xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 84f4764c22a32b319cb6fa7c8935b0d7977298cc)
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 		return true;
136 	default:
137 		break;
138 	}
139 	return false;
140 }
141 
142 static enum rte_flow_item_type
143 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
144 {
145 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
146 	uint16_t ether_type = 0;
147 	uint16_t ether_type_m;
148 	uint8_t ip_next_proto = 0;
149 	uint8_t ip_next_proto_m;
150 
151 	if (item == NULL || item->spec == NULL)
152 		return ret;
153 	switch (item->type) {
154 	case RTE_FLOW_ITEM_TYPE_ETH:
155 		if (item->mask)
156 			ether_type_m = ((const struct rte_flow_item_eth *)
157 						(item->mask))->type;
158 		else
159 			ether_type_m = rte_flow_item_eth_mask.type;
160 		if (ether_type_m != RTE_BE16(0xFFFF))
161 			break;
162 		ether_type = ((const struct rte_flow_item_eth *)
163 				(item->spec))->type;
164 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
165 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
166 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
167 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
168 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
169 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
170 		else
171 			ret = RTE_FLOW_ITEM_TYPE_END;
172 		break;
173 	case RTE_FLOW_ITEM_TYPE_VLAN:
174 		if (item->mask)
175 			ether_type_m = ((const struct rte_flow_item_vlan *)
176 						(item->mask))->inner_type;
177 		else
178 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
179 		if (ether_type_m != RTE_BE16(0xFFFF))
180 			break;
181 		ether_type = ((const struct rte_flow_item_vlan *)
182 				(item->spec))->inner_type;
183 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
184 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
185 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
186 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
187 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
188 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
189 		else
190 			ret = RTE_FLOW_ITEM_TYPE_END;
191 		break;
192 	case RTE_FLOW_ITEM_TYPE_IPV4:
193 		if (item->mask)
194 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
195 					(item->mask))->hdr.next_proto_id;
196 		else
197 			ip_next_proto_m =
198 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
199 		if (ip_next_proto_m != 0xFF)
200 			break;
201 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
202 				(item->spec))->hdr.next_proto_id;
203 		if (ip_next_proto == IPPROTO_UDP)
204 			ret = RTE_FLOW_ITEM_TYPE_UDP;
205 		else if (ip_next_proto == IPPROTO_TCP)
206 			ret = RTE_FLOW_ITEM_TYPE_TCP;
207 		else if (ip_next_proto == IPPROTO_IP)
208 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
209 		else if (ip_next_proto == IPPROTO_IPV6)
210 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
211 		else
212 			ret = RTE_FLOW_ITEM_TYPE_END;
213 		break;
214 	case RTE_FLOW_ITEM_TYPE_IPV6:
215 		if (item->mask)
216 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
217 						(item->mask))->hdr.proto;
218 		else
219 			ip_next_proto_m =
220 				rte_flow_item_ipv6_mask.hdr.proto;
221 		if (ip_next_proto_m != 0xFF)
222 			break;
223 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
224 				(item->spec))->hdr.proto;
225 		if (ip_next_proto == IPPROTO_UDP)
226 			ret = RTE_FLOW_ITEM_TYPE_UDP;
227 		else if (ip_next_proto == IPPROTO_TCP)
228 			ret = RTE_FLOW_ITEM_TYPE_TCP;
229 		else if (ip_next_proto == IPPROTO_IP)
230 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
231 		else if (ip_next_proto == IPPROTO_IPV6)
232 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
233 		else
234 			ret = RTE_FLOW_ITEM_TYPE_END;
235 		break;
236 	default:
237 		ret = RTE_FLOW_ITEM_TYPE_VOID;
238 		break;
239 	}
240 	return ret;
241 }
242 
243 #define MLX5_RSS_EXP_ELT_N 16
244 
245 /**
246  * Expand RSS flows into several possible flows according to the RSS hash
247  * fields requested and the driver capabilities.
248  *
249  * @param[out] buf
250  *   Buffer to store the result expansion.
251  * @param[in] size
252  *   Buffer size in bytes. If 0, @p buf can be NULL.
253  * @param[in] pattern
254  *   User flow pattern.
255  * @param[in] types
256  *   RSS types to expand (see ETH_RSS_* definitions).
257  * @param[in] graph
258  *   Input graph to expand @p pattern according to @p types.
259  * @param[in] graph_root_index
260  *   Index of root node in @p graph, typically 0.
261  *
262  * @return
263  *   A positive value representing the size of @p buf in bytes regardless of
264  *   @p size on success, a negative errno value otherwise and rte_errno is
265  *   set, the following errors are defined:
266  *
267  *   -E2BIG: graph-depth @p graph is too deep.
268  *   -EINVAL: @p size has not enough space for expanded pattern.
269  */
270 static int
271 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
272 		     const struct rte_flow_item *pattern, uint64_t types,
273 		     const struct mlx5_flow_expand_node graph[],
274 		     int graph_root_index)
275 {
276 	const struct rte_flow_item *item;
277 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
278 	const int *next_node;
279 	const int *stack[MLX5_RSS_EXP_ELT_N];
280 	int stack_pos = 0;
281 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
282 	unsigned int i;
283 	size_t lsize;
284 	size_t user_pattern_size = 0;
285 	void *addr = NULL;
286 	const struct mlx5_flow_expand_node *next = NULL;
287 	struct rte_flow_item missed_item;
288 	int missed = 0;
289 	int elt = 0;
290 	const struct rte_flow_item *last_item = NULL;
291 
292 	memset(&missed_item, 0, sizeof(missed_item));
293 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
294 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
295 	if (lsize > size)
296 		return -EINVAL;
297 	buf->entry[0].priority = 0;
298 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
299 	buf->entries = 0;
300 	addr = buf->entry[0].pattern;
301 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
302 		if (!mlx5_flow_is_rss_expandable_item(item)) {
303 			user_pattern_size += sizeof(*item);
304 			continue;
305 		}
306 		last_item = item;
307 		for (i = 0; node->next && node->next[i]; ++i) {
308 			next = &graph[node->next[i]];
309 			if (next->type == item->type)
310 				break;
311 		}
312 		if (next)
313 			node = next;
314 		user_pattern_size += sizeof(*item);
315 	}
316 	user_pattern_size += sizeof(*item); /* Handle END item. */
317 	lsize += user_pattern_size;
318 	if (lsize > size)
319 		return -EINVAL;
320 	/* Copy the user pattern in the first entry of the buffer. */
321 	rte_memcpy(addr, pattern, user_pattern_size);
322 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
323 	buf->entries = 1;
324 	/* Start expanding. */
325 	memset(flow_items, 0, sizeof(flow_items));
326 	user_pattern_size -= sizeof(*item);
327 	/*
328 	 * Check if the last valid item has spec set, need complete pattern,
329 	 * and the pattern can be used for expansion.
330 	 */
331 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
332 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
333 		/* Item type END indicates expansion is not required. */
334 		return lsize;
335 	}
336 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
337 		next = NULL;
338 		missed = 1;
339 		for (i = 0; node->next && node->next[i]; ++i) {
340 			next = &graph[node->next[i]];
341 			if (next->type == missed_item.type) {
342 				flow_items[0].type = missed_item.type;
343 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
344 				break;
345 			}
346 			next = NULL;
347 		}
348 	}
349 	if (next && missed) {
350 		elt = 2; /* missed item + item end. */
351 		node = next;
352 		lsize += elt * sizeof(*item) + user_pattern_size;
353 		if (lsize > size)
354 			return -EINVAL;
355 		if (node->rss_types & types) {
356 			buf->entry[buf->entries].priority = 1;
357 			buf->entry[buf->entries].pattern = addr;
358 			buf->entries++;
359 			rte_memcpy(addr, buf->entry[0].pattern,
360 				   user_pattern_size);
361 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
362 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
363 			addr = (void *)(((uintptr_t)addr) +
364 					elt * sizeof(*item));
365 		}
366 	}
367 	memset(flow_items, 0, sizeof(flow_items));
368 	next_node = node->next;
369 	stack[stack_pos] = next_node;
370 	node = next_node ? &graph[*next_node] : NULL;
371 	while (node) {
372 		flow_items[stack_pos].type = node->type;
373 		if (node->rss_types & types) {
374 			size_t n;
375 			/*
376 			 * compute the number of items to copy from the
377 			 * expansion and copy it.
378 			 * When the stack_pos is 0, there are 1 element in it,
379 			 * plus the addition END item.
380 			 */
381 			elt = stack_pos + 2;
382 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
383 			lsize += elt * sizeof(*item) + user_pattern_size;
384 			if (lsize > size)
385 				return -EINVAL;
386 			n = elt * sizeof(*item);
387 			buf->entry[buf->entries].priority =
388 				stack_pos + 1 + missed;
389 			buf->entry[buf->entries].pattern = addr;
390 			buf->entries++;
391 			rte_memcpy(addr, buf->entry[0].pattern,
392 				   user_pattern_size);
393 			addr = (void *)(((uintptr_t)addr) +
394 					user_pattern_size);
395 			rte_memcpy(addr, &missed_item,
396 				   missed * sizeof(*item));
397 			addr = (void *)(((uintptr_t)addr) +
398 				missed * sizeof(*item));
399 			rte_memcpy(addr, flow_items, n);
400 			addr = (void *)(((uintptr_t)addr) + n);
401 		}
402 		/* Go deeper. */
403 		if (!node->optional && node->next) {
404 			next_node = node->next;
405 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
406 				rte_errno = E2BIG;
407 				return -rte_errno;
408 			}
409 			stack[stack_pos] = next_node;
410 		} else if (*(next_node + 1)) {
411 			/* Follow up with the next possibility. */
412 			++next_node;
413 		} else {
414 			/* Move to the next path. */
415 			if (stack_pos)
416 				next_node = stack[--stack_pos];
417 			next_node++;
418 			stack[stack_pos] = next_node;
419 		}
420 		node = *next_node ? &graph[*next_node] : NULL;
421 	};
422 	return lsize;
423 }
424 
425 enum mlx5_expansion {
426 	MLX5_EXPANSION_ROOT,
427 	MLX5_EXPANSION_ROOT_OUTER,
428 	MLX5_EXPANSION_ROOT_ETH_VLAN,
429 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
430 	MLX5_EXPANSION_OUTER_ETH,
431 	MLX5_EXPANSION_OUTER_ETH_VLAN,
432 	MLX5_EXPANSION_OUTER_VLAN,
433 	MLX5_EXPANSION_OUTER_IPV4,
434 	MLX5_EXPANSION_OUTER_IPV4_UDP,
435 	MLX5_EXPANSION_OUTER_IPV4_TCP,
436 	MLX5_EXPANSION_OUTER_IPV6,
437 	MLX5_EXPANSION_OUTER_IPV6_UDP,
438 	MLX5_EXPANSION_OUTER_IPV6_TCP,
439 	MLX5_EXPANSION_VXLAN,
440 	MLX5_EXPANSION_VXLAN_GPE,
441 	MLX5_EXPANSION_GRE,
442 	MLX5_EXPANSION_NVGRE,
443 	MLX5_EXPANSION_GRE_KEY,
444 	MLX5_EXPANSION_MPLS,
445 	MLX5_EXPANSION_ETH,
446 	MLX5_EXPANSION_ETH_VLAN,
447 	MLX5_EXPANSION_VLAN,
448 	MLX5_EXPANSION_IPV4,
449 	MLX5_EXPANSION_IPV4_UDP,
450 	MLX5_EXPANSION_IPV4_TCP,
451 	MLX5_EXPANSION_IPV6,
452 	MLX5_EXPANSION_IPV6_UDP,
453 	MLX5_EXPANSION_IPV6_TCP,
454 };
455 
456 /** Supported expansion of items. */
457 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
458 	[MLX5_EXPANSION_ROOT] = {
459 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
460 						  MLX5_EXPANSION_IPV4,
461 						  MLX5_EXPANSION_IPV6),
462 		.type = RTE_FLOW_ITEM_TYPE_END,
463 	},
464 	[MLX5_EXPANSION_ROOT_OUTER] = {
465 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
466 						  MLX5_EXPANSION_OUTER_IPV4,
467 						  MLX5_EXPANSION_OUTER_IPV6),
468 		.type = RTE_FLOW_ITEM_TYPE_END,
469 	},
470 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
471 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
472 		.type = RTE_FLOW_ITEM_TYPE_END,
473 	},
474 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
475 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
476 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
477 		.type = RTE_FLOW_ITEM_TYPE_END,
478 	},
479 	[MLX5_EXPANSION_OUTER_ETH] = {
480 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
481 						  MLX5_EXPANSION_OUTER_IPV6),
482 		.type = RTE_FLOW_ITEM_TYPE_ETH,
483 		.rss_types = 0,
484 	},
485 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
486 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
487 		.type = RTE_FLOW_ITEM_TYPE_ETH,
488 		.rss_types = 0,
489 	},
490 	[MLX5_EXPANSION_OUTER_VLAN] = {
491 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
492 						  MLX5_EXPANSION_OUTER_IPV6),
493 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
494 	},
495 	[MLX5_EXPANSION_OUTER_IPV4] = {
496 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
497 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
498 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
499 			 MLX5_EXPANSION_GRE,
500 			 MLX5_EXPANSION_NVGRE,
501 			 MLX5_EXPANSION_IPV4,
502 			 MLX5_EXPANSION_IPV6),
503 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
504 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
505 			ETH_RSS_NONFRAG_IPV4_OTHER,
506 	},
507 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
508 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 						  MLX5_EXPANSION_VXLAN_GPE,
510 						  MLX5_EXPANSION_MPLS),
511 		.type = RTE_FLOW_ITEM_TYPE_UDP,
512 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
513 	},
514 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
515 		.type = RTE_FLOW_ITEM_TYPE_TCP,
516 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
517 	},
518 	[MLX5_EXPANSION_OUTER_IPV6] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
520 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
521 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
522 			 MLX5_EXPANSION_IPV4,
523 			 MLX5_EXPANSION_IPV6,
524 			 MLX5_EXPANSION_GRE,
525 			 MLX5_EXPANSION_NVGRE),
526 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
527 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
528 			ETH_RSS_NONFRAG_IPV6_OTHER,
529 	},
530 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
531 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
532 						  MLX5_EXPANSION_VXLAN_GPE,
533 						  MLX5_EXPANSION_MPLS),
534 		.type = RTE_FLOW_ITEM_TYPE_UDP,
535 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
536 	},
537 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
538 		.type = RTE_FLOW_ITEM_TYPE_TCP,
539 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
540 	},
541 	[MLX5_EXPANSION_VXLAN] = {
542 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
543 						  MLX5_EXPANSION_IPV4,
544 						  MLX5_EXPANSION_IPV6),
545 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
546 	},
547 	[MLX5_EXPANSION_VXLAN_GPE] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
549 						  MLX5_EXPANSION_IPV4,
550 						  MLX5_EXPANSION_IPV6),
551 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
552 	},
553 	[MLX5_EXPANSION_GRE] = {
554 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
555 						  MLX5_EXPANSION_IPV6,
556 						  MLX5_EXPANSION_GRE_KEY,
557 						  MLX5_EXPANSION_MPLS),
558 		.type = RTE_FLOW_ITEM_TYPE_GRE,
559 	},
560 	[MLX5_EXPANSION_GRE_KEY] = {
561 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
562 						  MLX5_EXPANSION_IPV6,
563 						  MLX5_EXPANSION_MPLS),
564 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
565 		.optional = 1,
566 	},
567 	[MLX5_EXPANSION_NVGRE] = {
568 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
569 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
570 	},
571 	[MLX5_EXPANSION_MPLS] = {
572 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
573 						  MLX5_EXPANSION_IPV6,
574 						  MLX5_EXPANSION_ETH),
575 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
576 	},
577 	[MLX5_EXPANSION_ETH] = {
578 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
579 						  MLX5_EXPANSION_IPV6),
580 		.type = RTE_FLOW_ITEM_TYPE_ETH,
581 	},
582 	[MLX5_EXPANSION_ETH_VLAN] = {
583 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
584 		.type = RTE_FLOW_ITEM_TYPE_ETH,
585 	},
586 	[MLX5_EXPANSION_VLAN] = {
587 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
588 						  MLX5_EXPANSION_IPV6),
589 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
590 	},
591 	[MLX5_EXPANSION_IPV4] = {
592 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
593 						  MLX5_EXPANSION_IPV4_TCP),
594 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
595 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
596 			ETH_RSS_NONFRAG_IPV4_OTHER,
597 	},
598 	[MLX5_EXPANSION_IPV4_UDP] = {
599 		.type = RTE_FLOW_ITEM_TYPE_UDP,
600 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
601 	},
602 	[MLX5_EXPANSION_IPV4_TCP] = {
603 		.type = RTE_FLOW_ITEM_TYPE_TCP,
604 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
605 	},
606 	[MLX5_EXPANSION_IPV6] = {
607 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
608 						  MLX5_EXPANSION_IPV6_TCP),
609 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
610 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
611 			ETH_RSS_NONFRAG_IPV6_OTHER,
612 	},
613 	[MLX5_EXPANSION_IPV6_UDP] = {
614 		.type = RTE_FLOW_ITEM_TYPE_UDP,
615 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
616 	},
617 	[MLX5_EXPANSION_IPV6_TCP] = {
618 		.type = RTE_FLOW_ITEM_TYPE_TCP,
619 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
620 	},
621 };
622 
623 static struct rte_flow_action_handle *
624 mlx5_action_handle_create(struct rte_eth_dev *dev,
625 			  const struct rte_flow_indir_action_conf *conf,
626 			  const struct rte_flow_action *action,
627 			  struct rte_flow_error *error);
628 static int mlx5_action_handle_destroy
629 				(struct rte_eth_dev *dev,
630 				 struct rte_flow_action_handle *handle,
631 				 struct rte_flow_error *error);
632 static int mlx5_action_handle_update
633 				(struct rte_eth_dev *dev,
634 				 struct rte_flow_action_handle *handle,
635 				 const void *update,
636 				 struct rte_flow_error *error);
637 static int mlx5_action_handle_query
638 				(struct rte_eth_dev *dev,
639 				 const struct rte_flow_action_handle *handle,
640 				 void *data,
641 				 struct rte_flow_error *error);
642 static int
643 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
644 		    struct rte_flow_tunnel *app_tunnel,
645 		    struct rte_flow_action **actions,
646 		    uint32_t *num_of_actions,
647 		    struct rte_flow_error *error);
648 static int
649 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
650 		       struct rte_flow_tunnel *app_tunnel,
651 		       struct rte_flow_item **items,
652 		       uint32_t *num_of_items,
653 		       struct rte_flow_error *error);
654 static int
655 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
656 			      struct rte_flow_item *pmd_items,
657 			      uint32_t num_items, struct rte_flow_error *err);
658 static int
659 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
660 				struct rte_flow_action *pmd_actions,
661 				uint32_t num_actions,
662 				struct rte_flow_error *err);
663 static int
664 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
665 				  struct rte_mbuf *m,
666 				  struct rte_flow_restore_info *info,
667 				  struct rte_flow_error *err);
668 
669 static const struct rte_flow_ops mlx5_flow_ops = {
670 	.validate = mlx5_flow_validate,
671 	.create = mlx5_flow_create,
672 	.destroy = mlx5_flow_destroy,
673 	.flush = mlx5_flow_flush,
674 	.isolate = mlx5_flow_isolate,
675 	.query = mlx5_flow_query,
676 	.dev_dump = mlx5_flow_dev_dump,
677 	.get_aged_flows = mlx5_flow_get_aged_flows,
678 	.action_handle_create = mlx5_action_handle_create,
679 	.action_handle_destroy = mlx5_action_handle_destroy,
680 	.action_handle_update = mlx5_action_handle_update,
681 	.action_handle_query = mlx5_action_handle_query,
682 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
683 	.tunnel_match = mlx5_flow_tunnel_match,
684 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
685 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
686 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
687 };
688 
689 /* Tunnel information. */
690 struct mlx5_flow_tunnel_info {
691 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
692 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
693 };
694 
695 static struct mlx5_flow_tunnel_info tunnels_info[] = {
696 	{
697 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
698 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
699 	},
700 	{
701 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
702 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
703 	},
704 	{
705 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
706 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
707 	},
708 	{
709 		.tunnel = MLX5_FLOW_LAYER_GRE,
710 		.ptype = RTE_PTYPE_TUNNEL_GRE,
711 	},
712 	{
713 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
714 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
715 	},
716 	{
717 		.tunnel = MLX5_FLOW_LAYER_MPLS,
718 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
719 	},
720 	{
721 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
722 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
723 	},
724 	{
725 		.tunnel = MLX5_FLOW_LAYER_IPIP,
726 		.ptype = RTE_PTYPE_TUNNEL_IP,
727 	},
728 	{
729 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
730 		.ptype = RTE_PTYPE_TUNNEL_IP,
731 	},
732 	{
733 		.tunnel = MLX5_FLOW_LAYER_GTP,
734 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
735 	},
736 };
737 
738 
739 
740 /**
741  * Translate tag ID to register.
742  *
743  * @param[in] dev
744  *   Pointer to the Ethernet device structure.
745  * @param[in] feature
746  *   The feature that request the register.
747  * @param[in] id
748  *   The request register ID.
749  * @param[out] error
750  *   Error description in case of any.
751  *
752  * @return
753  *   The request register on success, a negative errno
754  *   value otherwise and rte_errno is set.
755  */
756 int
757 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
758 		     enum mlx5_feature_name feature,
759 		     uint32_t id,
760 		     struct rte_flow_error *error)
761 {
762 	struct mlx5_priv *priv = dev->data->dev_private;
763 	struct mlx5_dev_config *config = &priv->config;
764 	enum modify_reg start_reg;
765 	bool skip_mtr_reg = false;
766 
767 	switch (feature) {
768 	case MLX5_HAIRPIN_RX:
769 		return REG_B;
770 	case MLX5_HAIRPIN_TX:
771 		return REG_A;
772 	case MLX5_METADATA_RX:
773 		switch (config->dv_xmeta_en) {
774 		case MLX5_XMETA_MODE_LEGACY:
775 			return REG_B;
776 		case MLX5_XMETA_MODE_META16:
777 			return REG_C_0;
778 		case MLX5_XMETA_MODE_META32:
779 			return REG_C_1;
780 		}
781 		break;
782 	case MLX5_METADATA_TX:
783 		return REG_A;
784 	case MLX5_METADATA_FDB:
785 		switch (config->dv_xmeta_en) {
786 		case MLX5_XMETA_MODE_LEGACY:
787 			return REG_NON;
788 		case MLX5_XMETA_MODE_META16:
789 			return REG_C_0;
790 		case MLX5_XMETA_MODE_META32:
791 			return REG_C_1;
792 		}
793 		break;
794 	case MLX5_FLOW_MARK:
795 		switch (config->dv_xmeta_en) {
796 		case MLX5_XMETA_MODE_LEGACY:
797 			return REG_NON;
798 		case MLX5_XMETA_MODE_META16:
799 			return REG_C_1;
800 		case MLX5_XMETA_MODE_META32:
801 			return REG_C_0;
802 		}
803 		break;
804 	case MLX5_MTR_ID:
805 		/*
806 		 * If meter color and meter id share one register, flow match
807 		 * should use the meter color register for match.
808 		 */
809 		if (priv->mtr_reg_share)
810 			return priv->mtr_color_reg;
811 		else
812 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
813 			       REG_C_3;
814 	case MLX5_MTR_COLOR:
815 	case MLX5_ASO_FLOW_HIT:
816 	case MLX5_ASO_CONNTRACK:
817 		/* All features use the same REG_C. */
818 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
819 		return priv->mtr_color_reg;
820 	case MLX5_COPY_MARK:
821 		/*
822 		 * Metadata COPY_MARK register using is in meter suffix sub
823 		 * flow while with meter. It's safe to share the same register.
824 		 */
825 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
826 	case MLX5_APP_TAG:
827 		/*
828 		 * If meter is enable, it will engage the register for color
829 		 * match and flow match. If meter color match is not using the
830 		 * REG_C_2, need to skip the REG_C_x be used by meter color
831 		 * match.
832 		 * If meter is disable, free to use all available registers.
833 		 */
834 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
835 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
836 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
837 		if (id > (uint32_t)(REG_C_7 - start_reg))
838 			return rte_flow_error_set(error, EINVAL,
839 						  RTE_FLOW_ERROR_TYPE_ITEM,
840 						  NULL, "invalid tag id");
841 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
842 			return rte_flow_error_set(error, ENOTSUP,
843 						  RTE_FLOW_ERROR_TYPE_ITEM,
844 						  NULL, "unsupported tag id");
845 		/*
846 		 * This case means meter is using the REG_C_x great than 2.
847 		 * Take care not to conflict with meter color REG_C_x.
848 		 * If the available index REG_C_y >= REG_C_x, skip the
849 		 * color register.
850 		 */
851 		if (skip_mtr_reg && config->flow_mreg_c
852 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
853 			if (id >= (uint32_t)(REG_C_7 - start_reg))
854 				return rte_flow_error_set(error, EINVAL,
855 						       RTE_FLOW_ERROR_TYPE_ITEM,
856 							NULL, "invalid tag id");
857 			if (config->flow_mreg_c
858 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
859 				return config->flow_mreg_c
860 					       [id + 1 + start_reg - REG_C_0];
861 			return rte_flow_error_set(error, ENOTSUP,
862 						  RTE_FLOW_ERROR_TYPE_ITEM,
863 						  NULL, "unsupported tag id");
864 		}
865 		return config->flow_mreg_c[id + start_reg - REG_C_0];
866 	}
867 	MLX5_ASSERT(false);
868 	return rte_flow_error_set(error, EINVAL,
869 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
870 				  NULL, "invalid feature name");
871 }
872 
873 /**
874  * Check extensive flow metadata register support.
875  *
876  * @param dev
877  *   Pointer to rte_eth_dev structure.
878  *
879  * @return
880  *   True if device supports extensive flow metadata register, otherwise false.
881  */
882 bool
883 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
884 {
885 	struct mlx5_priv *priv = dev->data->dev_private;
886 	struct mlx5_dev_config *config = &priv->config;
887 
888 	/*
889 	 * Having available reg_c can be regarded inclusively as supporting
890 	 * extensive flow metadata register, which could mean,
891 	 * - metadata register copy action by modify header.
892 	 * - 16 modify header actions is supported.
893 	 * - reg_c's are preserved across different domain (FDB and NIC) on
894 	 *   packet loopback by flow lookup miss.
895 	 */
896 	return config->flow_mreg_c[2] != REG_NON;
897 }
898 
899 /**
900  * Get the lowest priority.
901  *
902  * @param[in] dev
903  *   Pointer to the Ethernet device structure.
904  * @param[in] attributes
905  *   Pointer to device flow rule attributes.
906  *
907  * @return
908  *   The value of lowest priority of flow.
909  */
910 uint32_t
911 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
912 			  const struct rte_flow_attr *attr)
913 {
914 	struct mlx5_priv *priv = dev->data->dev_private;
915 
916 	if (!attr->group && !attr->transfer)
917 		return priv->config.flow_prio - 2;
918 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
919 }
920 
921 /**
922  * Calculate matcher priority of the flow.
923  *
924  * @param[in] dev
925  *   Pointer to the Ethernet device structure.
926  * @param[in] attr
927  *   Pointer to device flow rule attributes.
928  * @param[in] subpriority
929  *   The priority based on the items.
930  * @return
931  *   The matcher priority of the flow.
932  */
933 uint16_t
934 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
935 			  const struct rte_flow_attr *attr,
936 			  uint32_t subpriority)
937 {
938 	uint16_t priority = (uint16_t)attr->priority;
939 	struct mlx5_priv *priv = dev->data->dev_private;
940 
941 	if (!attr->group && !attr->transfer) {
942 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
943 			priority = priv->config.flow_prio - 1;
944 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
945 	}
946 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
947 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
948 	return priority * 3 + subpriority;
949 }
950 
951 /**
952  * Verify the @p item specifications (spec, last, mask) are compatible with the
953  * NIC capabilities.
954  *
955  * @param[in] item
956  *   Item specification.
957  * @param[in] mask
958  *   @p item->mask or flow default bit-masks.
959  * @param[in] nic_mask
960  *   Bit-masks covering supported fields by the NIC to compare with user mask.
961  * @param[in] size
962  *   Bit-masks size in bytes.
963  * @param[in] range_accepted
964  *   True if range of values is accepted for specific fields, false otherwise.
965  * @param[out] error
966  *   Pointer to error structure.
967  *
968  * @return
969  *   0 on success, a negative errno value otherwise and rte_errno is set.
970  */
971 int
972 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
973 			  const uint8_t *mask,
974 			  const uint8_t *nic_mask,
975 			  unsigned int size,
976 			  bool range_accepted,
977 			  struct rte_flow_error *error)
978 {
979 	unsigned int i;
980 
981 	MLX5_ASSERT(nic_mask);
982 	for (i = 0; i < size; ++i)
983 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
984 			return rte_flow_error_set(error, ENOTSUP,
985 						  RTE_FLOW_ERROR_TYPE_ITEM,
986 						  item,
987 						  "mask enables non supported"
988 						  " bits");
989 	if (!item->spec && (item->mask || item->last))
990 		return rte_flow_error_set(error, EINVAL,
991 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
992 					  "mask/last without a spec is not"
993 					  " supported");
994 	if (item->spec && item->last && !range_accepted) {
995 		uint8_t spec[size];
996 		uint8_t last[size];
997 		unsigned int i;
998 		int ret;
999 
1000 		for (i = 0; i < size; ++i) {
1001 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1002 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1003 		}
1004 		ret = memcmp(spec, last, size);
1005 		if (ret != 0)
1006 			return rte_flow_error_set(error, EINVAL,
1007 						  RTE_FLOW_ERROR_TYPE_ITEM,
1008 						  item,
1009 						  "range is not valid");
1010 	}
1011 	return 0;
1012 }
1013 
1014 /**
1015  * Adjust the hash fields according to the @p flow information.
1016  *
1017  * @param[in] dev_flow.
1018  *   Pointer to the mlx5_flow.
1019  * @param[in] tunnel
1020  *   1 when the hash field is for a tunnel item.
1021  * @param[in] layer_types
1022  *   ETH_RSS_* types.
1023  * @param[in] hash_fields
1024  *   Item hash fields.
1025  *
1026  * @return
1027  *   The hash fields that should be used.
1028  */
1029 uint64_t
1030 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1031 			    int tunnel __rte_unused, uint64_t layer_types,
1032 			    uint64_t hash_fields)
1033 {
1034 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1035 	int rss_request_inner = rss_desc->level >= 2;
1036 
1037 	/* Check RSS hash level for tunnel. */
1038 	if (tunnel && rss_request_inner)
1039 		hash_fields |= IBV_RX_HASH_INNER;
1040 	else if (tunnel || rss_request_inner)
1041 		return 0;
1042 #endif
1043 	/* Check if requested layer matches RSS hash fields. */
1044 	if (!(rss_desc->types & layer_types))
1045 		return 0;
1046 	return hash_fields;
1047 }
1048 
1049 /**
1050  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1051  * if several tunnel rules are used on this queue, the tunnel ptype will be
1052  * cleared.
1053  *
1054  * @param rxq_ctrl
1055  *   Rx queue to update.
1056  */
1057 static void
1058 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1059 {
1060 	unsigned int i;
1061 	uint32_t tunnel_ptype = 0;
1062 
1063 	/* Look up for the ptype to use. */
1064 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1065 		if (!rxq_ctrl->flow_tunnels_n[i])
1066 			continue;
1067 		if (!tunnel_ptype) {
1068 			tunnel_ptype = tunnels_info[i].ptype;
1069 		} else {
1070 			tunnel_ptype = 0;
1071 			break;
1072 		}
1073 	}
1074 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1075 }
1076 
1077 /**
1078  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1079  * flow.
1080  *
1081  * @param[in] dev
1082  *   Pointer to the Ethernet device structure.
1083  * @param[in] dev_handle
1084  *   Pointer to device flow handle structure.
1085  */
1086 void
1087 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1088 		       struct mlx5_flow_handle *dev_handle)
1089 {
1090 	struct mlx5_priv *priv = dev->data->dev_private;
1091 	const int mark = dev_handle->mark;
1092 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1093 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1094 	unsigned int i;
1095 
1096 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1097 		struct mlx5_hrxq *hrxq;
1098 
1099 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1100 			      dev_handle->rix_hrxq);
1101 		if (hrxq)
1102 			ind_tbl = hrxq->ind_table;
1103 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1104 		struct mlx5_shared_action_rss *shared_rss;
1105 
1106 		shared_rss = mlx5_ipool_get
1107 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1108 			 dev_handle->rix_srss);
1109 		if (shared_rss)
1110 			ind_tbl = shared_rss->ind_tbl;
1111 	}
1112 	if (!ind_tbl)
1113 		return;
1114 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1115 		int idx = ind_tbl->queues[i];
1116 		struct mlx5_rxq_ctrl *rxq_ctrl =
1117 			container_of((*priv->rxqs)[idx],
1118 				     struct mlx5_rxq_ctrl, rxq);
1119 
1120 		/*
1121 		 * To support metadata register copy on Tx loopback,
1122 		 * this must be always enabled (metadata may arive
1123 		 * from other port - not from local flows only.
1124 		 */
1125 		if (priv->config.dv_flow_en &&
1126 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1127 		    mlx5_flow_ext_mreg_supported(dev)) {
1128 			rxq_ctrl->rxq.mark = 1;
1129 			rxq_ctrl->flow_mark_n = 1;
1130 		} else if (mark) {
1131 			rxq_ctrl->rxq.mark = 1;
1132 			rxq_ctrl->flow_mark_n++;
1133 		}
1134 		if (tunnel) {
1135 			unsigned int j;
1136 
1137 			/* Increase the counter matching the flow. */
1138 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1139 				if ((tunnels_info[j].tunnel &
1140 				     dev_handle->layers) ==
1141 				    tunnels_info[j].tunnel) {
1142 					rxq_ctrl->flow_tunnels_n[j]++;
1143 					break;
1144 				}
1145 			}
1146 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1147 		}
1148 	}
1149 }
1150 
1151 /**
1152  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1153  *
1154  * @param[in] dev
1155  *   Pointer to the Ethernet device structure.
1156  * @param[in] flow
1157  *   Pointer to flow structure.
1158  */
1159 static void
1160 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1161 {
1162 	struct mlx5_priv *priv = dev->data->dev_private;
1163 	uint32_t handle_idx;
1164 	struct mlx5_flow_handle *dev_handle;
1165 
1166 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1167 		       handle_idx, dev_handle, next)
1168 		flow_drv_rxq_flags_set(dev, dev_handle);
1169 }
1170 
1171 /**
1172  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1173  * device flow if no other flow uses it with the same kind of request.
1174  *
1175  * @param dev
1176  *   Pointer to Ethernet device.
1177  * @param[in] dev_handle
1178  *   Pointer to the device flow handle structure.
1179  */
1180 static void
1181 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1182 			struct mlx5_flow_handle *dev_handle)
1183 {
1184 	struct mlx5_priv *priv = dev->data->dev_private;
1185 	const int mark = dev_handle->mark;
1186 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1187 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1188 	unsigned int i;
1189 
1190 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1191 		struct mlx5_hrxq *hrxq;
1192 
1193 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1194 			      dev_handle->rix_hrxq);
1195 		if (hrxq)
1196 			ind_tbl = hrxq->ind_table;
1197 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1198 		struct mlx5_shared_action_rss *shared_rss;
1199 
1200 		shared_rss = mlx5_ipool_get
1201 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1202 			 dev_handle->rix_srss);
1203 		if (shared_rss)
1204 			ind_tbl = shared_rss->ind_tbl;
1205 	}
1206 	if (!ind_tbl)
1207 		return;
1208 	MLX5_ASSERT(dev->data->dev_started);
1209 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1210 		int idx = ind_tbl->queues[i];
1211 		struct mlx5_rxq_ctrl *rxq_ctrl =
1212 			container_of((*priv->rxqs)[idx],
1213 				     struct mlx5_rxq_ctrl, rxq);
1214 
1215 		if (priv->config.dv_flow_en &&
1216 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1217 		    mlx5_flow_ext_mreg_supported(dev)) {
1218 			rxq_ctrl->rxq.mark = 1;
1219 			rxq_ctrl->flow_mark_n = 1;
1220 		} else if (mark) {
1221 			rxq_ctrl->flow_mark_n--;
1222 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1223 		}
1224 		if (tunnel) {
1225 			unsigned int j;
1226 
1227 			/* Decrease the counter matching the flow. */
1228 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1229 				if ((tunnels_info[j].tunnel &
1230 				     dev_handle->layers) ==
1231 				    tunnels_info[j].tunnel) {
1232 					rxq_ctrl->flow_tunnels_n[j]--;
1233 					break;
1234 				}
1235 			}
1236 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1237 		}
1238 	}
1239 }
1240 
1241 /**
1242  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1243  * @p flow if no other flow uses it with the same kind of request.
1244  *
1245  * @param dev
1246  *   Pointer to Ethernet device.
1247  * @param[in] flow
1248  *   Pointer to the flow.
1249  */
1250 static void
1251 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1252 {
1253 	struct mlx5_priv *priv = dev->data->dev_private;
1254 	uint32_t handle_idx;
1255 	struct mlx5_flow_handle *dev_handle;
1256 
1257 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1258 		       handle_idx, dev_handle, next)
1259 		flow_drv_rxq_flags_trim(dev, dev_handle);
1260 }
1261 
1262 /**
1263  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1264  *
1265  * @param dev
1266  *   Pointer to Ethernet device.
1267  */
1268 static void
1269 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1270 {
1271 	struct mlx5_priv *priv = dev->data->dev_private;
1272 	unsigned int i;
1273 
1274 	for (i = 0; i != priv->rxqs_n; ++i) {
1275 		struct mlx5_rxq_ctrl *rxq_ctrl;
1276 		unsigned int j;
1277 
1278 		if (!(*priv->rxqs)[i])
1279 			continue;
1280 		rxq_ctrl = container_of((*priv->rxqs)[i],
1281 					struct mlx5_rxq_ctrl, rxq);
1282 		rxq_ctrl->flow_mark_n = 0;
1283 		rxq_ctrl->rxq.mark = 0;
1284 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1285 			rxq_ctrl->flow_tunnels_n[j] = 0;
1286 		rxq_ctrl->rxq.tunnel = 0;
1287 	}
1288 }
1289 
1290 /**
1291  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1292  *
1293  * @param[in] dev
1294  *   Pointer to the Ethernet device structure.
1295  */
1296 void
1297 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1298 {
1299 	struct mlx5_priv *priv = dev->data->dev_private;
1300 	struct mlx5_rxq_data *data;
1301 	unsigned int i;
1302 
1303 	for (i = 0; i != priv->rxqs_n; ++i) {
1304 		if (!(*priv->rxqs)[i])
1305 			continue;
1306 		data = (*priv->rxqs)[i];
1307 		if (!rte_flow_dynf_metadata_avail()) {
1308 			data->dynf_meta = 0;
1309 			data->flow_meta_mask = 0;
1310 			data->flow_meta_offset = -1;
1311 			data->flow_meta_port_mask = 0;
1312 		} else {
1313 			data->dynf_meta = 1;
1314 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1315 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1316 			data->flow_meta_port_mask = (uint32_t)~0;
1317 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1318 				data->flow_meta_port_mask >>= 16;
1319 		}
1320 	}
1321 }
1322 
1323 /*
1324  * return a pointer to the desired action in the list of actions.
1325  *
1326  * @param[in] actions
1327  *   The list of actions to search the action in.
1328  * @param[in] action
1329  *   The action to find.
1330  *
1331  * @return
1332  *   Pointer to the action in the list, if found. NULL otherwise.
1333  */
1334 const struct rte_flow_action *
1335 mlx5_flow_find_action(const struct rte_flow_action *actions,
1336 		      enum rte_flow_action_type action)
1337 {
1338 	if (actions == NULL)
1339 		return NULL;
1340 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1341 		if (actions->type == action)
1342 			return actions;
1343 	return NULL;
1344 }
1345 
1346 /*
1347  * Validate the flag action.
1348  *
1349  * @param[in] action_flags
1350  *   Bit-fields that holds the actions detected until now.
1351  * @param[in] attr
1352  *   Attributes of flow that includes this action.
1353  * @param[out] error
1354  *   Pointer to error structure.
1355  *
1356  * @return
1357  *   0 on success, a negative errno value otherwise and rte_errno is set.
1358  */
1359 int
1360 mlx5_flow_validate_action_flag(uint64_t action_flags,
1361 			       const struct rte_flow_attr *attr,
1362 			       struct rte_flow_error *error)
1363 {
1364 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1365 		return rte_flow_error_set(error, EINVAL,
1366 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1367 					  "can't mark and flag in same flow");
1368 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1369 		return rte_flow_error_set(error, EINVAL,
1370 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1371 					  "can't have 2 flag"
1372 					  " actions in same flow");
1373 	if (attr->egress)
1374 		return rte_flow_error_set(error, ENOTSUP,
1375 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1376 					  "flag action not supported for "
1377 					  "egress");
1378 	return 0;
1379 }
1380 
1381 /*
1382  * Validate the mark action.
1383  *
1384  * @param[in] action
1385  *   Pointer to the queue action.
1386  * @param[in] action_flags
1387  *   Bit-fields that holds the actions detected until now.
1388  * @param[in] attr
1389  *   Attributes of flow that includes this action.
1390  * @param[out] error
1391  *   Pointer to error structure.
1392  *
1393  * @return
1394  *   0 on success, a negative errno value otherwise and rte_errno is set.
1395  */
1396 int
1397 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1398 			       uint64_t action_flags,
1399 			       const struct rte_flow_attr *attr,
1400 			       struct rte_flow_error *error)
1401 {
1402 	const struct rte_flow_action_mark *mark = action->conf;
1403 
1404 	if (!mark)
1405 		return rte_flow_error_set(error, EINVAL,
1406 					  RTE_FLOW_ERROR_TYPE_ACTION,
1407 					  action,
1408 					  "configuration cannot be null");
1409 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1410 		return rte_flow_error_set(error, EINVAL,
1411 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1412 					  &mark->id,
1413 					  "mark id must in 0 <= id < "
1414 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1415 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1416 		return rte_flow_error_set(error, EINVAL,
1417 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1418 					  "can't flag and mark in same flow");
1419 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1420 		return rte_flow_error_set(error, EINVAL,
1421 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1422 					  "can't have 2 mark actions in same"
1423 					  " flow");
1424 	if (attr->egress)
1425 		return rte_flow_error_set(error, ENOTSUP,
1426 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1427 					  "mark action not supported for "
1428 					  "egress");
1429 	return 0;
1430 }
1431 
1432 /*
1433  * Validate the drop action.
1434  *
1435  * @param[in] action_flags
1436  *   Bit-fields that holds the actions detected until now.
1437  * @param[in] attr
1438  *   Attributes of flow that includes this action.
1439  * @param[out] error
1440  *   Pointer to error structure.
1441  *
1442  * @return
1443  *   0 on success, a negative errno value otherwise and rte_errno is set.
1444  */
1445 int
1446 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1447 			       const struct rte_flow_attr *attr,
1448 			       struct rte_flow_error *error)
1449 {
1450 	if (attr->egress)
1451 		return rte_flow_error_set(error, ENOTSUP,
1452 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1453 					  "drop action not supported for "
1454 					  "egress");
1455 	return 0;
1456 }
1457 
1458 /*
1459  * Validate the queue action.
1460  *
1461  * @param[in] action
1462  *   Pointer to the queue action.
1463  * @param[in] action_flags
1464  *   Bit-fields that holds the actions detected until now.
1465  * @param[in] dev
1466  *   Pointer to the Ethernet device structure.
1467  * @param[in] attr
1468  *   Attributes of flow that includes this action.
1469  * @param[out] error
1470  *   Pointer to error structure.
1471  *
1472  * @return
1473  *   0 on success, a negative errno value otherwise and rte_errno is set.
1474  */
1475 int
1476 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1477 				uint64_t action_flags,
1478 				struct rte_eth_dev *dev,
1479 				const struct rte_flow_attr *attr,
1480 				struct rte_flow_error *error)
1481 {
1482 	struct mlx5_priv *priv = dev->data->dev_private;
1483 	const struct rte_flow_action_queue *queue = action->conf;
1484 
1485 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1486 		return rte_flow_error_set(error, EINVAL,
1487 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1488 					  "can't have 2 fate actions in"
1489 					  " same flow");
1490 	if (!priv->rxqs_n)
1491 		return rte_flow_error_set(error, EINVAL,
1492 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1493 					  NULL, "No Rx queues configured");
1494 	if (queue->index >= priv->rxqs_n)
1495 		return rte_flow_error_set(error, EINVAL,
1496 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1497 					  &queue->index,
1498 					  "queue index out of range");
1499 	if (!(*priv->rxqs)[queue->index])
1500 		return rte_flow_error_set(error, EINVAL,
1501 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1502 					  &queue->index,
1503 					  "queue is not configured");
1504 	if (attr->egress)
1505 		return rte_flow_error_set(error, ENOTSUP,
1506 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1507 					  "queue action not supported for "
1508 					  "egress");
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Validate the rss action.
1514  *
1515  * @param[in] dev
1516  *   Pointer to the Ethernet device structure.
1517  * @param[in] action
1518  *   Pointer to the queue action.
1519  * @param[out] error
1520  *   Pointer to error structure.
1521  *
1522  * @return
1523  *   0 on success, a negative errno value otherwise and rte_errno is set.
1524  */
1525 int
1526 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1527 			 const struct rte_flow_action *action,
1528 			 struct rte_flow_error *error)
1529 {
1530 	struct mlx5_priv *priv = dev->data->dev_private;
1531 	const struct rte_flow_action_rss *rss = action->conf;
1532 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1533 	unsigned int i;
1534 
1535 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1536 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1537 		return rte_flow_error_set(error, ENOTSUP,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 					  &rss->func,
1540 					  "RSS hash function not supported");
1541 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1542 	if (rss->level > 2)
1543 #else
1544 	if (rss->level > 1)
1545 #endif
1546 		return rte_flow_error_set(error, ENOTSUP,
1547 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1548 					  &rss->level,
1549 					  "tunnel RSS is not supported");
1550 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1551 	if (rss->key_len == 0 && rss->key != NULL)
1552 		return rte_flow_error_set(error, ENOTSUP,
1553 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1554 					  &rss->key_len,
1555 					  "RSS hash key length 0");
1556 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1559 					  &rss->key_len,
1560 					  "RSS hash key too small");
1561 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1562 		return rte_flow_error_set(error, ENOTSUP,
1563 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1564 					  &rss->key_len,
1565 					  "RSS hash key too large");
1566 	if (rss->queue_num > priv->config.ind_table_max_size)
1567 		return rte_flow_error_set(error, ENOTSUP,
1568 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 					  &rss->queue_num,
1570 					  "number of queues too large");
1571 	if (rss->types & MLX5_RSS_HF_MASK)
1572 		return rte_flow_error_set(error, ENOTSUP,
1573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1574 					  &rss->types,
1575 					  "some RSS protocols are not"
1576 					  " supported");
1577 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1578 	    !(rss->types & ETH_RSS_IP))
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1581 					  "L3 partial RSS requested but L3 RSS"
1582 					  " type not specified");
1583 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1584 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1585 		return rte_flow_error_set(error, EINVAL,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1587 					  "L4 partial RSS requested but L4 RSS"
1588 					  " type not specified");
1589 	if (!priv->rxqs_n)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 					  NULL, "No Rx queues configured");
1593 	if (!rss->queue_num)
1594 		return rte_flow_error_set(error, EINVAL,
1595 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1596 					  NULL, "No queues configured");
1597 	for (i = 0; i != rss->queue_num; ++i) {
1598 		struct mlx5_rxq_ctrl *rxq_ctrl;
1599 
1600 		if (rss->queue[i] >= priv->rxqs_n)
1601 			return rte_flow_error_set
1602 				(error, EINVAL,
1603 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1604 				 &rss->queue[i], "queue index out of range");
1605 		if (!(*priv->rxqs)[rss->queue[i]])
1606 			return rte_flow_error_set
1607 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1608 				 &rss->queue[i], "queue is not configured");
1609 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1610 					struct mlx5_rxq_ctrl, rxq);
1611 		if (i == 0)
1612 			rxq_type = rxq_ctrl->type;
1613 		if (rxq_type != rxq_ctrl->type)
1614 			return rte_flow_error_set
1615 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1616 				 &rss->queue[i],
1617 				 "combining hairpin and regular RSS queues is not supported");
1618 	}
1619 	return 0;
1620 }
1621 
1622 /*
1623  * Validate the rss action.
1624  *
1625  * @param[in] action
1626  *   Pointer to the queue action.
1627  * @param[in] action_flags
1628  *   Bit-fields that holds the actions detected until now.
1629  * @param[in] dev
1630  *   Pointer to the Ethernet device structure.
1631  * @param[in] attr
1632  *   Attributes of flow that includes this action.
1633  * @param[in] item_flags
1634  *   Items that were detected.
1635  * @param[out] error
1636  *   Pointer to error structure.
1637  *
1638  * @return
1639  *   0 on success, a negative errno value otherwise and rte_errno is set.
1640  */
1641 int
1642 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1643 			      uint64_t action_flags,
1644 			      struct rte_eth_dev *dev,
1645 			      const struct rte_flow_attr *attr,
1646 			      uint64_t item_flags,
1647 			      struct rte_flow_error *error)
1648 {
1649 	const struct rte_flow_action_rss *rss = action->conf;
1650 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1651 	int ret;
1652 
1653 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1654 		return rte_flow_error_set(error, EINVAL,
1655 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1656 					  "can't have 2 fate actions"
1657 					  " in same flow");
1658 	ret = mlx5_validate_action_rss(dev, action, error);
1659 	if (ret)
1660 		return ret;
1661 	if (attr->egress)
1662 		return rte_flow_error_set(error, ENOTSUP,
1663 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1664 					  "rss action not supported for "
1665 					  "egress");
1666 	if (rss->level > 1 && !tunnel)
1667 		return rte_flow_error_set(error, EINVAL,
1668 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1669 					  "inner RSS is not supported for "
1670 					  "non-tunnel flows");
1671 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1672 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1673 		return rte_flow_error_set(error, EINVAL,
1674 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1675 					  "RSS on eCPRI is not supported now");
1676 	}
1677 	return 0;
1678 }
1679 
1680 /*
1681  * Validate the default miss action.
1682  *
1683  * @param[in] action_flags
1684  *   Bit-fields that holds the actions detected until now.
1685  * @param[out] error
1686  *   Pointer to error structure.
1687  *
1688  * @return
1689  *   0 on success, a negative errno value otherwise and rte_errno is set.
1690  */
1691 int
1692 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1693 				const struct rte_flow_attr *attr,
1694 				struct rte_flow_error *error)
1695 {
1696 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1697 		return rte_flow_error_set(error, EINVAL,
1698 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1699 					  "can't have 2 fate actions in"
1700 					  " same flow");
1701 	if (attr->egress)
1702 		return rte_flow_error_set(error, ENOTSUP,
1703 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1704 					  "default miss action not supported "
1705 					  "for egress");
1706 	if (attr->group)
1707 		return rte_flow_error_set(error, ENOTSUP,
1708 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1709 					  "only group 0 is supported");
1710 	if (attr->transfer)
1711 		return rte_flow_error_set(error, ENOTSUP,
1712 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1713 					  NULL, "transfer is not supported");
1714 	return 0;
1715 }
1716 
1717 /*
1718  * Validate the count action.
1719  *
1720  * @param[in] dev
1721  *   Pointer to the Ethernet device structure.
1722  * @param[in] attr
1723  *   Attributes of flow that includes this action.
1724  * @param[out] error
1725  *   Pointer to error structure.
1726  *
1727  * @return
1728  *   0 on success, a negative errno value otherwise and rte_errno is set.
1729  */
1730 int
1731 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1732 				const struct rte_flow_attr *attr,
1733 				struct rte_flow_error *error)
1734 {
1735 	if (attr->egress)
1736 		return rte_flow_error_set(error, ENOTSUP,
1737 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1738 					  "count action not supported for "
1739 					  "egress");
1740 	return 0;
1741 }
1742 
1743 /*
1744  * Validate the ASO CT action.
1745  *
1746  * @param[in] dev
1747  *   Pointer to the Ethernet device structure.
1748  * @param[in] conntrack
1749  *   Pointer to the CT action profile.
1750  * @param[out] error
1751  *   Pointer to error structure.
1752  *
1753  * @return
1754  *   0 on success, a negative errno value otherwise and rte_errno is set.
1755  */
1756 int
1757 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1758 			const struct rte_flow_action_conntrack *conntrack,
1759 			struct rte_flow_error *error)
1760 {
1761 	RTE_SET_USED(dev);
1762 
1763 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1764 		return rte_flow_error_set(error, EINVAL,
1765 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1766 					  "Invalid CT state");
1767 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1768 		return rte_flow_error_set(error, EINVAL,
1769 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1770 					  "Invalid last TCP packet flag");
1771 	return 0;
1772 }
1773 
1774 /**
1775  * Verify the @p attributes will be correctly understood by the NIC and store
1776  * them in the @p flow if everything is correct.
1777  *
1778  * @param[in] dev
1779  *   Pointer to the Ethernet device structure.
1780  * @param[in] attributes
1781  *   Pointer to flow attributes
1782  * @param[out] error
1783  *   Pointer to error structure.
1784  *
1785  * @return
1786  *   0 on success, a negative errno value otherwise and rte_errno is set.
1787  */
1788 int
1789 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1790 			      const struct rte_flow_attr *attributes,
1791 			      struct rte_flow_error *error)
1792 {
1793 	struct mlx5_priv *priv = dev->data->dev_private;
1794 	uint32_t priority_max = priv->config.flow_prio - 1;
1795 
1796 	if (attributes->group)
1797 		return rte_flow_error_set(error, ENOTSUP,
1798 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1799 					  NULL, "groups is not supported");
1800 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1801 	    attributes->priority >= priority_max)
1802 		return rte_flow_error_set(error, ENOTSUP,
1803 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1804 					  NULL, "priority out of range");
1805 	if (attributes->egress)
1806 		return rte_flow_error_set(error, ENOTSUP,
1807 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1808 					  "egress is not supported");
1809 	if (attributes->transfer && !priv->config.dv_esw_en)
1810 		return rte_flow_error_set(error, ENOTSUP,
1811 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1812 					  NULL, "transfer is not supported");
1813 	if (!attributes->ingress)
1814 		return rte_flow_error_set(error, EINVAL,
1815 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1816 					  NULL,
1817 					  "ingress attribute is mandatory");
1818 	return 0;
1819 }
1820 
1821 /**
1822  * Validate ICMP6 item.
1823  *
1824  * @param[in] item
1825  *   Item specification.
1826  * @param[in] item_flags
1827  *   Bit-fields that holds the items detected until now.
1828  * @param[in] ext_vlan_sup
1829  *   Whether extended VLAN features are supported or not.
1830  * @param[out] error
1831  *   Pointer to error structure.
1832  *
1833  * @return
1834  *   0 on success, a negative errno value otherwise and rte_errno is set.
1835  */
1836 int
1837 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1838 			       uint64_t item_flags,
1839 			       uint8_t target_protocol,
1840 			       struct rte_flow_error *error)
1841 {
1842 	const struct rte_flow_item_icmp6 *mask = item->mask;
1843 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1844 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1845 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1846 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1847 				      MLX5_FLOW_LAYER_OUTER_L4;
1848 	int ret;
1849 
1850 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1851 		return rte_flow_error_set(error, EINVAL,
1852 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 					  "protocol filtering not compatible"
1854 					  " with ICMP6 layer");
1855 	if (!(item_flags & l3m))
1856 		return rte_flow_error_set(error, EINVAL,
1857 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1858 					  "IPv6 is mandatory to filter on"
1859 					  " ICMP6");
1860 	if (item_flags & l4m)
1861 		return rte_flow_error_set(error, EINVAL,
1862 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1863 					  "multiple L4 layers not supported");
1864 	if (!mask)
1865 		mask = &rte_flow_item_icmp6_mask;
1866 	ret = mlx5_flow_item_acceptable
1867 		(item, (const uint8_t *)mask,
1868 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1869 		 sizeof(struct rte_flow_item_icmp6),
1870 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1871 	if (ret < 0)
1872 		return ret;
1873 	return 0;
1874 }
1875 
1876 /**
1877  * Validate ICMP item.
1878  *
1879  * @param[in] item
1880  *   Item specification.
1881  * @param[in] item_flags
1882  *   Bit-fields that holds the items detected until now.
1883  * @param[out] error
1884  *   Pointer to error structure.
1885  *
1886  * @return
1887  *   0 on success, a negative errno value otherwise and rte_errno is set.
1888  */
1889 int
1890 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1891 			     uint64_t item_flags,
1892 			     uint8_t target_protocol,
1893 			     struct rte_flow_error *error)
1894 {
1895 	const struct rte_flow_item_icmp *mask = item->mask;
1896 	const struct rte_flow_item_icmp nic_mask = {
1897 		.hdr.icmp_type = 0xff,
1898 		.hdr.icmp_code = 0xff,
1899 		.hdr.icmp_ident = RTE_BE16(0xffff),
1900 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1901 	};
1902 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1903 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1904 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1905 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1906 				      MLX5_FLOW_LAYER_OUTER_L4;
1907 	int ret;
1908 
1909 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1910 		return rte_flow_error_set(error, EINVAL,
1911 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1912 					  "protocol filtering not compatible"
1913 					  " with ICMP layer");
1914 	if (!(item_flags & l3m))
1915 		return rte_flow_error_set(error, EINVAL,
1916 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1917 					  "IPv4 is mandatory to filter"
1918 					  " on ICMP");
1919 	if (item_flags & l4m)
1920 		return rte_flow_error_set(error, EINVAL,
1921 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1922 					  "multiple L4 layers not supported");
1923 	if (!mask)
1924 		mask = &nic_mask;
1925 	ret = mlx5_flow_item_acceptable
1926 		(item, (const uint8_t *)mask,
1927 		 (const uint8_t *)&nic_mask,
1928 		 sizeof(struct rte_flow_item_icmp),
1929 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1930 	if (ret < 0)
1931 		return ret;
1932 	return 0;
1933 }
1934 
1935 /**
1936  * Validate Ethernet item.
1937  *
1938  * @param[in] item
1939  *   Item specification.
1940  * @param[in] item_flags
1941  *   Bit-fields that holds the items detected until now.
1942  * @param[out] error
1943  *   Pointer to error structure.
1944  *
1945  * @return
1946  *   0 on success, a negative errno value otherwise and rte_errno is set.
1947  */
1948 int
1949 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1950 			    uint64_t item_flags, bool ext_vlan_sup,
1951 			    struct rte_flow_error *error)
1952 {
1953 	const struct rte_flow_item_eth *mask = item->mask;
1954 	const struct rte_flow_item_eth nic_mask = {
1955 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1956 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1957 		.type = RTE_BE16(0xffff),
1958 		.has_vlan = ext_vlan_sup ? 1 : 0,
1959 	};
1960 	int ret;
1961 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1962 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1963 				       MLX5_FLOW_LAYER_OUTER_L2;
1964 
1965 	if (item_flags & ethm)
1966 		return rte_flow_error_set(error, ENOTSUP,
1967 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1968 					  "multiple L2 layers not supported");
1969 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1970 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1971 		return rte_flow_error_set(error, EINVAL,
1972 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 					  "L2 layer should not follow "
1974 					  "L3 layers");
1975 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1976 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1977 		return rte_flow_error_set(error, EINVAL,
1978 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1979 					  "L2 layer should not follow VLAN");
1980 	if (!mask)
1981 		mask = &rte_flow_item_eth_mask;
1982 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1983 					(const uint8_t *)&nic_mask,
1984 					sizeof(struct rte_flow_item_eth),
1985 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1986 	return ret;
1987 }
1988 
1989 /**
1990  * Validate VLAN item.
1991  *
1992  * @param[in] item
1993  *   Item specification.
1994  * @param[in] item_flags
1995  *   Bit-fields that holds the items detected until now.
1996  * @param[in] dev
1997  *   Ethernet device flow is being created on.
1998  * @param[out] error
1999  *   Pointer to error structure.
2000  *
2001  * @return
2002  *   0 on success, a negative errno value otherwise and rte_errno is set.
2003  */
2004 int
2005 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2006 			     uint64_t item_flags,
2007 			     struct rte_eth_dev *dev,
2008 			     struct rte_flow_error *error)
2009 {
2010 	const struct rte_flow_item_vlan *spec = item->spec;
2011 	const struct rte_flow_item_vlan *mask = item->mask;
2012 	const struct rte_flow_item_vlan nic_mask = {
2013 		.tci = RTE_BE16(UINT16_MAX),
2014 		.inner_type = RTE_BE16(UINT16_MAX),
2015 	};
2016 	uint16_t vlan_tag = 0;
2017 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2018 	int ret;
2019 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2020 					MLX5_FLOW_LAYER_INNER_L4) :
2021 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2022 					MLX5_FLOW_LAYER_OUTER_L4);
2023 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2024 					MLX5_FLOW_LAYER_OUTER_VLAN;
2025 
2026 	if (item_flags & vlanm)
2027 		return rte_flow_error_set(error, EINVAL,
2028 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2029 					  "multiple VLAN layers not supported");
2030 	else if ((item_flags & l34m) != 0)
2031 		return rte_flow_error_set(error, EINVAL,
2032 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2033 					  "VLAN cannot follow L3/L4 layer");
2034 	if (!mask)
2035 		mask = &rte_flow_item_vlan_mask;
2036 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2037 					(const uint8_t *)&nic_mask,
2038 					sizeof(struct rte_flow_item_vlan),
2039 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2040 	if (ret)
2041 		return ret;
2042 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2043 		struct mlx5_priv *priv = dev->data->dev_private;
2044 
2045 		if (priv->vmwa_context) {
2046 			/*
2047 			 * Non-NULL context means we have a virtual machine
2048 			 * and SR-IOV enabled, we have to create VLAN interface
2049 			 * to make hypervisor to setup E-Switch vport
2050 			 * context correctly. We avoid creating the multiple
2051 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2052 			 */
2053 			return rte_flow_error_set(error, EINVAL,
2054 						  RTE_FLOW_ERROR_TYPE_ITEM,
2055 						  item,
2056 						  "VLAN tag mask is not"
2057 						  " supported in virtual"
2058 						  " environment");
2059 		}
2060 	}
2061 	if (spec) {
2062 		vlan_tag = spec->tci;
2063 		vlan_tag &= mask->tci;
2064 	}
2065 	/*
2066 	 * From verbs perspective an empty VLAN is equivalent
2067 	 * to a packet without VLAN layer.
2068 	 */
2069 	if (!vlan_tag)
2070 		return rte_flow_error_set(error, EINVAL,
2071 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2072 					  item->spec,
2073 					  "VLAN cannot be empty");
2074 	return 0;
2075 }
2076 
2077 /**
2078  * Validate IPV4 item.
2079  *
2080  * @param[in] item
2081  *   Item specification.
2082  * @param[in] item_flags
2083  *   Bit-fields that holds the items detected until now.
2084  * @param[in] last_item
2085  *   Previous validated item in the pattern items.
2086  * @param[in] ether_type
2087  *   Type in the ethernet layer header (including dot1q).
2088  * @param[in] acc_mask
2089  *   Acceptable mask, if NULL default internal default mask
2090  *   will be used to check whether item fields are supported.
2091  * @param[in] range_accepted
2092  *   True if range of values is accepted for specific fields, false otherwise.
2093  * @param[out] error
2094  *   Pointer to error structure.
2095  *
2096  * @return
2097  *   0 on success, a negative errno value otherwise and rte_errno is set.
2098  */
2099 int
2100 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2101 			     uint64_t item_flags,
2102 			     uint64_t last_item,
2103 			     uint16_t ether_type,
2104 			     const struct rte_flow_item_ipv4 *acc_mask,
2105 			     bool range_accepted,
2106 			     struct rte_flow_error *error)
2107 {
2108 	const struct rte_flow_item_ipv4 *mask = item->mask;
2109 	const struct rte_flow_item_ipv4 *spec = item->spec;
2110 	const struct rte_flow_item_ipv4 nic_mask = {
2111 		.hdr = {
2112 			.src_addr = RTE_BE32(0xffffffff),
2113 			.dst_addr = RTE_BE32(0xffffffff),
2114 			.type_of_service = 0xff,
2115 			.next_proto_id = 0xff,
2116 		},
2117 	};
2118 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2119 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2120 				      MLX5_FLOW_LAYER_OUTER_L3;
2121 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2122 				      MLX5_FLOW_LAYER_OUTER_L4;
2123 	int ret;
2124 	uint8_t next_proto = 0xFF;
2125 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2126 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2127 				  MLX5_FLOW_LAYER_INNER_VLAN);
2128 
2129 	if ((last_item & l2_vlan) && ether_type &&
2130 	    ether_type != RTE_ETHER_TYPE_IPV4)
2131 		return rte_flow_error_set(error, EINVAL,
2132 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2133 					  "IPv4 cannot follow L2/VLAN layer "
2134 					  "which ether type is not IPv4");
2135 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2136 		if (mask && spec)
2137 			next_proto = mask->hdr.next_proto_id &
2138 				     spec->hdr.next_proto_id;
2139 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2140 			return rte_flow_error_set(error, EINVAL,
2141 						  RTE_FLOW_ERROR_TYPE_ITEM,
2142 						  item,
2143 						  "multiple tunnel "
2144 						  "not supported");
2145 	}
2146 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2147 		return rte_flow_error_set(error, EINVAL,
2148 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2149 					  "wrong tunnel type - IPv6 specified "
2150 					  "but IPv4 item provided");
2151 	if (item_flags & l3m)
2152 		return rte_flow_error_set(error, ENOTSUP,
2153 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2154 					  "multiple L3 layers not supported");
2155 	else if (item_flags & l4m)
2156 		return rte_flow_error_set(error, EINVAL,
2157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 					  "L3 cannot follow an L4 layer.");
2159 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2160 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2161 		return rte_flow_error_set(error, EINVAL,
2162 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2163 					  "L3 cannot follow an NVGRE layer.");
2164 	if (!mask)
2165 		mask = &rte_flow_item_ipv4_mask;
2166 	else if (mask->hdr.next_proto_id != 0 &&
2167 		 mask->hdr.next_proto_id != 0xff)
2168 		return rte_flow_error_set(error, EINVAL,
2169 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2170 					  "partial mask is not supported"
2171 					  " for protocol");
2172 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2173 					acc_mask ? (const uint8_t *)acc_mask
2174 						 : (const uint8_t *)&nic_mask,
2175 					sizeof(struct rte_flow_item_ipv4),
2176 					range_accepted, error);
2177 	if (ret < 0)
2178 		return ret;
2179 	return 0;
2180 }
2181 
2182 /**
2183  * Validate IPV6 item.
2184  *
2185  * @param[in] item
2186  *   Item specification.
2187  * @param[in] item_flags
2188  *   Bit-fields that holds the items detected until now.
2189  * @param[in] last_item
2190  *   Previous validated item in the pattern items.
2191  * @param[in] ether_type
2192  *   Type in the ethernet layer header (including dot1q).
2193  * @param[in] acc_mask
2194  *   Acceptable mask, if NULL default internal default mask
2195  *   will be used to check whether item fields are supported.
2196  * @param[out] error
2197  *   Pointer to error structure.
2198  *
2199  * @return
2200  *   0 on success, a negative errno value otherwise and rte_errno is set.
2201  */
2202 int
2203 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2204 			     uint64_t item_flags,
2205 			     uint64_t last_item,
2206 			     uint16_t ether_type,
2207 			     const struct rte_flow_item_ipv6 *acc_mask,
2208 			     struct rte_flow_error *error)
2209 {
2210 	const struct rte_flow_item_ipv6 *mask = item->mask;
2211 	const struct rte_flow_item_ipv6 *spec = item->spec;
2212 	const struct rte_flow_item_ipv6 nic_mask = {
2213 		.hdr = {
2214 			.src_addr =
2215 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2216 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2217 			.dst_addr =
2218 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2219 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2220 			.vtc_flow = RTE_BE32(0xffffffff),
2221 			.proto = 0xff,
2222 		},
2223 	};
2224 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2225 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2226 				      MLX5_FLOW_LAYER_OUTER_L3;
2227 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2228 				      MLX5_FLOW_LAYER_OUTER_L4;
2229 	int ret;
2230 	uint8_t next_proto = 0xFF;
2231 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2232 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2233 				  MLX5_FLOW_LAYER_INNER_VLAN);
2234 
2235 	if ((last_item & l2_vlan) && ether_type &&
2236 	    ether_type != RTE_ETHER_TYPE_IPV6)
2237 		return rte_flow_error_set(error, EINVAL,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "IPv6 cannot follow L2/VLAN layer "
2240 					  "which ether type is not IPv6");
2241 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2242 		next_proto = spec->hdr.proto;
2243 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2244 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2245 			return rte_flow_error_set(error, EINVAL,
2246 						  RTE_FLOW_ERROR_TYPE_ITEM,
2247 						  item,
2248 						  "multiple tunnel "
2249 						  "not supported");
2250 	}
2251 	if (next_proto == IPPROTO_HOPOPTS  ||
2252 	    next_proto == IPPROTO_ROUTING  ||
2253 	    next_proto == IPPROTO_FRAGMENT ||
2254 	    next_proto == IPPROTO_ESP	   ||
2255 	    next_proto == IPPROTO_AH	   ||
2256 	    next_proto == IPPROTO_DSTOPTS)
2257 		return rte_flow_error_set(error, EINVAL,
2258 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2259 					  "IPv6 proto (next header) should "
2260 					  "not be set as extension header");
2261 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2262 		return rte_flow_error_set(error, EINVAL,
2263 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2264 					  "wrong tunnel type - IPv4 specified "
2265 					  "but IPv6 item provided");
2266 	if (item_flags & l3m)
2267 		return rte_flow_error_set(error, ENOTSUP,
2268 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2269 					  "multiple L3 layers not supported");
2270 	else if (item_flags & l4m)
2271 		return rte_flow_error_set(error, EINVAL,
2272 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2273 					  "L3 cannot follow an L4 layer.");
2274 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2275 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2276 		return rte_flow_error_set(error, EINVAL,
2277 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2278 					  "L3 cannot follow an NVGRE layer.");
2279 	if (!mask)
2280 		mask = &rte_flow_item_ipv6_mask;
2281 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2282 					acc_mask ? (const uint8_t *)acc_mask
2283 						 : (const uint8_t *)&nic_mask,
2284 					sizeof(struct rte_flow_item_ipv6),
2285 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2286 	if (ret < 0)
2287 		return ret;
2288 	return 0;
2289 }
2290 
2291 /**
2292  * Validate UDP item.
2293  *
2294  * @param[in] item
2295  *   Item specification.
2296  * @param[in] item_flags
2297  *   Bit-fields that holds the items detected until now.
2298  * @param[in] target_protocol
2299  *   The next protocol in the previous item.
2300  * @param[in] flow_mask
2301  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2302  * @param[out] error
2303  *   Pointer to error structure.
2304  *
2305  * @return
2306  *   0 on success, a negative errno value otherwise and rte_errno is set.
2307  */
2308 int
2309 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2310 			    uint64_t item_flags,
2311 			    uint8_t target_protocol,
2312 			    struct rte_flow_error *error)
2313 {
2314 	const struct rte_flow_item_udp *mask = item->mask;
2315 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2316 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2317 				      MLX5_FLOW_LAYER_OUTER_L3;
2318 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2319 				      MLX5_FLOW_LAYER_OUTER_L4;
2320 	int ret;
2321 
2322 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2323 		return rte_flow_error_set(error, EINVAL,
2324 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2325 					  "protocol filtering not compatible"
2326 					  " with UDP layer");
2327 	if (!(item_flags & l3m))
2328 		return rte_flow_error_set(error, EINVAL,
2329 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2330 					  "L3 is mandatory to filter on L4");
2331 	if (item_flags & l4m)
2332 		return rte_flow_error_set(error, EINVAL,
2333 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2334 					  "multiple L4 layers not supported");
2335 	if (!mask)
2336 		mask = &rte_flow_item_udp_mask;
2337 	ret = mlx5_flow_item_acceptable
2338 		(item, (const uint8_t *)mask,
2339 		 (const uint8_t *)&rte_flow_item_udp_mask,
2340 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2341 		 error);
2342 	if (ret < 0)
2343 		return ret;
2344 	return 0;
2345 }
2346 
2347 /**
2348  * Validate TCP item.
2349  *
2350  * @param[in] item
2351  *   Item specification.
2352  * @param[in] item_flags
2353  *   Bit-fields that holds the items detected until now.
2354  * @param[in] target_protocol
2355  *   The next protocol in the previous item.
2356  * @param[out] error
2357  *   Pointer to error structure.
2358  *
2359  * @return
2360  *   0 on success, a negative errno value otherwise and rte_errno is set.
2361  */
2362 int
2363 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2364 			    uint64_t item_flags,
2365 			    uint8_t target_protocol,
2366 			    const struct rte_flow_item_tcp *flow_mask,
2367 			    struct rte_flow_error *error)
2368 {
2369 	const struct rte_flow_item_tcp *mask = item->mask;
2370 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2371 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2372 				      MLX5_FLOW_LAYER_OUTER_L3;
2373 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2374 				      MLX5_FLOW_LAYER_OUTER_L4;
2375 	int ret;
2376 
2377 	MLX5_ASSERT(flow_mask);
2378 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2379 		return rte_flow_error_set(error, EINVAL,
2380 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2381 					  "protocol filtering not compatible"
2382 					  " with TCP layer");
2383 	if (!(item_flags & l3m))
2384 		return rte_flow_error_set(error, EINVAL,
2385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2386 					  "L3 is mandatory to filter on L4");
2387 	if (item_flags & l4m)
2388 		return rte_flow_error_set(error, EINVAL,
2389 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2390 					  "multiple L4 layers not supported");
2391 	if (!mask)
2392 		mask = &rte_flow_item_tcp_mask;
2393 	ret = mlx5_flow_item_acceptable
2394 		(item, (const uint8_t *)mask,
2395 		 (const uint8_t *)flow_mask,
2396 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2397 		 error);
2398 	if (ret < 0)
2399 		return ret;
2400 	return 0;
2401 }
2402 
2403 /**
2404  * Validate VXLAN item.
2405  *
2406  * @param[in] item
2407  *   Item specification.
2408  * @param[in] item_flags
2409  *   Bit-fields that holds the items detected until now.
2410  * @param[in] target_protocol
2411  *   The next protocol in the previous item.
2412  * @param[out] error
2413  *   Pointer to error structure.
2414  *
2415  * @return
2416  *   0 on success, a negative errno value otherwise and rte_errno is set.
2417  */
2418 int
2419 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2420 			      uint64_t item_flags,
2421 			      struct rte_flow_error *error)
2422 {
2423 	const struct rte_flow_item_vxlan *spec = item->spec;
2424 	const struct rte_flow_item_vxlan *mask = item->mask;
2425 	int ret;
2426 	union vni {
2427 		uint32_t vlan_id;
2428 		uint8_t vni[4];
2429 	} id = { .vlan_id = 0, };
2430 
2431 
2432 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2433 		return rte_flow_error_set(error, ENOTSUP,
2434 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2435 					  "multiple tunnel layers not"
2436 					  " supported");
2437 	/*
2438 	 * Verify only UDPv4 is present as defined in
2439 	 * https://tools.ietf.org/html/rfc7348
2440 	 */
2441 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2442 		return rte_flow_error_set(error, EINVAL,
2443 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2444 					  "no outer UDP layer found");
2445 	if (!mask)
2446 		mask = &rte_flow_item_vxlan_mask;
2447 	ret = mlx5_flow_item_acceptable
2448 		(item, (const uint8_t *)mask,
2449 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2450 		 sizeof(struct rte_flow_item_vxlan),
2451 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2452 	if (ret < 0)
2453 		return ret;
2454 	if (spec) {
2455 		memcpy(&id.vni[1], spec->vni, 3);
2456 		memcpy(&id.vni[1], mask->vni, 3);
2457 	}
2458 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2459 		return rte_flow_error_set(error, ENOTSUP,
2460 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 					  "VXLAN tunnel must be fully defined");
2462 	return 0;
2463 }
2464 
2465 /**
2466  * Validate VXLAN_GPE item.
2467  *
2468  * @param[in] item
2469  *   Item specification.
2470  * @param[in] item_flags
2471  *   Bit-fields that holds the items detected until now.
2472  * @param[in] priv
2473  *   Pointer to the private data structure.
2474  * @param[in] target_protocol
2475  *   The next protocol in the previous item.
2476  * @param[out] error
2477  *   Pointer to error structure.
2478  *
2479  * @return
2480  *   0 on success, a negative errno value otherwise and rte_errno is set.
2481  */
2482 int
2483 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2484 				  uint64_t item_flags,
2485 				  struct rte_eth_dev *dev,
2486 				  struct rte_flow_error *error)
2487 {
2488 	struct mlx5_priv *priv = dev->data->dev_private;
2489 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2490 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2491 	int ret;
2492 	union vni {
2493 		uint32_t vlan_id;
2494 		uint8_t vni[4];
2495 	} id = { .vlan_id = 0, };
2496 
2497 	if (!priv->config.l3_vxlan_en)
2498 		return rte_flow_error_set(error, ENOTSUP,
2499 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2500 					  "L3 VXLAN is not enabled by device"
2501 					  " parameter and/or not configured in"
2502 					  " firmware");
2503 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2504 		return rte_flow_error_set(error, ENOTSUP,
2505 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2506 					  "multiple tunnel layers not"
2507 					  " supported");
2508 	/*
2509 	 * Verify only UDPv4 is present as defined in
2510 	 * https://tools.ietf.org/html/rfc7348
2511 	 */
2512 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2513 		return rte_flow_error_set(error, EINVAL,
2514 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2515 					  "no outer UDP layer found");
2516 	if (!mask)
2517 		mask = &rte_flow_item_vxlan_gpe_mask;
2518 	ret = mlx5_flow_item_acceptable
2519 		(item, (const uint8_t *)mask,
2520 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2521 		 sizeof(struct rte_flow_item_vxlan_gpe),
2522 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2523 	if (ret < 0)
2524 		return ret;
2525 	if (spec) {
2526 		if (spec->protocol)
2527 			return rte_flow_error_set(error, ENOTSUP,
2528 						  RTE_FLOW_ERROR_TYPE_ITEM,
2529 						  item,
2530 						  "VxLAN-GPE protocol"
2531 						  " not supported");
2532 		memcpy(&id.vni[1], spec->vni, 3);
2533 		memcpy(&id.vni[1], mask->vni, 3);
2534 	}
2535 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2536 		return rte_flow_error_set(error, ENOTSUP,
2537 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2538 					  "VXLAN-GPE tunnel must be fully"
2539 					  " defined");
2540 	return 0;
2541 }
2542 /**
2543  * Validate GRE Key item.
2544  *
2545  * @param[in] item
2546  *   Item specification.
2547  * @param[in] item_flags
2548  *   Bit flags to mark detected items.
2549  * @param[in] gre_item
2550  *   Pointer to gre_item
2551  * @param[out] error
2552  *   Pointer to error structure.
2553  *
2554  * @return
2555  *   0 on success, a negative errno value otherwise and rte_errno is set.
2556  */
2557 int
2558 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2559 				uint64_t item_flags,
2560 				const struct rte_flow_item *gre_item,
2561 				struct rte_flow_error *error)
2562 {
2563 	const rte_be32_t *mask = item->mask;
2564 	int ret = 0;
2565 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2566 	const struct rte_flow_item_gre *gre_spec;
2567 	const struct rte_flow_item_gre *gre_mask;
2568 
2569 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2570 		return rte_flow_error_set(error, ENOTSUP,
2571 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2572 					  "Multiple GRE key not support");
2573 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2574 		return rte_flow_error_set(error, ENOTSUP,
2575 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2576 					  "No preceding GRE header");
2577 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2578 		return rte_flow_error_set(error, ENOTSUP,
2579 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2580 					  "GRE key following a wrong item");
2581 	gre_mask = gre_item->mask;
2582 	if (!gre_mask)
2583 		gre_mask = &rte_flow_item_gre_mask;
2584 	gre_spec = gre_item->spec;
2585 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2586 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2587 		return rte_flow_error_set(error, EINVAL,
2588 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2589 					  "Key bit must be on");
2590 
2591 	if (!mask)
2592 		mask = &gre_key_default_mask;
2593 	ret = mlx5_flow_item_acceptable
2594 		(item, (const uint8_t *)mask,
2595 		 (const uint8_t *)&gre_key_default_mask,
2596 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2597 	return ret;
2598 }
2599 
2600 /**
2601  * Validate GRE item.
2602  *
2603  * @param[in] item
2604  *   Item specification.
2605  * @param[in] item_flags
2606  *   Bit flags to mark detected items.
2607  * @param[in] target_protocol
2608  *   The next protocol in the previous item.
2609  * @param[out] error
2610  *   Pointer to error structure.
2611  *
2612  * @return
2613  *   0 on success, a negative errno value otherwise and rte_errno is set.
2614  */
2615 int
2616 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2617 			    uint64_t item_flags,
2618 			    uint8_t target_protocol,
2619 			    struct rte_flow_error *error)
2620 {
2621 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2622 	const struct rte_flow_item_gre *mask = item->mask;
2623 	int ret;
2624 	const struct rte_flow_item_gre nic_mask = {
2625 		.c_rsvd0_ver = RTE_BE16(0xB000),
2626 		.protocol = RTE_BE16(UINT16_MAX),
2627 	};
2628 
2629 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2630 		return rte_flow_error_set(error, EINVAL,
2631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2632 					  "protocol filtering not compatible"
2633 					  " with this GRE layer");
2634 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2635 		return rte_flow_error_set(error, ENOTSUP,
2636 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2637 					  "multiple tunnel layers not"
2638 					  " supported");
2639 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2640 		return rte_flow_error_set(error, ENOTSUP,
2641 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2642 					  "L3 Layer is missing");
2643 	if (!mask)
2644 		mask = &rte_flow_item_gre_mask;
2645 	ret = mlx5_flow_item_acceptable
2646 		(item, (const uint8_t *)mask,
2647 		 (const uint8_t *)&nic_mask,
2648 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2649 		 error);
2650 	if (ret < 0)
2651 		return ret;
2652 #ifndef HAVE_MLX5DV_DR
2653 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2654 	if (spec && (spec->protocol & mask->protocol))
2655 		return rte_flow_error_set(error, ENOTSUP,
2656 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2657 					  "without MPLS support the"
2658 					  " specification cannot be used for"
2659 					  " filtering");
2660 #endif
2661 #endif
2662 	return 0;
2663 }
2664 
2665 /**
2666  * Validate Geneve item.
2667  *
2668  * @param[in] item
2669  *   Item specification.
2670  * @param[in] itemFlags
2671  *   Bit-fields that holds the items detected until now.
2672  * @param[in] enPriv
2673  *   Pointer to the private data structure.
2674  * @param[out] error
2675  *   Pointer to error structure.
2676  *
2677  * @return
2678  *   0 on success, a negative errno value otherwise and rte_errno is set.
2679  */
2680 
2681 int
2682 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2683 			       uint64_t item_flags,
2684 			       struct rte_eth_dev *dev,
2685 			       struct rte_flow_error *error)
2686 {
2687 	struct mlx5_priv *priv = dev->data->dev_private;
2688 	const struct rte_flow_item_geneve *spec = item->spec;
2689 	const struct rte_flow_item_geneve *mask = item->mask;
2690 	int ret;
2691 	uint16_t gbhdr;
2692 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2693 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2694 	const struct rte_flow_item_geneve nic_mask = {
2695 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2696 		.vni = "\xff\xff\xff",
2697 		.protocol = RTE_BE16(UINT16_MAX),
2698 	};
2699 
2700 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2701 		return rte_flow_error_set(error, ENOTSUP,
2702 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2703 					  "L3 Geneve is not enabled by device"
2704 					  " parameter and/or not configured in"
2705 					  " firmware");
2706 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2707 		return rte_flow_error_set(error, ENOTSUP,
2708 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2709 					  "multiple tunnel layers not"
2710 					  " supported");
2711 	/*
2712 	 * Verify only UDPv4 is present as defined in
2713 	 * https://tools.ietf.org/html/rfc7348
2714 	 */
2715 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2716 		return rte_flow_error_set(error, EINVAL,
2717 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2718 					  "no outer UDP layer found");
2719 	if (!mask)
2720 		mask = &rte_flow_item_geneve_mask;
2721 	ret = mlx5_flow_item_acceptable
2722 				  (item, (const uint8_t *)mask,
2723 				   (const uint8_t *)&nic_mask,
2724 				   sizeof(struct rte_flow_item_geneve),
2725 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2726 	if (ret)
2727 		return ret;
2728 	if (spec) {
2729 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2730 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2731 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2732 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2733 			return rte_flow_error_set(error, ENOTSUP,
2734 						  RTE_FLOW_ERROR_TYPE_ITEM,
2735 						  item,
2736 						  "Geneve protocol unsupported"
2737 						  " fields are being used");
2738 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2739 			return rte_flow_error_set
2740 					(error, ENOTSUP,
2741 					 RTE_FLOW_ERROR_TYPE_ITEM,
2742 					 item,
2743 					 "Unsupported Geneve options length");
2744 	}
2745 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2746 		return rte_flow_error_set
2747 				    (error, ENOTSUP,
2748 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2749 				     "Geneve tunnel must be fully defined");
2750 	return 0;
2751 }
2752 
2753 /**
2754  * Validate Geneve TLV option item.
2755  *
2756  * @param[in] item
2757  *   Item specification.
2758  * @param[in] last_item
2759  *   Previous validated item in the pattern items.
2760  * @param[in] geneve_item
2761  *   Previous GENEVE item specification.
2762  * @param[in] dev
2763  *   Pointer to the rte_eth_dev structure.
2764  * @param[out] error
2765  *   Pointer to error structure.
2766  *
2767  * @return
2768  *   0 on success, a negative errno value otherwise and rte_errno is set.
2769  */
2770 int
2771 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2772 				   uint64_t last_item,
2773 				   const struct rte_flow_item *geneve_item,
2774 				   struct rte_eth_dev *dev,
2775 				   struct rte_flow_error *error)
2776 {
2777 	struct mlx5_priv *priv = dev->data->dev_private;
2778 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2779 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2780 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2781 	uint8_t data_max_supported =
2782 			hca_attr->max_geneve_tlv_option_data_len * 4;
2783 	struct mlx5_dev_config *config = &priv->config;
2784 	const struct rte_flow_item_geneve *geneve_spec;
2785 	const struct rte_flow_item_geneve *geneve_mask;
2786 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2787 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2788 	unsigned int i;
2789 	unsigned int data_len;
2790 	uint8_t tlv_option_len;
2791 	uint16_t optlen_m, optlen_v;
2792 	const struct rte_flow_item_geneve_opt full_mask = {
2793 		.option_class = RTE_BE16(0xffff),
2794 		.option_type = 0xff,
2795 		.option_len = 0x1f,
2796 	};
2797 
2798 	if (!mask)
2799 		mask = &rte_flow_item_geneve_opt_mask;
2800 	if (!spec)
2801 		return rte_flow_error_set
2802 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2803 			"Geneve TLV opt class/type/length must be specified");
2804 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2805 		return rte_flow_error_set
2806 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 			"Geneve TLV opt length exceeeds the limit (31)");
2808 	/* Check if class type and length masks are full. */
2809 	if (full_mask.option_class != mask->option_class ||
2810 	    full_mask.option_type != mask->option_type ||
2811 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2812 		return rte_flow_error_set
2813 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2814 			"Geneve TLV opt class/type/length masks must be full");
2815 	/* Check if length is supported */
2816 	if ((uint32_t)spec->option_len >
2817 			config->hca_attr.max_geneve_tlv_option_data_len)
2818 		return rte_flow_error_set
2819 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2820 			"Geneve TLV opt length not supported");
2821 	if (config->hca_attr.max_geneve_tlv_options > 1)
2822 		DRV_LOG(DEBUG,
2823 			"max_geneve_tlv_options supports more than 1 option");
2824 	/* Check GENEVE item preceding. */
2825 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2826 		return rte_flow_error_set
2827 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2828 			"Geneve opt item must be preceded with Geneve item");
2829 	geneve_spec = geneve_item->spec;
2830 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2831 					  &rte_flow_item_geneve_mask;
2832 	/* Check if GENEVE TLV option size doesn't exceed option length */
2833 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2834 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2835 		tlv_option_len = spec->option_len & mask->option_len;
2836 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2837 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2838 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2839 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2840 		if ((optlen_v & optlen_m) <= tlv_option_len)
2841 			return rte_flow_error_set
2842 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2843 				 "GENEVE TLV option length exceeds optlen");
2844 	}
2845 	/* Check if length is 0 or data is 0. */
2846 	if (spec->data == NULL || spec->option_len == 0)
2847 		return rte_flow_error_set
2848 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2849 			"Geneve TLV opt with zero data/length not supported");
2850 	/* Check not all data & mask are 0. */
2851 	data_len = spec->option_len * 4;
2852 	if (mask->data == NULL) {
2853 		for (i = 0; i < data_len; i++)
2854 			if (spec->data[i])
2855 				break;
2856 		if (i == data_len)
2857 			return rte_flow_error_set(error, ENOTSUP,
2858 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 				"Can't match on Geneve option data 0");
2860 	} else {
2861 		for (i = 0; i < data_len; i++)
2862 			if (spec->data[i] & mask->data[i])
2863 				break;
2864 		if (i == data_len)
2865 			return rte_flow_error_set(error, ENOTSUP,
2866 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2867 				"Can't match on Geneve option data and mask 0");
2868 		/* Check data mask supported. */
2869 		for (i = data_max_supported; i < data_len ; i++)
2870 			if (mask->data[i])
2871 				return rte_flow_error_set(error, ENOTSUP,
2872 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2873 					"Data mask is of unsupported size");
2874 	}
2875 	/* Check GENEVE option is supported in NIC. */
2876 	if (!config->hca_attr.geneve_tlv_opt)
2877 		return rte_flow_error_set
2878 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2879 			"Geneve TLV opt not supported");
2880 	/* Check if we already have geneve option with different type/class. */
2881 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2882 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2883 	if (geneve_opt_resource != NULL)
2884 		if (geneve_opt_resource->option_class != spec->option_class ||
2885 		    geneve_opt_resource->option_type != spec->option_type ||
2886 		    geneve_opt_resource->length != spec->option_len) {
2887 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2888 			return rte_flow_error_set(error, ENOTSUP,
2889 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2890 				"Only one Geneve TLV option supported");
2891 		}
2892 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2893 	return 0;
2894 }
2895 
2896 /**
2897  * Validate MPLS item.
2898  *
2899  * @param[in] dev
2900  *   Pointer to the rte_eth_dev structure.
2901  * @param[in] item
2902  *   Item specification.
2903  * @param[in] item_flags
2904  *   Bit-fields that holds the items detected until now.
2905  * @param[in] prev_layer
2906  *   The protocol layer indicated in previous item.
2907  * @param[out] error
2908  *   Pointer to error structure.
2909  *
2910  * @return
2911  *   0 on success, a negative errno value otherwise and rte_errno is set.
2912  */
2913 int
2914 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2915 			     const struct rte_flow_item *item __rte_unused,
2916 			     uint64_t item_flags __rte_unused,
2917 			     uint64_t prev_layer __rte_unused,
2918 			     struct rte_flow_error *error)
2919 {
2920 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2921 	const struct rte_flow_item_mpls *mask = item->mask;
2922 	struct mlx5_priv *priv = dev->data->dev_private;
2923 	int ret;
2924 
2925 	if (!priv->config.mpls_en)
2926 		return rte_flow_error_set(error, ENOTSUP,
2927 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2928 					  "MPLS not supported or"
2929 					  " disabled in firmware"
2930 					  " configuration.");
2931 	/* MPLS over UDP, GRE is allowed */
2932 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2933 			    MLX5_FLOW_LAYER_GRE |
2934 			    MLX5_FLOW_LAYER_GRE_KEY)))
2935 		return rte_flow_error_set(error, EINVAL,
2936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2937 					  "protocol filtering not compatible"
2938 					  " with MPLS layer");
2939 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2940 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2941 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2942 		return rte_flow_error_set(error, ENOTSUP,
2943 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2944 					  "multiple tunnel layers not"
2945 					  " supported");
2946 	if (!mask)
2947 		mask = &rte_flow_item_mpls_mask;
2948 	ret = mlx5_flow_item_acceptable
2949 		(item, (const uint8_t *)mask,
2950 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2951 		 sizeof(struct rte_flow_item_mpls),
2952 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2953 	if (ret < 0)
2954 		return ret;
2955 	return 0;
2956 #else
2957 	return rte_flow_error_set(error, ENOTSUP,
2958 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2959 				  "MPLS is not supported by Verbs, please"
2960 				  " update.");
2961 #endif
2962 }
2963 
2964 /**
2965  * Validate NVGRE item.
2966  *
2967  * @param[in] item
2968  *   Item specification.
2969  * @param[in] item_flags
2970  *   Bit flags to mark detected items.
2971  * @param[in] target_protocol
2972  *   The next protocol in the previous item.
2973  * @param[out] error
2974  *   Pointer to error structure.
2975  *
2976  * @return
2977  *   0 on success, a negative errno value otherwise and rte_errno is set.
2978  */
2979 int
2980 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2981 			      uint64_t item_flags,
2982 			      uint8_t target_protocol,
2983 			      struct rte_flow_error *error)
2984 {
2985 	const struct rte_flow_item_nvgre *mask = item->mask;
2986 	int ret;
2987 
2988 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2989 		return rte_flow_error_set(error, EINVAL,
2990 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2991 					  "protocol filtering not compatible"
2992 					  " with this GRE layer");
2993 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2994 		return rte_flow_error_set(error, ENOTSUP,
2995 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2996 					  "multiple tunnel layers not"
2997 					  " supported");
2998 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2999 		return rte_flow_error_set(error, ENOTSUP,
3000 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3001 					  "L3 Layer is missing");
3002 	if (!mask)
3003 		mask = &rte_flow_item_nvgre_mask;
3004 	ret = mlx5_flow_item_acceptable
3005 		(item, (const uint8_t *)mask,
3006 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3007 		 sizeof(struct rte_flow_item_nvgre),
3008 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3009 	if (ret < 0)
3010 		return ret;
3011 	return 0;
3012 }
3013 
3014 /**
3015  * Validate eCPRI item.
3016  *
3017  * @param[in] item
3018  *   Item specification.
3019  * @param[in] item_flags
3020  *   Bit-fields that holds the items detected until now.
3021  * @param[in] last_item
3022  *   Previous validated item in the pattern items.
3023  * @param[in] ether_type
3024  *   Type in the ethernet layer header (including dot1q).
3025  * @param[in] acc_mask
3026  *   Acceptable mask, if NULL default internal default mask
3027  *   will be used to check whether item fields are supported.
3028  * @param[out] error
3029  *   Pointer to error structure.
3030  *
3031  * @return
3032  *   0 on success, a negative errno value otherwise and rte_errno is set.
3033  */
3034 int
3035 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3036 			      uint64_t item_flags,
3037 			      uint64_t last_item,
3038 			      uint16_t ether_type,
3039 			      const struct rte_flow_item_ecpri *acc_mask,
3040 			      struct rte_flow_error *error)
3041 {
3042 	const struct rte_flow_item_ecpri *mask = item->mask;
3043 	const struct rte_flow_item_ecpri nic_mask = {
3044 		.hdr = {
3045 			.common = {
3046 				.u32 =
3047 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3048 					.type = 0xFF,
3049 					}).u32),
3050 			},
3051 			.dummy[0] = 0xFFFFFFFF,
3052 		},
3053 	};
3054 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3055 					MLX5_FLOW_LAYER_OUTER_VLAN);
3056 	struct rte_flow_item_ecpri mask_lo;
3057 
3058 	if (!(last_item & outer_l2_vlan) &&
3059 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3060 		return rte_flow_error_set(error, EINVAL,
3061 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3062 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3063 	if ((last_item & outer_l2_vlan) && ether_type &&
3064 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3065 		return rte_flow_error_set(error, EINVAL,
3066 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3067 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3068 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3069 		return rte_flow_error_set(error, EINVAL,
3070 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3071 					  "eCPRI with tunnel is not supported right now");
3072 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3073 		return rte_flow_error_set(error, ENOTSUP,
3074 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3075 					  "multiple L3 layers not supported");
3076 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3077 		return rte_flow_error_set(error, EINVAL,
3078 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3079 					  "eCPRI cannot coexist with a TCP layer");
3080 	/* In specification, eCPRI could be over UDP layer. */
3081 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3082 		return rte_flow_error_set(error, EINVAL,
3083 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3084 					  "eCPRI over UDP layer is not yet supported right now");
3085 	/* Mask for type field in common header could be zero. */
3086 	if (!mask)
3087 		mask = &rte_flow_item_ecpri_mask;
3088 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3089 	/* Input mask is in big-endian format. */
3090 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3091 		return rte_flow_error_set(error, EINVAL,
3092 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3093 					  "partial mask is not supported for protocol");
3094 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3095 		return rte_flow_error_set(error, EINVAL,
3096 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3097 					  "message header mask must be after a type mask");
3098 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3099 					 acc_mask ? (const uint8_t *)acc_mask
3100 						  : (const uint8_t *)&nic_mask,
3101 					 sizeof(struct rte_flow_item_ecpri),
3102 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3103 }
3104 
3105 /**
3106  * Release resource related QUEUE/RSS action split.
3107  *
3108  * @param dev
3109  *   Pointer to Ethernet device.
3110  * @param flow
3111  *   Flow to release id's from.
3112  */
3113 static void
3114 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3115 			     struct rte_flow *flow)
3116 {
3117 	struct mlx5_priv *priv = dev->data->dev_private;
3118 	uint32_t handle_idx;
3119 	struct mlx5_flow_handle *dev_handle;
3120 
3121 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3122 		       handle_idx, dev_handle, next)
3123 		if (dev_handle->split_flow_id &&
3124 		    !dev_handle->is_meter_flow_id)
3125 			mlx5_ipool_free(priv->sh->ipool
3126 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3127 					dev_handle->split_flow_id);
3128 }
3129 
3130 static int
3131 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3132 		   const struct rte_flow_attr *attr __rte_unused,
3133 		   const struct rte_flow_item items[] __rte_unused,
3134 		   const struct rte_flow_action actions[] __rte_unused,
3135 		   bool external __rte_unused,
3136 		   int hairpin __rte_unused,
3137 		   struct rte_flow_error *error)
3138 {
3139 	return rte_flow_error_set(error, ENOTSUP,
3140 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3141 }
3142 
3143 static struct mlx5_flow *
3144 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3145 		  const struct rte_flow_attr *attr __rte_unused,
3146 		  const struct rte_flow_item items[] __rte_unused,
3147 		  const struct rte_flow_action actions[] __rte_unused,
3148 		  struct rte_flow_error *error)
3149 {
3150 	rte_flow_error_set(error, ENOTSUP,
3151 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3152 	return NULL;
3153 }
3154 
3155 static int
3156 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3157 		    struct mlx5_flow *dev_flow __rte_unused,
3158 		    const struct rte_flow_attr *attr __rte_unused,
3159 		    const struct rte_flow_item items[] __rte_unused,
3160 		    const struct rte_flow_action actions[] __rte_unused,
3161 		    struct rte_flow_error *error)
3162 {
3163 	return rte_flow_error_set(error, ENOTSUP,
3164 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3165 }
3166 
3167 static int
3168 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3169 		struct rte_flow *flow __rte_unused,
3170 		struct rte_flow_error *error)
3171 {
3172 	return rte_flow_error_set(error, ENOTSUP,
3173 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3174 }
3175 
3176 static void
3177 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3178 		 struct rte_flow *flow __rte_unused)
3179 {
3180 }
3181 
3182 static void
3183 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3184 		  struct rte_flow *flow __rte_unused)
3185 {
3186 }
3187 
3188 static int
3189 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3190 		struct rte_flow *flow __rte_unused,
3191 		const struct rte_flow_action *actions __rte_unused,
3192 		void *data __rte_unused,
3193 		struct rte_flow_error *error)
3194 {
3195 	return rte_flow_error_set(error, ENOTSUP,
3196 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3197 }
3198 
3199 static int
3200 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3201 		      uint32_t domains __rte_unused,
3202 		      uint32_t flags __rte_unused)
3203 {
3204 	return 0;
3205 }
3206 
3207 /* Void driver to protect from null pointer reference. */
3208 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3209 	.validate = flow_null_validate,
3210 	.prepare = flow_null_prepare,
3211 	.translate = flow_null_translate,
3212 	.apply = flow_null_apply,
3213 	.remove = flow_null_remove,
3214 	.destroy = flow_null_destroy,
3215 	.query = flow_null_query,
3216 	.sync_domain = flow_null_sync_domain,
3217 };
3218 
3219 /**
3220  * Select flow driver type according to flow attributes and device
3221  * configuration.
3222  *
3223  * @param[in] dev
3224  *   Pointer to the dev structure.
3225  * @param[in] attr
3226  *   Pointer to the flow attributes.
3227  *
3228  * @return
3229  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3230  */
3231 static enum mlx5_flow_drv_type
3232 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3233 {
3234 	struct mlx5_priv *priv = dev->data->dev_private;
3235 	/* The OS can determine first a specific flow type (DV, VERBS) */
3236 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3237 
3238 	if (type != MLX5_FLOW_TYPE_MAX)
3239 		return type;
3240 	/* If no OS specific type - continue with DV/VERBS selection */
3241 	if (attr->transfer && priv->config.dv_esw_en)
3242 		type = MLX5_FLOW_TYPE_DV;
3243 	if (!attr->transfer)
3244 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3245 						 MLX5_FLOW_TYPE_VERBS;
3246 	return type;
3247 }
3248 
3249 #define flow_get_drv_ops(type) flow_drv_ops[type]
3250 
3251 /**
3252  * Flow driver validation API. This abstracts calling driver specific functions.
3253  * The type of flow driver is determined according to flow attributes.
3254  *
3255  * @param[in] dev
3256  *   Pointer to the dev structure.
3257  * @param[in] attr
3258  *   Pointer to the flow attributes.
3259  * @param[in] items
3260  *   Pointer to the list of items.
3261  * @param[in] actions
3262  *   Pointer to the list of actions.
3263  * @param[in] external
3264  *   This flow rule is created by request external to PMD.
3265  * @param[in] hairpin
3266  *   Number of hairpin TX actions, 0 means classic flow.
3267  * @param[out] error
3268  *   Pointer to the error structure.
3269  *
3270  * @return
3271  *   0 on success, a negative errno value otherwise and rte_errno is set.
3272  */
3273 static inline int
3274 flow_drv_validate(struct rte_eth_dev *dev,
3275 		  const struct rte_flow_attr *attr,
3276 		  const struct rte_flow_item items[],
3277 		  const struct rte_flow_action actions[],
3278 		  bool external, int hairpin, struct rte_flow_error *error)
3279 {
3280 	const struct mlx5_flow_driver_ops *fops;
3281 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3282 
3283 	fops = flow_get_drv_ops(type);
3284 	return fops->validate(dev, attr, items, actions, external,
3285 			      hairpin, error);
3286 }
3287 
3288 /**
3289  * Flow driver preparation API. This abstracts calling driver specific
3290  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3291  * calculates the size of memory required for device flow, allocates the memory,
3292  * initializes the device flow and returns the pointer.
3293  *
3294  * @note
3295  *   This function initializes device flow structure such as dv or verbs in
3296  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3297  *   rest. For example, adding returning device flow to flow->dev_flow list and
3298  *   setting backward reference to the flow should be done out of this function.
3299  *   layers field is not filled either.
3300  *
3301  * @param[in] dev
3302  *   Pointer to the dev structure.
3303  * @param[in] attr
3304  *   Pointer to the flow attributes.
3305  * @param[in] items
3306  *   Pointer to the list of items.
3307  * @param[in] actions
3308  *   Pointer to the list of actions.
3309  * @param[in] flow_idx
3310  *   This memory pool index to the flow.
3311  * @param[out] error
3312  *   Pointer to the error structure.
3313  *
3314  * @return
3315  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3316  */
3317 static inline struct mlx5_flow *
3318 flow_drv_prepare(struct rte_eth_dev *dev,
3319 		 const struct rte_flow *flow,
3320 		 const struct rte_flow_attr *attr,
3321 		 const struct rte_flow_item items[],
3322 		 const struct rte_flow_action actions[],
3323 		 uint32_t flow_idx,
3324 		 struct rte_flow_error *error)
3325 {
3326 	const struct mlx5_flow_driver_ops *fops;
3327 	enum mlx5_flow_drv_type type = flow->drv_type;
3328 	struct mlx5_flow *mlx5_flow = NULL;
3329 
3330 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3331 	fops = flow_get_drv_ops(type);
3332 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3333 	if (mlx5_flow)
3334 		mlx5_flow->flow_idx = flow_idx;
3335 	return mlx5_flow;
3336 }
3337 
3338 /**
3339  * Flow driver translation API. This abstracts calling driver specific
3340  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3341  * translates a generic flow into a driver flow. flow_drv_prepare() must
3342  * precede.
3343  *
3344  * @note
3345  *   dev_flow->layers could be filled as a result of parsing during translation
3346  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3347  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3348  *   flow->actions could be overwritten even though all the expanded dev_flows
3349  *   have the same actions.
3350  *
3351  * @param[in] dev
3352  *   Pointer to the rte dev structure.
3353  * @param[in, out] dev_flow
3354  *   Pointer to the mlx5 flow.
3355  * @param[in] attr
3356  *   Pointer to the flow attributes.
3357  * @param[in] items
3358  *   Pointer to the list of items.
3359  * @param[in] actions
3360  *   Pointer to the list of actions.
3361  * @param[out] error
3362  *   Pointer to the error structure.
3363  *
3364  * @return
3365  *   0 on success, a negative errno value otherwise and rte_errno is set.
3366  */
3367 static inline int
3368 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3369 		   const struct rte_flow_attr *attr,
3370 		   const struct rte_flow_item items[],
3371 		   const struct rte_flow_action actions[],
3372 		   struct rte_flow_error *error)
3373 {
3374 	const struct mlx5_flow_driver_ops *fops;
3375 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3376 
3377 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3378 	fops = flow_get_drv_ops(type);
3379 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3380 }
3381 
3382 /**
3383  * Flow driver apply API. This abstracts calling driver specific functions.
3384  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3385  * translated driver flows on to device. flow_drv_translate() must precede.
3386  *
3387  * @param[in] dev
3388  *   Pointer to Ethernet device structure.
3389  * @param[in, out] flow
3390  *   Pointer to flow structure.
3391  * @param[out] error
3392  *   Pointer to error structure.
3393  *
3394  * @return
3395  *   0 on success, a negative errno value otherwise and rte_errno is set.
3396  */
3397 static inline int
3398 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3399 	       struct rte_flow_error *error)
3400 {
3401 	const struct mlx5_flow_driver_ops *fops;
3402 	enum mlx5_flow_drv_type type = flow->drv_type;
3403 
3404 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3405 	fops = flow_get_drv_ops(type);
3406 	return fops->apply(dev, flow, error);
3407 }
3408 
3409 /**
3410  * Flow driver destroy API. This abstracts calling driver specific functions.
3411  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3412  * on device and releases resources of the flow.
3413  *
3414  * @param[in] dev
3415  *   Pointer to Ethernet device.
3416  * @param[in, out] flow
3417  *   Pointer to flow structure.
3418  */
3419 static inline void
3420 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3421 {
3422 	const struct mlx5_flow_driver_ops *fops;
3423 	enum mlx5_flow_drv_type type = flow->drv_type;
3424 
3425 	flow_mreg_split_qrss_release(dev, flow);
3426 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3427 	fops = flow_get_drv_ops(type);
3428 	fops->destroy(dev, flow);
3429 }
3430 
3431 /**
3432  * Flow driver find RSS policy tbl API. This abstracts calling driver
3433  * specific functions. Parent flow (rte_flow) should have driver
3434  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3435  *
3436  * @param[in] dev
3437  *   Pointer to Ethernet device.
3438  * @param[in, out] flow
3439  *   Pointer to flow structure.
3440  * @param[in] policy
3441  *   Pointer to meter policy table.
3442  * @param[in] rss_desc
3443  *   Pointer to rss_desc
3444  */
3445 static struct mlx5_flow_meter_sub_policy *
3446 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3447 		struct rte_flow *flow,
3448 		struct mlx5_flow_meter_policy *policy,
3449 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3450 {
3451 	const struct mlx5_flow_driver_ops *fops;
3452 	enum mlx5_flow_drv_type type = flow->drv_type;
3453 
3454 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3455 	fops = flow_get_drv_ops(type);
3456 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3457 }
3458 
3459 /**
3460  * Get RSS action from the action list.
3461  *
3462  * @param[in] dev
3463  *   Pointer to Ethernet device.
3464  * @param[in] actions
3465  *   Pointer to the list of actions.
3466  * @param[in] flow
3467  *   Parent flow structure pointer.
3468  *
3469  * @return
3470  *   Pointer to the RSS action if exist, else return NULL.
3471  */
3472 static const struct rte_flow_action_rss*
3473 flow_get_rss_action(struct rte_eth_dev *dev,
3474 		    const struct rte_flow_action actions[])
3475 {
3476 	struct mlx5_priv *priv = dev->data->dev_private;
3477 	const struct rte_flow_action_rss *rss = NULL;
3478 
3479 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3480 		switch (actions->type) {
3481 		case RTE_FLOW_ACTION_TYPE_RSS:
3482 			rss = actions->conf;
3483 			break;
3484 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3485 		{
3486 			const struct rte_flow_action_sample *sample =
3487 								actions->conf;
3488 			const struct rte_flow_action *act = sample->actions;
3489 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3490 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3491 					rss = act->conf;
3492 			break;
3493 		}
3494 		case RTE_FLOW_ACTION_TYPE_METER:
3495 		{
3496 			uint32_t mtr_idx;
3497 			struct mlx5_flow_meter_info *fm;
3498 			struct mlx5_flow_meter_policy *policy;
3499 			const struct rte_flow_action_meter *mtr = actions->conf;
3500 
3501 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3502 			if (fm) {
3503 				policy = mlx5_flow_meter_policy_find(dev,
3504 						fm->policy_id, NULL);
3505 				if (policy && policy->is_rss)
3506 					rss =
3507 				policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3508 			}
3509 			break;
3510 		}
3511 		default:
3512 			break;
3513 		}
3514 	}
3515 	return rss;
3516 }
3517 
3518 /**
3519  * Get ASO age action by index.
3520  *
3521  * @param[in] dev
3522  *   Pointer to the Ethernet device structure.
3523  * @param[in] age_idx
3524  *   Index to the ASO age action.
3525  *
3526  * @return
3527  *   The specified ASO age action.
3528  */
3529 struct mlx5_aso_age_action*
3530 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3531 {
3532 	uint16_t pool_idx = age_idx & UINT16_MAX;
3533 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3534 	struct mlx5_priv *priv = dev->data->dev_private;
3535 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3536 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3537 
3538 	return &pool->actions[offset - 1];
3539 }
3540 
3541 /* maps indirect action to translated direct in some actions array */
3542 struct mlx5_translated_action_handle {
3543 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3544 	int index; /**< Index in related array of rte_flow_action. */
3545 };
3546 
3547 /**
3548  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3549  * direct action if translation possible.
3550  * This functionality used to run same execution path for both direct and
3551  * indirect actions on flow create. All necessary preparations for indirect
3552  * action handling should be performed on *handle* actions list returned
3553  * from this call.
3554  *
3555  * @param[in] dev
3556  *   Pointer to Ethernet device.
3557  * @param[in] actions
3558  *   List of actions to translate.
3559  * @param[out] handle
3560  *   List to store translated indirect action object handles.
3561  * @param[in, out] indir_n
3562  *   Size of *handle* array. On return should be updated with number of
3563  *   indirect actions retrieved from the *actions* list.
3564  * @param[out] translated_actions
3565  *   List of actions where all indirect actions were translated to direct
3566  *   if possible. NULL if no translation took place.
3567  * @param[out] error
3568  *   Pointer to the error structure.
3569  *
3570  * @return
3571  *   0 on success, a negative errno value otherwise and rte_errno is set.
3572  */
3573 static int
3574 flow_action_handles_translate(struct rte_eth_dev *dev,
3575 			      const struct rte_flow_action actions[],
3576 			      struct mlx5_translated_action_handle *handle,
3577 			      int *indir_n,
3578 			      struct rte_flow_action **translated_actions,
3579 			      struct rte_flow_error *error)
3580 {
3581 	struct mlx5_priv *priv = dev->data->dev_private;
3582 	struct rte_flow_action *translated = NULL;
3583 	size_t actions_size;
3584 	int n;
3585 	int copied_n = 0;
3586 	struct mlx5_translated_action_handle *handle_end = NULL;
3587 
3588 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3589 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3590 			continue;
3591 		if (copied_n == *indir_n) {
3592 			return rte_flow_error_set
3593 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3594 				 NULL, "too many shared actions");
3595 		}
3596 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3597 			   sizeof(actions[n].conf));
3598 		handle[copied_n].index = n;
3599 		copied_n++;
3600 	}
3601 	n++;
3602 	*indir_n = copied_n;
3603 	if (!copied_n)
3604 		return 0;
3605 	actions_size = sizeof(struct rte_flow_action) * n;
3606 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3607 	if (!translated) {
3608 		rte_errno = ENOMEM;
3609 		return -ENOMEM;
3610 	}
3611 	memcpy(translated, actions, actions_size);
3612 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3613 		struct mlx5_shared_action_rss *shared_rss;
3614 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3615 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3616 		uint32_t idx = act_idx &
3617 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3618 
3619 		switch (type) {
3620 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3621 			shared_rss = mlx5_ipool_get
3622 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3623 			translated[handle->index].type =
3624 				RTE_FLOW_ACTION_TYPE_RSS;
3625 			translated[handle->index].conf =
3626 				&shared_rss->origin;
3627 			break;
3628 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3629 			translated[handle->index].type =
3630 						(enum rte_flow_action_type)
3631 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3632 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3633 			break;
3634 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3635 			if (priv->sh->flow_hit_aso_en) {
3636 				translated[handle->index].type =
3637 					(enum rte_flow_action_type)
3638 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3639 				translated[handle->index].conf =
3640 							 (void *)(uintptr_t)idx;
3641 				break;
3642 			}
3643 			/* Fall-through */
3644 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3645 			if (priv->sh->ct_aso_en) {
3646 				translated[handle->index].type =
3647 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3648 				translated[handle->index].conf =
3649 							 (void *)(uintptr_t)idx;
3650 				break;
3651 			}
3652 			/* Fall-through */
3653 		default:
3654 			mlx5_free(translated);
3655 			return rte_flow_error_set
3656 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3657 				 NULL, "invalid indirect action type");
3658 		}
3659 	}
3660 	*translated_actions = translated;
3661 	return 0;
3662 }
3663 
3664 /**
3665  * Get Shared RSS action from the action list.
3666  *
3667  * @param[in] dev
3668  *   Pointer to Ethernet device.
3669  * @param[in] shared
3670  *   Pointer to the list of actions.
3671  * @param[in] shared_n
3672  *   Actions list length.
3673  *
3674  * @return
3675  *   The MLX5 RSS action ID if exists, otherwise return 0.
3676  */
3677 static uint32_t
3678 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3679 			   struct mlx5_translated_action_handle *handle,
3680 			   int shared_n)
3681 {
3682 	struct mlx5_translated_action_handle *handle_end;
3683 	struct mlx5_priv *priv = dev->data->dev_private;
3684 	struct mlx5_shared_action_rss *shared_rss;
3685 
3686 
3687 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3688 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3689 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3690 		uint32_t idx = act_idx &
3691 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3692 		switch (type) {
3693 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3694 			shared_rss = mlx5_ipool_get
3695 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3696 									   idx);
3697 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3698 					   __ATOMIC_RELAXED);
3699 			return idx;
3700 		default:
3701 			break;
3702 		}
3703 	}
3704 	return 0;
3705 }
3706 
3707 static unsigned int
3708 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3709 {
3710 	const struct rte_flow_item *item;
3711 	unsigned int has_vlan = 0;
3712 
3713 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3714 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3715 			has_vlan = 1;
3716 			break;
3717 		}
3718 	}
3719 	if (has_vlan)
3720 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3721 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3722 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3723 			       MLX5_EXPANSION_ROOT_OUTER;
3724 }
3725 
3726 /**
3727  *  Get layer flags from the prefix flow.
3728  *
3729  *  Some flows may be split to several subflows, the prefix subflow gets the
3730  *  match items and the suffix sub flow gets the actions.
3731  *  Some actions need the user defined match item flags to get the detail for
3732  *  the action.
3733  *  This function helps the suffix flow to get the item layer flags from prefix
3734  *  subflow.
3735  *
3736  * @param[in] dev_flow
3737  *   Pointer the created preifx subflow.
3738  *
3739  * @return
3740  *   The layers get from prefix subflow.
3741  */
3742 static inline uint64_t
3743 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3744 {
3745 	uint64_t layers = 0;
3746 
3747 	/*
3748 	 * Layers bits could be localization, but usually the compiler will
3749 	 * help to do the optimization work for source code.
3750 	 * If no decap actions, use the layers directly.
3751 	 */
3752 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3753 		return dev_flow->handle->layers;
3754 	/* Convert L3 layers with decap action. */
3755 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3756 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3757 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3758 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3759 	/* Convert L4 layers with decap action.  */
3760 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3761 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3762 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3763 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3764 	return layers;
3765 }
3766 
3767 /**
3768  * Get metadata split action information.
3769  *
3770  * @param[in] actions
3771  *   Pointer to the list of actions.
3772  * @param[out] qrss
3773  *   Pointer to the return pointer.
3774  * @param[out] qrss_type
3775  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3776  *   if no QUEUE/RSS is found.
3777  * @param[out] encap_idx
3778  *   Pointer to the index of the encap action if exists, otherwise the last
3779  *   action index.
3780  *
3781  * @return
3782  *   Total number of actions.
3783  */
3784 static int
3785 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3786 				       const struct rte_flow_action **qrss,
3787 				       int *encap_idx)
3788 {
3789 	const struct rte_flow_action_raw_encap *raw_encap;
3790 	int actions_n = 0;
3791 	int raw_decap_idx = -1;
3792 
3793 	*encap_idx = -1;
3794 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3795 		switch (actions->type) {
3796 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3797 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3798 			*encap_idx = actions_n;
3799 			break;
3800 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3801 			raw_decap_idx = actions_n;
3802 			break;
3803 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3804 			raw_encap = actions->conf;
3805 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3806 				*encap_idx = raw_decap_idx != -1 ?
3807 						      raw_decap_idx : actions_n;
3808 			break;
3809 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3810 		case RTE_FLOW_ACTION_TYPE_RSS:
3811 			*qrss = actions;
3812 			break;
3813 		default:
3814 			break;
3815 		}
3816 		actions_n++;
3817 	}
3818 	if (*encap_idx == -1)
3819 		*encap_idx = actions_n;
3820 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3821 	return actions_n + 1;
3822 }
3823 
3824 /**
3825  * Check if the action will change packet.
3826  *
3827  * @param dev
3828  *   Pointer to Ethernet device.
3829  * @param[in] type
3830  *   action type.
3831  *
3832  * @return
3833  *   true if action will change packet, false otherwise.
3834  */
3835 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3836 					  enum rte_flow_action_type type)
3837 {
3838 	struct mlx5_priv *priv = dev->data->dev_private;
3839 
3840 	switch (type) {
3841 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3842 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3843 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3844 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3845 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3846 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3847 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3848 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3849 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3850 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
3851 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3852 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3853 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3854 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3855 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3856 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3857 	case RTE_FLOW_ACTION_TYPE_SET_META:
3858 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
3859 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3860 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3861 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3862 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3863 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3864 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3865 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3866 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3867 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3868 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3869 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3870 		return true;
3871 	case RTE_FLOW_ACTION_TYPE_FLAG:
3872 	case RTE_FLOW_ACTION_TYPE_MARK:
3873 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3874 			return true;
3875 		else
3876 			return false;
3877 	default:
3878 		return false;
3879 	}
3880 }
3881 
3882 /**
3883  * Check meter action from the action list.
3884  *
3885  * @param dev
3886  *   Pointer to Ethernet device.
3887  * @param[in] actions
3888  *   Pointer to the list of actions.
3889  * @param[out] has_mtr
3890  *   Pointer to the meter exist flag.
3891  * @param[out] has_modify
3892  *   Pointer to the flag showing there's packet change action.
3893  * @param[out] meter_id
3894  *   Pointer to the meter id.
3895  *
3896  * @return
3897  *   Total number of actions.
3898  */
3899 static int
3900 flow_check_meter_action(struct rte_eth_dev *dev,
3901 			const struct rte_flow_action actions[],
3902 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3903 {
3904 	const struct rte_flow_action_meter *mtr = NULL;
3905 	int actions_n = 0;
3906 
3907 	MLX5_ASSERT(has_mtr);
3908 	*has_mtr = false;
3909 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3910 		switch (actions->type) {
3911 		case RTE_FLOW_ACTION_TYPE_METER:
3912 			mtr = actions->conf;
3913 			*meter_id = mtr->mtr_id;
3914 			*has_mtr = true;
3915 			break;
3916 		default:
3917 			break;
3918 		}
3919 		if (!*has_mtr)
3920 			*has_modify |= flow_check_modify_action_type(dev,
3921 								actions->type);
3922 		actions_n++;
3923 	}
3924 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3925 	return actions_n + 1;
3926 }
3927 
3928 /**
3929  * Check if the flow should be split due to hairpin.
3930  * The reason for the split is that in current HW we can't
3931  * support encap and push-vlan on Rx, so if a flow contains
3932  * these actions we move it to Tx.
3933  *
3934  * @param dev
3935  *   Pointer to Ethernet device.
3936  * @param[in] attr
3937  *   Flow rule attributes.
3938  * @param[in] actions
3939  *   Associated actions (list terminated by the END action).
3940  *
3941  * @return
3942  *   > 0 the number of actions and the flow should be split,
3943  *   0 when no split required.
3944  */
3945 static int
3946 flow_check_hairpin_split(struct rte_eth_dev *dev,
3947 			 const struct rte_flow_attr *attr,
3948 			 const struct rte_flow_action actions[])
3949 {
3950 	int queue_action = 0;
3951 	int action_n = 0;
3952 	int split = 0;
3953 	const struct rte_flow_action_queue *queue;
3954 	const struct rte_flow_action_rss *rss;
3955 	const struct rte_flow_action_raw_encap *raw_encap;
3956 	const struct rte_eth_hairpin_conf *conf;
3957 
3958 	if (!attr->ingress)
3959 		return 0;
3960 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3961 		switch (actions->type) {
3962 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3963 			queue = actions->conf;
3964 			if (queue == NULL)
3965 				return 0;
3966 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3967 			if (conf == NULL || conf->tx_explicit != 0)
3968 				return 0;
3969 			queue_action = 1;
3970 			action_n++;
3971 			break;
3972 		case RTE_FLOW_ACTION_TYPE_RSS:
3973 			rss = actions->conf;
3974 			if (rss == NULL || rss->queue_num == 0)
3975 				return 0;
3976 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3977 			if (conf == NULL || conf->tx_explicit != 0)
3978 				return 0;
3979 			queue_action = 1;
3980 			action_n++;
3981 			break;
3982 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3983 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3984 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3985 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3986 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3987 			split++;
3988 			action_n++;
3989 			break;
3990 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3991 			raw_encap = actions->conf;
3992 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3993 				split++;
3994 			action_n++;
3995 			break;
3996 		default:
3997 			action_n++;
3998 			break;
3999 		}
4000 	}
4001 	if (split && queue_action)
4002 		return action_n;
4003 	return 0;
4004 }
4005 
4006 /* Declare flow create/destroy prototype in advance. */
4007 static uint32_t
4008 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4009 		 const struct rte_flow_attr *attr,
4010 		 const struct rte_flow_item items[],
4011 		 const struct rte_flow_action actions[],
4012 		 bool external, struct rte_flow_error *error);
4013 
4014 static void
4015 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4016 		  uint32_t flow_idx);
4017 
4018 int
4019 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
4020 		      struct mlx5_hlist_entry *entry,
4021 		      uint64_t key, void *cb_ctx __rte_unused)
4022 {
4023 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4024 		container_of(entry, typeof(*mcp_res), hlist_ent);
4025 
4026 	return mcp_res->mark_id != key;
4027 }
4028 
4029 struct mlx5_hlist_entry *
4030 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
4031 		       void *cb_ctx)
4032 {
4033 	struct rte_eth_dev *dev = list->ctx;
4034 	struct mlx5_priv *priv = dev->data->dev_private;
4035 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4036 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4037 	struct rte_flow_error *error = ctx->error;
4038 	uint32_t idx = 0;
4039 	int ret;
4040 	uint32_t mark_id = key;
4041 	struct rte_flow_attr attr = {
4042 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4043 		.ingress = 1,
4044 	};
4045 	struct mlx5_rte_flow_item_tag tag_spec = {
4046 		.data = mark_id,
4047 	};
4048 	struct rte_flow_item items[] = {
4049 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4050 	};
4051 	struct rte_flow_action_mark ftag = {
4052 		.id = mark_id,
4053 	};
4054 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4055 		.dst = REG_B,
4056 		.src = REG_NON,
4057 	};
4058 	struct rte_flow_action_jump jump = {
4059 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4060 	};
4061 	struct rte_flow_action actions[] = {
4062 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4063 	};
4064 
4065 	/* Fill the register fileds in the flow. */
4066 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4067 	if (ret < 0)
4068 		return NULL;
4069 	tag_spec.id = ret;
4070 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4071 	if (ret < 0)
4072 		return NULL;
4073 	cp_mreg.src = ret;
4074 	/* Provide the full width of FLAG specific value. */
4075 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4076 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4077 	/* Build a new flow. */
4078 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4079 		items[0] = (struct rte_flow_item){
4080 			.type = (enum rte_flow_item_type)
4081 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4082 			.spec = &tag_spec,
4083 		};
4084 		items[1] = (struct rte_flow_item){
4085 			.type = RTE_FLOW_ITEM_TYPE_END,
4086 		};
4087 		actions[0] = (struct rte_flow_action){
4088 			.type = (enum rte_flow_action_type)
4089 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4090 			.conf = &ftag,
4091 		};
4092 		actions[1] = (struct rte_flow_action){
4093 			.type = (enum rte_flow_action_type)
4094 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4095 			.conf = &cp_mreg,
4096 		};
4097 		actions[2] = (struct rte_flow_action){
4098 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4099 			.conf = &jump,
4100 		};
4101 		actions[3] = (struct rte_flow_action){
4102 			.type = RTE_FLOW_ACTION_TYPE_END,
4103 		};
4104 	} else {
4105 		/* Default rule, wildcard match. */
4106 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4107 		items[0] = (struct rte_flow_item){
4108 			.type = RTE_FLOW_ITEM_TYPE_END,
4109 		};
4110 		actions[0] = (struct rte_flow_action){
4111 			.type = (enum rte_flow_action_type)
4112 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4113 			.conf = &cp_mreg,
4114 		};
4115 		actions[1] = (struct rte_flow_action){
4116 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4117 			.conf = &jump,
4118 		};
4119 		actions[2] = (struct rte_flow_action){
4120 			.type = RTE_FLOW_ACTION_TYPE_END,
4121 		};
4122 	}
4123 	/* Build a new entry. */
4124 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4125 	if (!mcp_res) {
4126 		rte_errno = ENOMEM;
4127 		return NULL;
4128 	}
4129 	mcp_res->idx = idx;
4130 	mcp_res->mark_id = mark_id;
4131 	/*
4132 	 * The copy Flows are not included in any list. There
4133 	 * ones are referenced from other Flows and can not
4134 	 * be applied, removed, deleted in ardbitrary order
4135 	 * by list traversing.
4136 	 */
4137 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4138 					 actions, false, error);
4139 	if (!mcp_res->rix_flow) {
4140 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4141 		return NULL;
4142 	}
4143 	return &mcp_res->hlist_ent;
4144 }
4145 
4146 /**
4147  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4148  *
4149  * As mark_id is unique, if there's already a registered flow for the mark_id,
4150  * return by increasing the reference counter of the resource. Otherwise, create
4151  * the resource (mcp_res) and flow.
4152  *
4153  * Flow looks like,
4154  *   - If ingress port is ANY and reg_c[1] is mark_id,
4155  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4156  *
4157  * For default flow (zero mark_id), flow is like,
4158  *   - If ingress port is ANY,
4159  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4160  *
4161  * @param dev
4162  *   Pointer to Ethernet device.
4163  * @param mark_id
4164  *   ID of MARK action, zero means default flow for META.
4165  * @param[out] error
4166  *   Perform verbose error reporting if not NULL.
4167  *
4168  * @return
4169  *   Associated resource on success, NULL otherwise and rte_errno is set.
4170  */
4171 static struct mlx5_flow_mreg_copy_resource *
4172 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4173 			  struct rte_flow_error *error)
4174 {
4175 	struct mlx5_priv *priv = dev->data->dev_private;
4176 	struct mlx5_hlist_entry *entry;
4177 	struct mlx5_flow_cb_ctx ctx = {
4178 		.dev = dev,
4179 		.error = error,
4180 	};
4181 
4182 	/* Check if already registered. */
4183 	MLX5_ASSERT(priv->mreg_cp_tbl);
4184 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4185 	if (!entry)
4186 		return NULL;
4187 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4188 			    hlist_ent);
4189 }
4190 
4191 void
4192 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4193 {
4194 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4195 		container_of(entry, typeof(*mcp_res), hlist_ent);
4196 	struct rte_eth_dev *dev = list->ctx;
4197 	struct mlx5_priv *priv = dev->data->dev_private;
4198 
4199 	MLX5_ASSERT(mcp_res->rix_flow);
4200 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4201 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4202 }
4203 
4204 /**
4205  * Release flow in RX_CP_TBL.
4206  *
4207  * @param dev
4208  *   Pointer to Ethernet device.
4209  * @flow
4210  *   Parent flow for wich copying is provided.
4211  */
4212 static void
4213 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4214 			  struct rte_flow *flow)
4215 {
4216 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4217 	struct mlx5_priv *priv = dev->data->dev_private;
4218 
4219 	if (!flow->rix_mreg_copy)
4220 		return;
4221 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4222 				 flow->rix_mreg_copy);
4223 	if (!mcp_res || !priv->mreg_cp_tbl)
4224 		return;
4225 	MLX5_ASSERT(mcp_res->rix_flow);
4226 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4227 	flow->rix_mreg_copy = 0;
4228 }
4229 
4230 /**
4231  * Remove the default copy action from RX_CP_TBL.
4232  *
4233  * This functions is called in the mlx5_dev_start(). No thread safe
4234  * is guaranteed.
4235  *
4236  * @param dev
4237  *   Pointer to Ethernet device.
4238  */
4239 static void
4240 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4241 {
4242 	struct mlx5_hlist_entry *entry;
4243 	struct mlx5_priv *priv = dev->data->dev_private;
4244 
4245 	/* Check if default flow is registered. */
4246 	if (!priv->mreg_cp_tbl)
4247 		return;
4248 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4249 				  MLX5_DEFAULT_COPY_ID, NULL);
4250 	if (!entry)
4251 		return;
4252 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4253 }
4254 
4255 /**
4256  * Add the default copy action in in RX_CP_TBL.
4257  *
4258  * This functions is called in the mlx5_dev_start(). No thread safe
4259  * is guaranteed.
4260  *
4261  * @param dev
4262  *   Pointer to Ethernet device.
4263  * @param[out] error
4264  *   Perform verbose error reporting if not NULL.
4265  *
4266  * @return
4267  *   0 for success, negative value otherwise and rte_errno is set.
4268  */
4269 static int
4270 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4271 				  struct rte_flow_error *error)
4272 {
4273 	struct mlx5_priv *priv = dev->data->dev_private;
4274 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4275 
4276 	/* Check whether extensive metadata feature is engaged. */
4277 	if (!priv->config.dv_flow_en ||
4278 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4279 	    !mlx5_flow_ext_mreg_supported(dev) ||
4280 	    !priv->sh->dv_regc0_mask)
4281 		return 0;
4282 	/*
4283 	 * Add default mreg copy flow may be called multiple time, but
4284 	 * only be called once in stop. Avoid register it twice.
4285 	 */
4286 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4287 		return 0;
4288 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4289 	if (!mcp_res)
4290 		return -rte_errno;
4291 	return 0;
4292 }
4293 
4294 /**
4295  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4296  *
4297  * All the flow having Q/RSS action should be split by
4298  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4299  * performs the following,
4300  *   - CQE->flow_tag := reg_c[1] (MARK)
4301  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4302  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4303  * but there should be a flow per each MARK ID set by MARK action.
4304  *
4305  * For the aforementioned reason, if there's a MARK action in flow's action
4306  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4307  * the MARK ID to CQE's flow_tag like,
4308  *   - If reg_c[1] is mark_id,
4309  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4310  *
4311  * For SET_META action which stores value in reg_c[0], as the destination is
4312  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4313  * MARK ID means the default flow. The default flow looks like,
4314  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4315  *
4316  * @param dev
4317  *   Pointer to Ethernet device.
4318  * @param flow
4319  *   Pointer to flow structure.
4320  * @param[in] actions
4321  *   Pointer to the list of actions.
4322  * @param[out] error
4323  *   Perform verbose error reporting if not NULL.
4324  *
4325  * @return
4326  *   0 on success, negative value otherwise and rte_errno is set.
4327  */
4328 static int
4329 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4330 			    struct rte_flow *flow,
4331 			    const struct rte_flow_action *actions,
4332 			    struct rte_flow_error *error)
4333 {
4334 	struct mlx5_priv *priv = dev->data->dev_private;
4335 	struct mlx5_dev_config *config = &priv->config;
4336 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4337 	const struct rte_flow_action_mark *mark;
4338 
4339 	/* Check whether extensive metadata feature is engaged. */
4340 	if (!config->dv_flow_en ||
4341 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4342 	    !mlx5_flow_ext_mreg_supported(dev) ||
4343 	    !priv->sh->dv_regc0_mask)
4344 		return 0;
4345 	/* Find MARK action. */
4346 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4347 		switch (actions->type) {
4348 		case RTE_FLOW_ACTION_TYPE_FLAG:
4349 			mcp_res = flow_mreg_add_copy_action
4350 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4351 			if (!mcp_res)
4352 				return -rte_errno;
4353 			flow->rix_mreg_copy = mcp_res->idx;
4354 			return 0;
4355 		case RTE_FLOW_ACTION_TYPE_MARK:
4356 			mark = (const struct rte_flow_action_mark *)
4357 				actions->conf;
4358 			mcp_res =
4359 				flow_mreg_add_copy_action(dev, mark->id, error);
4360 			if (!mcp_res)
4361 				return -rte_errno;
4362 			flow->rix_mreg_copy = mcp_res->idx;
4363 			return 0;
4364 		default:
4365 			break;
4366 		}
4367 	}
4368 	return 0;
4369 }
4370 
4371 #define MLX5_MAX_SPLIT_ACTIONS 24
4372 #define MLX5_MAX_SPLIT_ITEMS 24
4373 
4374 /**
4375  * Split the hairpin flow.
4376  * Since HW can't support encap and push-vlan on Rx, we move these
4377  * actions to Tx.
4378  * If the count action is after the encap then we also
4379  * move the count action. in this case the count will also measure
4380  * the outer bytes.
4381  *
4382  * @param dev
4383  *   Pointer to Ethernet device.
4384  * @param[in] actions
4385  *   Associated actions (list terminated by the END action).
4386  * @param[out] actions_rx
4387  *   Rx flow actions.
4388  * @param[out] actions_tx
4389  *   Tx flow actions..
4390  * @param[out] pattern_tx
4391  *   The pattern items for the Tx flow.
4392  * @param[out] flow_id
4393  *   The flow ID connected to this flow.
4394  *
4395  * @return
4396  *   0 on success.
4397  */
4398 static int
4399 flow_hairpin_split(struct rte_eth_dev *dev,
4400 		   const struct rte_flow_action actions[],
4401 		   struct rte_flow_action actions_rx[],
4402 		   struct rte_flow_action actions_tx[],
4403 		   struct rte_flow_item pattern_tx[],
4404 		   uint32_t flow_id)
4405 {
4406 	const struct rte_flow_action_raw_encap *raw_encap;
4407 	const struct rte_flow_action_raw_decap *raw_decap;
4408 	struct mlx5_rte_flow_action_set_tag *set_tag;
4409 	struct rte_flow_action *tag_action;
4410 	struct mlx5_rte_flow_item_tag *tag_item;
4411 	struct rte_flow_item *item;
4412 	char *addr;
4413 	int encap = 0;
4414 
4415 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4416 		switch (actions->type) {
4417 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4418 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4419 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4420 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4421 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4422 			rte_memcpy(actions_tx, actions,
4423 			       sizeof(struct rte_flow_action));
4424 			actions_tx++;
4425 			break;
4426 		case RTE_FLOW_ACTION_TYPE_COUNT:
4427 			if (encap) {
4428 				rte_memcpy(actions_tx, actions,
4429 					   sizeof(struct rte_flow_action));
4430 				actions_tx++;
4431 			} else {
4432 				rte_memcpy(actions_rx, actions,
4433 					   sizeof(struct rte_flow_action));
4434 				actions_rx++;
4435 			}
4436 			break;
4437 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4438 			raw_encap = actions->conf;
4439 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4440 				memcpy(actions_tx, actions,
4441 				       sizeof(struct rte_flow_action));
4442 				actions_tx++;
4443 				encap = 1;
4444 			} else {
4445 				rte_memcpy(actions_rx, actions,
4446 					   sizeof(struct rte_flow_action));
4447 				actions_rx++;
4448 			}
4449 			break;
4450 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4451 			raw_decap = actions->conf;
4452 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4453 				memcpy(actions_tx, actions,
4454 				       sizeof(struct rte_flow_action));
4455 				actions_tx++;
4456 			} else {
4457 				rte_memcpy(actions_rx, actions,
4458 					   sizeof(struct rte_flow_action));
4459 				actions_rx++;
4460 			}
4461 			break;
4462 		default:
4463 			rte_memcpy(actions_rx, actions,
4464 				   sizeof(struct rte_flow_action));
4465 			actions_rx++;
4466 			break;
4467 		}
4468 	}
4469 	/* Add set meta action and end action for the Rx flow. */
4470 	tag_action = actions_rx;
4471 	tag_action->type = (enum rte_flow_action_type)
4472 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4473 	actions_rx++;
4474 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4475 	actions_rx++;
4476 	set_tag = (void *)actions_rx;
4477 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4478 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4479 		.data = flow_id,
4480 	};
4481 	MLX5_ASSERT(set_tag->id > REG_NON);
4482 	tag_action->conf = set_tag;
4483 	/* Create Tx item list. */
4484 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4485 	addr = (void *)&pattern_tx[2];
4486 	item = pattern_tx;
4487 	item->type = (enum rte_flow_item_type)
4488 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4489 	tag_item = (void *)addr;
4490 	tag_item->data = flow_id;
4491 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4492 	MLX5_ASSERT(set_tag->id > REG_NON);
4493 	item->spec = tag_item;
4494 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4495 	tag_item = (void *)addr;
4496 	tag_item->data = UINT32_MAX;
4497 	tag_item->id = UINT16_MAX;
4498 	item->mask = tag_item;
4499 	item->last = NULL;
4500 	item++;
4501 	item->type = RTE_FLOW_ITEM_TYPE_END;
4502 	return 0;
4503 }
4504 
4505 /**
4506  * The last stage of splitting chain, just creates the subflow
4507  * without any modification.
4508  *
4509  * @param[in] dev
4510  *   Pointer to Ethernet device.
4511  * @param[in] flow
4512  *   Parent flow structure pointer.
4513  * @param[in, out] sub_flow
4514  *   Pointer to return the created subflow, may be NULL.
4515  * @param[in] attr
4516  *   Flow rule attributes.
4517  * @param[in] items
4518  *   Pattern specification (list terminated by the END pattern item).
4519  * @param[in] actions
4520  *   Associated actions (list terminated by the END action).
4521  * @param[in] flow_split_info
4522  *   Pointer to flow split info structure.
4523  * @param[out] error
4524  *   Perform verbose error reporting if not NULL.
4525  * @return
4526  *   0 on success, negative value otherwise
4527  */
4528 static int
4529 flow_create_split_inner(struct rte_eth_dev *dev,
4530 			struct rte_flow *flow,
4531 			struct mlx5_flow **sub_flow,
4532 			const struct rte_flow_attr *attr,
4533 			const struct rte_flow_item items[],
4534 			const struct rte_flow_action actions[],
4535 			struct mlx5_flow_split_info *flow_split_info,
4536 			struct rte_flow_error *error)
4537 {
4538 	struct mlx5_flow *dev_flow;
4539 
4540 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4541 				    flow_split_info->flow_idx, error);
4542 	if (!dev_flow)
4543 		return -rte_errno;
4544 	dev_flow->flow = flow;
4545 	dev_flow->external = flow_split_info->external;
4546 	dev_flow->skip_scale = flow_split_info->skip_scale;
4547 	/* Subflow object was created, we must include one in the list. */
4548 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4549 		      dev_flow->handle, next);
4550 	/*
4551 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4552 	 * flow may need some user defined item layer flags, and pass the
4553 	 * Metadate rxq mark flag to suffix flow as well.
4554 	 */
4555 	if (flow_split_info->prefix_layers)
4556 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4557 	if (flow_split_info->prefix_mark)
4558 		dev_flow->handle->mark = 1;
4559 	if (sub_flow)
4560 		*sub_flow = dev_flow;
4561 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4562 	dev_flow->dv.table_id = flow_split_info->table_id;
4563 #endif
4564 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4565 }
4566 
4567 /**
4568  * Get the sub policy of a meter.
4569  *
4570  * @param[in] dev
4571  *   Pointer to Ethernet device.
4572  * @param[in] flow
4573  *   Parent flow structure pointer.
4574  * @param[in] policy_id;
4575  *   Meter Policy id.
4576  * @param[in] attr
4577  *   Flow rule attributes.
4578  * @param[in] items
4579  *   Pattern specification (list terminated by the END pattern item).
4580  * @param[out] error
4581  *   Perform verbose error reporting if not NULL.
4582  *
4583  * @return
4584  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4585  */
4586 static struct mlx5_flow_meter_sub_policy *
4587 get_meter_sub_policy(struct rte_eth_dev *dev,
4588 		     struct rte_flow *flow,
4589 		     uint32_t policy_id,
4590 		     const struct rte_flow_attr *attr,
4591 		     const struct rte_flow_item items[],
4592 		     struct rte_flow_error *error)
4593 {
4594 	struct mlx5_flow_meter_policy *policy;
4595 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4596 
4597 	policy = mlx5_flow_meter_policy_find(dev, policy_id, NULL);
4598 	if (!policy) {
4599 		rte_flow_error_set(error, EINVAL,
4600 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4601 				   "Failed to find Meter Policy.");
4602 		goto exit;
4603 	}
4604 	if (policy->is_rss ||
4605 		(policy->is_queue &&
4606 	!policy->sub_policys[MLX5_MTR_DOMAIN_INGRESS][0]->rix_hrxq[0])) {
4607 		struct mlx5_flow_workspace *wks =
4608 				mlx5_flow_get_thread_workspace();
4609 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4610 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4611 		uint32_t i;
4612 
4613 		MLX5_ASSERT(wks);
4614 		/**
4615 		 * This is a tmp dev_flow,
4616 		 * no need to register any matcher for it in translate.
4617 		 */
4618 		wks->skip_matcher_reg = 1;
4619 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4620 			struct mlx5_flow dev_flow = {0};
4621 			struct mlx5_flow_handle dev_handle = { {0} };
4622 
4623 			if (policy->is_rss) {
4624 				const void *rss_act =
4625 					policy->act_cnt[i].rss->conf;
4626 				struct rte_flow_action rss_actions[2] = {
4627 					[0] = {
4628 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4629 					.conf = rss_act
4630 					},
4631 					[1] = {
4632 					.type = RTE_FLOW_ACTION_TYPE_END,
4633 					.conf = NULL
4634 					}
4635 				};
4636 
4637 				dev_flow.handle = &dev_handle;
4638 				dev_flow.ingress = attr->ingress;
4639 				dev_flow.flow = flow;
4640 				dev_flow.external = 0;
4641 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4642 				dev_flow.dv.transfer = attr->transfer;
4643 #endif
4644 				/**
4645 				 * Translate RSS action to get rss hash fields.
4646 				 */
4647 				if (flow_drv_translate(dev, &dev_flow, attr,
4648 						items, rss_actions, error))
4649 					goto exit;
4650 				rss_desc_v[i] = wks->rss_desc;
4651 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4652 				rss_desc_v[i].hash_fields =
4653 						dev_flow.hash_fields;
4654 				rss_desc_v[i].queue_num =
4655 						rss_desc_v[i].hash_fields ?
4656 						rss_desc_v[i].queue_num : 1;
4657 				rss_desc_v[i].tunnel =
4658 					!!(dev_flow.handle->layers &
4659 					MLX5_FLOW_LAYER_TUNNEL);
4660 			} else {
4661 				/* This is queue action. */
4662 				rss_desc_v[i] = wks->rss_desc;
4663 				rss_desc_v[i].key_len = 0;
4664 				rss_desc_v[i].hash_fields = 0;
4665 				rss_desc_v[i].queue =
4666 					&policy->act_cnt[i].queue;
4667 				rss_desc_v[i].queue_num = 1;
4668 			}
4669 			rss_desc[i] = &rss_desc_v[i];
4670 		}
4671 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4672 						flow, policy, rss_desc);
4673 	} else {
4674 		enum mlx5_meter_domain mtr_domain =
4675 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4676 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4677 					MLX5_MTR_DOMAIN_INGRESS;
4678 		sub_policy = policy->sub_policys[mtr_domain][0];
4679 	}
4680 	if (!sub_policy) {
4681 		rte_flow_error_set(error, EINVAL,
4682 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4683 			"Failed to get meter sub-policy.");
4684 		goto exit;
4685 	}
4686 exit:
4687 	return sub_policy;
4688 }
4689 
4690 /**
4691  * Split the meter flow.
4692  *
4693  * As meter flow will split to three sub flow, other than meter
4694  * action, the other actions make sense to only meter accepts
4695  * the packet. If it need to be dropped, no other additional
4696  * actions should be take.
4697  *
4698  * One kind of special action which decapsulates the L3 tunnel
4699  * header will be in the prefix sub flow, as not to take the
4700  * L3 tunnel header into account.
4701  *
4702  * @param[in] dev
4703  *   Pointer to Ethernet device.
4704  * @param[in] flow
4705  *   Parent flow structure pointer.
4706  * @param[in] fm
4707  *   Pointer to flow meter structure.
4708  * @param[in] attr
4709  *   Flow rule attributes.
4710  * @param[in] items
4711  *   Pattern specification (list terminated by the END pattern item).
4712  * @param[out] sfx_items
4713  *   Suffix flow match items (list terminated by the END pattern item).
4714  * @param[in] actions
4715  *   Associated actions (list terminated by the END action).
4716  * @param[out] actions_sfx
4717  *   Suffix flow actions.
4718  * @param[out] actions_pre
4719  *   Prefix flow actions.
4720  * @param[out] mtr_flow_id
4721  *   Pointer to meter flow id.
4722  * @param[out] error
4723  *   Perform verbose error reporting if not NULL.
4724  *
4725  * @return
4726  *   0 on success, a negative errno value otherwise and rte_errno is set.
4727  */
4728 static int
4729 flow_meter_split_prep(struct rte_eth_dev *dev,
4730 		      struct rte_flow *flow,
4731 		      struct mlx5_flow_meter_info *fm,
4732 		      const struct rte_flow_attr *attr,
4733 		      const struct rte_flow_item items[],
4734 		      struct rte_flow_item sfx_items[],
4735 		      const struct rte_flow_action actions[],
4736 		      struct rte_flow_action actions_sfx[],
4737 		      struct rte_flow_action actions_pre[],
4738 		      uint32_t *mtr_flow_id,
4739 		      struct rte_flow_error *error)
4740 {
4741 	struct mlx5_priv *priv = dev->data->dev_private;
4742 	struct rte_flow_action *tag_action = NULL;
4743 	struct rte_flow_item *tag_item;
4744 	struct mlx5_rte_flow_action_set_tag *set_tag;
4745 	const struct rte_flow_action_raw_encap *raw_encap;
4746 	const struct rte_flow_action_raw_decap *raw_decap;
4747 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4748 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4749 	uint32_t tag_id = 0;
4750 	struct rte_flow_item *vlan_item_dst = NULL;
4751 	const struct rte_flow_item *vlan_item_src = NULL;
4752 	struct rte_flow_action *hw_mtr_action;
4753 	struct rte_flow_action *action_pre_head = NULL;
4754 	int32_t flow_src_port = priv->representor_id;
4755 	bool mtr_first;
4756 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4757 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4758 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4759 	uint32_t flow_id = 0;
4760 	uint32_t flow_id_reversed = 0;
4761 	uint8_t flow_id_bits = 0;
4762 	int shift;
4763 
4764 	/* Prepare the suffix subflow items. */
4765 	tag_item = sfx_items++;
4766 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4767 		struct mlx5_priv *port_priv;
4768 		const struct rte_flow_item_port_id *pid_v;
4769 		int item_type = items->type;
4770 
4771 		switch (item_type) {
4772 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4773 			pid_v = items->spec;
4774 			MLX5_ASSERT(pid_v);
4775 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
4776 			if (!port_priv)
4777 				return rte_flow_error_set(error,
4778 						rte_errno,
4779 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
4780 						pid_v,
4781 						"Failed to get port info.");
4782 			flow_src_port = port_priv->representor_id;
4783 			memcpy(sfx_items, items, sizeof(*sfx_items));
4784 			sfx_items++;
4785 			break;
4786 		case RTE_FLOW_ITEM_TYPE_VLAN:
4787 			/* Determine if copy vlan item below. */
4788 			vlan_item_src = items;
4789 			vlan_item_dst = sfx_items++;
4790 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
4791 			break;
4792 		default:
4793 			break;
4794 		}
4795 	}
4796 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4797 	sfx_items++;
4798 	mtr_first = priv->sh->meter_aso_en &&
4799 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
4800 	/* For ASO meter, meter must be before tag in TX direction. */
4801 	if (mtr_first) {
4802 		action_pre_head = actions_pre++;
4803 		/* Leave space for tag action. */
4804 		tag_action = actions_pre++;
4805 	}
4806 	/* Prepare the actions for prefix and suffix flow. */
4807 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4808 		struct rte_flow_action *action_cur = NULL;
4809 
4810 		switch (actions->type) {
4811 		case RTE_FLOW_ACTION_TYPE_METER:
4812 			if (mtr_first) {
4813 				action_cur = action_pre_head;
4814 			} else {
4815 				/* Leave space for tag action. */
4816 				tag_action = actions_pre++;
4817 				action_cur = actions_pre++;
4818 			}
4819 			break;
4820 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4821 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4822 			action_cur = actions_pre++;
4823 			break;
4824 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4825 			raw_encap = actions->conf;
4826 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4827 				action_cur = actions_pre++;
4828 			break;
4829 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4830 			raw_decap = actions->conf;
4831 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4832 				action_cur = actions_pre++;
4833 			break;
4834 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4835 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4836 			if (vlan_item_dst && vlan_item_src) {
4837 				memcpy(vlan_item_dst, vlan_item_src,
4838 					sizeof(*vlan_item_dst));
4839 				/*
4840 				 * Convert to internal match item, it is used
4841 				 * for vlan push and set vid.
4842 				 */
4843 				vlan_item_dst->type = (enum rte_flow_item_type)
4844 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4845 			}
4846 			break;
4847 		default:
4848 			break;
4849 		}
4850 		if (!action_cur)
4851 			action_cur = (fm->def_policy) ?
4852 					actions_sfx++ : actions_pre++;
4853 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4854 	}
4855 	/* Add end action to the actions. */
4856 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4857 	if (priv->sh->meter_aso_en) {
4858 		/**
4859 		 * For ASO meter, need to add an extra jump action explicitly,
4860 		 * to jump from meter to policer table.
4861 		 */
4862 		struct mlx5_flow_meter_sub_policy *sub_policy;
4863 		struct mlx5_flow_tbl_data_entry *tbl_data;
4864 
4865 		if (!fm->def_policy) {
4866 			sub_policy = get_meter_sub_policy(dev, flow,
4867 							  fm->policy_id, attr,
4868 							  items, error);
4869 			if (!sub_policy)
4870 				return -rte_errno;
4871 		} else {
4872 			enum mlx5_meter_domain mtr_domain =
4873 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4874 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4875 					MLX5_MTR_DOMAIN_INGRESS;
4876 
4877 			sub_policy =
4878 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4879 		}
4880 		tbl_data = container_of(sub_policy->tbl_rsc,
4881 					struct mlx5_flow_tbl_data_entry, tbl);
4882 		hw_mtr_action = actions_pre++;
4883 		hw_mtr_action->type = (enum rte_flow_action_type)
4884 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4885 		hw_mtr_action->conf = tbl_data->jump.action;
4886 	}
4887 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4888 	actions_pre++;
4889 	if (!tag_action)
4890 		return rte_flow_error_set(error, ENOMEM,
4891 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4892 					"No tag action space.");
4893 	if (!mtr_flow_id) {
4894 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4895 		goto exit;
4896 	}
4897 	/* Only default-policy Meter creates mtr flow id. */
4898 	if (fm->def_policy) {
4899 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4900 		if (!tag_id)
4901 			return rte_flow_error_set(error, ENOMEM,
4902 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4903 					"Failed to allocate meter flow id.");
4904 		flow_id = tag_id - 1;
4905 		flow_id_bits = (!flow_id) ? 1 :
4906 				(MLX5_REG_BITS - __builtin_clz(flow_id));
4907 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4908 		    mtr_reg_bits) {
4909 			mlx5_ipool_free(fm->flow_ipool, tag_id);
4910 			return rte_flow_error_set(error, EINVAL,
4911 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4912 					"Meter flow id exceeds max limit.");
4913 		}
4914 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4915 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4916 	}
4917 	/* Build tag actions and items for meter_id/meter flow_id. */
4918 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4919 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4920 	tag_item_mask = tag_item_spec + 1;
4921 	/* Both flow_id and meter_id share the same register. */
4922 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4923 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4924 							    0, error),
4925 		.offset = mtr_id_offset,
4926 		.length = mtr_reg_bits,
4927 		.data = flow->meter,
4928 	};
4929 	/*
4930 	 * The color Reg bits used by flow_id are growing from
4931 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4932 	 */
4933 	for (shift = 0; shift < flow_id_bits; shift++)
4934 		flow_id_reversed = (flow_id_reversed << 1) |
4935 				((flow_id >> shift) & 0x1);
4936 	set_tag->data |=
4937 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4938 	tag_item_spec->id = set_tag->id;
4939 	tag_item_spec->data = set_tag->data << mtr_id_offset;
4940 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4941 	tag_action->type = (enum rte_flow_action_type)
4942 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4943 	tag_action->conf = set_tag;
4944 	tag_item->type = (enum rte_flow_item_type)
4945 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4946 	tag_item->spec = tag_item_spec;
4947 	tag_item->last = NULL;
4948 	tag_item->mask = tag_item_mask;
4949 exit:
4950 	if (mtr_flow_id)
4951 		*mtr_flow_id = tag_id;
4952 	return 0;
4953 }
4954 
4955 /**
4956  * Split action list having QUEUE/RSS for metadata register copy.
4957  *
4958  * Once Q/RSS action is detected in user's action list, the flow action
4959  * should be split in order to copy metadata registers, which will happen in
4960  * RX_CP_TBL like,
4961  *   - CQE->flow_tag := reg_c[1] (MARK)
4962  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4963  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4964  * This is because the last action of each flow must be a terminal action
4965  * (QUEUE, RSS or DROP).
4966  *
4967  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4968  * stored and kept in the mlx5_flow structure per each sub_flow.
4969  *
4970  * The Q/RSS action is replaced with,
4971  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4972  * And the following JUMP action is added at the end,
4973  *   - JUMP, to RX_CP_TBL.
4974  *
4975  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4976  * flow_create_split_metadata() routine. The flow will look like,
4977  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4978  *
4979  * @param dev
4980  *   Pointer to Ethernet device.
4981  * @param[out] split_actions
4982  *   Pointer to store split actions to jump to CP_TBL.
4983  * @param[in] actions
4984  *   Pointer to the list of original flow actions.
4985  * @param[in] qrss
4986  *   Pointer to the Q/RSS action.
4987  * @param[in] actions_n
4988  *   Number of original actions.
4989  * @param[out] error
4990  *   Perform verbose error reporting if not NULL.
4991  *
4992  * @return
4993  *   non-zero unique flow_id on success, otherwise 0 and
4994  *   error/rte_error are set.
4995  */
4996 static uint32_t
4997 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4998 			  struct rte_flow_action *split_actions,
4999 			  const struct rte_flow_action *actions,
5000 			  const struct rte_flow_action *qrss,
5001 			  int actions_n, struct rte_flow_error *error)
5002 {
5003 	struct mlx5_priv *priv = dev->data->dev_private;
5004 	struct mlx5_rte_flow_action_set_tag *set_tag;
5005 	struct rte_flow_action_jump *jump;
5006 	const int qrss_idx = qrss - actions;
5007 	uint32_t flow_id = 0;
5008 	int ret = 0;
5009 
5010 	/*
5011 	 * Given actions will be split
5012 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5013 	 * - Add jump to mreg CP_TBL.
5014 	 * As a result, there will be one more action.
5015 	 */
5016 	++actions_n;
5017 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5018 	set_tag = (void *)(split_actions + actions_n);
5019 	/*
5020 	 * If tag action is not set to void(it means we are not the meter
5021 	 * suffix flow), add the tag action. Since meter suffix flow already
5022 	 * has the tag added.
5023 	 */
5024 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
5025 		/*
5026 		 * Allocate the new subflow ID. This one is unique within
5027 		 * device and not shared with representors. Otherwise,
5028 		 * we would have to resolve multi-thread access synch
5029 		 * issue. Each flow on the shared device is appended
5030 		 * with source vport identifier, so the resulting
5031 		 * flows will be unique in the shared (by master and
5032 		 * representors) domain even if they have coinciding
5033 		 * IDs.
5034 		 */
5035 		mlx5_ipool_malloc(priv->sh->ipool
5036 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5037 		if (!flow_id)
5038 			return rte_flow_error_set(error, ENOMEM,
5039 						  RTE_FLOW_ERROR_TYPE_ACTION,
5040 						  NULL, "can't allocate id "
5041 						  "for split Q/RSS subflow");
5042 		/* Internal SET_TAG action to set flow ID. */
5043 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5044 			.data = flow_id,
5045 		};
5046 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5047 		if (ret < 0)
5048 			return ret;
5049 		set_tag->id = ret;
5050 		/* Construct new actions array. */
5051 		/* Replace QUEUE/RSS action. */
5052 		split_actions[qrss_idx] = (struct rte_flow_action){
5053 			.type = (enum rte_flow_action_type)
5054 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5055 			.conf = set_tag,
5056 		};
5057 	}
5058 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5059 	jump = (void *)(set_tag + 1);
5060 	*jump = (struct rte_flow_action_jump){
5061 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5062 	};
5063 	split_actions[actions_n - 2] = (struct rte_flow_action){
5064 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5065 		.conf = jump,
5066 	};
5067 	split_actions[actions_n - 1] = (struct rte_flow_action){
5068 		.type = RTE_FLOW_ACTION_TYPE_END,
5069 	};
5070 	return flow_id;
5071 }
5072 
5073 /**
5074  * Extend the given action list for Tx metadata copy.
5075  *
5076  * Copy the given action list to the ext_actions and add flow metadata register
5077  * copy action in order to copy reg_a set by WQE to reg_c[0].
5078  *
5079  * @param[out] ext_actions
5080  *   Pointer to the extended action list.
5081  * @param[in] actions
5082  *   Pointer to the list of actions.
5083  * @param[in] actions_n
5084  *   Number of actions in the list.
5085  * @param[out] error
5086  *   Perform verbose error reporting if not NULL.
5087  * @param[in] encap_idx
5088  *   The encap action inndex.
5089  *
5090  * @return
5091  *   0 on success, negative value otherwise
5092  */
5093 static int
5094 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5095 		       struct rte_flow_action *ext_actions,
5096 		       const struct rte_flow_action *actions,
5097 		       int actions_n, struct rte_flow_error *error,
5098 		       int encap_idx)
5099 {
5100 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5101 		(struct mlx5_flow_action_copy_mreg *)
5102 			(ext_actions + actions_n + 1);
5103 	int ret;
5104 
5105 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5106 	if (ret < 0)
5107 		return ret;
5108 	cp_mreg->dst = ret;
5109 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5110 	if (ret < 0)
5111 		return ret;
5112 	cp_mreg->src = ret;
5113 	if (encap_idx != 0)
5114 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5115 	if (encap_idx == actions_n - 1) {
5116 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5117 			.type = (enum rte_flow_action_type)
5118 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5119 			.conf = cp_mreg,
5120 		};
5121 		ext_actions[actions_n] = (struct rte_flow_action){
5122 			.type = RTE_FLOW_ACTION_TYPE_END,
5123 		};
5124 	} else {
5125 		ext_actions[encap_idx] = (struct rte_flow_action){
5126 			.type = (enum rte_flow_action_type)
5127 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5128 			.conf = cp_mreg,
5129 		};
5130 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5131 				sizeof(*ext_actions) * (actions_n - encap_idx));
5132 	}
5133 	return 0;
5134 }
5135 
5136 /**
5137  * Check the match action from the action list.
5138  *
5139  * @param[in] actions
5140  *   Pointer to the list of actions.
5141  * @param[in] attr
5142  *   Flow rule attributes.
5143  * @param[in] action
5144  *   The action to be check if exist.
5145  * @param[out] match_action_pos
5146  *   Pointer to the position of the matched action if exists, otherwise is -1.
5147  * @param[out] qrss_action_pos
5148  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5149  * @param[out] modify_after_mirror
5150  *   Pointer to the flag of modify action after FDB mirroring.
5151  *
5152  * @return
5153  *   > 0 the total number of actions.
5154  *   0 if not found match action in action list.
5155  */
5156 static int
5157 flow_check_match_action(const struct rte_flow_action actions[],
5158 			const struct rte_flow_attr *attr,
5159 			enum rte_flow_action_type action,
5160 			int *match_action_pos, int *qrss_action_pos,
5161 			int *modify_after_mirror)
5162 {
5163 	const struct rte_flow_action_sample *sample;
5164 	int actions_n = 0;
5165 	uint32_t ratio = 0;
5166 	int sub_type = 0;
5167 	int flag = 0;
5168 	int fdb_mirror = 0;
5169 
5170 	*match_action_pos = -1;
5171 	*qrss_action_pos = -1;
5172 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5173 		if (actions->type == action) {
5174 			flag = 1;
5175 			*match_action_pos = actions_n;
5176 		}
5177 		switch (actions->type) {
5178 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5179 		case RTE_FLOW_ACTION_TYPE_RSS:
5180 			*qrss_action_pos = actions_n;
5181 			break;
5182 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5183 			sample = actions->conf;
5184 			ratio = sample->ratio;
5185 			sub_type = ((const struct rte_flow_action *)
5186 					(sample->actions))->type;
5187 			if (ratio == 1 && attr->transfer)
5188 				fdb_mirror = 1;
5189 			break;
5190 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5191 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5192 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5193 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5194 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5195 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5196 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5197 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5198 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5199 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5200 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5201 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5202 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5203 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5204 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5205 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5206 		case RTE_FLOW_ACTION_TYPE_FLAG:
5207 		case RTE_FLOW_ACTION_TYPE_MARK:
5208 		case RTE_FLOW_ACTION_TYPE_SET_META:
5209 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5210 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5211 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5212 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5213 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5214 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5215 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5216 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5217 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5218 		case RTE_FLOW_ACTION_TYPE_METER:
5219 			if (fdb_mirror)
5220 				*modify_after_mirror = 1;
5221 			break;
5222 		default:
5223 			break;
5224 		}
5225 		actions_n++;
5226 	}
5227 	if (flag && fdb_mirror && !*modify_after_mirror) {
5228 		/* FDB mirroring uses the destination array to implement
5229 		 * instead of FLOW_SAMPLER object.
5230 		 */
5231 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5232 			flag = 0;
5233 	}
5234 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5235 	return flag ? actions_n + 1 : 0;
5236 }
5237 
5238 #define SAMPLE_SUFFIX_ITEM 2
5239 
5240 /**
5241  * Split the sample flow.
5242  *
5243  * As sample flow will split to two sub flow, sample flow with
5244  * sample action, the other actions will move to new suffix flow.
5245  *
5246  * Also add unique tag id with tag action in the sample flow,
5247  * the same tag id will be as match in the suffix flow.
5248  *
5249  * @param dev
5250  *   Pointer to Ethernet device.
5251  * @param[in] add_tag
5252  *   Add extra tag action flag.
5253  * @param[out] sfx_items
5254  *   Suffix flow match items (list terminated by the END pattern item).
5255  * @param[in] actions
5256  *   Associated actions (list terminated by the END action).
5257  * @param[out] actions_sfx
5258  *   Suffix flow actions.
5259  * @param[out] actions_pre
5260  *   Prefix flow actions.
5261  * @param[in] actions_n
5262  *  The total number of actions.
5263  * @param[in] sample_action_pos
5264  *   The sample action position.
5265  * @param[in] qrss_action_pos
5266  *   The Queue/RSS action position.
5267  * @param[in] jump_table
5268  *   Add extra jump action flag.
5269  * @param[out] error
5270  *   Perform verbose error reporting if not NULL.
5271  *
5272  * @return
5273  *   0 on success, or unique flow_id, a negative errno value
5274  *   otherwise and rte_errno is set.
5275  */
5276 static int
5277 flow_sample_split_prep(struct rte_eth_dev *dev,
5278 		       int add_tag,
5279 		       struct rte_flow_item sfx_items[],
5280 		       const struct rte_flow_action actions[],
5281 		       struct rte_flow_action actions_sfx[],
5282 		       struct rte_flow_action actions_pre[],
5283 		       int actions_n,
5284 		       int sample_action_pos,
5285 		       int qrss_action_pos,
5286 		       int jump_table,
5287 		       struct rte_flow_error *error)
5288 {
5289 	struct mlx5_priv *priv = dev->data->dev_private;
5290 	struct mlx5_rte_flow_action_set_tag *set_tag;
5291 	struct mlx5_rte_flow_item_tag *tag_spec;
5292 	struct mlx5_rte_flow_item_tag *tag_mask;
5293 	struct rte_flow_action_jump *jump_action;
5294 	uint32_t tag_id = 0;
5295 	int index;
5296 	int append_index = 0;
5297 	int ret;
5298 
5299 	if (sample_action_pos < 0)
5300 		return rte_flow_error_set(error, EINVAL,
5301 					  RTE_FLOW_ERROR_TYPE_ACTION,
5302 					  NULL, "invalid position of sample "
5303 					  "action in list");
5304 	/* Prepare the actions for prefix and suffix flow. */
5305 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5306 		index = qrss_action_pos;
5307 		/* Put the preceding the Queue/RSS action into prefix flow. */
5308 		if (index != 0)
5309 			memcpy(actions_pre, actions,
5310 			       sizeof(struct rte_flow_action) * index);
5311 		/* Put others preceding the sample action into prefix flow. */
5312 		if (sample_action_pos > index + 1)
5313 			memcpy(actions_pre + index, actions + index + 1,
5314 			       sizeof(struct rte_flow_action) *
5315 			       (sample_action_pos - index - 1));
5316 		index = sample_action_pos - 1;
5317 		/* Put Queue/RSS action into Suffix flow. */
5318 		memcpy(actions_sfx, actions + qrss_action_pos,
5319 		       sizeof(struct rte_flow_action));
5320 		actions_sfx++;
5321 	} else {
5322 		index = sample_action_pos;
5323 		if (index != 0)
5324 			memcpy(actions_pre, actions,
5325 			       sizeof(struct rte_flow_action) * index);
5326 	}
5327 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5328 	 * For CX6DX and above, metadata registers Cx preserve their value,
5329 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5330 	 */
5331 	if (add_tag) {
5332 		/* Prepare the prefix tag action. */
5333 		append_index++;
5334 		set_tag = (void *)(actions_pre + actions_n + append_index);
5335 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5336 		if (ret < 0)
5337 			return ret;
5338 		mlx5_ipool_malloc(priv->sh->ipool
5339 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5340 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5341 			.id = ret,
5342 			.data = tag_id,
5343 		};
5344 		/* Prepare the suffix subflow items. */
5345 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5346 		tag_spec->data = tag_id;
5347 		tag_spec->id = set_tag->id;
5348 		tag_mask = tag_spec + 1;
5349 		tag_mask->data = UINT32_MAX;
5350 		sfx_items[0] = (struct rte_flow_item){
5351 			.type = (enum rte_flow_item_type)
5352 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5353 			.spec = tag_spec,
5354 			.last = NULL,
5355 			.mask = tag_mask,
5356 		};
5357 		sfx_items[1] = (struct rte_flow_item){
5358 			.type = (enum rte_flow_item_type)
5359 				RTE_FLOW_ITEM_TYPE_END,
5360 		};
5361 		/* Prepare the tag action in prefix subflow. */
5362 		actions_pre[index++] =
5363 			(struct rte_flow_action){
5364 			.type = (enum rte_flow_action_type)
5365 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5366 			.conf = set_tag,
5367 		};
5368 	}
5369 	memcpy(actions_pre + index, actions + sample_action_pos,
5370 	       sizeof(struct rte_flow_action));
5371 	index += 1;
5372 	/* For the modify action after the sample action in E-Switch mirroring,
5373 	 * Add the extra jump action in prefix subflow and jump into the next
5374 	 * table, then do the modify action in the new table.
5375 	 */
5376 	if (jump_table) {
5377 		/* Prepare the prefix jump action. */
5378 		append_index++;
5379 		jump_action = (void *)(actions_pre + actions_n + append_index);
5380 		jump_action->group = jump_table;
5381 		actions_pre[index++] =
5382 			(struct rte_flow_action){
5383 			.type = (enum rte_flow_action_type)
5384 				RTE_FLOW_ACTION_TYPE_JUMP,
5385 			.conf = jump_action,
5386 		};
5387 	}
5388 	actions_pre[index] = (struct rte_flow_action){
5389 		.type = (enum rte_flow_action_type)
5390 			RTE_FLOW_ACTION_TYPE_END,
5391 	};
5392 	/* Put the actions after sample into Suffix flow. */
5393 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5394 	       sizeof(struct rte_flow_action) *
5395 	       (actions_n - sample_action_pos - 1));
5396 	return tag_id;
5397 }
5398 
5399 /**
5400  * The splitting for metadata feature.
5401  *
5402  * - Q/RSS action on NIC Rx should be split in order to pass by
5403  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5404  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5405  *
5406  * - All the actions on NIC Tx should have a mreg copy action to
5407  *   copy reg_a from WQE to reg_c[0].
5408  *
5409  * @param dev
5410  *   Pointer to Ethernet device.
5411  * @param[in] flow
5412  *   Parent flow structure pointer.
5413  * @param[in] attr
5414  *   Flow rule attributes.
5415  * @param[in] items
5416  *   Pattern specification (list terminated by the END pattern item).
5417  * @param[in] actions
5418  *   Associated actions (list terminated by the END action).
5419  * @param[in] flow_split_info
5420  *   Pointer to flow split info structure.
5421  * @param[out] error
5422  *   Perform verbose error reporting if not NULL.
5423  * @return
5424  *   0 on success, negative value otherwise
5425  */
5426 static int
5427 flow_create_split_metadata(struct rte_eth_dev *dev,
5428 			   struct rte_flow *flow,
5429 			   const struct rte_flow_attr *attr,
5430 			   const struct rte_flow_item items[],
5431 			   const struct rte_flow_action actions[],
5432 			   struct mlx5_flow_split_info *flow_split_info,
5433 			   struct rte_flow_error *error)
5434 {
5435 	struct mlx5_priv *priv = dev->data->dev_private;
5436 	struct mlx5_dev_config *config = &priv->config;
5437 	const struct rte_flow_action *qrss = NULL;
5438 	struct rte_flow_action *ext_actions = NULL;
5439 	struct mlx5_flow *dev_flow = NULL;
5440 	uint32_t qrss_id = 0;
5441 	int mtr_sfx = 0;
5442 	size_t act_size;
5443 	int actions_n;
5444 	int encap_idx;
5445 	int ret;
5446 
5447 	/* Check whether extensive metadata feature is engaged. */
5448 	if (!config->dv_flow_en ||
5449 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5450 	    !mlx5_flow_ext_mreg_supported(dev))
5451 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5452 					       actions, flow_split_info, error);
5453 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5454 							   &encap_idx);
5455 	if (qrss) {
5456 		/* Exclude hairpin flows from splitting. */
5457 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5458 			const struct rte_flow_action_queue *queue;
5459 
5460 			queue = qrss->conf;
5461 			if (mlx5_rxq_get_type(dev, queue->index) ==
5462 			    MLX5_RXQ_TYPE_HAIRPIN)
5463 				qrss = NULL;
5464 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5465 			const struct rte_flow_action_rss *rss;
5466 
5467 			rss = qrss->conf;
5468 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5469 			    MLX5_RXQ_TYPE_HAIRPIN)
5470 				qrss = NULL;
5471 		}
5472 	}
5473 	if (qrss) {
5474 		/* Check if it is in meter suffix table. */
5475 		mtr_sfx = attr->group == (attr->transfer ?
5476 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5477 			  MLX5_FLOW_TABLE_LEVEL_METER);
5478 		/*
5479 		 * Q/RSS action on NIC Rx should be split in order to pass by
5480 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5481 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5482 		 */
5483 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5484 			   sizeof(struct rte_flow_action_set_tag) +
5485 			   sizeof(struct rte_flow_action_jump);
5486 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5487 					  SOCKET_ID_ANY);
5488 		if (!ext_actions)
5489 			return rte_flow_error_set(error, ENOMEM,
5490 						  RTE_FLOW_ERROR_TYPE_ACTION,
5491 						  NULL, "no memory to split "
5492 						  "metadata flow");
5493 		/*
5494 		 * If we are the suffix flow of meter, tag already exist.
5495 		 * Set the tag action to void.
5496 		 */
5497 		if (mtr_sfx)
5498 			ext_actions[qrss - actions].type =
5499 						RTE_FLOW_ACTION_TYPE_VOID;
5500 		else
5501 			ext_actions[qrss - actions].type =
5502 						(enum rte_flow_action_type)
5503 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5504 		/*
5505 		 * Create the new actions list with removed Q/RSS action
5506 		 * and appended set tag and jump to register copy table
5507 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5508 		 * in advance, because it is needed for set tag action.
5509 		 */
5510 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5511 						    qrss, actions_n, error);
5512 		if (!mtr_sfx && !qrss_id) {
5513 			ret = -rte_errno;
5514 			goto exit;
5515 		}
5516 	} else if (attr->egress && !attr->transfer) {
5517 		/*
5518 		 * All the actions on NIC Tx should have a metadata register
5519 		 * copy action to copy reg_a from WQE to reg_c[meta]
5520 		 */
5521 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5522 			   sizeof(struct mlx5_flow_action_copy_mreg);
5523 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5524 					  SOCKET_ID_ANY);
5525 		if (!ext_actions)
5526 			return rte_flow_error_set(error, ENOMEM,
5527 						  RTE_FLOW_ERROR_TYPE_ACTION,
5528 						  NULL, "no memory to split "
5529 						  "metadata flow");
5530 		/* Create the action list appended with copy register. */
5531 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5532 					     actions_n, error, encap_idx);
5533 		if (ret < 0)
5534 			goto exit;
5535 	}
5536 	/* Add the unmodified original or prefix subflow. */
5537 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5538 				      items, ext_actions ? ext_actions :
5539 				      actions, flow_split_info, error);
5540 	if (ret < 0)
5541 		goto exit;
5542 	MLX5_ASSERT(dev_flow);
5543 	if (qrss) {
5544 		const struct rte_flow_attr q_attr = {
5545 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5546 			.ingress = 1,
5547 		};
5548 		/* Internal PMD action to set register. */
5549 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5550 			.data = qrss_id,
5551 			.id = REG_NON,
5552 		};
5553 		struct rte_flow_item q_items[] = {
5554 			{
5555 				.type = (enum rte_flow_item_type)
5556 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5557 				.spec = &q_tag_spec,
5558 				.last = NULL,
5559 				.mask = NULL,
5560 			},
5561 			{
5562 				.type = RTE_FLOW_ITEM_TYPE_END,
5563 			},
5564 		};
5565 		struct rte_flow_action q_actions[] = {
5566 			{
5567 				.type = qrss->type,
5568 				.conf = qrss->conf,
5569 			},
5570 			{
5571 				.type = RTE_FLOW_ACTION_TYPE_END,
5572 			},
5573 		};
5574 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5575 
5576 		/*
5577 		 * Configure the tag item only if there is no meter subflow.
5578 		 * Since tag is already marked in the meter suffix subflow
5579 		 * we can just use the meter suffix items as is.
5580 		 */
5581 		if (qrss_id) {
5582 			/* Not meter subflow. */
5583 			MLX5_ASSERT(!mtr_sfx);
5584 			/*
5585 			 * Put unique id in prefix flow due to it is destroyed
5586 			 * after suffix flow and id will be freed after there
5587 			 * is no actual flows with this id and identifier
5588 			 * reallocation becomes possible (for example, for
5589 			 * other flows in other threads).
5590 			 */
5591 			dev_flow->handle->split_flow_id = qrss_id;
5592 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5593 						   error);
5594 			if (ret < 0)
5595 				goto exit;
5596 			q_tag_spec.id = ret;
5597 		}
5598 		dev_flow = NULL;
5599 		/* Add suffix subflow to execute Q/RSS. */
5600 		flow_split_info->prefix_layers = layers;
5601 		flow_split_info->prefix_mark = 0;
5602 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5603 					      &q_attr, mtr_sfx ? items :
5604 					      q_items, q_actions,
5605 					      flow_split_info, error);
5606 		if (ret < 0)
5607 			goto exit;
5608 		/* qrss ID should be freed if failed. */
5609 		qrss_id = 0;
5610 		MLX5_ASSERT(dev_flow);
5611 	}
5612 
5613 exit:
5614 	/*
5615 	 * We do not destroy the partially created sub_flows in case of error.
5616 	 * These ones are included into parent flow list and will be destroyed
5617 	 * by flow_drv_destroy.
5618 	 */
5619 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5620 			qrss_id);
5621 	mlx5_free(ext_actions);
5622 	return ret;
5623 }
5624 
5625 /**
5626  * Create meter internal drop flow with the original pattern.
5627  *
5628  * @param dev
5629  *   Pointer to Ethernet device.
5630  * @param[in] flow
5631  *   Parent flow structure pointer.
5632  * @param[in] attr
5633  *   Flow rule attributes.
5634  * @param[in] items
5635  *   Pattern specification (list terminated by the END pattern item).
5636  * @param[in] flow_split_info
5637  *   Pointer to flow split info structure.
5638  * @param[in] fm
5639  *   Pointer to flow meter structure.
5640  * @param[out] error
5641  *   Perform verbose error reporting if not NULL.
5642  * @return
5643  *   0 on success, negative value otherwise
5644  */
5645 static uint32_t
5646 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5647 			struct rte_flow *flow,
5648 			const struct rte_flow_attr *attr,
5649 			const struct rte_flow_item items[],
5650 			struct mlx5_flow_split_info *flow_split_info,
5651 			struct mlx5_flow_meter_info *fm,
5652 			struct rte_flow_error *error)
5653 {
5654 	struct mlx5_flow *dev_flow = NULL;
5655 	struct rte_flow_attr drop_attr = *attr;
5656 	struct rte_flow_action drop_actions[3];
5657 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5658 
5659 	MLX5_ASSERT(fm->drop_cnt);
5660 	drop_actions[0].type =
5661 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5662 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5663 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5664 	drop_actions[1].conf = NULL;
5665 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5666 	drop_actions[2].conf = NULL;
5667 	drop_split_info.external = false;
5668 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5669 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5670 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5671 	return flow_create_split_inner(dev, flow, &dev_flow,
5672 				&drop_attr, items, drop_actions,
5673 				&drop_split_info, error);
5674 }
5675 
5676 /**
5677  * The splitting for meter feature.
5678  *
5679  * - The meter flow will be split to two flows as prefix and
5680  *   suffix flow. The packets make sense only it pass the prefix
5681  *   meter action.
5682  *
5683  * - Reg_C_5 is used for the packet to match betweend prefix and
5684  *   suffix flow.
5685  *
5686  * @param dev
5687  *   Pointer to Ethernet device.
5688  * @param[in] flow
5689  *   Parent flow structure pointer.
5690  * @param[in] attr
5691  *   Flow rule attributes.
5692  * @param[in] items
5693  *   Pattern specification (list terminated by the END pattern item).
5694  * @param[in] actions
5695  *   Associated actions (list terminated by the END action).
5696  * @param[in] flow_split_info
5697  *   Pointer to flow split info structure.
5698  * @param[out] error
5699  *   Perform verbose error reporting if not NULL.
5700  * @return
5701  *   0 on success, negative value otherwise
5702  */
5703 static int
5704 flow_create_split_meter(struct rte_eth_dev *dev,
5705 			struct rte_flow *flow,
5706 			const struct rte_flow_attr *attr,
5707 			const struct rte_flow_item items[],
5708 			const struct rte_flow_action actions[],
5709 			struct mlx5_flow_split_info *flow_split_info,
5710 			struct rte_flow_error *error)
5711 {
5712 	struct mlx5_priv *priv = dev->data->dev_private;
5713 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5714 	struct rte_flow_action *sfx_actions = NULL;
5715 	struct rte_flow_action *pre_actions = NULL;
5716 	struct rte_flow_item *sfx_items = NULL;
5717 	struct mlx5_flow *dev_flow = NULL;
5718 	struct rte_flow_attr sfx_attr = *attr;
5719 	struct mlx5_flow_meter_info *fm = NULL;
5720 	uint8_t skip_scale_restore;
5721 	bool has_mtr = false;
5722 	bool has_modify = false;
5723 	bool set_mtr_reg = true;
5724 	uint32_t meter_id = 0;
5725 	uint32_t mtr_idx = 0;
5726 	uint32_t mtr_flow_id = 0;
5727 	size_t act_size;
5728 	size_t item_size;
5729 	int actions_n = 0;
5730 	int ret = 0;
5731 
5732 	if (priv->mtr_en)
5733 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5734 						    &has_modify, &meter_id);
5735 	if (has_mtr) {
5736 		if (flow->meter) {
5737 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5738 			if (!fm)
5739 				return rte_flow_error_set(error, EINVAL,
5740 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5741 						NULL, "Meter not found.");
5742 		} else {
5743 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5744 			if (!fm)
5745 				return rte_flow_error_set(error, EINVAL,
5746 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5747 						NULL, "Meter not found.");
5748 			ret = mlx5_flow_meter_attach(priv, fm,
5749 						     &sfx_attr, error);
5750 			if (ret)
5751 				return -rte_errno;
5752 			flow->meter = mtr_idx;
5753 		}
5754 		MLX5_ASSERT(wks);
5755 		wks->fm = fm;
5756 		/*
5757 		 * If it isn't default-policy Meter, and
5758 		 * 1. There's no action in flow to change
5759 		 *    packet (modify/encap/decap etc.), OR
5760 		 * 2. No drop count needed for this meter.
5761 		 * no need to use regC to save meter id anymore.
5762 		 */
5763 		if (!fm->def_policy && (!has_modify || !fm->drop_cnt))
5764 			set_mtr_reg = false;
5765 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5766 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5767 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5768 		/* Suffix items: tag, vlan, port id, end. */
5769 #define METER_SUFFIX_ITEM 4
5770 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5771 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5772 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5773 					  0, SOCKET_ID_ANY);
5774 		if (!sfx_actions)
5775 			return rte_flow_error_set(error, ENOMEM,
5776 						  RTE_FLOW_ERROR_TYPE_ACTION,
5777 						  NULL, "no memory to split "
5778 						  "meter flow");
5779 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5780 			     act_size);
5781 		/* There's no suffix flow for meter of non-default policy. */
5782 		if (!fm->def_policy)
5783 			pre_actions = sfx_actions + 1;
5784 		else
5785 			pre_actions = sfx_actions + actions_n;
5786 		ret = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5787 					    items, sfx_items, actions,
5788 					    sfx_actions, pre_actions,
5789 					    (set_mtr_reg ? &mtr_flow_id : NULL),
5790 					    error);
5791 		if (ret) {
5792 			ret = -rte_errno;
5793 			goto exit;
5794 		}
5795 		/* Add the prefix subflow. */
5796 		flow_split_info->prefix_mark = 0;
5797 		skip_scale_restore = flow_split_info->skip_scale;
5798 		flow_split_info->skip_scale |=
5799 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5800 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5801 					      attr, items, pre_actions,
5802 					      flow_split_info, error);
5803 		flow_split_info->skip_scale = skip_scale_restore;
5804 		if (ret) {
5805 			if (mtr_flow_id)
5806 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5807 			ret = -rte_errno;
5808 			goto exit;
5809 		}
5810 		if (mtr_flow_id) {
5811 			dev_flow->handle->split_flow_id = mtr_flow_id;
5812 			dev_flow->handle->is_meter_flow_id = 1;
5813 		}
5814 		if (!fm->def_policy) {
5815 			if (!set_mtr_reg && fm->drop_cnt)
5816 				ret =
5817 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5818 							&sfx_attr, items,
5819 							flow_split_info,
5820 							fm, error);
5821 			goto exit;
5822 		}
5823 		/* Setting the sfx group atrr. */
5824 		sfx_attr.group = sfx_attr.transfer ?
5825 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5826 				 MLX5_FLOW_TABLE_LEVEL_METER;
5827 		flow_split_info->prefix_layers =
5828 				flow_get_prefix_layer_flags(dev_flow);
5829 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5830 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5831 	}
5832 	/* Add the prefix subflow. */
5833 	ret = flow_create_split_metadata(dev, flow,
5834 					 &sfx_attr, sfx_items ?
5835 					 sfx_items : items,
5836 					 sfx_actions ? sfx_actions : actions,
5837 					 flow_split_info, error);
5838 exit:
5839 	if (sfx_actions)
5840 		mlx5_free(sfx_actions);
5841 	return ret;
5842 }
5843 
5844 /**
5845  * The splitting for sample feature.
5846  *
5847  * Once Sample action is detected in the action list, the flow actions should
5848  * be split into prefix sub flow and suffix sub flow.
5849  *
5850  * The original items remain in the prefix sub flow, all actions preceding the
5851  * sample action and the sample action itself will be copied to the prefix
5852  * sub flow, the actions following the sample action will be copied to the
5853  * suffix sub flow, Queue action always be located in the suffix sub flow.
5854  *
5855  * In order to make the packet from prefix sub flow matches with suffix sub
5856  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5857  * flow uses tag item with the unique flow id.
5858  *
5859  * @param dev
5860  *   Pointer to Ethernet device.
5861  * @param[in] flow
5862  *   Parent flow structure pointer.
5863  * @param[in] attr
5864  *   Flow rule attributes.
5865  * @param[in] items
5866  *   Pattern specification (list terminated by the END pattern item).
5867  * @param[in] actions
5868  *   Associated actions (list terminated by the END action).
5869  * @param[in] flow_split_info
5870  *   Pointer to flow split info structure.
5871  * @param[out] error
5872  *   Perform verbose error reporting if not NULL.
5873  * @return
5874  *   0 on success, negative value otherwise
5875  */
5876 static int
5877 flow_create_split_sample(struct rte_eth_dev *dev,
5878 			 struct rte_flow *flow,
5879 			 const struct rte_flow_attr *attr,
5880 			 const struct rte_flow_item items[],
5881 			 const struct rte_flow_action actions[],
5882 			 struct mlx5_flow_split_info *flow_split_info,
5883 			 struct rte_flow_error *error)
5884 {
5885 	struct mlx5_priv *priv = dev->data->dev_private;
5886 	struct rte_flow_action *sfx_actions = NULL;
5887 	struct rte_flow_action *pre_actions = NULL;
5888 	struct rte_flow_item *sfx_items = NULL;
5889 	struct mlx5_flow *dev_flow = NULL;
5890 	struct rte_flow_attr sfx_attr = *attr;
5891 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5892 	struct mlx5_flow_dv_sample_resource *sample_res;
5893 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5894 	struct mlx5_flow_tbl_resource *sfx_tbl;
5895 #endif
5896 	size_t act_size;
5897 	size_t item_size;
5898 	uint32_t fdb_tx = 0;
5899 	int32_t tag_id = 0;
5900 	int actions_n = 0;
5901 	int sample_action_pos;
5902 	int qrss_action_pos;
5903 	int add_tag = 0;
5904 	int modify_after_mirror = 0;
5905 	uint16_t jump_table = 0;
5906 	const uint32_t next_ft_step = 1;
5907 	int ret = 0;
5908 
5909 	if (priv->sampler_en)
5910 		actions_n = flow_check_match_action(actions, attr,
5911 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5912 					&sample_action_pos, &qrss_action_pos,
5913 					&modify_after_mirror);
5914 	if (actions_n) {
5915 		/* The prefix actions must includes sample, tag, end. */
5916 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5917 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5918 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5919 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5920 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5921 					  item_size), 0, SOCKET_ID_ANY);
5922 		if (!sfx_actions)
5923 			return rte_flow_error_set(error, ENOMEM,
5924 						  RTE_FLOW_ERROR_TYPE_ACTION,
5925 						  NULL, "no memory to split "
5926 						  "sample flow");
5927 		/* The representor_id is UINT16_MAX for uplink. */
5928 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
5929 		/*
5930 		 * When reg_c_preserve is set, metadata registers Cx preserve
5931 		 * their value even through packet duplication.
5932 		 */
5933 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5934 		if (add_tag)
5935 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5936 					+ act_size);
5937 		if (modify_after_mirror)
5938 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5939 				     next_ft_step;
5940 		pre_actions = sfx_actions + actions_n;
5941 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5942 						actions, sfx_actions,
5943 						pre_actions, actions_n,
5944 						sample_action_pos,
5945 						qrss_action_pos, jump_table,
5946 						error);
5947 		if (tag_id < 0 || (add_tag && !tag_id)) {
5948 			ret = -rte_errno;
5949 			goto exit;
5950 		}
5951 		if (modify_after_mirror)
5952 			flow_split_info->skip_scale =
5953 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5954 		/* Add the prefix subflow. */
5955 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5956 					      items, pre_actions,
5957 					      flow_split_info, error);
5958 		if (ret) {
5959 			ret = -rte_errno;
5960 			goto exit;
5961 		}
5962 		dev_flow->handle->split_flow_id = tag_id;
5963 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5964 		if (!modify_after_mirror) {
5965 			/* Set the sfx group attr. */
5966 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5967 						dev_flow->dv.sample_res;
5968 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5969 						sample_res->normal_path_tbl;
5970 			sfx_tbl_data = container_of(sfx_tbl,
5971 						struct mlx5_flow_tbl_data_entry,
5972 						tbl);
5973 			sfx_attr.group = sfx_attr.transfer ?
5974 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5975 		} else {
5976 			MLX5_ASSERT(attr->transfer);
5977 			sfx_attr.group = jump_table;
5978 		}
5979 		flow_split_info->prefix_layers =
5980 				flow_get_prefix_layer_flags(dev_flow);
5981 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5982 		/* Suffix group level already be scaled with factor, set
5983 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5984 		 * again in translation.
5985 		 */
5986 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5987 #endif
5988 	}
5989 	/* Add the suffix subflow. */
5990 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5991 				      sfx_items ? sfx_items : items,
5992 				      sfx_actions ? sfx_actions : actions,
5993 				      flow_split_info, error);
5994 exit:
5995 	if (sfx_actions)
5996 		mlx5_free(sfx_actions);
5997 	return ret;
5998 }
5999 
6000 /**
6001  * Split the flow to subflow set. The splitters might be linked
6002  * in the chain, like this:
6003  * flow_create_split_outer() calls:
6004  *   flow_create_split_meter() calls:
6005  *     flow_create_split_metadata(meter_subflow_0) calls:
6006  *       flow_create_split_inner(metadata_subflow_0)
6007  *       flow_create_split_inner(metadata_subflow_1)
6008  *       flow_create_split_inner(metadata_subflow_2)
6009  *     flow_create_split_metadata(meter_subflow_1) calls:
6010  *       flow_create_split_inner(metadata_subflow_0)
6011  *       flow_create_split_inner(metadata_subflow_1)
6012  *       flow_create_split_inner(metadata_subflow_2)
6013  *
6014  * This provide flexible way to add new levels of flow splitting.
6015  * The all of successfully created subflows are included to the
6016  * parent flow dev_flow list.
6017  *
6018  * @param dev
6019  *   Pointer to Ethernet device.
6020  * @param[in] flow
6021  *   Parent flow structure pointer.
6022  * @param[in] attr
6023  *   Flow rule attributes.
6024  * @param[in] items
6025  *   Pattern specification (list terminated by the END pattern item).
6026  * @param[in] actions
6027  *   Associated actions (list terminated by the END action).
6028  * @param[in] flow_split_info
6029  *   Pointer to flow split info structure.
6030  * @param[out] error
6031  *   Perform verbose error reporting if not NULL.
6032  * @return
6033  *   0 on success, negative value otherwise
6034  */
6035 static int
6036 flow_create_split_outer(struct rte_eth_dev *dev,
6037 			struct rte_flow *flow,
6038 			const struct rte_flow_attr *attr,
6039 			const struct rte_flow_item items[],
6040 			const struct rte_flow_action actions[],
6041 			struct mlx5_flow_split_info *flow_split_info,
6042 			struct rte_flow_error *error)
6043 {
6044 	int ret;
6045 
6046 	ret = flow_create_split_sample(dev, flow, attr, items,
6047 				       actions, flow_split_info, error);
6048 	MLX5_ASSERT(ret <= 0);
6049 	return ret;
6050 }
6051 
6052 static inline struct mlx5_flow_tunnel *
6053 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6054 {
6055 	struct mlx5_flow_tunnel *tunnel;
6056 
6057 #pragma GCC diagnostic push
6058 #pragma GCC diagnostic ignored "-Wcast-qual"
6059 	tunnel = (typeof(tunnel))flow->tunnel;
6060 #pragma GCC diagnostic pop
6061 
6062 	return tunnel;
6063 }
6064 
6065 /**
6066  * Adjust flow RSS workspace if needed.
6067  *
6068  * @param wks
6069  *   Pointer to thread flow work space.
6070  * @param rss_desc
6071  *   Pointer to RSS descriptor.
6072  * @param[in] nrssq_num
6073  *   New RSS queue number.
6074  *
6075  * @return
6076  *   0 on success, -1 otherwise and rte_errno is set.
6077  */
6078 static int
6079 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6080 			  struct mlx5_flow_rss_desc *rss_desc,
6081 			  uint32_t nrssq_num)
6082 {
6083 	if (likely(nrssq_num <= wks->rssq_num))
6084 		return 0;
6085 	rss_desc->queue = realloc(rss_desc->queue,
6086 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6087 	if (!rss_desc->queue) {
6088 		rte_errno = ENOMEM;
6089 		return -1;
6090 	}
6091 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6092 	return 0;
6093 }
6094 
6095 /**
6096  * Create a flow and add it to @p list.
6097  *
6098  * @param dev
6099  *   Pointer to Ethernet device.
6100  * @param list
6101  *   Pointer to a TAILQ flow list. If this parameter NULL,
6102  *   no list insertion occurred, flow is just created,
6103  *   this is caller's responsibility to track the
6104  *   created flow.
6105  * @param[in] attr
6106  *   Flow rule attributes.
6107  * @param[in] items
6108  *   Pattern specification (list terminated by the END pattern item).
6109  * @param[in] actions
6110  *   Associated actions (list terminated by the END action).
6111  * @param[in] external
6112  *   This flow rule is created by request external to PMD.
6113  * @param[out] error
6114  *   Perform verbose error reporting if not NULL.
6115  *
6116  * @return
6117  *   A flow index on success, 0 otherwise and rte_errno is set.
6118  */
6119 static uint32_t
6120 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
6121 		 const struct rte_flow_attr *attr,
6122 		 const struct rte_flow_item items[],
6123 		 const struct rte_flow_action original_actions[],
6124 		 bool external, struct rte_flow_error *error)
6125 {
6126 	struct mlx5_priv *priv = dev->data->dev_private;
6127 	struct rte_flow *flow = NULL;
6128 	struct mlx5_flow *dev_flow;
6129 	const struct rte_flow_action_rss *rss = NULL;
6130 	struct mlx5_translated_action_handle
6131 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6132 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6133 	union {
6134 		struct mlx5_flow_expand_rss buf;
6135 		uint8_t buffer[2048];
6136 	} expand_buffer;
6137 	union {
6138 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6139 		uint8_t buffer[2048];
6140 	} actions_rx;
6141 	union {
6142 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6143 		uint8_t buffer[2048];
6144 	} actions_hairpin_tx;
6145 	union {
6146 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6147 		uint8_t buffer[2048];
6148 	} items_tx;
6149 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6150 	struct mlx5_flow_rss_desc *rss_desc;
6151 	const struct rte_flow_action *p_actions_rx;
6152 	uint32_t i;
6153 	uint32_t idx = 0;
6154 	int hairpin_flow;
6155 	struct rte_flow_attr attr_tx = { .priority = 0 };
6156 	const struct rte_flow_action *actions;
6157 	struct rte_flow_action *translated_actions = NULL;
6158 	struct mlx5_flow_tunnel *tunnel;
6159 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6160 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6161 	struct mlx5_flow_split_info flow_split_info = {
6162 		.external = !!external,
6163 		.skip_scale = 0,
6164 		.flow_idx = 0,
6165 		.prefix_mark = 0,
6166 		.prefix_layers = 0,
6167 		.table_id = 0
6168 	};
6169 	int ret;
6170 
6171 	MLX5_ASSERT(wks);
6172 	rss_desc = &wks->rss_desc;
6173 	ret = flow_action_handles_translate(dev, original_actions,
6174 					    indir_actions,
6175 					    &indir_actions_n,
6176 					    &translated_actions, error);
6177 	if (ret < 0) {
6178 		MLX5_ASSERT(translated_actions == NULL);
6179 		return 0;
6180 	}
6181 	actions = translated_actions ? translated_actions : original_actions;
6182 	p_actions_rx = actions;
6183 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6184 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6185 				external, hairpin_flow, error);
6186 	if (ret < 0)
6187 		goto error_before_hairpin_split;
6188 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6189 	if (!flow) {
6190 		rte_errno = ENOMEM;
6191 		goto error_before_hairpin_split;
6192 	}
6193 	if (hairpin_flow > 0) {
6194 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6195 			rte_errno = EINVAL;
6196 			goto error_before_hairpin_split;
6197 		}
6198 		flow_hairpin_split(dev, actions, actions_rx.actions,
6199 				   actions_hairpin_tx.actions, items_tx.items,
6200 				   idx);
6201 		p_actions_rx = actions_rx.actions;
6202 	}
6203 	flow_split_info.flow_idx = idx;
6204 	flow->drv_type = flow_get_drv_type(dev, attr);
6205 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6206 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6207 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6208 	/* RSS Action only works on NIC RX domain */
6209 	if (attr->ingress && !attr->transfer)
6210 		rss = flow_get_rss_action(dev, p_actions_rx);
6211 	if (rss) {
6212 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6213 			return 0;
6214 		/*
6215 		 * The following information is required by
6216 		 * mlx5_flow_hashfields_adjust() in advance.
6217 		 */
6218 		rss_desc->level = rss->level;
6219 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6220 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6221 	}
6222 	flow->dev_handles = 0;
6223 	if (rss && rss->types) {
6224 		unsigned int graph_root;
6225 
6226 		graph_root = find_graph_root(items, rss->level);
6227 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6228 					   items, rss->types,
6229 					   mlx5_support_expansion, graph_root);
6230 		MLX5_ASSERT(ret > 0 &&
6231 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6232 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6233 			for (i = 0; i < buf->entries; ++i)
6234 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6235 		}
6236 	} else {
6237 		buf->entries = 1;
6238 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6239 	}
6240 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6241 						      indir_actions_n);
6242 	for (i = 0; i < buf->entries; ++i) {
6243 		/* Initialize flow split data. */
6244 		flow_split_info.prefix_layers = 0;
6245 		flow_split_info.prefix_mark = 0;
6246 		flow_split_info.skip_scale = 0;
6247 		/*
6248 		 * The splitter may create multiple dev_flows,
6249 		 * depending on configuration. In the simplest
6250 		 * case it just creates unmodified original flow.
6251 		 */
6252 		ret = flow_create_split_outer(dev, flow, attr,
6253 					      buf->entry[i].pattern,
6254 					      p_actions_rx, &flow_split_info,
6255 					      error);
6256 		if (ret < 0)
6257 			goto error;
6258 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6259 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6260 							   p_actions_rx,
6261 							   idx,
6262 							   wks->flows[0].tunnel,
6263 							   &default_miss_ctx,
6264 							   error);
6265 			if (ret < 0) {
6266 				mlx5_free(default_miss_ctx.queue);
6267 				goto error;
6268 			}
6269 		}
6270 	}
6271 	/* Create the tx flow. */
6272 	if (hairpin_flow) {
6273 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6274 		attr_tx.ingress = 0;
6275 		attr_tx.egress = 1;
6276 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6277 					 actions_hairpin_tx.actions,
6278 					 idx, error);
6279 		if (!dev_flow)
6280 			goto error;
6281 		dev_flow->flow = flow;
6282 		dev_flow->external = 0;
6283 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6284 			      dev_flow->handle, next);
6285 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6286 					 items_tx.items,
6287 					 actions_hairpin_tx.actions, error);
6288 		if (ret < 0)
6289 			goto error;
6290 	}
6291 	/*
6292 	 * Update the metadata register copy table. If extensive
6293 	 * metadata feature is enabled and registers are supported
6294 	 * we might create the extra rte_flow for each unique
6295 	 * MARK/FLAG action ID.
6296 	 *
6297 	 * The table is updated for ingress Flows only, because
6298 	 * the egress Flows belong to the different device and
6299 	 * copy table should be updated in peer NIC Rx domain.
6300 	 */
6301 	if (attr->ingress &&
6302 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6303 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6304 		if (ret)
6305 			goto error;
6306 	}
6307 	/*
6308 	 * If the flow is external (from application) OR device is started,
6309 	 * OR mreg discover, then apply immediately.
6310 	 */
6311 	if (external || dev->data->dev_started ||
6312 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6313 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6314 		ret = flow_drv_apply(dev, flow, error);
6315 		if (ret < 0)
6316 			goto error;
6317 	}
6318 	if (list) {
6319 		rte_spinlock_lock(&priv->flow_list_lock);
6320 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6321 			     flow, next);
6322 		rte_spinlock_unlock(&priv->flow_list_lock);
6323 	}
6324 	flow_rxq_flags_set(dev, flow);
6325 	rte_free(translated_actions);
6326 	tunnel = flow_tunnel_from_rule(wks->flows);
6327 	if (tunnel) {
6328 		flow->tunnel = 1;
6329 		flow->tunnel_id = tunnel->tunnel_id;
6330 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6331 		mlx5_free(default_miss_ctx.queue);
6332 	}
6333 	mlx5_flow_pop_thread_workspace();
6334 	return idx;
6335 error:
6336 	MLX5_ASSERT(flow);
6337 	ret = rte_errno; /* Save rte_errno before cleanup. */
6338 	flow_mreg_del_copy_action(dev, flow);
6339 	flow_drv_destroy(dev, flow);
6340 	if (rss_desc->shared_rss)
6341 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6342 			mlx5_ipool_get
6343 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6344 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6345 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6346 	rte_errno = ret; /* Restore rte_errno. */
6347 	ret = rte_errno;
6348 	rte_errno = ret;
6349 	mlx5_flow_pop_thread_workspace();
6350 error_before_hairpin_split:
6351 	rte_free(translated_actions);
6352 	return 0;
6353 }
6354 
6355 /**
6356  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6357  * incoming packets to table 1.
6358  *
6359  * Other flow rules, requested for group n, will be created in
6360  * e-switch table n+1.
6361  * Jump action to e-switch group n will be created to group n+1.
6362  *
6363  * Used when working in switchdev mode, to utilise advantages of table 1
6364  * and above.
6365  *
6366  * @param dev
6367  *   Pointer to Ethernet device.
6368  *
6369  * @return
6370  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6371  */
6372 struct rte_flow *
6373 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6374 {
6375 	const struct rte_flow_attr attr = {
6376 		.group = 0,
6377 		.priority = 0,
6378 		.ingress = 1,
6379 		.egress = 0,
6380 		.transfer = 1,
6381 	};
6382 	const struct rte_flow_item pattern = {
6383 		.type = RTE_FLOW_ITEM_TYPE_END,
6384 	};
6385 	struct rte_flow_action_jump jump = {
6386 		.group = 1,
6387 	};
6388 	const struct rte_flow_action actions[] = {
6389 		{
6390 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6391 			.conf = &jump,
6392 		},
6393 		{
6394 			.type = RTE_FLOW_ACTION_TYPE_END,
6395 		},
6396 	};
6397 	struct mlx5_priv *priv = dev->data->dev_private;
6398 	struct rte_flow_error error;
6399 
6400 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6401 						   &attr, &pattern,
6402 						   actions, false, &error);
6403 }
6404 
6405 /**
6406  * Validate a flow supported by the NIC.
6407  *
6408  * @see rte_flow_validate()
6409  * @see rte_flow_ops
6410  */
6411 int
6412 mlx5_flow_validate(struct rte_eth_dev *dev,
6413 		   const struct rte_flow_attr *attr,
6414 		   const struct rte_flow_item items[],
6415 		   const struct rte_flow_action original_actions[],
6416 		   struct rte_flow_error *error)
6417 {
6418 	int hairpin_flow;
6419 	struct mlx5_translated_action_handle
6420 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6421 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6422 	const struct rte_flow_action *actions;
6423 	struct rte_flow_action *translated_actions = NULL;
6424 	int ret = flow_action_handles_translate(dev, original_actions,
6425 						indir_actions,
6426 						&indir_actions_n,
6427 						&translated_actions, error);
6428 
6429 	if (ret)
6430 		return ret;
6431 	actions = translated_actions ? translated_actions : original_actions;
6432 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6433 	ret = flow_drv_validate(dev, attr, items, actions,
6434 				true, hairpin_flow, error);
6435 	rte_free(translated_actions);
6436 	return ret;
6437 }
6438 
6439 /**
6440  * Create a flow.
6441  *
6442  * @see rte_flow_create()
6443  * @see rte_flow_ops
6444  */
6445 struct rte_flow *
6446 mlx5_flow_create(struct rte_eth_dev *dev,
6447 		 const struct rte_flow_attr *attr,
6448 		 const struct rte_flow_item items[],
6449 		 const struct rte_flow_action actions[],
6450 		 struct rte_flow_error *error)
6451 {
6452 	struct mlx5_priv *priv = dev->data->dev_private;
6453 
6454 	/*
6455 	 * If the device is not started yet, it is not allowed to created a
6456 	 * flow from application. PMD default flows and traffic control flows
6457 	 * are not affected.
6458 	 */
6459 	if (unlikely(!dev->data->dev_started)) {
6460 		DRV_LOG(DEBUG, "port %u is not started when "
6461 			"inserting a flow", dev->data->port_id);
6462 		rte_flow_error_set(error, ENODEV,
6463 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6464 				   NULL,
6465 				   "port not started");
6466 		return NULL;
6467 	}
6468 
6469 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6470 				  attr, items, actions, true, error);
6471 }
6472 
6473 /**
6474  * Destroy a flow in a list.
6475  *
6476  * @param dev
6477  *   Pointer to Ethernet device.
6478  * @param list
6479  *   Pointer to the Indexed flow list. If this parameter NULL,
6480  *   there is no flow removal from the list. Be noted that as
6481  *   flow is add to the indexed list, memory of the indexed
6482  *   list points to maybe changed as flow destroyed.
6483  * @param[in] flow_idx
6484  *   Index of flow to destroy.
6485  */
6486 static void
6487 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6488 		  uint32_t flow_idx)
6489 {
6490 	struct mlx5_priv *priv = dev->data->dev_private;
6491 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6492 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
6493 
6494 	if (!flow)
6495 		return;
6496 	/*
6497 	 * Update RX queue flags only if port is started, otherwise it is
6498 	 * already clean.
6499 	 */
6500 	if (dev->data->dev_started)
6501 		flow_rxq_flags_trim(dev, flow);
6502 	flow_drv_destroy(dev, flow);
6503 	if (list) {
6504 		rte_spinlock_lock(&priv->flow_list_lock);
6505 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6506 			     flow_idx, flow, next);
6507 		rte_spinlock_unlock(&priv->flow_list_lock);
6508 	}
6509 	if (flow->tunnel) {
6510 		struct mlx5_flow_tunnel *tunnel;
6511 
6512 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6513 		RTE_VERIFY(tunnel);
6514 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6515 			mlx5_flow_tunnel_free(dev, tunnel);
6516 	}
6517 	flow_mreg_del_copy_action(dev, flow);
6518 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6519 }
6520 
6521 /**
6522  * Destroy all flows.
6523  *
6524  * @param dev
6525  *   Pointer to Ethernet device.
6526  * @param list
6527  *   Pointer to the Indexed flow list.
6528  * @param active
6529  *   If flushing is called avtively.
6530  */
6531 void
6532 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6533 {
6534 	uint32_t num_flushed = 0;
6535 
6536 	while (*list) {
6537 		flow_list_destroy(dev, list, *list);
6538 		num_flushed++;
6539 	}
6540 	if (active) {
6541 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6542 			dev->data->port_id, num_flushed);
6543 	}
6544 }
6545 
6546 /**
6547  * Stop all default actions for flows.
6548  *
6549  * @param dev
6550  *   Pointer to Ethernet device.
6551  */
6552 void
6553 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6554 {
6555 	flow_mreg_del_default_copy_action(dev);
6556 	flow_rxq_flags_clear(dev);
6557 }
6558 
6559 /**
6560  * Start all default actions for flows.
6561  *
6562  * @param dev
6563  *   Pointer to Ethernet device.
6564  * @return
6565  *   0 on success, a negative errno value otherwise and rte_errno is set.
6566  */
6567 int
6568 mlx5_flow_start_default(struct rte_eth_dev *dev)
6569 {
6570 	struct rte_flow_error error;
6571 
6572 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6573 	return flow_mreg_add_default_copy_action(dev, &error);
6574 }
6575 
6576 /**
6577  * Release key of thread specific flow workspace data.
6578  */
6579 void
6580 flow_release_workspace(void *data)
6581 {
6582 	struct mlx5_flow_workspace *wks = data;
6583 	struct mlx5_flow_workspace *next;
6584 
6585 	while (wks) {
6586 		next = wks->next;
6587 		free(wks->rss_desc.queue);
6588 		free(wks);
6589 		wks = next;
6590 	}
6591 }
6592 
6593 /**
6594  * Get thread specific current flow workspace.
6595  *
6596  * @return pointer to thread specific flow workspace data, NULL on error.
6597  */
6598 struct mlx5_flow_workspace*
6599 mlx5_flow_get_thread_workspace(void)
6600 {
6601 	struct mlx5_flow_workspace *data;
6602 
6603 	data = mlx5_flow_os_get_specific_workspace();
6604 	MLX5_ASSERT(data && data->inuse);
6605 	if (!data || !data->inuse)
6606 		DRV_LOG(ERR, "flow workspace not initialized.");
6607 	return data;
6608 }
6609 
6610 /**
6611  * Allocate and init new flow workspace.
6612  *
6613  * @return pointer to flow workspace data, NULL on error.
6614  */
6615 static struct mlx5_flow_workspace*
6616 flow_alloc_thread_workspace(void)
6617 {
6618 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6619 
6620 	if (!data) {
6621 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6622 			"memory.");
6623 		return NULL;
6624 	}
6625 	data->rss_desc.queue = calloc(1,
6626 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6627 	if (!data->rss_desc.queue)
6628 		goto err;
6629 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6630 	return data;
6631 err:
6632 	if (data->rss_desc.queue)
6633 		free(data->rss_desc.queue);
6634 	free(data);
6635 	return NULL;
6636 }
6637 
6638 /**
6639  * Get new thread specific flow workspace.
6640  *
6641  * If current workspace inuse, create new one and set as current.
6642  *
6643  * @return pointer to thread specific flow workspace data, NULL on error.
6644  */
6645 static struct mlx5_flow_workspace*
6646 mlx5_flow_push_thread_workspace(void)
6647 {
6648 	struct mlx5_flow_workspace *curr;
6649 	struct mlx5_flow_workspace *data;
6650 
6651 	curr = mlx5_flow_os_get_specific_workspace();
6652 	if (!curr) {
6653 		data = flow_alloc_thread_workspace();
6654 		if (!data)
6655 			return NULL;
6656 	} else if (!curr->inuse) {
6657 		data = curr;
6658 	} else if (curr->next) {
6659 		data = curr->next;
6660 	} else {
6661 		data = flow_alloc_thread_workspace();
6662 		if (!data)
6663 			return NULL;
6664 		curr->next = data;
6665 		data->prev = curr;
6666 	}
6667 	data->inuse = 1;
6668 	data->flow_idx = 0;
6669 	/* Set as current workspace */
6670 	if (mlx5_flow_os_set_specific_workspace(data))
6671 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6672 	return data;
6673 }
6674 
6675 /**
6676  * Close current thread specific flow workspace.
6677  *
6678  * If previous workspace available, set it as current.
6679  *
6680  * @return pointer to thread specific flow workspace data, NULL on error.
6681  */
6682 static void
6683 mlx5_flow_pop_thread_workspace(void)
6684 {
6685 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6686 
6687 	if (!data)
6688 		return;
6689 	if (!data->inuse) {
6690 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6691 		return;
6692 	}
6693 	data->inuse = 0;
6694 	if (!data->prev)
6695 		return;
6696 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6697 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6698 }
6699 
6700 /**
6701  * Verify the flow list is empty
6702  *
6703  * @param dev
6704  *  Pointer to Ethernet device.
6705  *
6706  * @return the number of flows not released.
6707  */
6708 int
6709 mlx5_flow_verify(struct rte_eth_dev *dev)
6710 {
6711 	struct mlx5_priv *priv = dev->data->dev_private;
6712 	struct rte_flow *flow;
6713 	uint32_t idx;
6714 	int ret = 0;
6715 
6716 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6717 		      flow, next) {
6718 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6719 			dev->data->port_id, (void *)flow);
6720 		++ret;
6721 	}
6722 	return ret;
6723 }
6724 
6725 /**
6726  * Enable default hairpin egress flow.
6727  *
6728  * @param dev
6729  *   Pointer to Ethernet device.
6730  * @param queue
6731  *   The queue index.
6732  *
6733  * @return
6734  *   0 on success, a negative errno value otherwise and rte_errno is set.
6735  */
6736 int
6737 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6738 			    uint32_t queue)
6739 {
6740 	struct mlx5_priv *priv = dev->data->dev_private;
6741 	const struct rte_flow_attr attr = {
6742 		.egress = 1,
6743 		.priority = 0,
6744 	};
6745 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6746 		.queue = queue,
6747 	};
6748 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6749 		.queue = UINT32_MAX,
6750 	};
6751 	struct rte_flow_item items[] = {
6752 		{
6753 			.type = (enum rte_flow_item_type)
6754 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6755 			.spec = &queue_spec,
6756 			.last = NULL,
6757 			.mask = &queue_mask,
6758 		},
6759 		{
6760 			.type = RTE_FLOW_ITEM_TYPE_END,
6761 		},
6762 	};
6763 	struct rte_flow_action_jump jump = {
6764 		.group = MLX5_HAIRPIN_TX_TABLE,
6765 	};
6766 	struct rte_flow_action actions[2];
6767 	uint32_t flow_idx;
6768 	struct rte_flow_error error;
6769 
6770 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6771 	actions[0].conf = &jump;
6772 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6773 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6774 				&attr, items, actions, false, &error);
6775 	if (!flow_idx) {
6776 		DRV_LOG(DEBUG,
6777 			"Failed to create ctrl flow: rte_errno(%d),"
6778 			" type(%d), message(%s)",
6779 			rte_errno, error.type,
6780 			error.message ? error.message : " (no stated reason)");
6781 		return -rte_errno;
6782 	}
6783 	return 0;
6784 }
6785 
6786 /**
6787  * Enable a control flow configured from the control plane.
6788  *
6789  * @param dev
6790  *   Pointer to Ethernet device.
6791  * @param eth_spec
6792  *   An Ethernet flow spec to apply.
6793  * @param eth_mask
6794  *   An Ethernet flow mask to apply.
6795  * @param vlan_spec
6796  *   A VLAN flow spec to apply.
6797  * @param vlan_mask
6798  *   A VLAN flow mask to apply.
6799  *
6800  * @return
6801  *   0 on success, a negative errno value otherwise and rte_errno is set.
6802  */
6803 int
6804 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6805 		    struct rte_flow_item_eth *eth_spec,
6806 		    struct rte_flow_item_eth *eth_mask,
6807 		    struct rte_flow_item_vlan *vlan_spec,
6808 		    struct rte_flow_item_vlan *vlan_mask)
6809 {
6810 	struct mlx5_priv *priv = dev->data->dev_private;
6811 	const struct rte_flow_attr attr = {
6812 		.ingress = 1,
6813 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6814 	};
6815 	struct rte_flow_item items[] = {
6816 		{
6817 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6818 			.spec = eth_spec,
6819 			.last = NULL,
6820 			.mask = eth_mask,
6821 		},
6822 		{
6823 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6824 					      RTE_FLOW_ITEM_TYPE_END,
6825 			.spec = vlan_spec,
6826 			.last = NULL,
6827 			.mask = vlan_mask,
6828 		},
6829 		{
6830 			.type = RTE_FLOW_ITEM_TYPE_END,
6831 		},
6832 	};
6833 	uint16_t queue[priv->reta_idx_n];
6834 	struct rte_flow_action_rss action_rss = {
6835 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6836 		.level = 0,
6837 		.types = priv->rss_conf.rss_hf,
6838 		.key_len = priv->rss_conf.rss_key_len,
6839 		.queue_num = priv->reta_idx_n,
6840 		.key = priv->rss_conf.rss_key,
6841 		.queue = queue,
6842 	};
6843 	struct rte_flow_action actions[] = {
6844 		{
6845 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6846 			.conf = &action_rss,
6847 		},
6848 		{
6849 			.type = RTE_FLOW_ACTION_TYPE_END,
6850 		},
6851 	};
6852 	uint32_t flow_idx;
6853 	struct rte_flow_error error;
6854 	unsigned int i;
6855 
6856 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6857 		return 0;
6858 	}
6859 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6860 		action_rss.types = 0;
6861 	for (i = 0; i != priv->reta_idx_n; ++i)
6862 		queue[i] = (*priv->reta_idx)[i];
6863 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6864 				&attr, items, actions, false, &error);
6865 	if (!flow_idx)
6866 		return -rte_errno;
6867 	return 0;
6868 }
6869 
6870 /**
6871  * Enable a flow control configured from the control plane.
6872  *
6873  * @param dev
6874  *   Pointer to Ethernet device.
6875  * @param eth_spec
6876  *   An Ethernet flow spec to apply.
6877  * @param eth_mask
6878  *   An Ethernet flow mask to apply.
6879  *
6880  * @return
6881  *   0 on success, a negative errno value otherwise and rte_errno is set.
6882  */
6883 int
6884 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6885 	       struct rte_flow_item_eth *eth_spec,
6886 	       struct rte_flow_item_eth *eth_mask)
6887 {
6888 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6889 }
6890 
6891 /**
6892  * Create default miss flow rule matching lacp traffic
6893  *
6894  * @param dev
6895  *   Pointer to Ethernet device.
6896  * @param eth_spec
6897  *   An Ethernet flow spec to apply.
6898  *
6899  * @return
6900  *   0 on success, a negative errno value otherwise and rte_errno is set.
6901  */
6902 int
6903 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6904 {
6905 	struct mlx5_priv *priv = dev->data->dev_private;
6906 	/*
6907 	 * The LACP matching is done by only using ether type since using
6908 	 * a multicast dst mac causes kernel to give low priority to this flow.
6909 	 */
6910 	static const struct rte_flow_item_eth lacp_spec = {
6911 		.type = RTE_BE16(0x8809),
6912 	};
6913 	static const struct rte_flow_item_eth lacp_mask = {
6914 		.type = 0xffff,
6915 	};
6916 	const struct rte_flow_attr attr = {
6917 		.ingress = 1,
6918 	};
6919 	struct rte_flow_item items[] = {
6920 		{
6921 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6922 			.spec = &lacp_spec,
6923 			.mask = &lacp_mask,
6924 		},
6925 		{
6926 			.type = RTE_FLOW_ITEM_TYPE_END,
6927 		},
6928 	};
6929 	struct rte_flow_action actions[] = {
6930 		{
6931 			.type = (enum rte_flow_action_type)
6932 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6933 		},
6934 		{
6935 			.type = RTE_FLOW_ACTION_TYPE_END,
6936 		},
6937 	};
6938 	struct rte_flow_error error;
6939 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6940 				&attr, items, actions, false, &error);
6941 
6942 	if (!flow_idx)
6943 		return -rte_errno;
6944 	return 0;
6945 }
6946 
6947 /**
6948  * Destroy a flow.
6949  *
6950  * @see rte_flow_destroy()
6951  * @see rte_flow_ops
6952  */
6953 int
6954 mlx5_flow_destroy(struct rte_eth_dev *dev,
6955 		  struct rte_flow *flow,
6956 		  struct rte_flow_error *error __rte_unused)
6957 {
6958 	struct mlx5_priv *priv = dev->data->dev_private;
6959 
6960 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6961 	return 0;
6962 }
6963 
6964 /**
6965  * Destroy all flows.
6966  *
6967  * @see rte_flow_flush()
6968  * @see rte_flow_ops
6969  */
6970 int
6971 mlx5_flow_flush(struct rte_eth_dev *dev,
6972 		struct rte_flow_error *error __rte_unused)
6973 {
6974 	struct mlx5_priv *priv = dev->data->dev_private;
6975 
6976 	mlx5_flow_list_flush(dev, &priv->flows, false);
6977 	return 0;
6978 }
6979 
6980 /**
6981  * Isolated mode.
6982  *
6983  * @see rte_flow_isolate()
6984  * @see rte_flow_ops
6985  */
6986 int
6987 mlx5_flow_isolate(struct rte_eth_dev *dev,
6988 		  int enable,
6989 		  struct rte_flow_error *error)
6990 {
6991 	struct mlx5_priv *priv = dev->data->dev_private;
6992 
6993 	if (dev->data->dev_started) {
6994 		rte_flow_error_set(error, EBUSY,
6995 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6996 				   NULL,
6997 				   "port must be stopped first");
6998 		return -rte_errno;
6999 	}
7000 	priv->isolated = !!enable;
7001 	if (enable)
7002 		dev->dev_ops = &mlx5_dev_ops_isolate;
7003 	else
7004 		dev->dev_ops = &mlx5_dev_ops;
7005 
7006 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7007 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7008 
7009 	return 0;
7010 }
7011 
7012 /**
7013  * Query a flow.
7014  *
7015  * @see rte_flow_query()
7016  * @see rte_flow_ops
7017  */
7018 static int
7019 flow_drv_query(struct rte_eth_dev *dev,
7020 	       uint32_t flow_idx,
7021 	       const struct rte_flow_action *actions,
7022 	       void *data,
7023 	       struct rte_flow_error *error)
7024 {
7025 	struct mlx5_priv *priv = dev->data->dev_private;
7026 	const struct mlx5_flow_driver_ops *fops;
7027 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
7028 					       [MLX5_IPOOL_RTE_FLOW],
7029 					       flow_idx);
7030 	enum mlx5_flow_drv_type ftype;
7031 
7032 	if (!flow) {
7033 		return rte_flow_error_set(error, ENOENT,
7034 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7035 			  NULL,
7036 			  "invalid flow handle");
7037 	}
7038 	ftype = flow->drv_type;
7039 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7040 	fops = flow_get_drv_ops(ftype);
7041 
7042 	return fops->query(dev, flow, actions, data, error);
7043 }
7044 
7045 /**
7046  * Query a flow.
7047  *
7048  * @see rte_flow_query()
7049  * @see rte_flow_ops
7050  */
7051 int
7052 mlx5_flow_query(struct rte_eth_dev *dev,
7053 		struct rte_flow *flow,
7054 		const struct rte_flow_action *actions,
7055 		void *data,
7056 		struct rte_flow_error *error)
7057 {
7058 	int ret;
7059 
7060 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7061 			     error);
7062 	if (ret < 0)
7063 		return ret;
7064 	return 0;
7065 }
7066 
7067 /**
7068  * Get rte_flow callbacks.
7069  *
7070  * @param dev
7071  *   Pointer to Ethernet device structure.
7072  * @param ops
7073  *   Pointer to operation-specific structure.
7074  *
7075  * @return 0
7076  */
7077 int
7078 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7079 		  const struct rte_flow_ops **ops)
7080 {
7081 	*ops = &mlx5_flow_ops;
7082 	return 0;
7083 }
7084 
7085 /**
7086  * Validate meter policy actions.
7087  * Dispatcher for action type specific validation.
7088  *
7089  * @param[in] dev
7090  *   Pointer to the Ethernet device structure.
7091  * @param[in] action
7092  *   The meter policy action object to validate.
7093  * @param[in] attr
7094  *   Attributes of flow to determine steering domain.
7095  * @param[out] is_rss
7096  *   Is RSS or not.
7097  * @param[out] domain_bitmap
7098  *   Domain bitmap.
7099  * @param[out] is_def_policy
7100  *   Is default policy or not.
7101  * @param[out] error
7102  *   Perform verbose error reporting if not NULL. Initialized in case of
7103  *   error only.
7104  *
7105  * @return
7106  *   0 on success, otherwise negative errno value.
7107  */
7108 int
7109 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7110 			const struct rte_flow_action *actions[RTE_COLORS],
7111 			struct rte_flow_attr *attr,
7112 			bool *is_rss,
7113 			uint8_t *domain_bitmap,
7114 			bool *is_def_policy,
7115 			struct rte_mtr_error *error)
7116 {
7117 	const struct mlx5_flow_driver_ops *fops;
7118 
7119 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7120 	return fops->validate_mtr_acts(dev, actions, attr,
7121 			is_rss, domain_bitmap, is_def_policy, error);
7122 }
7123 
7124 /**
7125  * Destroy the meter table set.
7126  *
7127  * @param[in] dev
7128  *   Pointer to Ethernet device.
7129  * @param[in] mtr_policy
7130  *   Meter policy struct.
7131  */
7132 void
7133 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7134 		      struct mlx5_flow_meter_policy *mtr_policy)
7135 {
7136 	const struct mlx5_flow_driver_ops *fops;
7137 
7138 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7139 	fops->destroy_mtr_acts(dev, mtr_policy);
7140 }
7141 
7142 /**
7143  * Create policy action, lock free,
7144  * (mutex should be acquired by caller).
7145  * Dispatcher for action type specific call.
7146  *
7147  * @param[in] dev
7148  *   Pointer to the Ethernet device structure.
7149  * @param[in] mtr_policy
7150  *   Meter policy struct.
7151  * @param[in] action
7152  *   Action specification used to create meter actions.
7153  * @param[out] error
7154  *   Perform verbose error reporting if not NULL. Initialized in case of
7155  *   error only.
7156  *
7157  * @return
7158  *   0 on success, otherwise negative errno value.
7159  */
7160 int
7161 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7162 		      struct mlx5_flow_meter_policy *mtr_policy,
7163 		      const struct rte_flow_action *actions[RTE_COLORS],
7164 		      struct rte_mtr_error *error)
7165 {
7166 	const struct mlx5_flow_driver_ops *fops;
7167 
7168 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7169 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7170 }
7171 
7172 /**
7173  * Create policy rules, lock free,
7174  * (mutex should be acquired by caller).
7175  * Dispatcher for action type specific call.
7176  *
7177  * @param[in] dev
7178  *   Pointer to the Ethernet device structure.
7179  * @param[in] mtr_policy
7180  *   Meter policy struct.
7181  *
7182  * @return
7183  *   0 on success, -1 otherwise.
7184  */
7185 int
7186 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7187 			     struct mlx5_flow_meter_policy *mtr_policy)
7188 {
7189 	const struct mlx5_flow_driver_ops *fops;
7190 
7191 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7192 	return fops->create_policy_rules(dev, mtr_policy);
7193 }
7194 
7195 /**
7196  * Destroy policy rules, lock free,
7197  * (mutex should be acquired by caller).
7198  * Dispatcher for action type specific call.
7199  *
7200  * @param[in] dev
7201  *   Pointer to the Ethernet device structure.
7202  * @param[in] mtr_policy
7203  *   Meter policy struct.
7204  */
7205 void
7206 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7207 			     struct mlx5_flow_meter_policy *mtr_policy)
7208 {
7209 	const struct mlx5_flow_driver_ops *fops;
7210 
7211 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7212 	fops->destroy_policy_rules(dev, mtr_policy);
7213 }
7214 
7215 /**
7216  * Destroy the default policy table set.
7217  *
7218  * @param[in] dev
7219  *   Pointer to Ethernet device.
7220  */
7221 void
7222 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7223 {
7224 	const struct mlx5_flow_driver_ops *fops;
7225 
7226 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7227 	fops->destroy_def_policy(dev);
7228 }
7229 
7230 /**
7231  * Destroy the default policy table set.
7232  *
7233  * @param[in] dev
7234  *   Pointer to Ethernet device.
7235  *
7236  * @return
7237  *   0 on success, -1 otherwise.
7238  */
7239 int
7240 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7241 {
7242 	const struct mlx5_flow_driver_ops *fops;
7243 
7244 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7245 	return fops->create_def_policy(dev);
7246 }
7247 
7248 /**
7249  * Create the needed meter and suffix tables.
7250  *
7251  * @param[in] dev
7252  *   Pointer to Ethernet device.
7253  *
7254  * @return
7255  *   0 on success, -1 otherwise.
7256  */
7257 int
7258 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7259 			struct mlx5_flow_meter_info *fm,
7260 			uint32_t mtr_idx,
7261 			uint8_t domain_bitmap)
7262 {
7263 	const struct mlx5_flow_driver_ops *fops;
7264 
7265 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7266 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7267 }
7268 
7269 /**
7270  * Destroy the meter table set.
7271  *
7272  * @param[in] dev
7273  *   Pointer to Ethernet device.
7274  * @param[in] tbl
7275  *   Pointer to the meter table set.
7276  */
7277 void
7278 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7279 			   struct mlx5_flow_meter_info *fm)
7280 {
7281 	const struct mlx5_flow_driver_ops *fops;
7282 
7283 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7284 	fops->destroy_mtr_tbls(dev, fm);
7285 }
7286 
7287 /**
7288  * Destroy the global meter drop table.
7289  *
7290  * @param[in] dev
7291  *   Pointer to Ethernet device.
7292  */
7293 void
7294 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7295 {
7296 	const struct mlx5_flow_driver_ops *fops;
7297 
7298 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7299 	fops->destroy_mtr_drop_tbls(dev);
7300 }
7301 
7302 /**
7303  * Destroy the sub policy table with RX queue.
7304  *
7305  * @param[in] dev
7306  *   Pointer to Ethernet device.
7307  * @param[in] mtr_policy
7308  *   Pointer to meter policy table.
7309  */
7310 void
7311 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7312 		struct mlx5_flow_meter_policy *mtr_policy)
7313 {
7314 	const struct mlx5_flow_driver_ops *fops;
7315 
7316 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7317 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7318 }
7319 
7320 /**
7321  * Allocate the needed aso flow meter id.
7322  *
7323  * @param[in] dev
7324  *   Pointer to Ethernet device.
7325  *
7326  * @return
7327  *   Index to aso flow meter on success, NULL otherwise.
7328  */
7329 uint32_t
7330 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7331 {
7332 	const struct mlx5_flow_driver_ops *fops;
7333 
7334 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7335 	return fops->create_meter(dev);
7336 }
7337 
7338 /**
7339  * Free the aso flow meter id.
7340  *
7341  * @param[in] dev
7342  *   Pointer to Ethernet device.
7343  * @param[in] mtr_idx
7344  *  Index to aso flow meter to be free.
7345  *
7346  * @return
7347  *   0 on success.
7348  */
7349 void
7350 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7351 {
7352 	const struct mlx5_flow_driver_ops *fops;
7353 
7354 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7355 	fops->free_meter(dev, mtr_idx);
7356 }
7357 
7358 /**
7359  * Allocate a counter.
7360  *
7361  * @param[in] dev
7362  *   Pointer to Ethernet device structure.
7363  *
7364  * @return
7365  *   Index to allocated counter  on success, 0 otherwise.
7366  */
7367 uint32_t
7368 mlx5_counter_alloc(struct rte_eth_dev *dev)
7369 {
7370 	const struct mlx5_flow_driver_ops *fops;
7371 	struct rte_flow_attr attr = { .transfer = 0 };
7372 
7373 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7374 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7375 		return fops->counter_alloc(dev);
7376 	}
7377 	DRV_LOG(ERR,
7378 		"port %u counter allocate is not supported.",
7379 		 dev->data->port_id);
7380 	return 0;
7381 }
7382 
7383 /**
7384  * Free a counter.
7385  *
7386  * @param[in] dev
7387  *   Pointer to Ethernet device structure.
7388  * @param[in] cnt
7389  *   Index to counter to be free.
7390  */
7391 void
7392 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7393 {
7394 	const struct mlx5_flow_driver_ops *fops;
7395 	struct rte_flow_attr attr = { .transfer = 0 };
7396 
7397 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7398 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7399 		fops->counter_free(dev, cnt);
7400 		return;
7401 	}
7402 	DRV_LOG(ERR,
7403 		"port %u counter free is not supported.",
7404 		 dev->data->port_id);
7405 }
7406 
7407 /**
7408  * Query counter statistics.
7409  *
7410  * @param[in] dev
7411  *   Pointer to Ethernet device structure.
7412  * @param[in] cnt
7413  *   Index to counter to query.
7414  * @param[in] clear
7415  *   Set to clear counter statistics.
7416  * @param[out] pkts
7417  *   The counter hits packets number to save.
7418  * @param[out] bytes
7419  *   The counter hits bytes number to save.
7420  *
7421  * @return
7422  *   0 on success, a negative errno value otherwise.
7423  */
7424 int
7425 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7426 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7427 {
7428 	const struct mlx5_flow_driver_ops *fops;
7429 	struct rte_flow_attr attr = { .transfer = 0 };
7430 
7431 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7432 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7433 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7434 	}
7435 	DRV_LOG(ERR,
7436 		"port %u counter query is not supported.",
7437 		 dev->data->port_id);
7438 	return -ENOTSUP;
7439 }
7440 
7441 /**
7442  * Allocate a new memory for the counter values wrapped by all the needed
7443  * management.
7444  *
7445  * @param[in] sh
7446  *   Pointer to mlx5_dev_ctx_shared object.
7447  *
7448  * @return
7449  *   0 on success, a negative errno value otherwise.
7450  */
7451 static int
7452 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7453 {
7454 	struct mlx5_devx_mkey_attr mkey_attr;
7455 	struct mlx5_counter_stats_mem_mng *mem_mng;
7456 	volatile struct flow_counter_stats *raw_data;
7457 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7458 	int size = (sizeof(struct flow_counter_stats) *
7459 			MLX5_COUNTERS_PER_POOL +
7460 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7461 			sizeof(struct mlx5_counter_stats_mem_mng);
7462 	size_t pgsize = rte_mem_page_size();
7463 	uint8_t *mem;
7464 	int i;
7465 
7466 	if (pgsize == (size_t)-1) {
7467 		DRV_LOG(ERR, "Failed to get mem page size");
7468 		rte_errno = ENOMEM;
7469 		return -ENOMEM;
7470 	}
7471 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7472 	if (!mem) {
7473 		rte_errno = ENOMEM;
7474 		return -ENOMEM;
7475 	}
7476 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7477 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7478 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7479 						 IBV_ACCESS_LOCAL_WRITE);
7480 	if (!mem_mng->umem) {
7481 		rte_errno = errno;
7482 		mlx5_free(mem);
7483 		return -rte_errno;
7484 	}
7485 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7486 	mkey_attr.addr = (uintptr_t)mem;
7487 	mkey_attr.size = size;
7488 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7489 	mkey_attr.pd = sh->pdn;
7490 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7491 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7492 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7493 	if (!mem_mng->dm) {
7494 		mlx5_os_umem_dereg(mem_mng->umem);
7495 		rte_errno = errno;
7496 		mlx5_free(mem);
7497 		return -rte_errno;
7498 	}
7499 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7500 	raw_data = (volatile struct flow_counter_stats *)mem;
7501 	for (i = 0; i < raws_n; ++i) {
7502 		mem_mng->raws[i].mem_mng = mem_mng;
7503 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7504 	}
7505 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7506 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7507 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7508 				 next);
7509 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7510 	sh->cmng.mem_mng = mem_mng;
7511 	return 0;
7512 }
7513 
7514 /**
7515  * Set the statistic memory to the new counter pool.
7516  *
7517  * @param[in] sh
7518  *   Pointer to mlx5_dev_ctx_shared object.
7519  * @param[in] pool
7520  *   Pointer to the pool to set the statistic memory.
7521  *
7522  * @return
7523  *   0 on success, a negative errno value otherwise.
7524  */
7525 static int
7526 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7527 			       struct mlx5_flow_counter_pool *pool)
7528 {
7529 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7530 	/* Resize statistic memory once used out. */
7531 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7532 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7533 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7534 		return -1;
7535 	}
7536 	rte_spinlock_lock(&pool->sl);
7537 	pool->raw = cmng->mem_mng->raws + pool->index %
7538 		    MLX5_CNT_CONTAINER_RESIZE;
7539 	rte_spinlock_unlock(&pool->sl);
7540 	pool->raw_hw = NULL;
7541 	return 0;
7542 }
7543 
7544 #define MLX5_POOL_QUERY_FREQ_US 1000000
7545 
7546 /**
7547  * Set the periodic procedure for triggering asynchronous batch queries for all
7548  * the counter pools.
7549  *
7550  * @param[in] sh
7551  *   Pointer to mlx5_dev_ctx_shared object.
7552  */
7553 void
7554 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7555 {
7556 	uint32_t pools_n, us;
7557 
7558 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7559 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7560 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7561 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7562 		sh->cmng.query_thread_on = 0;
7563 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7564 	} else {
7565 		sh->cmng.query_thread_on = 1;
7566 	}
7567 }
7568 
7569 /**
7570  * The periodic procedure for triggering asynchronous batch queries for all the
7571  * counter pools. This function is probably called by the host thread.
7572  *
7573  * @param[in] arg
7574  *   The parameter for the alarm process.
7575  */
7576 void
7577 mlx5_flow_query_alarm(void *arg)
7578 {
7579 	struct mlx5_dev_ctx_shared *sh = arg;
7580 	int ret;
7581 	uint16_t pool_index = sh->cmng.pool_index;
7582 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7583 	struct mlx5_flow_counter_pool *pool;
7584 	uint16_t n_valid;
7585 
7586 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7587 		goto set_alarm;
7588 	rte_spinlock_lock(&cmng->pool_update_sl);
7589 	pool = cmng->pools[pool_index];
7590 	n_valid = cmng->n_valid;
7591 	rte_spinlock_unlock(&cmng->pool_update_sl);
7592 	/* Set the statistic memory to the new created pool. */
7593 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7594 		goto set_alarm;
7595 	if (pool->raw_hw)
7596 		/* There is a pool query in progress. */
7597 		goto set_alarm;
7598 	pool->raw_hw =
7599 		LIST_FIRST(&sh->cmng.free_stat_raws);
7600 	if (!pool->raw_hw)
7601 		/* No free counter statistics raw memory. */
7602 		goto set_alarm;
7603 	/*
7604 	 * Identify the counters released between query trigger and query
7605 	 * handle more efficiently. The counter released in this gap period
7606 	 * should wait for a new round of query as the new arrived packets
7607 	 * will not be taken into account.
7608 	 */
7609 	pool->query_gen++;
7610 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7611 					       MLX5_COUNTERS_PER_POOL,
7612 					       NULL, NULL,
7613 					       pool->raw_hw->mem_mng->dm->id,
7614 					       (void *)(uintptr_t)
7615 					       pool->raw_hw->data,
7616 					       sh->devx_comp,
7617 					       (uint64_t)(uintptr_t)pool);
7618 	if (ret) {
7619 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7620 			" %d", pool->min_dcs->id);
7621 		pool->raw_hw = NULL;
7622 		goto set_alarm;
7623 	}
7624 	LIST_REMOVE(pool->raw_hw, next);
7625 	sh->cmng.pending_queries++;
7626 	pool_index++;
7627 	if (pool_index >= n_valid)
7628 		pool_index = 0;
7629 set_alarm:
7630 	sh->cmng.pool_index = pool_index;
7631 	mlx5_set_query_alarm(sh);
7632 }
7633 
7634 /**
7635  * Check and callback event for new aged flow in the counter pool
7636  *
7637  * @param[in] sh
7638  *   Pointer to mlx5_dev_ctx_shared object.
7639  * @param[in] pool
7640  *   Pointer to Current counter pool.
7641  */
7642 static void
7643 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7644 		   struct mlx5_flow_counter_pool *pool)
7645 {
7646 	struct mlx5_priv *priv;
7647 	struct mlx5_flow_counter *cnt;
7648 	struct mlx5_age_info *age_info;
7649 	struct mlx5_age_param *age_param;
7650 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7651 	struct mlx5_counter_stats_raw *prev = pool->raw;
7652 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7653 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7654 	uint16_t expected = AGE_CANDIDATE;
7655 	uint32_t i;
7656 
7657 	pool->time_of_last_age_check = curr_time;
7658 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7659 		cnt = MLX5_POOL_GET_CNT(pool, i);
7660 		age_param = MLX5_CNT_TO_AGE(cnt);
7661 		if (__atomic_load_n(&age_param->state,
7662 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7663 			continue;
7664 		if (cur->data[i].hits != prev->data[i].hits) {
7665 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7666 					 __ATOMIC_RELAXED);
7667 			continue;
7668 		}
7669 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7670 				       time_delta,
7671 				       __ATOMIC_RELAXED) <= age_param->timeout)
7672 			continue;
7673 		/**
7674 		 * Hold the lock first, or if between the
7675 		 * state AGE_TMOUT and tailq operation the
7676 		 * release happened, the release procedure
7677 		 * may delete a non-existent tailq node.
7678 		 */
7679 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7680 		age_info = GET_PORT_AGE_INFO(priv);
7681 		rte_spinlock_lock(&age_info->aged_sl);
7682 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7683 						AGE_TMOUT, false,
7684 						__ATOMIC_RELAXED,
7685 						__ATOMIC_RELAXED)) {
7686 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7687 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7688 		}
7689 		rte_spinlock_unlock(&age_info->aged_sl);
7690 	}
7691 	mlx5_age_event_prepare(sh);
7692 }
7693 
7694 /**
7695  * Handler for the HW respond about ready values from an asynchronous batch
7696  * query. This function is probably called by the host thread.
7697  *
7698  * @param[in] sh
7699  *   The pointer to the shared device context.
7700  * @param[in] async_id
7701  *   The Devx async ID.
7702  * @param[in] status
7703  *   The status of the completion.
7704  */
7705 void
7706 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7707 				  uint64_t async_id, int status)
7708 {
7709 	struct mlx5_flow_counter_pool *pool =
7710 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7711 	struct mlx5_counter_stats_raw *raw_to_free;
7712 	uint8_t query_gen = pool->query_gen ^ 1;
7713 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7714 	enum mlx5_counter_type cnt_type =
7715 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7716 				MLX5_COUNTER_TYPE_ORIGIN;
7717 
7718 	if (unlikely(status)) {
7719 		raw_to_free = pool->raw_hw;
7720 	} else {
7721 		raw_to_free = pool->raw;
7722 		if (pool->is_aged)
7723 			mlx5_flow_aging_check(sh, pool);
7724 		rte_spinlock_lock(&pool->sl);
7725 		pool->raw = pool->raw_hw;
7726 		rte_spinlock_unlock(&pool->sl);
7727 		/* Be sure the new raw counters data is updated in memory. */
7728 		rte_io_wmb();
7729 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7730 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7731 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7732 				     &pool->counters[query_gen], next);
7733 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7734 		}
7735 	}
7736 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7737 	pool->raw_hw = NULL;
7738 	sh->cmng.pending_queries--;
7739 }
7740 
7741 static int
7742 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7743 		    const struct flow_grp_info *grp_info,
7744 		    struct rte_flow_error *error)
7745 {
7746 	if (grp_info->transfer && grp_info->external &&
7747 	    grp_info->fdb_def_rule) {
7748 		if (group == UINT32_MAX)
7749 			return rte_flow_error_set
7750 						(error, EINVAL,
7751 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7752 						 NULL,
7753 						 "group index not supported");
7754 		*table = group + 1;
7755 	} else {
7756 		*table = group;
7757 	}
7758 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7759 	return 0;
7760 }
7761 
7762 /**
7763  * Translate the rte_flow group index to HW table value.
7764  *
7765  * If tunnel offload is disabled, all group ids converted to flow table
7766  * id using the standard method.
7767  * If tunnel offload is enabled, group id can be converted using the
7768  * standard or tunnel conversion method. Group conversion method
7769  * selection depends on flags in `grp_info` parameter:
7770  * - Internal (grp_info.external == 0) groups conversion uses the
7771  *   standard method.
7772  * - Group ids in JUMP action converted with the tunnel conversion.
7773  * - Group id in rule attribute conversion depends on a rule type and
7774  *   group id value:
7775  *   ** non zero group attributes converted with the tunnel method
7776  *   ** zero group attribute in non-tunnel rule is converted using the
7777  *      standard method - there's only one root table
7778  *   ** zero group attribute in steer tunnel rule is converted with the
7779  *      standard method - single root table
7780  *   ** zero group attribute in match tunnel rule is a special OvS
7781  *      case: that value is used for portability reasons. That group
7782  *      id is converted with the tunnel conversion method.
7783  *
7784  * @param[in] dev
7785  *   Port device
7786  * @param[in] tunnel
7787  *   PMD tunnel offload object
7788  * @param[in] group
7789  *   rte_flow group index value.
7790  * @param[out] table
7791  *   HW table value.
7792  * @param[in] grp_info
7793  *   flags used for conversion
7794  * @param[out] error
7795  *   Pointer to error structure.
7796  *
7797  * @return
7798  *   0 on success, a negative errno value otherwise and rte_errno is set.
7799  */
7800 int
7801 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7802 			 const struct mlx5_flow_tunnel *tunnel,
7803 			 uint32_t group, uint32_t *table,
7804 			 const struct flow_grp_info *grp_info,
7805 			 struct rte_flow_error *error)
7806 {
7807 	int ret;
7808 	bool standard_translation;
7809 
7810 	if (!grp_info->skip_scale && grp_info->external &&
7811 	    group < MLX5_MAX_TABLES_EXTERNAL)
7812 		group *= MLX5_FLOW_TABLE_FACTOR;
7813 	if (is_tunnel_offload_active(dev)) {
7814 		standard_translation = !grp_info->external ||
7815 					grp_info->std_tbl_fix;
7816 	} else {
7817 		standard_translation = true;
7818 	}
7819 	DRV_LOG(DEBUG,
7820 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7821 		dev->data->port_id, group, grp_info->transfer,
7822 		grp_info->external, grp_info->fdb_def_rule,
7823 		standard_translation ? "STANDARD" : "TUNNEL");
7824 	if (standard_translation)
7825 		ret = flow_group_to_table(dev->data->port_id, group, table,
7826 					  grp_info, error);
7827 	else
7828 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7829 						      table, error);
7830 
7831 	return ret;
7832 }
7833 
7834 /**
7835  * Discover availability of metadata reg_c's.
7836  *
7837  * Iteratively use test flows to check availability.
7838  *
7839  * @param[in] dev
7840  *   Pointer to the Ethernet device structure.
7841  *
7842  * @return
7843  *   0 on success, a negative errno value otherwise and rte_errno is set.
7844  */
7845 int
7846 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7847 {
7848 	struct mlx5_priv *priv = dev->data->dev_private;
7849 	struct mlx5_dev_config *config = &priv->config;
7850 	enum modify_reg idx;
7851 	int n = 0;
7852 
7853 	/* reg_c[0] and reg_c[1] are reserved. */
7854 	config->flow_mreg_c[n++] = REG_C_0;
7855 	config->flow_mreg_c[n++] = REG_C_1;
7856 	/* Discover availability of other reg_c's. */
7857 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7858 		struct rte_flow_attr attr = {
7859 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7860 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7861 			.ingress = 1,
7862 		};
7863 		struct rte_flow_item items[] = {
7864 			[0] = {
7865 				.type = RTE_FLOW_ITEM_TYPE_END,
7866 			},
7867 		};
7868 		struct rte_flow_action actions[] = {
7869 			[0] = {
7870 				.type = (enum rte_flow_action_type)
7871 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7872 				.conf = &(struct mlx5_flow_action_copy_mreg){
7873 					.src = REG_C_1,
7874 					.dst = idx,
7875 				},
7876 			},
7877 			[1] = {
7878 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7879 				.conf = &(struct rte_flow_action_jump){
7880 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7881 				},
7882 			},
7883 			[2] = {
7884 				.type = RTE_FLOW_ACTION_TYPE_END,
7885 			},
7886 		};
7887 		uint32_t flow_idx;
7888 		struct rte_flow *flow;
7889 		struct rte_flow_error error;
7890 
7891 		if (!config->dv_flow_en)
7892 			break;
7893 		/* Create internal flow, validation skips copy action. */
7894 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7895 					    actions, false, &error);
7896 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7897 				      flow_idx);
7898 		if (!flow)
7899 			continue;
7900 		config->flow_mreg_c[n++] = idx;
7901 		flow_list_destroy(dev, NULL, flow_idx);
7902 	}
7903 	for (; n < MLX5_MREG_C_NUM; ++n)
7904 		config->flow_mreg_c[n] = REG_NON;
7905 	return 0;
7906 }
7907 
7908 int
7909 save_dump_file(const uint8_t *data, uint32_t size,
7910 	uint32_t type, uint32_t id, void *arg, FILE *file)
7911 {
7912 	char line[BUF_SIZE];
7913 	uint32_t out = 0;
7914 	uint32_t k;
7915 	uint32_t actions_num;
7916 	struct rte_flow_query_count *count;
7917 
7918 	memset(line, 0, BUF_SIZE);
7919 	switch (type) {
7920 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
7921 		actions_num = *(uint32_t *)(arg);
7922 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
7923 				type, id, actions_num);
7924 		break;
7925 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
7926 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
7927 				type, id);
7928 		break;
7929 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
7930 		count = (struct rte_flow_query_count *)arg;
7931 		fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
7932 				id, count->hits, count->bytes);
7933 		return 0;
7934 	default:
7935 		return -1;
7936 	}
7937 
7938 	for (k = 0; k < size; k++) {
7939 		/* Make sure we do not overrun the line buffer length. */
7940 		if (out >= BUF_SIZE - 4) {
7941 			line[out] = '\0';
7942 			break;
7943 		}
7944 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
7945 				(data[k]) & 0xff);
7946 	}
7947 	fprintf(file, "%s\n", line);
7948 	return 0;
7949 }
7950 
7951 int
7952 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
7953 	struct rte_flow_query_count *count, struct rte_flow_error *error)
7954 {
7955 	struct rte_flow_action action[2];
7956 	enum mlx5_flow_drv_type ftype;
7957 	const struct mlx5_flow_driver_ops *fops;
7958 
7959 	if (!flow) {
7960 		return rte_flow_error_set(error, ENOENT,
7961 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7962 				NULL,
7963 				"invalid flow handle");
7964 	}
7965 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
7966 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
7967 	if (flow->counter) {
7968 		memset(count, 0, sizeof(struct rte_flow_query_count));
7969 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
7970 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
7971 						ftype < MLX5_FLOW_TYPE_MAX);
7972 		fops = flow_get_drv_ops(ftype);
7973 		return fops->query(dev, flow, action, count, error);
7974 	}
7975 	return -1;
7976 }
7977 
7978 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7979 /**
7980  * Dump flow ipool data to file
7981  *
7982  * @param[in] dev
7983  *   The pointer to Ethernet device.
7984  * @param[in] file
7985  *   A pointer to a file for output.
7986  * @param[out] error
7987  *   Perform verbose error reporting if not NULL. PMDs initialize this
7988  *   structure in case of error only.
7989  * @return
7990  *   0 on success, a negative value otherwise.
7991  */
7992 int
7993 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
7994 	struct rte_flow *flow, FILE *file,
7995 	struct rte_flow_error *error)
7996 {
7997 	struct mlx5_priv *priv = dev->data->dev_private;
7998 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
7999 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8000 	uint32_t handle_idx;
8001 	struct mlx5_flow_handle *dh;
8002 	struct rte_flow_query_count count;
8003 	uint32_t actions_num;
8004 	const uint8_t *data;
8005 	size_t size;
8006 	uint32_t id;
8007 	uint32_t type;
8008 
8009 	if (!flow) {
8010 		return rte_flow_error_set(error, ENOENT,
8011 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8012 			NULL,
8013 			"invalid flow handle");
8014 	}
8015 	handle_idx = flow->dev_handles;
8016 	while (handle_idx) {
8017 		dh = mlx5_ipool_get(priv->sh->ipool
8018 			[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8019 		if (!dh)
8020 			continue;
8021 		handle_idx = dh->next.next;
8022 		id = (uint32_t)(uintptr_t)dh->drv_flow;
8023 
8024 		/* query counter */
8025 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8026 		if (!mlx5_flow_query_counter(dev, flow, &count, error))
8027 			save_dump_file(NULL, 0, type,
8028 					id, (void *)&count, file);
8029 
8030 		/* Get modify_hdr and encap_decap buf from ipools. */
8031 		encap_decap = NULL;
8032 		modify_hdr = dh->dvh.modify_hdr;
8033 
8034 		if (dh->dvh.rix_encap_decap) {
8035 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8036 						[MLX5_IPOOL_DECAP_ENCAP],
8037 						dh->dvh.rix_encap_decap);
8038 		}
8039 		if (modify_hdr) {
8040 			data = (const uint8_t *)modify_hdr->actions;
8041 			size = (size_t)(modify_hdr->actions_num) * 8;
8042 			actions_num = modify_hdr->actions_num;
8043 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8044 			save_dump_file(data, size, type, id,
8045 					(void *)(&actions_num), file);
8046 		}
8047 		if (encap_decap) {
8048 			data = encap_decap->buf;
8049 			size = encap_decap->size;
8050 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8051 			save_dump_file(data, size, type,
8052 						id, NULL, file);
8053 		}
8054 	}
8055 	return 0;
8056 }
8057 #endif
8058 
8059 /**
8060  * Dump flow raw hw data to file
8061  *
8062  * @param[in] dev
8063  *    The pointer to Ethernet device.
8064  * @param[in] file
8065  *   A pointer to a file for output.
8066  * @param[out] error
8067  *   Perform verbose error reporting if not NULL. PMDs initialize this
8068  *   structure in case of error only.
8069  * @return
8070  *   0 on success, a nagative value otherwise.
8071  */
8072 int
8073 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8074 		   FILE *file,
8075 		   struct rte_flow_error *error __rte_unused)
8076 {
8077 	struct mlx5_priv *priv = dev->data->dev_private;
8078 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8079 	uint32_t handle_idx;
8080 	int ret;
8081 	struct mlx5_flow_handle *dh;
8082 	struct rte_flow *flow;
8083 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8084 	uint32_t idx;
8085 #endif
8086 
8087 	if (!priv->config.dv_flow_en) {
8088 		if (fputs("device dv flow disabled\n", file) <= 0)
8089 			return -errno;
8090 		return -ENOTSUP;
8091 	}
8092 
8093 	/* dump all */
8094 	if (!flow_idx) {
8095 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8096 		ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
8097 						priv->flows, idx, flow, next)
8098 			mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8099 #endif
8100 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8101 					sh->rx_domain,
8102 					sh->tx_domain, file);
8103 	}
8104 	/* dump one */
8105 	flow = mlx5_ipool_get(priv->sh->ipool
8106 			[MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
8107 	if (!flow)
8108 		return -ENOENT;
8109 
8110 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8111 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8112 #endif
8113 	handle_idx = flow->dev_handles;
8114 	while (handle_idx) {
8115 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8116 				handle_idx);
8117 		if (!dh)
8118 			return -ENOENT;
8119 		if (dh->drv_flow) {
8120 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8121 					file);
8122 			if (ret)
8123 				return -ENOENT;
8124 		}
8125 		handle_idx = dh->next.next;
8126 	}
8127 	return 0;
8128 }
8129 
8130 /**
8131  * Get aged-out flows.
8132  *
8133  * @param[in] dev
8134  *   Pointer to the Ethernet device structure.
8135  * @param[in] context
8136  *   The address of an array of pointers to the aged-out flows contexts.
8137  * @param[in] nb_countexts
8138  *   The length of context array pointers.
8139  * @param[out] error
8140  *   Perform verbose error reporting if not NULL. Initialized in case of
8141  *   error only.
8142  *
8143  * @return
8144  *   how many contexts get in success, otherwise negative errno value.
8145  *   if nb_contexts is 0, return the amount of all aged contexts.
8146  *   if nb_contexts is not 0 , return the amount of aged flows reported
8147  *   in the context array.
8148  */
8149 int
8150 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8151 			uint32_t nb_contexts, struct rte_flow_error *error)
8152 {
8153 	const struct mlx5_flow_driver_ops *fops;
8154 	struct rte_flow_attr attr = { .transfer = 0 };
8155 
8156 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8157 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8158 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8159 						    error);
8160 	}
8161 	DRV_LOG(ERR,
8162 		"port %u get aged flows is not supported.",
8163 		 dev->data->port_id);
8164 	return -ENOTSUP;
8165 }
8166 
8167 /* Wrapper for driver action_validate op callback */
8168 static int
8169 flow_drv_action_validate(struct rte_eth_dev *dev,
8170 			 const struct rte_flow_indir_action_conf *conf,
8171 			 const struct rte_flow_action *action,
8172 			 const struct mlx5_flow_driver_ops *fops,
8173 			 struct rte_flow_error *error)
8174 {
8175 	static const char err_msg[] = "indirect action validation unsupported";
8176 
8177 	if (!fops->action_validate) {
8178 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8179 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8180 				   NULL, err_msg);
8181 		return -rte_errno;
8182 	}
8183 	return fops->action_validate(dev, conf, action, error);
8184 }
8185 
8186 /**
8187  * Destroys the shared action by handle.
8188  *
8189  * @param dev
8190  *   Pointer to Ethernet device structure.
8191  * @param[in] handle
8192  *   Handle for the indirect action object to be destroyed.
8193  * @param[out] error
8194  *   Perform verbose error reporting if not NULL. PMDs initialize this
8195  *   structure in case of error only.
8196  *
8197  * @return
8198  *   0 on success, a negative errno value otherwise and rte_errno is set.
8199  *
8200  * @note: wrapper for driver action_create op callback.
8201  */
8202 static int
8203 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8204 			   struct rte_flow_action_handle *handle,
8205 			   struct rte_flow_error *error)
8206 {
8207 	static const char err_msg[] = "indirect action destruction unsupported";
8208 	struct rte_flow_attr attr = { .transfer = 0 };
8209 	const struct mlx5_flow_driver_ops *fops =
8210 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8211 
8212 	if (!fops->action_destroy) {
8213 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8214 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8215 				   NULL, err_msg);
8216 		return -rte_errno;
8217 	}
8218 	return fops->action_destroy(dev, handle, error);
8219 }
8220 
8221 /* Wrapper for driver action_destroy op callback */
8222 static int
8223 flow_drv_action_update(struct rte_eth_dev *dev,
8224 		       struct rte_flow_action_handle *handle,
8225 		       const void *update,
8226 		       const struct mlx5_flow_driver_ops *fops,
8227 		       struct rte_flow_error *error)
8228 {
8229 	static const char err_msg[] = "indirect action update unsupported";
8230 
8231 	if (!fops->action_update) {
8232 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8233 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8234 				   NULL, err_msg);
8235 		return -rte_errno;
8236 	}
8237 	return fops->action_update(dev, handle, update, error);
8238 }
8239 
8240 /* Wrapper for driver action_destroy op callback */
8241 static int
8242 flow_drv_action_query(struct rte_eth_dev *dev,
8243 		      const struct rte_flow_action_handle *handle,
8244 		      void *data,
8245 		      const struct mlx5_flow_driver_ops *fops,
8246 		      struct rte_flow_error *error)
8247 {
8248 	static const char err_msg[] = "indirect action query unsupported";
8249 
8250 	if (!fops->action_query) {
8251 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8252 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8253 				   NULL, err_msg);
8254 		return -rte_errno;
8255 	}
8256 	return fops->action_query(dev, handle, data, error);
8257 }
8258 
8259 /**
8260  * Create indirect action for reuse in multiple flow rules.
8261  *
8262  * @param dev
8263  *   Pointer to Ethernet device structure.
8264  * @param conf
8265  *   Pointer to indirect action object configuration.
8266  * @param[in] action
8267  *   Action configuration for indirect action object creation.
8268  * @param[out] error
8269  *   Perform verbose error reporting if not NULL. PMDs initialize this
8270  *   structure in case of error only.
8271  * @return
8272  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8273  */
8274 static struct rte_flow_action_handle *
8275 mlx5_action_handle_create(struct rte_eth_dev *dev,
8276 			  const struct rte_flow_indir_action_conf *conf,
8277 			  const struct rte_flow_action *action,
8278 			  struct rte_flow_error *error)
8279 {
8280 	static const char err_msg[] = "indirect action creation unsupported";
8281 	struct rte_flow_attr attr = { .transfer = 0 };
8282 	const struct mlx5_flow_driver_ops *fops =
8283 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8284 
8285 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8286 		return NULL;
8287 	if (!fops->action_create) {
8288 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8289 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8290 				   NULL, err_msg);
8291 		return NULL;
8292 	}
8293 	return fops->action_create(dev, conf, action, error);
8294 }
8295 
8296 /**
8297  * Updates inplace the indirect action configuration pointed by *handle*
8298  * with the configuration provided as *update* argument.
8299  * The update of the indirect action configuration effects all flow rules
8300  * reusing the action via handle.
8301  *
8302  * @param dev
8303  *   Pointer to Ethernet device structure.
8304  * @param[in] handle
8305  *   Handle for the indirect action to be updated.
8306  * @param[in] update
8307  *   Action specification used to modify the action pointed by handle.
8308  *   *update* could be of same type with the action pointed by the *handle*
8309  *   handle argument, or some other structures like a wrapper, depending on
8310  *   the indirect action type.
8311  * @param[out] error
8312  *   Perform verbose error reporting if not NULL. PMDs initialize this
8313  *   structure in case of error only.
8314  *
8315  * @return
8316  *   0 on success, a negative errno value otherwise and rte_errno is set.
8317  */
8318 static int
8319 mlx5_action_handle_update(struct rte_eth_dev *dev,
8320 		struct rte_flow_action_handle *handle,
8321 		const void *update,
8322 		struct rte_flow_error *error)
8323 {
8324 	struct rte_flow_attr attr = { .transfer = 0 };
8325 	const struct mlx5_flow_driver_ops *fops =
8326 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8327 	int ret;
8328 
8329 	ret = flow_drv_action_validate(dev, NULL,
8330 			(const struct rte_flow_action *)update, fops, error);
8331 	if (ret)
8332 		return ret;
8333 	return flow_drv_action_update(dev, handle, update, fops,
8334 				      error);
8335 }
8336 
8337 /**
8338  * Query the indirect action by handle.
8339  *
8340  * This function allows retrieving action-specific data such as counters.
8341  * Data is gathered by special action which may be present/referenced in
8342  * more than one flow rule definition.
8343  *
8344  * see @RTE_FLOW_ACTION_TYPE_COUNT
8345  *
8346  * @param dev
8347  *   Pointer to Ethernet device structure.
8348  * @param[in] handle
8349  *   Handle for the indirect action to query.
8350  * @param[in, out] data
8351  *   Pointer to storage for the associated query data type.
8352  * @param[out] error
8353  *   Perform verbose error reporting if not NULL. PMDs initialize this
8354  *   structure in case of error only.
8355  *
8356  * @return
8357  *   0 on success, a negative errno value otherwise and rte_errno is set.
8358  */
8359 static int
8360 mlx5_action_handle_query(struct rte_eth_dev *dev,
8361 			 const struct rte_flow_action_handle *handle,
8362 			 void *data,
8363 			 struct rte_flow_error *error)
8364 {
8365 	struct rte_flow_attr attr = { .transfer = 0 };
8366 	const struct mlx5_flow_driver_ops *fops =
8367 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8368 
8369 	return flow_drv_action_query(dev, handle, data, fops, error);
8370 }
8371 
8372 /**
8373  * Destroy all indirect actions (shared RSS).
8374  *
8375  * @param dev
8376  *   Pointer to Ethernet device.
8377  *
8378  * @return
8379  *   0 on success, a negative errno value otherwise and rte_errno is set.
8380  */
8381 int
8382 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8383 {
8384 	struct rte_flow_error error;
8385 	struct mlx5_priv *priv = dev->data->dev_private;
8386 	struct mlx5_shared_action_rss *shared_rss;
8387 	int ret = 0;
8388 	uint32_t idx;
8389 
8390 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8391 		      priv->rss_shared_actions, idx, shared_rss, next) {
8392 		ret |= mlx5_action_handle_destroy(dev,
8393 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8394 	}
8395 	return ret;
8396 }
8397 
8398 #ifndef HAVE_MLX5DV_DR
8399 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8400 #else
8401 #define MLX5_DOMAIN_SYNC_FLOW \
8402 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8403 #endif
8404 
8405 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8406 {
8407 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8408 	const struct mlx5_flow_driver_ops *fops;
8409 	int ret;
8410 	struct rte_flow_attr attr = { .transfer = 0 };
8411 
8412 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8413 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8414 	if (ret > 0)
8415 		ret = -ret;
8416 	return ret;
8417 }
8418 
8419 const struct mlx5_flow_tunnel *
8420 mlx5_get_tof(const struct rte_flow_item *item,
8421 	     const struct rte_flow_action *action,
8422 	     enum mlx5_tof_rule_type *rule_type)
8423 {
8424 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8425 		if (item->type == (typeof(item->type))
8426 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8427 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8428 			return flow_items_to_tunnel(item);
8429 		}
8430 	}
8431 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8432 		if (action->type == (typeof(action->type))
8433 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8434 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8435 			return flow_actions_to_tunnel(action);
8436 		}
8437 	}
8438 	return NULL;
8439 }
8440 
8441 /**
8442  * tunnel offload functionalilty is defined for DV environment only
8443  */
8444 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8445 __extension__
8446 union tunnel_offload_mark {
8447 	uint32_t val;
8448 	struct {
8449 		uint32_t app_reserve:8;
8450 		uint32_t table_id:15;
8451 		uint32_t transfer:1;
8452 		uint32_t _unused_:8;
8453 	};
8454 };
8455 
8456 static bool
8457 mlx5_access_tunnel_offload_db
8458 	(struct rte_eth_dev *dev,
8459 	 bool (*match)(struct rte_eth_dev *,
8460 		       struct mlx5_flow_tunnel *, const void *),
8461 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8462 	 void (*miss)(struct rte_eth_dev *, void *),
8463 	 void *ctx, bool lock_op);
8464 
8465 static int
8466 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8467 			     struct rte_flow *flow,
8468 			     const struct rte_flow_attr *attr,
8469 			     const struct rte_flow_action *app_actions,
8470 			     uint32_t flow_idx,
8471 			     const struct mlx5_flow_tunnel *tunnel,
8472 			     struct tunnel_default_miss_ctx *ctx,
8473 			     struct rte_flow_error *error)
8474 {
8475 	struct mlx5_priv *priv = dev->data->dev_private;
8476 	struct mlx5_flow *dev_flow;
8477 	struct rte_flow_attr miss_attr = *attr;
8478 	const struct rte_flow_item miss_items[2] = {
8479 		{
8480 			.type = RTE_FLOW_ITEM_TYPE_ETH,
8481 			.spec = NULL,
8482 			.last = NULL,
8483 			.mask = NULL
8484 		},
8485 		{
8486 			.type = RTE_FLOW_ITEM_TYPE_END,
8487 			.spec = NULL,
8488 			.last = NULL,
8489 			.mask = NULL
8490 		}
8491 	};
8492 	union tunnel_offload_mark mark_id;
8493 	struct rte_flow_action_mark miss_mark;
8494 	struct rte_flow_action miss_actions[3] = {
8495 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8496 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
8497 	};
8498 	const struct rte_flow_action_jump *jump_data;
8499 	uint32_t i, flow_table = 0; /* prevent compilation warning */
8500 	struct flow_grp_info grp_info = {
8501 		.external = 1,
8502 		.transfer = attr->transfer,
8503 		.fdb_def_rule = !!priv->fdb_def_rule,
8504 		.std_tbl_fix = 0,
8505 	};
8506 	int ret;
8507 
8508 	if (!attr->transfer) {
8509 		uint32_t q_size;
8510 
8511 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8512 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8513 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8514 					 0, SOCKET_ID_ANY);
8515 		if (!ctx->queue)
8516 			return rte_flow_error_set
8517 				(error, ENOMEM,
8518 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8519 				NULL, "invalid default miss RSS");
8520 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8521 		ctx->action_rss.level = 0,
8522 		ctx->action_rss.types = priv->rss_conf.rss_hf,
8523 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8524 		ctx->action_rss.queue_num = priv->reta_idx_n,
8525 		ctx->action_rss.key = priv->rss_conf.rss_key,
8526 		ctx->action_rss.queue = ctx->queue;
8527 		if (!priv->reta_idx_n || !priv->rxqs_n)
8528 			return rte_flow_error_set
8529 				(error, EINVAL,
8530 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8531 				NULL, "invalid port configuration");
8532 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8533 			ctx->action_rss.types = 0;
8534 		for (i = 0; i != priv->reta_idx_n; ++i)
8535 			ctx->queue[i] = (*priv->reta_idx)[i];
8536 	} else {
8537 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8538 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8539 	}
8540 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8541 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8542 	jump_data = app_actions->conf;
8543 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8544 	miss_attr.group = jump_data->group;
8545 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8546 				       &flow_table, &grp_info, error);
8547 	if (ret)
8548 		return rte_flow_error_set(error, EINVAL,
8549 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8550 					  NULL, "invalid tunnel id");
8551 	mark_id.app_reserve = 0;
8552 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8553 	mark_id.transfer = !!attr->transfer;
8554 	mark_id._unused_ = 0;
8555 	miss_mark.id = mark_id.val;
8556 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8557 				    miss_items, miss_actions, flow_idx, error);
8558 	if (!dev_flow)
8559 		return -rte_errno;
8560 	dev_flow->flow = flow;
8561 	dev_flow->external = true;
8562 	dev_flow->tunnel = tunnel;
8563 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8564 	/* Subflow object was created, we must include one in the list. */
8565 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8566 		      dev_flow->handle, next);
8567 	DRV_LOG(DEBUG,
8568 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8569 		dev->data->port_id, tunnel->app_tunnel.type,
8570 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8571 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8572 				  miss_actions, error);
8573 	if (!ret)
8574 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8575 						  error);
8576 
8577 	return ret;
8578 }
8579 
8580 static const struct mlx5_flow_tbl_data_entry  *
8581 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8582 {
8583 	struct mlx5_priv *priv = dev->data->dev_private;
8584 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8585 	struct mlx5_hlist_entry *he;
8586 	union tunnel_offload_mark mbits = { .val = mark };
8587 	union mlx5_flow_tbl_key table_key = {
8588 		{
8589 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
8590 			.id = 0,
8591 			.reserved = 0,
8592 			.dummy = 0,
8593 			.is_fdb = !!mbits.transfer,
8594 			.is_egress = 0,
8595 		}
8596 	};
8597 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8598 	return he ?
8599 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8600 }
8601 
8602 static void
8603 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8604 				   struct mlx5_hlist_entry *entry)
8605 {
8606 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8607 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8608 
8609 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8610 			tunnel_flow_tbl_to_id(tte->flow_table));
8611 	mlx5_free(tte);
8612 }
8613 
8614 static int
8615 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8616 				  struct mlx5_hlist_entry *entry,
8617 				  uint64_t key, void *cb_ctx __rte_unused)
8618 {
8619 	union tunnel_tbl_key tbl = {
8620 		.val = key,
8621 	};
8622 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8623 
8624 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8625 }
8626 
8627 static struct mlx5_hlist_entry *
8628 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8629 				   void *ctx __rte_unused)
8630 {
8631 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8632 	struct tunnel_tbl_entry *tte;
8633 	union tunnel_tbl_key tbl = {
8634 		.val = key,
8635 	};
8636 
8637 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8638 			  sizeof(*tte), 0,
8639 			  SOCKET_ID_ANY);
8640 	if (!tte)
8641 		goto err;
8642 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8643 			  &tte->flow_table);
8644 	if (tte->flow_table >= MLX5_MAX_TABLES) {
8645 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8646 			tte->flow_table);
8647 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8648 				tte->flow_table);
8649 		goto err;
8650 	} else if (!tte->flow_table) {
8651 		goto err;
8652 	}
8653 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8654 	tte->tunnel_id = tbl.tunnel_id;
8655 	tte->group = tbl.group;
8656 	return &tte->hash;
8657 err:
8658 	if (tte)
8659 		mlx5_free(tte);
8660 	return NULL;
8661 }
8662 
8663 static uint32_t
8664 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8665 				const struct mlx5_flow_tunnel *tunnel,
8666 				uint32_t group, uint32_t *table,
8667 				struct rte_flow_error *error)
8668 {
8669 	struct mlx5_hlist_entry *he;
8670 	struct tunnel_tbl_entry *tte;
8671 	union tunnel_tbl_key key = {
8672 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8673 		.group = group
8674 	};
8675 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8676 	struct mlx5_hlist *group_hash;
8677 
8678 	group_hash = tunnel ? tunnel->groups : thub->groups;
8679 	he = mlx5_hlist_register(group_hash, key.val, NULL);
8680 	if (!he)
8681 		return rte_flow_error_set(error, EINVAL,
8682 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8683 					  NULL,
8684 					  "tunnel group index not supported");
8685 	tte = container_of(he, typeof(*tte), hash);
8686 	*table = tte->flow_table;
8687 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8688 		dev->data->port_id, key.tunnel_id, group, *table);
8689 	return 0;
8690 }
8691 
8692 static void
8693 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8694 		      struct mlx5_flow_tunnel *tunnel)
8695 {
8696 	struct mlx5_priv *priv = dev->data->dev_private;
8697 	struct mlx5_indexed_pool *ipool;
8698 
8699 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8700 		dev->data->port_id, tunnel->tunnel_id);
8701 	LIST_REMOVE(tunnel, chain);
8702 	mlx5_hlist_destroy(tunnel->groups);
8703 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8704 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
8705 }
8706 
8707 static bool
8708 mlx5_access_tunnel_offload_db
8709 	(struct rte_eth_dev *dev,
8710 	 bool (*match)(struct rte_eth_dev *,
8711 		       struct mlx5_flow_tunnel *, const void *),
8712 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8713 	 void (*miss)(struct rte_eth_dev *, void *),
8714 	 void *ctx, bool lock_op)
8715 {
8716 	bool verdict = false;
8717 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8718 	struct mlx5_flow_tunnel *tunnel;
8719 
8720 	rte_spinlock_lock(&thub->sl);
8721 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8722 		verdict = match(dev, tunnel, (const void *)ctx);
8723 		if (verdict)
8724 			break;
8725 	}
8726 	if (!lock_op)
8727 		rte_spinlock_unlock(&thub->sl);
8728 	if (verdict && hit)
8729 		hit(dev, tunnel, ctx);
8730 	if (!verdict && miss)
8731 		miss(dev, ctx);
8732 	if (lock_op)
8733 		rte_spinlock_unlock(&thub->sl);
8734 
8735 	return verdict;
8736 }
8737 
8738 struct tunnel_db_find_tunnel_id_ctx {
8739 	uint32_t tunnel_id;
8740 	struct mlx5_flow_tunnel *tunnel;
8741 };
8742 
8743 static bool
8744 find_tunnel_id_match(struct rte_eth_dev *dev,
8745 		     struct mlx5_flow_tunnel *tunnel, const void *x)
8746 {
8747 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8748 
8749 	RTE_SET_USED(dev);
8750 	return tunnel->tunnel_id == ctx->tunnel_id;
8751 }
8752 
8753 static void
8754 find_tunnel_id_hit(struct rte_eth_dev *dev,
8755 		   struct mlx5_flow_tunnel *tunnel, void *x)
8756 {
8757 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8758 	RTE_SET_USED(dev);
8759 	ctx->tunnel = tunnel;
8760 }
8761 
8762 static struct mlx5_flow_tunnel *
8763 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8764 {
8765 	struct tunnel_db_find_tunnel_id_ctx ctx = {
8766 		.tunnel_id = id,
8767 	};
8768 
8769 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8770 				      find_tunnel_id_hit, NULL, &ctx, true);
8771 
8772 	return ctx.tunnel;
8773 }
8774 
8775 static struct mlx5_flow_tunnel *
8776 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8777 			  const struct rte_flow_tunnel *app_tunnel)
8778 {
8779 	struct mlx5_priv *priv = dev->data->dev_private;
8780 	struct mlx5_indexed_pool *ipool;
8781 	struct mlx5_flow_tunnel *tunnel;
8782 	uint32_t id;
8783 
8784 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8785 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
8786 	if (!tunnel)
8787 		return NULL;
8788 	if (id >= MLX5_MAX_TUNNELS) {
8789 		mlx5_ipool_free(ipool, id);
8790 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8791 		return NULL;
8792 	}
8793 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8794 					   mlx5_flow_tunnel_grp2tbl_create_cb,
8795 					   mlx5_flow_tunnel_grp2tbl_match_cb,
8796 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
8797 	if (!tunnel->groups) {
8798 		mlx5_ipool_free(ipool, id);
8799 		return NULL;
8800 	}
8801 	tunnel->groups->ctx = priv->sh;
8802 	/* initiate new PMD tunnel */
8803 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8804 	tunnel->tunnel_id = id;
8805 	tunnel->action.type = (typeof(tunnel->action.type))
8806 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8807 	tunnel->action.conf = tunnel;
8808 	tunnel->item.type = (typeof(tunnel->item.type))
8809 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8810 	tunnel->item.spec = tunnel;
8811 	tunnel->item.last = NULL;
8812 	tunnel->item.mask = NULL;
8813 
8814 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8815 		dev->data->port_id, tunnel->tunnel_id);
8816 
8817 	return tunnel;
8818 }
8819 
8820 struct tunnel_db_get_tunnel_ctx {
8821 	const struct rte_flow_tunnel *app_tunnel;
8822 	struct mlx5_flow_tunnel *tunnel;
8823 };
8824 
8825 static bool get_tunnel_match(struct rte_eth_dev *dev,
8826 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8827 {
8828 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
8829 
8830 	RTE_SET_USED(dev);
8831 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8832 		       sizeof(*ctx->app_tunnel));
8833 }
8834 
8835 static void get_tunnel_hit(struct rte_eth_dev *dev,
8836 			   struct mlx5_flow_tunnel *tunnel, void *x)
8837 {
8838 	/* called under tunnel spinlock protection */
8839 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8840 
8841 	RTE_SET_USED(dev);
8842 	tunnel->refctn++;
8843 	ctx->tunnel = tunnel;
8844 }
8845 
8846 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8847 {
8848 	/* called under tunnel spinlock protection */
8849 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8850 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8851 
8852 	rte_spinlock_unlock(&thub->sl);
8853 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8854 	rte_spinlock_lock(&thub->sl);
8855 	if (ctx->tunnel) {
8856 		ctx->tunnel->refctn = 1;
8857 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8858 	}
8859 }
8860 
8861 
8862 static int
8863 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8864 		     const struct rte_flow_tunnel *app_tunnel,
8865 		     struct mlx5_flow_tunnel **tunnel)
8866 {
8867 	struct tunnel_db_get_tunnel_ctx ctx = {
8868 		.app_tunnel = app_tunnel,
8869 	};
8870 
8871 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8872 				      get_tunnel_miss, &ctx, true);
8873 	*tunnel = ctx.tunnel;
8874 	return ctx.tunnel ? 0 : -ENOMEM;
8875 }
8876 
8877 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8878 {
8879 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8880 
8881 	if (!thub)
8882 		return;
8883 	if (!LIST_EMPTY(&thub->tunnels))
8884 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
8885 	mlx5_hlist_destroy(thub->groups);
8886 	mlx5_free(thub);
8887 }
8888 
8889 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8890 {
8891 	int err;
8892 	struct mlx5_flow_tunnel_hub *thub;
8893 
8894 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8895 			   0, SOCKET_ID_ANY);
8896 	if (!thub)
8897 		return -ENOMEM;
8898 	LIST_INIT(&thub->tunnels);
8899 	rte_spinlock_init(&thub->sl);
8900 	thub->groups = mlx5_hlist_create("flow groups",
8901 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
8902 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8903 					 mlx5_flow_tunnel_grp2tbl_match_cb,
8904 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
8905 	if (!thub->groups) {
8906 		err = -rte_errno;
8907 		goto err;
8908 	}
8909 	thub->groups->ctx = sh;
8910 	sh->tunnel_hub = thub;
8911 
8912 	return 0;
8913 
8914 err:
8915 	if (thub->groups)
8916 		mlx5_hlist_destroy(thub->groups);
8917 	if (thub)
8918 		mlx5_free(thub);
8919 	return err;
8920 }
8921 
8922 static inline bool
8923 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8924 			  struct rte_flow_tunnel *tunnel,
8925 			  const char *err_msg)
8926 {
8927 	err_msg = NULL;
8928 	if (!is_tunnel_offload_active(dev)) {
8929 		err_msg = "tunnel offload was not activated";
8930 		goto out;
8931 	} else if (!tunnel) {
8932 		err_msg = "no application tunnel";
8933 		goto out;
8934 	}
8935 
8936 	switch (tunnel->type) {
8937 	default:
8938 		err_msg = "unsupported tunnel type";
8939 		goto out;
8940 	case RTE_FLOW_ITEM_TYPE_VXLAN:
8941 		break;
8942 	}
8943 
8944 out:
8945 	return !err_msg;
8946 }
8947 
8948 static int
8949 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8950 		    struct rte_flow_tunnel *app_tunnel,
8951 		    struct rte_flow_action **actions,
8952 		    uint32_t *num_of_actions,
8953 		    struct rte_flow_error *error)
8954 {
8955 	int ret;
8956 	struct mlx5_flow_tunnel *tunnel;
8957 	const char *err_msg = NULL;
8958 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8959 
8960 	if (!verdict)
8961 		return rte_flow_error_set(error, EINVAL,
8962 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8963 					  err_msg);
8964 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8965 	if (ret < 0) {
8966 		return rte_flow_error_set(error, ret,
8967 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8968 					  "failed to initialize pmd tunnel");
8969 	}
8970 	*actions = &tunnel->action;
8971 	*num_of_actions = 1;
8972 	return 0;
8973 }
8974 
8975 static int
8976 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8977 		       struct rte_flow_tunnel *app_tunnel,
8978 		       struct rte_flow_item **items,
8979 		       uint32_t *num_of_items,
8980 		       struct rte_flow_error *error)
8981 {
8982 	int ret;
8983 	struct mlx5_flow_tunnel *tunnel;
8984 	const char *err_msg = NULL;
8985 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8986 
8987 	if (!verdict)
8988 		return rte_flow_error_set(error, EINVAL,
8989 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8990 					  err_msg);
8991 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8992 	if (ret < 0) {
8993 		return rte_flow_error_set(error, ret,
8994 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8995 					  "failed to initialize pmd tunnel");
8996 	}
8997 	*items = &tunnel->item;
8998 	*num_of_items = 1;
8999 	return 0;
9000 }
9001 
9002 struct tunnel_db_element_release_ctx {
9003 	struct rte_flow_item *items;
9004 	struct rte_flow_action *actions;
9005 	uint32_t num_elements;
9006 	struct rte_flow_error *error;
9007 	int ret;
9008 };
9009 
9010 static bool
9011 tunnel_element_release_match(struct rte_eth_dev *dev,
9012 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9013 {
9014 	const struct tunnel_db_element_release_ctx *ctx = x;
9015 
9016 	RTE_SET_USED(dev);
9017 	if (ctx->num_elements != 1)
9018 		return false;
9019 	else if (ctx->items)
9020 		return ctx->items == &tunnel->item;
9021 	else if (ctx->actions)
9022 		return ctx->actions == &tunnel->action;
9023 
9024 	return false;
9025 }
9026 
9027 static void
9028 tunnel_element_release_hit(struct rte_eth_dev *dev,
9029 			   struct mlx5_flow_tunnel *tunnel, void *x)
9030 {
9031 	struct tunnel_db_element_release_ctx *ctx = x;
9032 	ctx->ret = 0;
9033 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9034 		mlx5_flow_tunnel_free(dev, tunnel);
9035 }
9036 
9037 static void
9038 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9039 {
9040 	struct tunnel_db_element_release_ctx *ctx = x;
9041 	RTE_SET_USED(dev);
9042 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9043 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9044 				      "invalid argument");
9045 }
9046 
9047 static int
9048 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9049 		       struct rte_flow_item *pmd_items,
9050 		       uint32_t num_items, struct rte_flow_error *err)
9051 {
9052 	struct tunnel_db_element_release_ctx ctx = {
9053 		.items = pmd_items,
9054 		.actions = NULL,
9055 		.num_elements = num_items,
9056 		.error = err,
9057 	};
9058 
9059 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9060 				      tunnel_element_release_hit,
9061 				      tunnel_element_release_miss, &ctx, false);
9062 
9063 	return ctx.ret;
9064 }
9065 
9066 static int
9067 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9068 			 struct rte_flow_action *pmd_actions,
9069 			 uint32_t num_actions, struct rte_flow_error *err)
9070 {
9071 	struct tunnel_db_element_release_ctx ctx = {
9072 		.items = NULL,
9073 		.actions = pmd_actions,
9074 		.num_elements = num_actions,
9075 		.error = err,
9076 	};
9077 
9078 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9079 				      tunnel_element_release_hit,
9080 				      tunnel_element_release_miss, &ctx, false);
9081 
9082 	return ctx.ret;
9083 }
9084 
9085 static int
9086 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9087 				  struct rte_mbuf *m,
9088 				  struct rte_flow_restore_info *info,
9089 				  struct rte_flow_error *err)
9090 {
9091 	uint64_t ol_flags = m->ol_flags;
9092 	const struct mlx5_flow_tbl_data_entry *tble;
9093 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
9094 
9095 	if (!is_tunnel_offload_active(dev)) {
9096 		info->flags = 0;
9097 		return 0;
9098 	}
9099 
9100 	if ((ol_flags & mask) != mask)
9101 		goto err;
9102 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9103 	if (!tble) {
9104 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9105 			dev->data->port_id, m->hash.fdir.hi);
9106 		goto err;
9107 	}
9108 	MLX5_ASSERT(tble->tunnel);
9109 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9110 	info->group_id = tble->group_id;
9111 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9112 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9113 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9114 
9115 	return 0;
9116 
9117 err:
9118 	return rte_flow_error_set(err, EINVAL,
9119 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9120 				  "failed to get restore info");
9121 }
9122 
9123 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9124 static int
9125 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9126 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9127 			   __rte_unused struct rte_flow_action **actions,
9128 			   __rte_unused uint32_t *num_of_actions,
9129 			   __rte_unused struct rte_flow_error *error)
9130 {
9131 	return -ENOTSUP;
9132 }
9133 
9134 static int
9135 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9136 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9137 		       __rte_unused struct rte_flow_item **items,
9138 		       __rte_unused uint32_t *num_of_items,
9139 		       __rte_unused struct rte_flow_error *error)
9140 {
9141 	return -ENOTSUP;
9142 }
9143 
9144 static int
9145 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9146 			      __rte_unused struct rte_flow_item *pmd_items,
9147 			      __rte_unused uint32_t num_items,
9148 			      __rte_unused struct rte_flow_error *err)
9149 {
9150 	return -ENOTSUP;
9151 }
9152 
9153 static int
9154 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9155 				__rte_unused struct rte_flow_action *pmd_action,
9156 				__rte_unused uint32_t num_actions,
9157 				__rte_unused struct rte_flow_error *err)
9158 {
9159 	return -ENOTSUP;
9160 }
9161 
9162 static int
9163 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9164 				  __rte_unused struct rte_mbuf *m,
9165 				  __rte_unused struct rte_flow_restore_info *i,
9166 				  __rte_unused struct rte_flow_error *err)
9167 {
9168 	return -ENOTSUP;
9169 }
9170 
9171 static int
9172 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9173 			     __rte_unused struct rte_flow *flow,
9174 			     __rte_unused const struct rte_flow_attr *attr,
9175 			     __rte_unused const struct rte_flow_action *actions,
9176 			     __rte_unused uint32_t flow_idx,
9177 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9178 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9179 			     __rte_unused struct rte_flow_error *error)
9180 {
9181 	return -ENOTSUP;
9182 }
9183 
9184 static struct mlx5_flow_tunnel *
9185 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9186 		    __rte_unused uint32_t id)
9187 {
9188 	return NULL;
9189 }
9190 
9191 static void
9192 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9193 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9194 {
9195 }
9196 
9197 static uint32_t
9198 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9199 				__rte_unused const struct mlx5_flow_tunnel *t,
9200 				__rte_unused uint32_t group,
9201 				__rte_unused uint32_t *table,
9202 				struct rte_flow_error *error)
9203 {
9204 	return rte_flow_error_set(error, ENOTSUP,
9205 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9206 				  "tunnel offload requires DV support");
9207 }
9208 
9209 void
9210 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9211 			__rte_unused  uint16_t port_id)
9212 {
9213 }
9214 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9215 
9216 static void
9217 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9218 {
9219 	int ret;
9220 	struct rte_flow_error error;
9221 
9222 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9223 		char *item_name;
9224 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9225 				    sizeof(item_name),
9226 				    (void *)(uintptr_t)item->type, &error);
9227 		if (ret > 0)
9228 			printf("%s ", item_name);
9229 		else
9230 			printf("%d\n", (int)item->type);
9231 	}
9232 	printf("END\n");
9233 }
9234