9285 {
9286 	return -ENOTSUP;
9287 }
9288 
9289 static int
9290 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9291 			     __rte_unused struct rte_flow *flow,
9292 			     __rte_unused const struct rte_flow_attr *attr,
9293 			     __rte_unused const struct rte_flow_action *actions,
9294 			     __rte_unused uint32_t flow_idx,
9295 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9296 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9297 			     __rte_unused struct rte_flow_error *error)
9298 {
9299 	return -ENOTSUP;
9300 }
9301 
9302 static struct mlx5_flow_tunnel *
9303 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9304 		    __rte_unused uint32_t id)
9305 {
9306 	return NULL;
9307 }
9308 
9309 static void
9310 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9311 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9312 {
9313 }
9314 
9315 static uint32_t
9316 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9317 				__rte_unused const struct mlx5_flow_tunnel *t,
9318 				__rte_unused uint32_t group,
9319 				__rte_unused uint32_t *table,
9320 				struct rte_flow_error *error)
9321 {
9322 	return rte_flow_error_set(error, ENOTSUP,
9323 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9324 				  "tunnel offload requires DV support");
9325 }
9326 
9327 void
9328 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9329 			__rte_unused  uint16_t port_id)
9330 {
9331 }
9332 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9333 
9334 static void
9335 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9336 {
9337 	int ret;
9338 	struct rte_flow_error error;
9339 
9340 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9341 		char *item_name;
9342 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9343 				    sizeof(item_name),
9344 				    (void *)(uintptr_t)item->type, &error);
9345 		if (ret > 0)
9346 			printf("%s ", item_name);
9347 		else
9348 			printf("%d\n", (int)item->type);
9349 	}
9350 	printf("END\n");
9351 }
9352